(48) (human cognitive processing) alexander ziem frames of understanding in text and discourse...

Frames of Understanding in Text and Discourse

Volume 48

Frames of Understanding in Text and Discourse. Theoretical foundations and descriptive applicationsby Alexander Ziem

Editorial BoardBogusław BierwiaczonekJan Dlugosz University, Czestochowa, Poland / Higher School of Labour Safety Management, KatowiceMario BrdarJosip Juraj Strossmayer University, Croatia

Barbara DancygierUniversity of British ColumbiaN.J. EnfieldMax Planck Institute for Psycholinguistics, Nijmegen & Radboud University NijmegenElisabeth Engberg-PedersenUniversity of Copenhagen

Ad FoolenRadboud University NijmegenRaymond W. Gibbs, Jr.University of California at Santa CruzRachel GioraTel Aviv University

Elżbieta GórskaUniversity of WarsawMartin HilpertUniversity of Neuchâtel

Zoltán KövecsesEötvös Loránd University, HungaryTeenie MatlockUniversity of California at Merced

Carita ParadisLund UniversityGünter RaddenUniversity of HamburgFrancisco José Ruiz de Mendoza IbáñezUniversity of La Rioja

Doris SchönefeldUniversity of LeipzigDebra ZiegelerUniversity of Paris III

Human Cognitive Processing (HCP)Cognitive Foundations of Language Structure and UseThis book series is a forum for interdisciplinary research on the grammatical struc-ture, semantic organization, and communicative function of language(s), and their anchoring in human cognitive faculties.

For an overview of all books published in this series, please see http://benjamins.com/catalog/hcp

Editors

Klaus-Uwe PantherNanjing Normal University & University of Hamburg

Linda L. ThornburgNanjing Normal University

Frames of Understanding in Text and DiscourseTheoretical foundations and descriptive applications

Alexander ZiemHeinrich Heine University, Düsseldorf

John Benjamins Publishing CompanyAmsterdam / Philadelphia

Library of Congress Cataloging-in-Publication Data

Ziem, Alexander.[Frames und sprachliches Wissen. English] Frames of Understanding in Text and Discourse : Theoretical foundations and descriptive

applications / Alexander Ziem.p. cm. (Human Cognitive Processing, issn 1387-6724 ; v. 48)Translated from the German title Frames und sprachliches Wissen, 2008.Includes bibliographical references and index.1. Frames (Linguistics) 2. Semantics--Psychological aspects. 3. Cognition. I. Schwerin,

Catherine, editor. II. Title.P325.5.F72Z5413 2014401’.43--dc23 2014020746isbn 978 90 272 4664 6 (Hb ; alk. paper)isbn 978 90 272 6964 5 (Eb)

© 2014 – John Benjamins B.V.No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher.

John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The NetherlandsJohn Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

8 TM The paper used in this publication meets the minimum requirements of the American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

The translation of this work was funded by Geisteswissenschaften International – Translation Funding for Work in the Humanities and Social Sciences from Germany, a joint initiative of the Fritz Thyssen Foundation, the German Federal Foreign Office, the collecting society VG WORT and the Börsenverein des Deutschen Buchhandels (German Publishers & Booksellers Association).

Alexander Ziem, Frames und sprachliches Wissen.© Walter de Gruyter GmbH Berlin/Boston. All rights reserved.

Translated by Catherine Schwerin.

In memory of Chuck Fillmore

Table of contents

Preface xi

Introduction 1

chapter 1 Semantic interest in frames 51.1 Evidence for frames 51.2 Frames in research 10 1.2.1 The development of frame research 11 1.2.2 Frames and other representation formats 171.3 Frames in cognitive science 29 1.3.1 Cognition, representation, categorization 29 1.3.2 Positions in cognitive theory 36 1.3.3 Frames in modularist and holistic approaches 43

chapter 2 Cognitive theory and semantic issues 492.1 Holism vs. Modularism: an example 502.2 Modularism 55 2.2.1 Two-level semantics (M. Bierwisch) 56 2.2.2 Frame semantics vs. two-level semantics: some issues 61 2.2.3 Example analyses 65 2.2.4 Three-level semantics (M. Schwarz) 772.3 Holism 87 2.3.1 Meaning as conceptualization 88 2.3.2 Language as conceptualization (R. Langacker vs. R. Jackendoff) 91

chapter 3 The holistic paradigm 993.1 Are linguistic and conceptual knowledge distinct entities? 101 3.1.1 Essence vs. accidence? 107 3.1.2 Synthetic vs. analytic truths? 109 3.1.3 Culture vs. language? 111 3.1.4 Semantics vs. pragmatics? 1133.2 The “space of understanding” (C. Demmerling) 121

viii Frames of Understanding in Text and Discourse

3.3 The postulate of U-relevance 128 3.3.1 Busse’s explicative semantics 129 3.3.2 Approaches in psycholinguistic research

on language-processing 135 3.3.3 Comparison of knowledge types 143

chapter 4 Semiotic issues 1474.1 Linguistic signs as constructions 150 4.1.1 The symbolic principle in construction grammar

and Cognitive Grammar 151 4.1.2 What are constructions and symbolic units? 153 4.1.3 Constructions in the “space of understanding” 1644.2 Frames and symbolic units 168 4.2.1 Conventional vs. contextual aspects of meaning (R. Langacker) 170 4.2.2 Are “situations” and “backgrounds” elements of semantic units?

(J. Zlatev) 179 4.2.3 Are “scenes” elements of semantic units? (C. Fillmore) 1884.3 Relations 196 4.3.1 Evoked and invoked frames (C. Fillmore) 197 4.3.2 Meaning potentials (J. Allwood) 202

chapter 5 Frames as schemata 2115.1 Categorization 2125.2 Schemata 218 5.2.1 Schemata as representational formats of non-specific modality 220 5.2.2 Shared features of frames and schemata 2275.3 Frames as schemata: example analysis 233

chapter 6The structural constituents of frames 2436.1 Issues 2436.2 Reference 247 6.2.1 Frames as a projection area of referentiality 248 6.2.2 Every word evokes a frame 2526.3 Predication potential: slots 256 6.3.1 What are slots? 257 6.3.2 Hyperonym type reduction: determining slots 264 6.3.3 Example analysis 2746.4 Explicit predications: fillers 280 6.4.1 When are predications explicit? 281 6.4.2 Linguistic manifestations 284

Table of contents ix

6.5 Implicit predications: default values 289 6.5.1 Recurrent schema instantiations: token and type frequency 292 6.5.2 “Cognitive trails” as phenomena of the third kind 300 6.5.3 Type frequency: an example 307

chapter 7Frames in discourse: the financial investors as locusts metaphor 3157.1 Preliminaries 316 7.1.1 Frames as an instrument of corpus-based analysis 316 7.1.2 Cognitive and discourse-related aspects of metaphors 3227.2 The “capitalism debate” 332 7.2.1 Discourse and corpus 333 7.2.2 Investigation period, discourse development, research corpus 336 7.2.3 Locust: a basic discourse-semantic figure 3397.3 Methodological guidelines for the corpus-based analysis 349 7.3.1 Annotations 349 7.3.2 Predication analysis 352 7.3.3 Hyperonym type reduction 354 7.3.4 Classification of explicit predications 3597.4 Empirical results 362 7.4.1 The generic frame 362 7.4.2 The input frames locust/s and financial investor/s 365 7.4.3 The metaphor frame 3717.5 Frame semantics and discourse analysis: some conclusions 377

chapter 8Conclusions and future research perspectives 379

References 387

Name index 421

Subject index 423

Preface

This book attempts to examine the concept of semantic frames, one of the most impor-tant concepts of Cognitive Linguistics, and place it in the overall context of linguistics and cognitive science. Whereas frames are first and foremost relevant to semantic and semiotic issues, the explanatory power of the frame model presented here extends far beyond this. It is no less relevant to issues of morphology, syntax, and general linguistic theory. Frames constitute a fundamental component of a comprehensive theory of language understanding and grammar. In addition, employing frames as a (corpus-based) instrument for investigating knowledge-building processes in society makes them applicable to empirical research on “the way we think”. This has been the focal point of Fillmore’s so-called “frames of understanding”, an approach to semantics that Fillmore developed in the mid-1970s and early 1980s, that is, in between his well-known studies on “case grammar” and the on going FrameNet project. The concept of “frames of understanding” constitutes the starting point of this monograph.

The study presented in this book goes back to concepts that I have been develop-ing since 2004 and that were first published in German in the monograph “Frames und sprachliches Wissen” (de Gruyter, 2008). The present book is a translation of a revised version of this publication. I am very grateful to Catherine Schwerin, who has translated the book. Indeed, this was a hard job. Thanks, Catherine, for solving even seemingly unsolvable problems!

In 2010, the Academy of Sciences (“Akademie der Wissenschaften”) in Göttingen, Germany, awarded me the “Prize of the Humanities”. It was this award that motivated me to present ideas originally published in the aforementioned German book to a broader scientific community. It took a couple of years until this plan could be realized with the financial help of the Fritz Thyssen Foundation, VG Wort, the Börsenverein des Deutschen Buchhandels, and the Ministry of Foreign Affairs. They awarded my German book the “Prize for supporting translations of works in the humanities” (“Preis zur Förderung der Übersetzung geisteswissenschaftlicher Werke”). Thanks for the very generous financial support!

I am also very grateful to Klaus-Uwe Panther and Linda L. Thornburg, who ac-cepted the book for publication in the Benjamins series Human Cognitive Processing. They always provided instant support, regarding both technical and content-related issues.

It is virtually impossible to thank everybody who helped make the monograph a better publication. Nonetheless, I would like to mention explicitly Hans C. Boas, Heike Behrens, Dietrich Busse, Andreas Gardt, Ekkehard Felder, Roland Posner, Martin Reisigl, Sven Staffeldt, Ingo Warnke, Martin Wengeler, and my wife Andrea Ziem as

xii Frames of Understanding in Text and Discourse

well as the “FrameNetters” Collin Baker, Michael Ellsworth, and Miriam R. L. Petruck. During my academic work in Berlin, Basel, Düsseldorf and Berkeley you were indis-pensable supporters!

Perhaps the most vital influence on the considerations presented here goes back to Chuck Fillmore. I dedicate this book to Chuck who died on 13 February 2014. At this time, I had the privilege to work as a research fellow at FrameNet, Berkeley, California. Undoubtedly, Chuck was one of the most brilliant linguists ever. And he was much more than that. He was an ingenious fellow, a supporter of ideas, an attentive listener. His sense for details, his visionary mind, his wit and humour remain unique. Chuck is the “father” of many ideas presented in this book, even though I only had the op-portunity to meet him in person long after these ideas had been written up.

Alexander Ziem Berkeley/Düsseldorf April 2014

Introduction

If someone said, “It’s raining frogs”, your mind would swiftly fill with thoughts about the origins of those frogs, about what happens

to them when they hit the ground, about what could have caused that peculiar plague, and about whether or not the announcer

had gone mad. Yet the stimulus for all of this is just three words. How do our minds conceive such complex scenes from such sparse cues?

The additional details must come from memories and reasoning. (Minsky 1988: 244)

When we comprehend the meaning of a linguistic expression, we understand more than just this expression: we bring to mind an entire context of facts and knowledge. Understanding the example “It’s raining frogs” in Marvin Minsky’s quote is not too difficult, even though most of us have probably never experienced frogs falling from the sky. Even when we hear or read the sentence “It’s raining frogs” for the first time, these few words are enough to give us a rather precise notion of the event described. It would require no effort to come up with innumerable further details beyond the facets of knowledge mentioned by Minsky. We can imagine how large the falling frogs are (perhaps no bigger than your hand, certainly not as small as a fly or as big as a bear), what colour they could be (brownish or green, but not blue or pink), what body parts they have, and what their skin feels like. We also know that frogs typically live in or near water, but that, should they fall from the sky, they will be subject to the law of gravity just as other entities are as well. We can also imagine how many frogs are raining down and what sort of consequences the volume of falling frogs might have – and not just for the frogs. In order to give meaning to the unusual sentence, someone might hit upon the idea of interpreting the expression frog or raining as a metaphor. A whole realm of other assumptions would then come into play.

If words such as frogs or raining or complex expressions like It’s raining frogs evoke knowledge, this does not occur without our involvement. We, the text recipients, con-struct contexts in which the received expressions typically occur. To achieve this we draw on various background knowledge that we have at our disposal. The use of the word tip, for example, evokes a “scene” consisting of scenery (the restaurant), actors (guests, the waiter), certain props (a table, chairs, cutlery, etc.), a script (ordering food, eating, paying) and much more, because everyone knows under what circumstances and in which everyday situations tips are given. To take up an example from Karl Bühler (1934: 171–172), our everyday experience will tell us what the word radish means, since possible scenic contexts in which radishes prototypically occur are part of

2 Frames of Understanding in Text and Discourse

our experience. Bühler observes that reading or hearing the word radish will instantly transport the recipient to the table or the garden.

This book departs from the premise that knowledge that is employed to grasp the meaning of a linguistic expression demonstrates structures that can be precisely described in linguistic terms using so-called “frames”. Frames are conceptual knowl-edge units that linguistic expressions evoke. In other word, language users call up these frames from their memories to grasp the meaning of a linguistic expression. Accordingly, knowing what an expression means and how it is to be used means hav-ing a certain cognitive structure at one’s disposal that is conventionally associated with an expression.

Linguistic frame theory can be traced back to numerous works by the American lin-guist Charles Fillmore. He first spoke of “frame semantics” in the mid-1970s (Fillmore 1975, 1976a, 1977b, 1977a). In these early papers, Fillmore lays the groundwork for the conception of a so-called “understanding semantics” (short: “U-semantics”, Fillmore 1985). U-semantics aims at explicating all knowledge facets necessary to fully under-stand the meaning of a linguistic expression. Another frame concept closely related to U-semantics was realized in the FrameNet project launched in the mid-1990s at the International Computer Science Institute in Berkeley, USA.1 In contrast to Fillmore’s earlier work on frames, FrameNet’s goal is to document the combinatorial potential, respectively the syntactic and semantic valences, of each word in each of its senses. The conception of U-semantics, rather than the valency-oriented approach to frames motivating the FrameNet project, constitutes the starting point of the present study.

Thanks to Fillmore’s enormous endeavour, frame semantics has become an es-sential part of both Cognitive Linguistics and construction grammar. In recent years, frames have also increasingly gained popularity in the German-language literature through the works of Klaus-Peter Konerding (1993), Claudia Fraas (1996a), and Birte Lönneker (2003a). In this book, frame semantics is anchored in a cognitive paradigm, also taking into account the relevant German-language literature.

One of my aims is to link frame-semantic theorems directly to empirical analytical practice on the basis of Fillmore’s idea that frames are both cognitive structures for organizing our experience and at the same time analytical tools that can be imple-mented to analyse precisely these structures.2 Apart from regularly analysing numerous examples to illustrate the empirical relevance of the theoretical findings, I conduct a comprehensive corpus-based analysis in the final chapter. This is intended to dem-onstrate, by way of example, the use of this approach to the analysis of discourse and metaphor. In addition, the sample analyses serve overall to show the integrative char-acter of the semantic approach developed here. Many linguistic phenomena, including

1. Cf. https://framenet.icsi.berkeley.edu/fndrupal/, accessed on 10 April 2014.

2. Cf. Fillmore (1985: 232): “In addition to seeing frames as organizers of experience and tools for understanding, we must also see frames as tools for the description and explanation of lexical and grammatical meaning”.

Introduction 3

for instance metaphors, presuppositions, basic discourse-semantic figures, (indirect) anaphors and cataphors, along with argument patterns, language change, coherence building, aesthetic-semantic effects (demonstrated using Haruki Murakami’s novel South of the Border, West of the Sun as an example), and much more can be addressed and explained under the auspices of frame semantics using frame-semantic tools. This has the advantage of providing a unified terminological and methodological apparatus to describe and analyse such linguistic phenomena. An additional advantage is that employing the frame-semantic categories developed here makes it always clear under what theoretical linguistic and cognitive premises the respective analyses are being conducted. These premises will be elaborated in the first two chapters.

The key premise for this entire study is the postulate of the “relevance of un-derstanding”, in short “U-relevance”, following Fillmore (1985) (cf. the discussion in Section 3.3). This states that no aspect of meaning that is relevant to understanding may be excluded from the analysis simply on the basis that the methodological premises of the chosen theory settings do not allow the application of any background knowledge to be taken into account. Frame semantics is understood as a non-reductionist theory of linguistic meaning. A theory of meaning is reductionist if its methods or premises do not adequately extend to including aspects of meaning or even impede their inclu-sion. In keeping with this premise, the cognitive and linguistic structure of frames is precisely described and their fundamental semantic relevance is demonstrated using examples of their application.

This book is divided into seven chapters. Apart from introducing the research focus, Chapter 1 provides a brief overview of the research literature and describes the larger cognitive-theoretical context in which frames are addressed. It shows that frames play a role in widely differing cognitive science approaches; however, what is exactly understood by frames and what Frames can achieve within a particular approach varies depending on the cognitive model chosen.

Chapter 2 describes two basic positions, a modularist one and a holistic one. In the German context, authors such as Manfred Bierwisch and Monika Schwarz take up a modular semantic position. This chapter examines the extent of influence that the methodological framework exerts on the conception of frame semantics in their respective two- and three-level models. The same question applies to the discussion of holistic models, above all Ronald Langacker’s Cognitive Grammar and Ray Jackendoff ’s Conceptual Semantics.

Chapter 3 is devoted to the holistic paradigm. This begins with the question of whether it is possible to draw a distinction between semantic and general encyclopae-dic knowledge. Various distinguishing criteria are tested for their validity. The finding that there are no criteria that legitimize drawing a sharp distinction between linguistic and world knowledge (conceptual knowledge) leads me to a conceptual approach to semantics which states that linguistic signs are embedded in a “space of understand-ing”. This means that linguistic signs are inferentially constituted by drawing upon varied background knowledge. In this context, the chapter discusses Dietrich Busse’s


non-reductionist theory of meaning as well as the results of psycholinguistic inference research. These are both systematically correlated with each other.

At the basis of a conceptualist semantic model is a theory of signs and grammar proceeding from the assumption that the smallest units of language are form-meaning pairs or constructions. This “symbolic principle”, as I designate it, is the subject of Chapter 4. Proceeding from considerations of construction grammar and Cognitive Grammar, the central proposition is that frames structure the content side of linguistic expressions. This chapter draws on Fillmore’s frame theory as well as Jordan Zlatev’s concept of “situation” and Jens Allwood’s concept of “meaning potential” to precisely determine the structure of frames.

Chapter 5 takes up the central finding of the previous chapter, which is that frames represent schematic units, and anchors frames in cognitive science and schema theory. I examine to what extent frames are cognitive schemata and to what extent these concepts share common characteristics. This paves the way for the analysis of the structural constituents of frames, giving the chapter the character of an introduction to the one following.

Chapter 6 is concerned with the linguistic representational properties of frames. The claim is that frames consist of three structural constituents, namely slots, fillers, and default values. However, what linguistic form do these structural constituents take? What different linguistic variants can be distinguished from one another? How can frames, their slots, fillers, and default values be determined in texts? These questions are in the foreground.

The final chapter employs the frame theory I develop here as a tool for corpus linguistics. It examines how the conceptual structure of the metaphor locust has constituted itself within the discourse of the so-called “capitalism debate”. In spring 2005, the then SPD party chairman Franz Müntefering likened financial investors to “a swarm of locusts”, thus triggering a public controversy. In the meantime, the expression locust has established itself in German as a metaphor for ruthless financial investors. Set against the background of the capitalism debate, I demonstrate which facets of knowledge are relevant for the emergence of the metaphor’s meaning.

Finally, I recapitulate the most important findings of the individual chapters and conclude with a brief look at some of the perspectives this book provides for research, as well as where and in what form further studies could take up the insights shared here.

chapter 1

Semantic interest in frames

This chapter uses a number of examples to introduce the subject matter of this book. Then an overview of frame research demonstrates the variety of areas to which it can be applied. Particular attention is paid to the way it has been received in the German context, in particular in German linguistics. The third section highlights the larger cognitive science context in which frames are addressed.

1.1 Evidence for frames

Instead of providing technical definitions, theoretical problems, and methodological intricacies, I would like to present a few examples to introduce the subject matter, “real-life examples”, so to speak.

In the Chicago Tribune on 2 August 2004 stood the following:

(1) A 60-year-old Japanese passenger on a flight leaving O’Hare International Airport caused a bomb scare late Sunday afternoon when a passenger saw him write the words “suicide bomb” on a piece of paper and alerted authorities, police said. United Airlines Flight 1184, scheduled to leave for Columbus, Ohio, was on the runway around 5:30 p.m. when the pilot learned of the note, turned the plane around and taxied to a nearby gate. All 120 passengers were taken off the plane. But authorities soon learned that the Japanese national, who was on the plane on business, was only writing words he didn’t understand so he could look them up later with a dictionary, said O’Hare police Sgt. Philip Deerig. The man was released, police said, and allowed to re-board the flight, which left three hours late. (Chicago Tribune, 2 August 2004)

What had happened? Obviously the expression suicide bomb had awakened certain associations in the passenger seated next to the Japanese man who had written down the unfamiliar English words. Yet, it was much more than that, since they were not just random associations, such as how heavy a suicide bomb might be or what it was made of or who had made it. The expression suicide bomb had elicited particular knowledge about the object of reference, such as what the object indicated by the expression is used for, where it is typically used, who normally uses it and for what reasons, what consequences its “use” could have, etc. The knowledge associated with the expression by the Japanese man’s neighbour appears to have been based on an organizational structure. Suicide bomb evoked a frame that contains certain information, some more and some less relevant.


The activated frame caused a chain reaction with far-reaching consequences. It was triggered by two words, words that at the time of the occurrence were merely “empty signs” for the Japanese man, form without content. The Japanese man was not in a posi-tion to connect his vocabulary note with knowledge about bombs and suicide bombers, let alone to make a connection between suicide bomb and the security concerns that crystallized as a result of the events of 9/11 – perhaps it would not even have struck the attention of his neighbour prior to 9/11. In order to grasp the meaning of suicide bomb the Japanese man would have had to know the frame with which the expression is connected and have knowledge at his disposal that is anchored in his world experi-ence and is prerequisite in order to understand the expression. Beyond that, he would have had to know the conventions that this knowledge normally associates with the expression suicide bomb.

The knowledge that language users connect with the form side of a linguistic ex-pression is on the one hand conventionalized, but particularly because of this, it spans the language community and is diachronically variable. Both extremes are marked in Example (1): The expression suicide bomb does not evoke a frame in the mind of the Japanese man; it is not possible for him to construe a context that could motivate the use and understanding of the expression. By contrast, the expression suicide bomb activated a very differentiated frame in the flight crew, the pilot, the security staff, and the police.

Any scale of activation between these two extremes is possible. To illustrate this, compare the meaning of the word menu in the following joke:

(2) A cannibal is sitting in the restaurant of a luxury cruise ship. The waiter comes along and asks, “Would you care for the menu?” The cannibal replies, “No, thanks. Just bring me the passenger list.”

What distinguishes the meaning of the word menu from the understanding of the word passenger list as a menu? When the waiter uses the word menu, he connects it with a frame that entails a variety of communicatively relevant information. A menu is for him a “prop” in a restaurant, an aid offered by a restaurant to its guests to facilitate the ordering of food and drink. The word menu also obviously evokes a very similar frame in the cannibal’s mind. The menu also has informational value for him, but in keeping with his expectations, the menu will list completely different meals.3 These ex-pectations can be gleaned from the circumstance that, in contrast to the typical guests at a restaurant, he is a cannibal. Apart from a different concept of what constitutes a “meal”, the other features of what the cannibal and the waiter imagine in connection with the word menu differ very little from each other (the typical material form, the typical place it is found, its purpose, etc.). For the cannibal, the passenger list and the

3. This presupposes that cannibals exist and that they typically eat human flesh, both of which can be questioned. The joke merely activates an extremely simplistic typecasting of cannibals and makes this the subject matter of the punch line.

Chapter 1. Semantic interest in frames 7

menu are one and the same, whereas he must explicitly request the passenger list (and not the menu) from the waiter, since his knowledge of what constitutes a meal differs from conventional knowledge.

The third example is a caricature that appeared in the Rheinische Post newspaper:

(3) The caricature shows a beach on which two men are sitting on deck chairs next to each other. Both of them are wearing straw hats. While one has a newspaper open in front of him, the other is holding a cocktail glass in his hand. A hotel is visible in the background. Both men are looking at each other. One of them asks, “Neckermann?” And the other replies, “Ruhrgas.” (Rheinische Post, 4 August 2004)

As in the first example, we are dealing with only two words. In order to establish co-herence out of the minimalistically reduced communication between the two men, the readers must draw on all the information at their disposal. Neckermann evokes knowledge of a tour operator, and in view of the situative setting in which the com-munication takes place as depicted in the image, it seems likely that the inquirer intends to find out which organization made the travel arrangements for the man sitting next to him. However, the answer conflicts with what would be typically expected. Ruhrgas evokes a frame around an electric power utility. Ruhrgas is not a company that organ-izes travel. At first sight, it has nothing to do with travel at all. However, awareness of current political affairs motivates a possible interpretation: in the period during which the caricature appeared in the newspaper, the investor-owned electric utility service provider E.ON Ruhrgas was facing criticism because it had targeted politicians with invitations to tourist activities in order to exert more influence over them. This information flows into the activated ruhrgas frame. Likewise, the responding person becomes identifiable as a politician. What is more, the fact that he answers the question so matter-of-factly implies that it is common practice to finance politicians’ trips. All this information (and much more) is highly contextual and can only be gleaned on the basis of background knowledge. Nonetheless, it is relevant to understanding and flows into the activated ruhrgas frame.

The role of inferences can be illustrated with another example from advertising:

(4) Bündnis 19. Die Lights von Lucky Strike. ‘Alliance 19. Lucky Strike Lights’

Most German readers will respond as I did: On first reading (4), the reader is mis-led by the phonological similarity between Alliance 19 (Bündnis 19) and Alliance 90 (Bündnis 90) and overlooks the semantic difference between the two expressions. Alliance 90 evokes a frame concerning an alliance of GDR civic movements that merged with the Green Party to form what has since then been known as Alliance 90/ The Greens. It is known that “Lucky Strike” is a brand of cigarettes, whereas “Lights” is a particular product line. But what have GDR civic movements got to do with cigarettes? The fact that it is Alliance 19 and not Alliance 90 allows a conceptual integration of the


semantic units ‘Alliance 19’ and ‘Lucky Strike Lights’ via the indirect route of ‘cigarette packet’. A packet of Lucky Strike Lights cigarettes contains 19 cigarettes, and these 19 cigarettes metaphorically constitute an “alliance”.

Linguistic expressions can clearly be ascribed the function of cognitive stimuli. They prompt recipients to construct a conceptual unit. Small differences in the form side can lead to large differences in the content side (e.g. Alliance 19 vs. Alliance 90). At the same time, it is easy to manipulate previously established cognitive constructs (frames), i.e. to retrospectively modify particular features. Sometimes it is necessary to make such modifications in order to understand a sentence or phrase. Consider the following example:4

(5) Everyone was having such fun jumping off the tree into the swimming pool that we decided to put a little water into it.

Upon reading the main clause, we have a rather clear image of the swimming pool being referred to. The relevant information in the activated swimming pool frame does not have the character of individual, disjointed facts but appears rather as a configured unit. We would have difficulty listing the relevant characteristics of the swimming pool and would probably omit many since they are simply obvious (for instance the expected size and material characteristics of the swimming pool, the depth, and the water temperature). Nevertheless knowledge of this kind appears to have played a role when reading the main clause, since the subordinate clause forces us to semantically reinterpret the expression swimming pool. Clearly it was “wrong” to imagine a swimming pool full of water. Without the subordinate clause, the standard interpretation would however remain intact: the “swimming pool” would be con-ceptualized as a retaining basin filled with water. Clearly we ascribe predicates to the reference object (in this case, the ‘swimming pool’) without these predicates having to appear in the text.

By the same token, a text may contain predicates without a reference object to which these predicates would correlate. A further example from cigarette advertising illustrates this:

(6) Meckert nicht, hat nie Kopfschmerzen und sieht auch noch gut aus. ‘Doesn’t nag, never has a headache and even looks good‘

Even cases like this present no problems to understanding. Ignoring for the moment that (6) actually appears on a billboard against the background of the Lucky Strike emblem, we are faced with the task of ascribing the three predicates doesn’t nag, never has a headache, even looks good to a reference object that has yet to be deduced. Hence, these questions arise: Which word could take the unoccupied subject position of the sentence? Who typically never nags, typically never has a headache, and typically looks good? In contrast to what the advertisement suggests, the usually expected answer is

4. Cf. the extensive discussion of the example in Section 6.5.


not “Lucky Strike”. Due to our communicative experience and our social knowledge, we know in fact that chauvinists typically ascribe these predicates to the perfect woman. The humour of the advertisement is created by the activation of two competing frames, those of lucky strike and the perfect woman, as well as by the fact that typically human attributes are ascribed to a commodity.

The issues presented in these six examples illustrate some of the characteristics that seem to affect linguistic meanings: (i) the form side of a linguistic sign constitutes the starting point of a cognitive conceptualization process for the content side of the same linguistic sign; (ii) the knowledge (in the form of predicates) that flows into the content side, although variable and highly dependent on context, is not, however, arbitrary since it must have a conventional character (Neckermann vs. Ruhrgas, menu vs. passenger list); (iii) default information is only relevant to the interpretation of a linguistic expression as long as contradictory data is not given (for instance, as to what constitutes the concept of a ‘swimming pool’ or a ‘menu’); (iv) the majority of meaning-relevant knowledge is founded in the experience of language users (e.g. knowledge of suicide bombs in relation to terrorist attacks, knowledge of menus in eating conven-tions); (v) ultimately the conceptual meaning of a linguistic expression is the result of cognitive inferential processes such as conceptual blending, reinterpretation, and modification (perfect woman vs. Lucky Strike; the features of a swimming pool).

This book proceeds from the premise that a linguistic expression (such as suicide bomb, menu, Ruhrgas, Bündnis 19, swimming pool) evokes knowledge in the form of interrelated informational units whose structural cohesion can be explained using linguistic means. Or more precisely, the object of reference of a linguistic expression is a conceptual unit that is linguistically specified by means of predicates (in the sense of predicate logics) and can be further specified by means of predicates. Following Fillmore (and many other cognitive linguists who are rooted in the tradition of frame semantics), I call these conceptual units “frames”. Frames have “slots”, i.e. when con-ceptual objects are called up, they can be more precisely specified using various (but not arbitrary) knowledge features. Since frames in this sense constitute schematic units, then the predicative specifications of frames are called “values” or “instances”. Values (or instances) “occupy”, and hence specify, particular slots. Language users categorize values in particular slots of an activated frame to particularize a conceptual unit. For instance the word swimming pool evokes a frame within which slots like ‘size’, ‘material qualities’, ‘content’ and ‘depth’ are more precisely defined by values such as ‘six-by-five metres’, ‘blue ceramic tiles’, ‘chlorinated water’, and ‘two metres deep’.

In this book I distinguish between two types of values. On the one hand, values can be motivated by the given text base (i.e. by linguistically realized predicates). In (5) for example, the slot ‘use’ is more closely defined by the value ‘jump off the tree into the swimming pool’. A swimming pool may be put to many purposes, but only this particular form of use is specified. The specifications within such text-based frames are referred to as “fillers”.

On the other hand, values may also be inferred, i.e. “mentally added”, to under-stand linguistic expressions. Thus, in Example (5) we assume that the swimming pool


mentioned has particular material qualities, a particular size and a particular depth, etc. Such values added on the basis of the text are referred to as “default values”.

The three structural constituents of frames – slot, filler and default values – have been used since the beginnings of frame theory (e.g. in Minsky 1975) and have also entered more recent publications (such as Coulson 2001), although at times termino-logical divergences are evident.5 Regardless of the fact that it is not yet by any means clear what exactly slots, fillers, and default values are, what status they should have, how they can be identified, and which linguistic variants can be distinguished, in the next chapters I presuppose a merely rudimentary understanding of the structural constitu-ents of frames. These questions become particularly important when frames are used in corpus linguistic investigations. For this reason, this is dealt with in more detail in Chapter 6. However, in Chapter 6 it is not necessary to have read the previous chapters, so interested readers can read about the respective explanations there at any time.

1.2 Frames in research

The next two sections concern themselves with the research literature on frames in the context of epistemological interest in cognitive semantics. First, Section 2.1 provides a brief overview of research since the mid-1970s. The focus is on presenting the es-sential development phases within semantics, with particular attention being devoted to Charles Fillmore, the “founder” of frame semantics. Instead of extensively discuss-ing individual research positions, I prefer to highlight the development trends in the course of the disparate history of the frame concept and take into account the particular reception of the frame concept in the German context.6

In Section 2.2, I introduce – also with the same proposed brevity – alternative con-cepts to frames that have developed in the last 25 years in Anglo-American cognitive linguistics but have only found a very rudimentary reception in German linguistics. In particular these concern Fauconnier’s theory of “mental spaces”, Langacker’s concept of “cognitive domain”, and Lakoff ’s “ideal cognitive models”. The aim is to determine the relation of frames to these semantic representational formats.

5. Thus Minsky (1975) speaks of “default assignments”, whereas Coulson (2001) speaks of “default values”; cf. Section 6.5.

6. Konerding (1993: 20–77) provides an extensive critical discussion of individual research positions.


1.2.1 The development of frame research

Initially intensive work on frames took place exclusively in the United States, chiefly at universities on the West Coast. At the end of the 1970s, when the frame concept was gradually gaining attention in linguistic semantics, its use for artificial intelligence research was already being keenly discussed (cf. for instance Bobrow and Collins 1975; Minston 1975; Metzing 1979). Neighbouring disciplines such as cognitive psychol-ogy and sociology (Goffman 1974) were also already using frames and schemata as analytical categories to describe knowledge structures that allow humans to interpret their experiential data. What was new at this time was at best the label “frame”, since the basic idea behind it originated in the 1920s and 1930s: without the psychological concept of “gestalt” or the schemata concept developed by the psychologist Frederic Bartlett (1932) in his theory of remembering, these different frame-theoretical ap-proaches in the last quarter of the 20th century would not have gained a foothold to the extent that they actually did.7

The term “frame” gained entry into linguistics via Fillmore’s studies on verb va-lency, so-called “case grammar”. In 1968 Fillmore’s influential work The case for case was published, which can be considered as an important precursor to his later Frame Semantics. The motivation for speaking of frames in connection with verb valency lay in the observation that the syntactic function of individual sentence elements could not be comprehended solely on the basis of those elements that are realized in a sentence. The elements that are not realized have just as important a role, but are inherent in the corresponding case frame. Case frames also describe the semantic valency of words, regardless of which of the arguments “required” by the words are actually syntactically realized in the sentences. Consider (7) in this context:

(7) He bought the house.

In (7) only two deep cases of the verb buy are realized, that is the agentive and the objective; however, relevant to understanding the verb are at least the deep cases in-strumental (that is, the money that is used to pay) and dative (that is, the seller of the house). One of the aims of determining all the deep cases required by a verb was to gain insight into the lexico-semantic representation of the verb. On the other hand, the proposed hierarchy of the postulated deep cases was intended to serve as a tool for identifying the fundamental syntactic organizational structures of sentences.

For many years case grammar was widely received and expanded in various ways. However, it faced criticism for two open questions: At what point can it be assumed that the list of possible deep cases has been exhaustively completed? How can a hierarchy of deep cases be collated and tested? Fillmore (1971c, 1977d) had attempted to take these

7. On the relevance of the gestalt concept cf. Fillmore (1977a: 60), (1982a); also: Lakoff (1977); Liebert (1992: 14–27); Roos (2001: 131–138). I deal with Bartlett’s Schema Theory in Sections 5.1 and 5.2.1.


points of criticism into account in his revised versions of his original case grammar, but the principle objection that there were no reliable criteria for determining the deep cases of a verb could not be refuted.

This failure to provide a well-motivated list of case notions became a central objec-tion to the theory… Many writers have pointed out that one can always find both reasons for recognizing ever more refined distinctions and reasons for recognizing high-level generalization, concluding that there could be no theoretically justified way of coming up with as single linear list. (Fillmore 2003: 466)

In retrospect it was arguably this insight that the central problem could not be solved that was decisive for the development of the principles for a theory of meaning that no longer oriented itself towards the syntactic realization of arguments in the form of deep structures as was the case with case frames.

Fillmore first speaks of “frame semantics” in 1975. Without involving case frames in his considerations,8 his focus turned to the idea that there were schemata

…which link together as a system, which impose structure or coherence on some aspect of human experience, and which may contain elements which are simul-taneously parts of other such frameworks. (Fillmore 1975: 123)

Whereas case grammar was still rooted in the tradition of generative semantics due to its orientation towards deep structures contingent on valency, with his conception of frame semantics Fillmore increasingly distanced himself from this (cf. Fillmore 1976b: 22–23; 1982a: 112–119). Nonetheless, a general movement away from genera-tive theories only began to form in the course of the 1980s, when Fillmore and linguists like George Lakoff, Ronald Langacker, and Leonard Talmy began to develop cognitive approaches that ran counter to these theories, which in the meantime are well known under the terms construction grammar and Cognitive Grammar.

However, Fillmore only shed his ties to generative semantics step by step. The first step was to interpret case frames semantically as linguistic tools for characterizing ab-stract “scenes” or “situations”, “so that to understand the semantic structure of the verb it was necessary to understand the properties of such schematized scenes” (Fillmore 1982a: 115).9 Whereas Fillmore was initially convinced that it would be sufficient to describe a schematized scene using the participant deep cases alone, in the course of his later studies he came to realize that many facets of knowledge relevant to under-standing could not be accounted for in this way. Hitherto unconsidered aspects affected the conditions that had to be fulfilled to adequately understand or use a linguistic

8. In a footnote Fillmore writes laconically that frames stand in the tradition of his own case grammar as well as the frame concept of Minsky and Goffman (Fillmore 1975: 130).

9. A detailed discussion of the relationship between the concepts of “deep case”, “scene”, “situ-ation”, and “frame” can be found in Section 4.2.3.


expression. The agenda was set by Fillmore’s (1985) advocation of “U-semantics”,10 which directed its focus on the data relevant to understanding an expression with-out already excluding (supposedly) negligible non-linguistic knowledge at the outset. Fillmore concluded that a very broad perspective is necessary to grasp data relevant to understanding, namely “a general account of the relation be tween linguistic texts, the contexts in which they are instanced, and the process and products of their in-terpretation” (Fillmore 1985: 222). In expanding the research focus in this manner, Fillmore finally discarded his sentence-oriented case grammar. It is the conception of U-semantics that motivates the frame-semantic model I attempt to develop in the present book. A major goal is to turn Fillmore’s rather “intuitive” heuristics into an analytical model that allows to apply U-semantic ideas and concepts to corpus data.

Fillmore’s conception of U-semantics is unique in many respects. Importantly, its analytical scope goes beyond the scope of both Case Grammar and the Berkeley FrameNet project (Busse 2012: 92–132, Ziem 2014). Unlike the latter, frames within U-semantics address not only valence patterns but also all knowledge facets that are relevant for fully capturing the meaning of a word or phrase.

One may argue as to what extent Fillmore’s concept of frame semantics was in-fluenced by work in AI research, particularly by Minsky’s (1975) recourse to frames and by Schank and Abelson’s (1977) Script Theory. What is certain, however, is that Fillmore owes a great deal to Bartlett’s (1932) Schema Theory, as do countless AI re-searchers. In addition, it is certain that Fillmore was already aware of Minsky’s ideas in 1975 and at the latest from 1982 knew of Schank and Abelson’s Script Theory (cf. Fillmore 1975: 124; 1982a: 137). The increasing preoccupation with frames in AI re-search in the mid-1970s also played such an important role for the development of frame semantics that it is worth reporting on some of its central ideas.

Apart from Minsky and Schank, a whole array of other researchers (like Benjamin Kuipers, Eugene Charniak, Patrick Hayes, Terry Winograd) were interested in the com-puter simulation of complex cognitive processes such as understanding texts. It soon became apparent that the ability to understand a text demanded far more knowledge than structurally oriented mainstream semantics gave linguistic expressions credit for. Distinguishing between purely linguistic knowledge (that is, knowledge of semantic features inherent to language) and general encyclopaedic knowledge proved to be a hindrance to the practical interests of AI research. The realization grew that, without considering the typical everyday knowledge that humans acquire and have learnt to use step by step during their socialization, any attempt at simulating the processes of understanding language would fail. Yet, in order to “implement” such conceptual knowledge (i.e. knowledge of situations, scripts, events, objects, etc.) on machines, it was first of all necessary to conduct some fundamental research. It was in this con-text that frame theory evolved as a model to represent human knowledge. I will not elaborate on the, in part, complex and technical proposals made by Minsky (1975) and

10. The “U” stands for “understanding” and is contrasted with “T-semantics”, which indicates a meaning-reductionist model oriented towards truth conditions.


Schank and Abelson (1977);11 instead I will introduce Minsky’s frame concept insofar as it is relevant to meaning-theoretical issues.

Minsky shares Fillmore’s understanding of frames as standardized formations of knowledge elements that are partially variable and change quickly, but are to some extent also resistant in nature. Minsky’s well-known definition is:

When one encounters a new situation… one selects from memory a substantial structure called a Frame. This is a remembered framework to be adapted to fit reality by changing details as necessary. A frame is a data-structure for represent-ing a stereotyped situation, like being in a certain kind of living room, or going to a child’s birthday party. Attached to each frame are several kinds of information. Some of this information is about how to use the frame. Some is about what one can expect to happen next. Some is about what to do if these expectations are not confirmed. (Minsky 1975: 212)

The fact that Minsky’s frame concept is rather vague – for example neither the status nor the quality of the frame elements are defined and it remains unclear what a “situa-tion” is and to what extent a frame adapts to this – seems to have increased its accept-ance rather than hindered it. At least in later approaches within semantics, numerous characteristics of frames can be found that Minsky had already highlighted. Consider, for instance, the network structure in which, according to Minsky, every frame is em-bedded and which every frame constructs itself because its elements (fillers, default values) also have the status of (sub-)frames. What is also generally recognized is the finding that the elements of a frame vary according to their stability and dynamics, as well as the basic premise that frames have a holistic character and originate on the basis of recurrent experience (and change on the basis of new experience). Although many details remain open, Minsky’s contribution can hardly be appreciated highly enough in terms of historical reception. Schank and Abelson (1977) also adopt Minsky’s frame conditions in their more activity-oriented Script Theory. The same conditions form the basis of Charniak’s (1976) attempts to model a system of frames using “painting” as an example.

Whereas studies on frames were readily received across disciplinary boarders, almost ten years passed after Fillmore’s early publications on frame semantics before a larger research group formed within linguistics. The workshop Round Table Discussion on Frame/Script Semantics organized by Victor Raskin for the 13th International Linguistics Congress in Tokyo in 1983 marked a certain turning point. The resultant collection of essays published two years later (Raskin 1985) documents that frame semantics was in the process of establishing itself as an independent field of research and that there was a growing interest in engaging with the issues it addressed.

At the end of the 1980s and beginning of the 1990s, a phase of general acceptance of frame theory began. A number of both English-speaking and German-speaking linguists took up frames as an empirical analytical tool, yet the auspices under which

11. Konerding (1993: 24–40) provides a compact summary.


this took place differed in each context. In the Anglo-American context, frame-based studies stood in close connection with construction grammar as developed by Fillmore and others at Berkeley (Fillmore, Kay, and O’Connor 1988; Kay 1997; Kay and Fillmore 1999). In individual cases frames were associated with Ronald Langacker’s (1987) Cognitive Grammar, which shared essential premises with construction grammar.12 In this theoretical framework, frames were employed, among others, in the areas of lexi-cology, lexicography, syntax, and pragmatics. The following are some salient examples.

Petruck followed lexicological research aims in her work on the linguistic concep-tualization of body parts in Hebrew (Petruck 1986 and 1995). Lexicographic studies were carried out by linguists such as Fillmore in collaboration with Sue Atkins, for in-stance on the lexeme risk, whose frame is characterized by two subframes chance and harm (Fillmore and Atkins 1992, 1994). Using larger text corpora, Fillmore and Atkins determined, among other things, the lexico-syntactic structures in which risk occurs. Such corpus linguistic analytical perspectives of frame structures were consistently and systematically continued in the large-scale FrameNet project at the International Computer Science Institute in Berkeley, USA (cf. Baker, Fillmore, and Lowe 1998). In the meantime, an immense number of publications have appeared in the context of the FrameNet project.13

In contrast, syntactical analyses stood in the foreground of O’Connor 1994 and Goldberg 1995. In the meantime, Goldberg’s construction grammar approach to the analysis of argument structures has become highly influential. Goldberg showed that argument structures can only be understood relative to background frames. Finally, some case studies have examined the conceptual content of linguistic expressions in specific domains of knowledge. For example, Botha (1996) and Tsohatzidis (1993) used a frame-analytical perspective to respectively examine specialist terms used in cricket and selected speech acts. The mentioned studies illustrate the extensive range of frame analysis; many other studies have not been listed here.14

In recent years a new trend in development has become evident. Deriving from his theory of so-called “mental spaces” (Fauconnier 1985), Gilles Fauconnier, in col-laboration with Mark Turner, developed a general theory of Conceptual Integration (“blending”) in the late 1990s (cf. for example Fauconnier and Turner 1998a, 1999); I elaborate on the relation between frames and mental spaces in the next chapter. This is interesting from a research-historical perspective because, since the end of the 1990s, the integration of frames (of a Fillmorean character) into the theory of conceptual integration has become increasingly topical. Thus, Coulson (2001) considers frame se-mantics and mental space theory as complementary approaches, illustrating this using numerous linguistic phenomena such as metaphors, compounds, scalar implicatures,

12. Cf. the explanations in Sections 1.3.2 and 4.1.1.

13. Cf. the bibliography on the official FrameNet website: http://framenet.icsi.berkeley.edu [last accessed: 10 April 2014].

14. Cf. the (similarly selective) literature overviews in Fillmore (2006) and Petruck (1996).


and conditional clauses. Similar considerations are found, for instance, in Fauconnier and Turner 1998b and Sweetser 1999. Another effort to connect frame semantics more tightly with cognitive operations and other representational formats (not only mental spaces, but also “image schemas”, cf. Watters 1996) is Baker’s 1999 dissertation. In ad-dition to conducting frame semantic analyses to determine the conceptual structure of the verb see, Baker also carried out lexicological studies and psycholinguistic experi-ments; mental spaces were also employed.

It is worth taking also the German research literature into account. Here, an in-tegrative perspective on frames such as the one mentioned above is not found. This is partially due to the fact that the German literature ignores the embedding of frames in construction grammar and Cognitive Grammar that occurs in Anglo-American literature. As a result, one fails to find references to related concepts such as Langacker’s “cognitive domains” or Fauconnier’s “mental spaces”. The fact that the specific needs of German research have steered engagement with frames in a biased direction is likely to be a further reason for this deficit in reception. The use of frames for strictly semantic enquiry only occurred at a very late stage. Certainly, Immo Wegner had already as-sociated frames with questions of lexical semantics in 1979, referring in particular to Minsky (1975) and Fillmore (1975), but this article remained largely ignored. Some years later, Wegner (1984, 1985) presented two further studies in which frames were employed largely for lexicographical purposes.15 Konerding’s (1993) linguistic ground-work in frame theory stands in this tradition. None of the mentioned studies contain any further reflections from the background of construction grammar or Cognitive Grammar.

The same applies for other monographs. In their corpus-based case studies, Fraas (1996a), Lönneker (2003a), and Klein and Meßner (1999) oriented themselves towards the operationalization proposal developed by Konerding.16 In terms of historical re-search, these studies have been extremely important, since they exemplify how linguis-tic frame analyses can be set to a great variety of purposes; however, they had no effect on cognitive theory or linguistics, because apart from employing frames empirically, they did not demonstrate the significance of the basic premises of frame theory for theories of meaning or even linguistics.

As for the range of application of frames, the German context is just as varied as the English-speaking one. It extends from the analysis of metaphors (Klein 2002a) and the already mentioned lexicographical and lexicological studies (Wegner 1984, 1985; Konerding 1993) through to argument analysis (Klein 2002b) and individual historical

15. Frames also found passing mention at this time in van Dijk (1980: 178) and Scherner (1984: 219). Von Polenz (1985: 130ff., 156) spoke with reference to Fillmore (1968) and (1977a) of “reference frames” and “predication frames”, cf. my critical remarks in Section 6.3.1.

16. A brief summary of the procedures and research aims of Fraas (1996a), Klein and Meißner (1999), and several other empirical contributions such as those of Holly (2001, 2002) and Klein (1999, 2002a, b) are found in Section 7.1.1.


semantic case studies (Blank 1999; Fraas 1996a). In addition there is Konerding’s (1996) frame analysis of speech act verbs and Müller’s very early (1982), but largely ignored, discourse analytical study. Suggestions were even made to include frames in the semiotics of literary texts (Müske 1991, 1992). Busse already discussed frames at the beginning of the 1990s, initially in the context of studies inspired by text linguis-tics (e.g. in 1991a). In his more recent studies, Busse (cf. 2005a, 2006, 2008a, 2009b) has developed the basic principles of a non-reductionist theory of meaning following Fillmore and Minsky. Since these abandon the traditional division between linguistic and conceptual knowledge, Busse (2003a) speaks of a “semantic epistemology”, which is the concept also subscribed to in this book. However, it would certainly be too soon to claim that a frame-based epistemology presents a new development trend.

The fact remains that in frame research the most obvious distinction between the German and Anglo-American contexts concerns the premises under which frames are addressed. These vary from case to case in German studies, since there is no com-mon (methodological, cognitive and/or linguistic) framework that unites the studies. The “normalizing” or even “synergizing” power of a shared paradigm (in the Kuhnian sense) is lacking. Thus, it comes to a form of “parcelling” of the individual contribu-tions. By contrast, frame studies in the Anglo-American context follow the guide-lines of conceptualist semantics (with the character of either construction grammar or Cognitive Grammar).

The lack of reception of conceptual approaches in the German context beyond frame theory itself gives me occasion to outline the most important representa-tional formats. Cognitive Grammar and construction grammar are introduced in Section 1.3.2 and further elaborated in Chapter 4.

1.2.2 Frames and other representation formats

The definitions of frames in the literature vary considerably,17 and while some can be seen as complementary, others have such different theoretical premises as their base that they consequently accent issues that are of no or very little relevance to a linguistic theory of meaning. In the meantime, in various disciplines such as linguistics, sociol-ogy, media studies, and AI research, the literature on frames is almost impossible to survey. The specific requirements of each discipline often give rise to varying defini-tions of what frames are, what they can do, and how they can be analytically employed (cf. the overviews in Fisher 1997; Konerding 1993: 23–63). In the following, with the exception of Minsky’s approach, I only elaborate on the formation of linguistic theory. In order to partially account for the lack of reception of conceptual approaches in

17. Cf. for instance Fillmore (1975); Minsky (1975); Charniak (1976: 362); van Dijk (1980: 233); Beaugrande and Dressler (1981).


German linguistics, as noted in the previous section, I will compare frames with other representational formats that have until now received little attention.

As mentioned, Fillmore understands frames as conceptual structures that both form the basis of the meaning of linguistic expressions and motivate their use. In one well-known passage he says:

By the word ‘frame’ I have in mind any system of concepts related in such a way that to understand any of them you have to understand the whole structure in which it fits; when one of the things in such a structure is introduced into a text, or into a conversation, all of the others are automatically made available. (Fillmore 1982a: 111)

A frame is thus always a unity of multiplicity. Taking up an example from Fillmore (1982a: 118), understanding a word like weekend presupposes a background frame that encompasses concepts like “week”, “day”, “Saturday”, “Sunday”, “work”, and “leisure”. The word weekend can only be understood against the background that a (cyclically recurring) week has seven days, five of which are workdays.

Fillmore (1985: 223) stresses that such conceptual structures had already been described by researchers such as Minsky (1975), yet at times using other terminology; for instance Lakoff (1983) speaks of “cognitive model” while Langacker (1984) uses “base”.18 Are “frames” (Minsky), “cognitive models” (Lakoff), and “bases” (Langacker) equivalent to Fillmore’s “frames”? If not, where is the difference?

Even within Fillmore’s own work, we find conceptually different albeit related conceptions of frames (for an extensive discussion cf. Busse 2012: 23–250; Ziem 2014). First, as mentioned earlier, case frames address so called “deep cases” that describe the semantic valency of verbs, adjectives and nouns (Fillmore 1968). By presuppos-ing a limited roster of assumingly universal semantic roles, case grammar subscribes to major assumptions of Transformational Grammar. In contrast, the conception of “frames of understanding” (Fillmore 1985) takes a much broader perspective; it focuses on any kind of knowledge that is relevant to fully understand a linguistic expression. In the following, I elaborate this idea in more detail. Second, the Berkeley FrameNet project aims at building up a lexicographic database documenting valence patterns of Lexical Units (word-meaning pairs). The research scope of FrameNet is thus also much broader than the one of Case Grammar. At the same time the valency-oriented approach in FrameNet forms only a part of the knowledge relevant for understanding an expression as targeted with “frames of understanding”.

Minsky (1975) shows to what extent frames describe event structures (e.g. those of a children’s birthday party), explain the function of an engine, or govern the process-ing of visual data (for instance of a cube or a room). Not least, Minsky demonstrates

18. In addition, Fillmore mentions Bartlett’s (1932) schema concept, his own concept of “scene” (in Fillmore 1977a), Lakoff and Johnson’s (1980) Gestalt concept, and many more. It is doubtful whether frames in Fillmore’s terms do in fact encompass the conceptual structures described by these terms. For the relation between the concepts of “frame” and “scene” see Section 4.2.3.


the relevance of frames for processes of understanding language (for instance for the understanding of storytelling and the conceptualization of lexical knowledge), and he even claims that frames are suitable for the representation of such abstract knowledge structures as paradigms in the Kuhnian sense.

According to Minsky, not only the understanding of language but perception in general is governed by highly complex cognitive processes that cannot be supported by the available perceptual data alone, since such a narrow data base is not sufficient, in Minsky’s view, to explain why what we perceive appears meaningful to us and is inte-grated into our “stream of consciousness” with such apparent ease (Minsky 1988: 257). The perceived coherence occurs only after taking recourse to further mentally added informational units. Coherence is accordingly the result of a successful correlation of perceptual data with these mental informational units.

Our idea is that each perceptual experience activates some structure that we’ll call frames – structures we’ve acquired in the course of previous experience. We all remember millions of frames, each representing some stereotyped situation like meeting a certain kind of person, being in a certain kind of room, or attending a certain kind of party. (Minsky 1988: 244)

By “frames” Minsky understands “a sort of skeleton, somewhat like an application form with many blanks or slots to be filled” (Minsky 1988: 245). Since many slots of a frame are already specified with default assumptions, frames are similar to stereotyped data structures saved in the long-term memory that govern different processes of under-standing.19 Whereas default assumptions represent the average anticipated number of knowledge features – that is, in sum they are equivalent to our stereotyped knowledge of the world – concrete experiential data are the slot fillers that cannot be implied or inferred, since they are perceptually “given”.

Frames come into play anywhere where there is an overlap between perceptual and experiential units. Minsky’s thesis is that this process is same whether it is a case of understanding language or visually perceiving objects. The fact that perceptual data activates frames means in both cases that it can be understood as a component of a more comprehensive knowledge structure in which “stereotypical situations” are represented.

Minsky makes this clear using a simple example. In order to understand the meaning of the word chair, it is first of all necessary to draw on a set of informa-tion that describes the different features we expect to be peculiar to a chair – that is, relative to our experience to date (Minsky 1988: 245). A chair has, for instance, four legs, a backrest, a horizontal seat, etc. However, these are assumptions that merely more closely define a typical chair. Should the contextually given information be more specific, these assumptions can be replaced at any time (thus, reference may

19. “Default assumptions fill our frames to represent what’s typical. As soon as you hear a word like ‘person’, ‘frog’, or ‘chair’, you assume the details of some ‘typical’ sort of person, frog, or chair. You do this not only with language, but with vision, too.” (Minsky 1988: 245)


more specifically be to a three-legged chair). The decisive factor here is that, analo-gous to the concept chair, every linguistic unit is semantically highly underspecified, while understanding only occurs once provisionally valid assumptions override this underspecification.

The situation is similar for visual perception of objects. In this case, frames “with many blanks or slots to be filled” (Minsky 1988: 245) play an equally important role. To be able to see a chair behind a table and recognize it as such, it is normally sufficient to see its backrest above the table surface. The fact that the table is occluding the view of the seat and legs of the chair is no obstacle. In a sense, we imagine the table as a transparent object. Yet, the backrest could just as easily be a wooden panel attached to the table. However nothing prompts us to make this atypical assumption, since we are accustomed to it being otherwise.

What makes Minsky’s frame theory different to Fillmore’s is the aspiration to show that frames are a general format for knowledge representation. Minsky uses frames in particular to explain phenomena of visual perception.20 Thus, to the best of my knowl-edge, this has been the only elaborate attempt to date to embed frames in a general cognitive theory. Minsky’s considerations are ground-breaking, but especially for this reason many questions remain unanswered (cf. Konerding 1993: 25): What status do slots and default values have in visual and linguistic frames? How do they manifest themselves phenomenally? Can Minsky’s postulates support theories of remembering and cognitive psychology? How can frames be implemented analytically?

Although Minsky’s approach raises more questions than it answers, it is the fore-runner of a methodological reorientation in cognitive science. Frames are – in the sense of the holistic paradigm (cf. Chapter 3) – modally non-specific representational formats, making a division between linguistic and non-linguistic knowledge prob-lematic. In addition, contrary to Chomsky’s modularistic approach to language and cognition, another basic assumption is that basic and general cognitive abilities (such as categorizing, schematizing, etc.) are equally valid for all modalities of perception.

Thus it can be concluded that Minsky used frames in an attempt to embrace quite different cognitive phenomena, which are summarized in Table 1. By contrast, the research focus of Fillmore’s frame theory is limited to conceptual structures that prove to be relevant to understanding linguistic expressions.

Unlike Minsky, Lakoff (1987: 68ff.) understands his theory of an “idealized cogni-tive model”21 (ICM) expressly as a contribution to linguistic semantics. However it is clear that ICM is by no means equivalent to those conceptual structures that Fillmore attempts to cover using frames. Like frames, ICMs are also model-like units of knowl-edge that do not reflect segments of reality but rather represent idealized representa-tions of recurrent experience. In this sense, Lakoff ’s main thesis is “that we organize

20. Cf. Minsky (1975: 211–230; 1977: 355–364; 1988: 244–259).

21. I prefer the term “idealized cognitive model”. Lakoff ’s (1987) theoretical concept expands on the approach outlined in Lakoff (1983).


our knowledge by means of structures called idealized cognitive models, or ICMs, and that category structures and prototype effects are by-products of that organization” (Lakoff 1987: 68). A few lines later, however, he states:

Each ICM is a complex structured whole, a gestalt, which uses four kinds of struc-turing principles: propositional structure, as in Fillmore’s frames image-schematic structure, as in Langacker’s cognitive grammar metaphoric mappings, as described in Lakoff and Johnson metonymic mappings, as described by Lakoff and JohnsonEach ICM, as used, structures a mental space, as described by Fauconnier. (Lakoff 1987: 68)

With ICM, Lakoff apparently attempts to encompass quite different facets of knowledge whose relevance to theories of meaning has been proven in a variety of approaches (such as Fillmore 1982a; Fauconnier 1985; Langacker 1987; Lakoff and Johnson 1980). ICM is thus an integrative model that attempts to account for the fact that conceptual knowledge is organized through structuring principles. Frames are only one structur-ing principle among many. When Lakoff assigns frames the function of capturing the “propositional structure” of an ICM, he means the elements (fillers and default val-ues) of an activated frame as well as the relation between the frame elements (Lakoff 1987: 285). Frames may be justifiably used in this manner because they have a gestalt-like character, which to some extent “dissolves” in propositional structures. Since the frame weekend encompasses, for example, concepts such as “day”, “Saturday”, and “Sunday”, propositions such as (8) and (9) can be constructed:

(8) A weekend consists of two days. (9) A weekend consists of the days Saturday and Sunday.

It is important, however, to note that Lakoff (1987: 284) proposes four additional struc-turing principles for propositional structures: (i) scenes and scripts (with reference presumably to Schank and Abelson 1977), (ii) feature bundles, (iii) taxonomies, and (iv) prototypically organized radial categories. However Lakoff does not explain what

Table 1. Frame types in Minsky (1975)

Frame types Examples

Perceptual Visual (such as the image of a cube)Visual-deictic (such as perspectives when entering a room)

Psychological FormsTechnological Function of an engineSociological Event types (such as birthday party, visit to restaurant)

Habits (such as components of social events)Historical-scientific Paradigms (in the Kuhnian sense)


exactly is to be understood by “feature bundles” and “taxonomies” or to what extent they are integral components of ICM. Conversely, he elaborates all the more on ICM consisting of organized categories and being able to change various prototypicality effects of the category structures (cf. the overview in Evans and Green 2006: 270ff.). However, no explanation as to the relationship between frames and prototypes is given. In fact, frames are not addressed any further, so the question of the extent to which they are really relevant to ICM ultimately remains unanswered.

The fact that frames are only one of five total options to propositionally structure ICM may partially account for the marginal role that frames play in Lakoff ’s theory, but in addition, propositionality itself is only one structuring principle among many. So-called “image schemas” constitute a second structuring principle. Lakoff understands image schemas as very abstract structures (like, for instance, the container schema) that evolve through experience and form the basis of conceptual structures (thus the container schema motivates the interpretation of prepositions like in or on).22 Two additional structuring principles concern the correspondences between conceptual structures. Thus, propositional or image schematic structures can be metaphorically transferred to other areas of experience, for instance, when mental or physical states can be conceived as being a container (as in brimming with rage or to be in trouble). Lakoff distinguishes metaphorical correspondences from metonymical ones. One of the latter occurs when part stands for whole, for instance, when a mother is stereotyped as a housewife and this stereotype stands for the whole category of “mother”. Finally, Lakoff mentions “symbolic” structures. They emerge whenever linguistic elements are associated with conceptual elements (Lakoff 1987: 289). Interestingly, Evans und Green (2006: 281) see frames as the ordering structures of these conceptual elements, although according to Lakoff, frames determine the propositional structure of ICM. Apparently there is no consensus concerning the functional definition of frames in ICM, probably also due to the fact that Lakoff ’s explanation remains very vague in this respect. A comparison between “propositional” and “symbolic structures” suggests that they should not be separated and that both describe inherent features of linguistic signs. Propositional structures apparently correlate to syntagmatic correspondences, whereas symbolic structures correlate to paradigmatic correspondences. Whether Lakoff would agree with this definition must remain open; Lakoff himself offers no alternative explanation.

The outcomes of the previous remarks are summarized in Table 2. Overall it re-mains unclear to what extent frames structure ICM, in what relation frames stand to other structuring principles, and whether frames really constitute an alternative

22. Cf. also Section 2.1. According to Lakoff, image schemas are the most important structuring principle of ICM. Lakoff (1987: 284) explains the role of image schemas as follows: “ICMs are typically quite complex structures, defined by image schemas of all the sorts… Some symbols in an ICM may be directly meaningful: the basic-level and image-schematic concepts. Other symbols are understood indirectly via their relationship to directly understood concepts. Such relation-ships are defined by the image schemas that structure the ICMs.” Cf. also Lakoff (1987: 81).


concept to mental spaces, scenes, and scripts, as Lakoff suggests. Although Lakoff ’s concept of ICM aims to integrate linguistic frames (in the Fillmorean sense) into a universal model, in a similar way to Minsky’s frame theory, the incorporation of so many theories and concepts appears to lead to additional terminological confusion. What are “feature bundles”? Aren’t “taxonomies” simply the prototypical organization of categories? What distinguishes “scenes” from “scripts”?23 Taylor is correct when he observes:

The terminology in this area is confusing, partly because different terms may be used by different authors to refer to what seems to be the same construct, or the same term may be used to refer to very different constructs. Furthermore, it is not at all clear that it is possible to make clean conceptual distinctions in this area. Nevertheless, I found the term ‘frame’ to be a useful theoretical term, denoting the knowledge network linking multiple domains associated with a given linguistic form. We can reserve the term ‘script’ for the temporal sequencing and causal rela-tions which link events and states within certain action frames. (Taylor 2003: 89)

After noting the correspondence between “frames” and “scripts” addressed by Taylor, I conclude by comparing frames with “mental spaces” and “cognitive domains”.

The term “domain” is a term used in Langacker’s (1987) Cognitive grammar”.24 In Langacker’s theory of meaning, “cognitive domains” are assigned the same central role

23. At one point, Lakoff (1987: 284) indicates that the expressions “scene” and “script” are syn-onymous. However, Sanford and Garrod (1981), who introduced both these terms, certainly distinguish between the two.

24. Langacker distinguishes between two types of “domains”: “abstract” and “basic” domains. The following discussion refers only to the former. “Basic domains” are very abstract conceptual units that cannot be traced back to other conceptual units. Examples of this are time, space, temperature, pain, and many others (Langacker 1987: 147–152). Matrix frames, which I in-troduce in Section 6.3.2 following Konerding (1993), are in this respect the counterpart to “basic domains”.

Table 2. ICM types and their manifestations according to Lakoff (cf. 1987: 284)

ICM types Type of structuring/Examples

Image schematic e.g. ContainerPropositional Propositional ICM (Fillmore’s “frames”, Fauconnier’s “mental spaces”)

Scenes, scriptsFeature bundlesTaxonomiesPrototypically organized categories

Metaphorical Projections (from a source domain onto a target domain)Metonymic esp. part-for-whole correspondencesSymbolic Frames (according to Evans and Green 2006: 281)


as frames are in Fillmore’s theory. Yet, both Langacker and Fillmore proceed from the same premise that linguistic meanings are purely conceptual structures and that they can generally only be understood against the background of superordinate knowledge structures. Langacker uses the term “cognitive domains” to describe such knowledge structures that are relevant to understanding. Domains are conceptual units of vary-ing degrees of abstraction. They provide the background information that makes un-derstanding a linguistic expression possible. Without reference to domains, linguistic meaning cannot be understood – not even in a rudimentary form.

[S]emantic units are characterized relative to cognitive domains… any concept or knowledge system can function as a domain for this purpose. Included as possible domains, consequently, are the conception of social relationship, of the speech situation, of the existence of various dialects, and so on. (Langacker 1987: 63; italics in bold in the original)

In this respect, domains are no different to frames. According to Fillmore (1985), ana-logue contextual and co-textual data along with general background knowledge and extra-linguistic factors motivate the understanding of an expression. In this respect the frame and domain concepts are equivalent. For example, the meaning of the expression thumb can only be understood relative to the domain or frame hand (and in turn the expression hand only relative to the domain or frame arm, etc.).

However, domains differ from frames in two respects. On the one hand, Langacker stresses much more highly than Fillmore the fact that every concept is specified with respect to several domains. These give rise to a so-called “domain matrix” (cf. Clausner and Croft 1999: 7). Thus, the concept “human”, for example, is determined relative to a multitude of very different domains, such as those of ability, size, physical consti-tution, etc. On the other hand, Langacker is more keenly interested in the structure and organization of our conceptual system; his aim, at least according to Evans and Green (2006: 231), is to achieve a “conceptual ontology”. By contrast, Fillmore strongly integrates grammatical structures (so-called “constructions”) in his analysis.

Further, whereas Evans and Green (2006: 231) maintain that Fillmore takes less account of the hierarchical organization of the conceptual system than Langacker, in my view neither Langacker nor Fillmore have paid enough attention to this aspect. In the meantime, however, Langacker 1987: 68, 147ff.; 1991a: 61–64) highlights the fact that each domain presupposes at least one hierarchically higher domain. This leads to conceptual hierarchies such as (10), in which the arrow is to be read as “presupposes”:

(10) (finger) knuckle → finger → hand → arm → body25

Nevertheless, Langacker provides no information as to what conceptual hierarchies may be used for or which empirical procedure is used to construct them. Fillmore had already recognized the relevance of conceptual hierarchies to theories of meaning prior

25. The domain body in turn presupposes the basic domain (see above) space.


to Langacker – in the form of what he calls “taxonomies” (cf. Fillmore 1982a: 132); similarly, however, he has not used them for analysis within frame semantics.26

Frames are structurally equivalent to domains in that both feature foreground and background dimensions. Langacker illustrates this distinction, which can be traced back to gestalt psychology (cf. Metzger 1923), using the example of hypotenuse. The conceptual pair foreground/background (or figure/ground) is reflected in Langacker’s contrast between a semantic “base” and a semantic “profile”. No one can understand the meaning of the word hypotenuse unless they know what a triangle and a right angle are (cf. Langacker 1988b: 59). Whereas these two cognitive domains constitute the base, the expression hypotenuse profiles a segment of the knowledge domain right triangle: it refers to the side of the triangle that lies opposite the right angle. When Fillmore (1985: 228) takes up the hypotenuse example, he is not so much highlighting the figure/ground distinction as the culturally specific background knowledge required for understanding. Nonetheless, Fillmore underlines the fact that activated frames give perspective to knowledge.

Sometimes the perspective which a word assigns is not a perspective on the cur-rent scene – something that might be visible in a pictorial representation of the scene – but is that of a much larger framework. Thus, the description of someone as a heretic presupposes an established religion, or a religious community which has a well-defined notion of doctrinal correctness. (Fillmore 1982a: 123)

Following Langacker’s profile/base distinction, the expression heretic profiles the prop-erty of a person against the background of knowledge of an established religion or religious community. At the very least, the domains (or frames) person and religion serve as bases.

In summary, it is clear that Langacker’s concept of cognitive domains is compatible with Fillmore’s concept in many fundamental aspects and there are even identifiable points of equivalence between the two. Differences derive in particular from the larger theoretical framework in which frames or domains are addressed. Domains are firmly anchored in Langacker’s Cognitive Grammar, whereas frames are an inherent part of Fillmore’s and Kay’s conception of construction grammar. In view of the observed similarities between domains and frames, it is, however, surprising that Langacker equates domains with Lakoff ’s ICM:

An abstract domain is essentially equivalent to what Lakoff … terms an ICM (for idealized cognitive model) and what others have variously called a frame, scene, schema, or even script (at least in some uses). (Langacker 1987: 150; italics in bold in the original)

After the previous analysis it should be apparent that this is hardly justifiable. Although frames and domains are similar in many respects, there are significant differences between domains and ICM as well as frames and ICM.

26. In Section 6.3 I show that conceptual hierarchies are essential, especially to frame semantics.


Finally I turn to Fauconnier’s theory of mental spaces. It is apparently unclear how mental spaces stand in relation to frames and ICM. While Lakoff (1987: 68) claims that mental spaces are structured by ICM, Fauconnier and Turner (1998a: 137) consider frames to be the structuring principle of mental spaces. The remainder of the discus-sion in this section places the focus on the extent to which frames and mental spaces may be seen as complementary approaches.

What are “mental spaces”? Fauconnier, who is considered the main representative of the theory of mental spaces, introduced the term “mental space” in 1985 to more strongly accentuate the cognitive process of meaning construction. Building on these considerations, Fauconnier has in the meantime developed a general theory of con-ceptual integration (“blending”) in collaboration with Turner (Fauconnier and Turner 1998a, 2002). Mental spaces are “partial structures that proliferate when we think and talk, allowing a fine-grained partitioning of our discourse and knowledge structures” (Fauconnier 1997: 11). Thus, mental spaces differ from domains and frames in that they only evolve during the act of understanding. Mental spaces do draw on data in our long-term memory, but they are constantly changing during a conversation or while reading a text.

Therefore, while the cognitive operations that are involved in the construction of new conceptual units are fundamental to mental spaces, the conceptual structures that are drawn upon in this process relate to cognitive domains and frames. It thus seems logical to understand frames in Sweetser’s use of the word as structures of mental spaces:

Mental spaces have internal structure which includes frame … structure; one could view Fillmore’s … frame as a rather schematic (partially-filled) and con-ventional mental space, or as a possible internal structural component of more filled-out mental spaces. (Sweetser 1999: 135)

While frames give cognitive access to relatively stable background knowledge, mental spaces additionally contain very detailed informational units relevant to the act of understanding.

According to Fauconnier, these meaning constructions are guided by two pro-cesses: by the construction of mental spaces on the one hand and by the correlation of selected elements in the constructed mental spaces on the other. Fauconnier names this process of correlation of elements “mapping”. A multitude of linguistic phenomena (like metaphor, metonymy, counterfactual references, knowledge modes) can only be explained by mapping.

Example (11) illustrates this:

(11) If Peter were John, Peter would be rich.

Peter and John are the two linguistic elements in Example (11) that prompt the con-struction of a mental space. In counterfactual references of this type, conceptual knowl-edge elements from both mental spaces are correlated with each other. An array of possible propositions relevant to understanding can be derived by correlating indi-vidual elements from both mental spaces. Figure 1 uses dashed lines to depict two mappings, adopting the notation form used by Fauconnier and Turner (1998b).


Mental space I: Peter Mental space II: John

Mental space III: Peter as John

d1d1

a2

a1’ a2’

b1’c1’

c1

b1 a1

Figure 1. Conceptual integration using the example If Peter were John, Peter would be rich

Mental space III allows the counterfactual supposition that Peter and John share the property of being wealthy. The conceptual property of John-being-wealthy (a2 in Figure 1) is mapped from mental space II into mental space III. The conceptual prop-erty of ‘Peter-not-being-wealthy’ (a1 in Figure 1) is also mapped from mental space I into mental space III and replaced by a2.27 Now a multitude of possible correlations of knowledge elements result from the construction of mental space III, for instance, that Peter, like John, owns a villa and plays golf every day (c1 in Figure 1), but also some – such as b1 – that do not belong to mental space II.28 It would be possible, for instance, to say If Peter were John, Peter would be rich and would have four children, although John in reality has none. The concept ‘being wealthy’ would be the departure point for further inferences that would belong to mental space III but would not correlate to

27. In order to make clear that the elements projected in mental space III occupy the same slot in mental space I and II (in this case, ‘financial status’) but are specified differently (in this case, ‘being wealthy’ vs. ‘not being wealthy’), Fauconnier and Turner link these elements in the graphical representation using a solid line (in this case, between a1 and a2); these are missing in Figure 1. In addition, a superordinate generic space that contains common facets of knowledge of both inputs is missing.

28. This is an insight of blending theory and thus not explicit in Fauconnier (1985). Every so-called “blended space” contains new, emergent elements that are not inherent to any of the participant inputs. As Fauconnier and Turner (2002) impressively bear witness to, this also ap-plies for a multitude of metaphors; cf. also Section 7.1.2.


mental space II. This could express that children cost a lot of money and for that reason Peter cannot “afford” four children although he would love to have some if he were rich.

Finally, there are conceptual elements that are in both mental space I and mental space II that are identical. Perhaps John and Peter both own a Harley Davidson (d1 in Figure 1) The element d1 is therefore not mapped into mental space III because the sentence If Peter were John, Peter would be rich and would own a Harley Davidson would not contain new insights. Peter already owns a Harley Davidson, so this would not change if he were as rich as John.

In contrast to Fauconnier and Turner (1998a, 2002), Fillmore (1975, 1977a) does not introduce the frame concept to mark the status of linguistically expressed knowl-edge (beliefs, wishes, hypothetical and counterfactual assumptions). His greater con-cern is that expressions with apparently less complex reference areas (like weekend, write, poison, right; cf. Fillmore 1977a, 1982a) can also only be understood in the context of schematically correlated knowledge. Both approaches can be seen as com-plementary to the extent that frames structure background knowledge that also flows into mental spaces (cf. Coulson 2001). In Example (11), frames concern relevant back-ground knowledge about Peter and John.

What is more, frame theory and the theory of mental spaces are both based on the same cognitive theoretical premises. Both stress the construction character of con-ceptual integration processes, and for both, linguistic meanings are purely conceptual units. The difference lies in the focus. While Fillmore’s frame theory concentrates on lexical meaning, mental spaces take account of the cognitive operations that guide the construal of conceptual structures.29

Overall it can be said that while Fillmore’s frame approach is largely compatible with both Langacker’s theory of cognitive domains and mental spaces, this does not apply to the same extent with Minsky’s frame approach or Lakoff ’s ICM. The two latter approaches lack any theoretical reflection on the fact that the many and varied descriptive models which are referenced may be compatible. It also remains unclear how cognitive phenomena (cf. Table 1 and 2) should be investigated within the pro-posed frame/ICM concept.

Nonetheless in my view it is important to attempt to explain the different cognitive phenomena within a unified model – as Minsky and Lakoff aimed to do. Such a model could either, as Lakoff suggests, unify various representational formats (like scripts, scenes, frames, domains, mental spaces, etc.) or, in Minsky’s sense of “frames”, provide a general format that can encompass various cognitive phenomena (cf. Ziem 2006c). At this point of time, the development of an integrative-holistic model is still only a distant goal in linguistic research.

29. These are, above all, mapping, blending, and compression; cf. Fauconnier and Turner (1999), (2000), (2002).


1.3 Frames in cognitive science

Whereas the previous two sections dealt with frames in the stricter sense, we now turn our attention to the universal research context of cognitive science. This section outlines the research focus of cognitive science with regard to frame-relevant enquiry. In linguistics there are a variety of distinct cognitive theoretical positions that are motivated by either a holistic or a modularist understanding of language. These posi-tions are presented in Section 3.2. Finally, the fact that there is very little clarity in the German context as to which cognitive model frames should be embedded in is the subject of Section 3.3.

1.3.1 Cognition, representation, categorization

Although the human mind has been the object of research since antiquity, and cogni-tive processes such as perceiving, thinking, and speaking have also been addressed and investigated for over 2,000 years, cognitive science in the stricter sense did not develop until the 1960s. At this time more and more researchers, initially in the area of psychology, abandoned the prevailing behaviourist paradigm (cf. Neisser 1967). Only a radical break with the positivist constraints of behaviourism and its understanding of the human mind as an unfathomable “black box” could bring the specific capacities of the human mind back into view. Although the so-called “cognitive turn” initially occurred within psychology, neighbouring disciplines such as AI, linguistics, and phi-losophy quickly adopted an interdisciplinary interest in mental processes such as the storage, processing, and application of information.

Chomsky’s early criticism of Behaviourism is commonly considered the precursor of cognitive linguistics. Chomsky demanded an explanatory linguistic theory whose goal was to investigate the mental-intellectual basis of the human language faculty (Chomsky 1965, 1975). Apart from this, at the same time other linguists were also pointing out the cognitive nature of language (cf. for instance Bransford, Barclay, and Franks 1972; Hörmann 1994).30

From the mid-1970s, the cognitive research paradigm established itself in other disciplines beyond the disciplinary borders of psychology and AI. In the following years, linguistic research began to be conducted along these lines. In this context, cog-nition denotes the set of all processes of understanding, or the processing of experience and information, irrespective of whether these are motivated by specific linguistic or other data. This broad understanding of cognition encompasses the human faculty of thinking and speaking as well as areas of psychological research on memory.

30. Gardner (1989) provides a comprehensive historical survey, in particular of the interdisci-plinary development of cognitive research.


In its broadest sense, the research focus of cognitive science is all the mental struc-tures and processes that contribute to the formation of human knowledge without simultaneously recurring to the philosophical-epistemological distinction of true/false.

In cognitive approaches, human cognition is seen as a system of mental structures and processes and described using models that take into account the complexity of mental activities…. cognitive models are based on the premise that cognitive processes are purposeful activities that cannot simply be described as being causal on the basis of associative mechanisms. Cognitive units and processes are to be understood as components of complex correlations and cannot be reduced to isolated, unstructured units. (Translated from Schwarz 1992b: 12)

Cognitive processes are by no means opaque in nature and consequently not em-pirically unfathomable; rather they form complex structural entities with their own reality, which can be described in psychological terms. Thus, apart from psychologi-cal plausibility, theoretical consistency and empirical explanatory adequacy are also significant criteria of a cognitive science approach. Although cognitive linguistics has been interested in the neurophysiological basis of language ability from the outset, the methodology required that cognitive representations of knowledge should on principle be able to be investigated independently of their neurophysiological premises. Thus, linguistic investigation was able to employ the tried and tested, empirically supported methods of experimental psychology (cf. Danks and Glucksberg 1980).

With this empirical orientation, cognitive linguistics takes account of its funda-mental epistemological conviction that knowledge representations of any form are not “mirrors of nature”, as the philosopher Richard Rorty (1987) put it. This long served as an implicit epistemological premise, for instance in 18th century Empiricism.31 Although our conceptual knowledge is substantially constructed from experiential data, this data is not given as isolated units, but rather occurs as elements embedded in more comprehensive knowledge units (semantic networks, cognitive models, schemas, domains, frames, etc.) and, in the construction of such representational knowledge structures, is subject to specific cognitive operations that aid its mental processing.

The knowledge that living creatures construct by means of cognition is realized in the form of mental representations… These mental representations are cogni-tive models of the objects and events that they refer to. Cognitive models have an important function for coping with complex objects, since they reduce them to their essential properties. (Translated from Rickheit and Strohner 1993: 15)

Cognitive models do not construe objects or events outside our own consciousness, but rather are products of abstraction that arise solely on the basis of our own active

31. Cf. the overview in Sinha (1999); for the philosophical debate cf. Putnam (1988), Rorty (1987). The philosopher Wilfrid Sellars (1999) speaks in this context of the “myth of the given”. Sensory data are not simply given, thus not a “mirror of nature”, but rather always embedded in a network of epistemological convictions, that is, more or less supported assumptions about the world.


cognitive-constructive operations. They are representational values, since they do not exist outside of our consciousness. The function of cognitive representations is there-fore of major importance because without representations, it is not possible to process information.

[A conceptual representation] stands at the centre of the information process-ing flow, with input from perceptual modules of differing kinds, and is centrally involved in memory, speech, planning, decision-making, actions, inductive infer-ences and much more besides. (Hampton and Moss 2003: 505)

Thus representations are specifically linguistic when the perceptual input is of a linguis-tic nature. These representations are individual in that they construct mental entities, and interindividual as far as the type of model itself is a result of a communicative practice (Harras, Herrmann, and Grabowski 1996: 13). The fact that cognitive repre-sentations are by no means independent of the prior social experiences in which they occurred and are constantly modified through them is of particular interest for frames, especially in connection with default values.

Accordingly, cognitive representations describe an epistemological relation. Although they may be motivated by perceptual data (that is, also by means of the perception of words, sentences, texts), their existence is relatively independent if this.

Representation is, first and foremost, something that stands for something else. In other words, it is some sort of model of the thing (or things) it represents. This description implies the existence of two related but functionally separate worlds: the represented world and the representing world. The job of the representing world is to reflect some aspects of the represented world in some fashion. (Palmer 1978: 262)

Although Palmer departs from a problematic two-world ontology in that he attests an objective and thus cognitively transcendent existence to a representatum, he draws attention to an important aspect of the model-like disposition of representations.32 His objectivist economy is not necessary in the case of language representations, since lin-guistic expressions are the communicative medium that in turn refers to other models as representata (cf. Jackendoff 1983: 23–40). For instance, if two people are discussing whether the death penalty is justified, they both refer to the cognitive model of what justice is (or should be). Put more generally, understanding an utterance means plac-ing the uttered words and sentences in connection with what they might mean, and that is not a process of approximating facticity, but rather a reconstruction of what was postulated. This initially requires much less reference to an extralinguistic entity than reference to a conceptual structure (or cognitive model, frame), even if it is a case of a simple noun that is grounded in the experiential world of a language user by the definite article (as in the Wednesday). Taylor observes: “It is important to bear in

32. Cf. Krämer’s (2001) study on a multitude of linguistic theoretical positions carried out under the aspect of a dualist ontology.


mind, however, that the referent of a grounded nominal is not some object out there in the external world, but an entity in a mental space” (Taylor 2002: 347). In addition the meaning of such a noun can only be gleaned within an activated cognitive model. This model features specific contours. As already outlined in Section 1.2.2, it is three dimensional insofar as it distinguishes a specific profile (the denotatum) from a specific cognitive domain. Thus, on the one hand, the noun Wednesday profiles a day within the cognitive domain “week”; on the other hand, the definite article the grounds the deno-tatum in the experiential world of the communicator to the extent that it temporally refers to a specific Wednesday. Gilles Fauconnier (1985) introduced the term “mental space” among other things to highlight this cognitive process of referentialization, at the same time conveying the multidimensional character of the denotation domain and in particular its cognitive-representational status. This is what Taylor is referring to in the above quotation.

Meanwhile, Palmer rightly stresses that there is no representational relationship between the object represented (representatum) and the physical phenomenon it is representing (represantans). Representata are rather the result of a controlled recon-struction process with perceptual data as its possible departure point. Such cognitive processing is relatively autonomous and is subject to specific principles (like those for categorization). Hörmann particularly emphasizes this:

Whoever perceives…, understands something. That means he understands in the physical substrate and all the way through it a meaning that goes beyond this substrate and establishes a correlation with the meaningful unit that is already available in the hearer. (Translated from Hörmann 1994: 198)

Thus a representation is that which is cognitively produced. It comprises previously stored sediments of past experiences, or more precisely, schematic representations of a higher degree of abstraction that are already stored in the long-term memory.

However, when percepts themselves emerge from cognitive-constructive pro-cesses, to what extent do they differ from representations? Since percepts are mental entities (and not objective entities external to our consciousness), it does not make sense to compare percepts to representations. Conversely, if they are not equivalent, then not every representation is perceptually motivated. Thus, we can use our imagina-tion to call to mind objects that have no correlates in our current world of perception. For this reason Monika Schwarz distinguishes between a so-called “internal” and “ex-ternal mode”. According to her, percepts are experienced in the external mode, whereas representations (as far as they are not based on perceptual data) are experienced in the internal mode.

The total of all percepts represent the world Wp, which is experienced as a real state in external mode. The set of all units of representational cognition in the internal mode that have referents in the perceptual world constitute the mental world model Wm. This world model is experienced in its mental character as an internal state of the organism. (Translated from Schwarz 1992a: 44; italics in bold in the original)


Schwarz introduces the criterion of manipulability to distinguish Wp from Wm. Representations that are not perceptually motivated can be altered (almost) at will, for instance, the colour, size, or other features of an imagined car can be changed. By contrast, perceptually motivated representations are to some extent subject to external constraints. Size, colour, etc. of a sensually perceived car are not arbitrarily modifiable. Nonetheless percepts remain cognitive constructions. For instance, which colour we perceive is dependent on the biological and physiological preconditions of our colour perception.

Cognitive representational entities (which include frames) do not have a static character, but rather change with each new experience (cf. Barsalou 1983, 1987, 1992a). Renate Bartsch (1998, 2002, 2005) expressly pointed out that cognition is first and foremost a process, namely one of “understanding or conceptualizing, and herewith the ability to conceptualize” (Bartsch 2005: 6). Thus, what is needed is a theory of conceptual structures that does not try to account for the factor of processuality after the fact, but rather understands it as an essential (sub-)component. Strictly speaking, concepts constitute inferential bases rather than representational formats. However, their representational status cannot be entirely denied. This can be shown with a sim-ple observation: In principle it presents no difficulty to explain what the understood or the conceptualized object consists of even during the process of understanding or conceptualizing.33 In Chapter 3, I broach this problem separately and, on the basis of studies in inference theory, propose making inferentiality the departure point of a linguistic frame theory.

The fact that cognition is a process that is always based on previously available (and consequently schematized) knowledge leads us to a further central property of conceptual representations. Whereas representations are constructed ad hoc when it is a matter of constructing a communicative unit of meaning (a “communicat”) from a linguistic expression, other conceptual entities retrieved from the language user’s memory play an essential role in this process. The activation of background knowledge leads to the emergence of new knowledge entities, that is, new representations. Every representational entity emerges as a complete structural framework that also contains, or can contain, elements that were not perceptually available at the beginning of the cognitive process.34

33. For this reason, Bartsch’s position seems to me to be too radical; in a consciously exagger-ated manner she claims that there are no concepts (Bartsch 2005: 3ff.). In contrast, cf. Dominiek Sandra (whose view I follow): “A mental representation seems, however, unavoidable in the synchronic semantic analyses. It is hard to believe that language users do not re cognize the links between usages that are … close to each other. If these analyses do not refer to mental representations (not their format but their content), it is hard to see what they could refer to.” (Sandra 1998: 370)

34. In the frame-semantic context I distinguish between the different structural constituents of frames (as representational formats of language-relevant knowledge), cf. Section 6.3 to 6.5.


The previously mentioned example Wednesday illustrates to what extent a lin-guistic expression evokes a concept (or a frame).35 For the comprehending recipient, this expression merely adopts a linguistic marking function for the interpretation of meaning-relevant knowledge (Busse 1997a; Langacker 1987: 163). Thus, each word serves to give the recipient access to semantic knowledge that has previously been conceptually structured and stored in the long-term memory. The retrieved concept “constitutes the content potential for semantic units, which then bind conceptual infor-mation to linguistic forms segment for segment” (translated from Schwarz 1994: 16).36

On the basis of Rosch’s empirical studies, a pivotal achievement of prototype the-ory consists in identifying the central function of cognitive categorization for processes of language understanding (Rosch 1977, 1978).37 Thus the meaning of the expression Wednesday can only be gleaned by referring to the superordinate cognitive domain week. At the same time, the two concepts of Wednesday and week stand in a part-whole relation, which is only constructed through the process of categorization or schema instantiation (cf. Figure 2). Any arbitrary example illustrates the fact that conceptual structures can be retraced to a cognitive process of integration consisting of setting different concepts in relation to each other. In this case, it is an act of categorization, or more precisely, of instantiating a concept in a superordinate schema that in turn consists of a network of concepts, namely of other days of the week. From the time of Rosch’s early studies on categorization, the human capacity for categorization has been a core area of study in cognitive research.

Week

Mon

day

Wed

nesd

ay

Thur

sday

Frid

ay

Satu

rday

Sund

ay

Tues

day

Figure 2. Categorization of the concept Wednesday in the knowledge domain (or frame) week (cf. Boas/Petruck 2003)

There are two possible research perspectives for a cognitively oriented linguistic theory. On the one hand, it can set itself the goal of identifying, on the basis of the empirical

35. On the relevance of lexico-cognitive analysis, cf. Cuyckens, Dirven, and Taylor (2003).

36. For the linguistic basis of the concept theory, cf. Chapter 6.

37. Taylor (2003) gives a comprehensive outline of the linguistic relevance of categorization processes. Keil (2003) provides an insight into the close and fundamental correlation between categorization and causality.


materials under investigation, those cognitive principles that govern the human con-ceptual system. More recently in Anglo-American linguistics, George Lakoff (1987) and Ronald Langacker (1987) have conducted pioneer work in this area. On the other hand, it can also investigate the conceptual structures themselves.38 This is illustrated particularly by Charles Fillmore’s frame theory (initially in Fillmore 1975).

Overall both perspectives present fewer alternatives than more complementary linguistic descriptions. After all, conceptual structures are, in fact, always the result of cognitive operations and, in turn, these always have the construction of a complex conceptual structure as a result. Due to the high complexity of the empirical data col-lection, it is nonetheless practical to engage with each research perspective separately. The present study places emphasis on the analysis of conceptual structures.

Therefore, subject to the premises of cognitive theory, the focus of research in lin-guistic semantics is the conceptual systemic correlation between different knowledge forms and elements that are relevant to understanding, as well as the modes of their connections with each other.39 Instantiated meanings are the result of selecting and combining relevant and conceptually bundled knowledge elements. In this process, it plays only a subordinate role whether the informational units are called up from the long-term memory or are linguistic or non-linguistic data from the working or short-term memory.40

In cognitive research, it is undisputed that the understanding of linguistic mean-ings constitutes a highly complex cognitive process that involves quite distinct types of knowledge. By contrast, it is highly controversial whether, in this process, different lev-els of semantic-conceptual knowledge representation should be distinguished from one another, that is, whether different types of knowledge (social knowledge, contextual knowledge, linguistic knowledge, etc.) are subject to different organizational structures. In one case, it is a matter of taking a holistic position, in the other a modularist one. I now elaborate on both basic positions in connection with frames.

38. This corresponds to Sandra’s (1998: 362) distinction between “mental predispositions of language”, i.e. cognitive principles of language understanding and “mental representations”.

39. I elaborate on the question of which knowledge forms are potentially relevant in Section 3.3.1. The modes of their connection are essentially of concern to those topic areas that are addressed within conceptual approaches as “construal operations”, “focal adjustment” and “conceptualiza-tion” (Langacker 1987), “frame-shifting”, “blending” (Coulson 2001), “imaging systems” (Talmy 1988) and “mapping” (Fauconnier 1990).

40. Schwarz (1994) limits the semantic research focus to the cognitive system stored in the long-term memory. However, this loses sight of a large proportion of knowledge that is potentially relevant to understanding. I see no necessity to exclude this area of knowledge; in fact, this seems to be an impingement strongly induced by theory, one which already operated in the previous distinction between competence and performance or in Schwarz’s case with the level-semantic model.


1.3.2 Positions in cognitive theory

Since the 1980s, a broad consensus has existed that linguistics and psychology con-stitute the cognitive research disciplines par excellence. What language is, what constitutes the language faculty, and according to what rules or principles linguistic utterances are produced (or understood) depends highly on how a speaker mentally processes, stores, and retrieves information. That is why language is (also) a cognitive phenomenon. Cognitive abilities (in the sense of knowing-how) develop largely in the first five years of life. Conceptual knowledge (in the sense of knowing-that) is constantly changing; the learning process is life long.

The human language faculty is perhaps the best example of how complex cogni-tive processes can be. Yet, this complexity is only seldom perceived by language users. Linguistic knowledge constitutes a stable knowledge system, whose functional princi-ples remain largely subconscious. Through constant language praxis, speaking and lan-guage understanding become ingrained processes. A language user knows intuitively when, for instance, a linguistic utterance is grammatically correct and what it means in a given context. Yet, it is difficult for the user to explain what this “knowledge” exactly consists of. Which cognitive prerequisites have to be fulfilled? What sort of knowledge is it that enables us to learn a language? What cognitive structures (and processes) contribute in what way to the success of linguistic processes of understanding? The fact that answers to these questions are by no means obvious legitimizes a cognitively oriented view of language.

One overall aim of linguistic analysis is to adequately describe the complexity of linguistic structures in their multidimensionality. Empirically seen, this proves to be an extremely complex and methodologically elaborate venture. The analytical processes and key premises within cognitive research are accordingly heterogeneous. The most fundamental difference concerns the issue of whether cognition is subdivided into autonomous modules (and these into submodules) or whether it is constituted by universal cognitive operations. A consensual decision for one or the other of these posi-tions is not to be expected in the near future in view of the close correlation between the (neuro-)psychological, (socio-)anthropological, and philosophical issues involved in this problem complex (Coulson 1995). Rather, the positions appear to be becoming more entrenched (cf. Dölling 2005; Ritter 2005; 2005; Taylor 1995).

Sucharowski’s assessment (1996: 158) that the modularity conception is gaining precedence applies at most to the German context. By contrast, in Anglo-American research there is a growing number of cognitive scientists who work within the holistic paradigm.41 It is interesting to note, however, that historically speaking the reception

41. Two established publications have affiliated themselves with a holistic position: the journal Cognitive Linguistics and the book series Cognitive Linguistics Research, which has been published by Mouton de Gruyter since the beginning of the 1990s and in the meantime encompasses over 40 books. In addition, the special edition Linguistic Review 22/2005 provides a recent overview of research.


in Germany – and in particular in German linguistics – was rather one-sided. Whereas generative grammar found a strong resonance as early as the 1970s, then was further semantically refined by Manfred Bierwisch, Ewald, Lang, and Dieter Wunderlich, and since then has a fixed place in the curriculum, the research programme of holistically oriented approaches was acknowledged only hesitantly at first and gained only a rather eclectic reception, if at all.42

Despite the holistic/modularistic divide, the following points remain largely undisputed:

– Human language is a “window” to cognition. Cognition is not a black box phe-nomenon; rather, through linguistic analysis empirically valid statements can be made about cognitive processes.

– Constructing and understanding linguistic expressions (recursively) constitutes a particularly complex cognitive ability that entails both procedural and represen-tational factors.

– From a procedural point of view, the analysis of linguistic understanding and of what determines this achievement of understanding is the focus of the investigation.

– Psychological factors such as the capacity and structure of the human memory as well as the methodological principle of cognitive economy must be taken into account (since a limited number of cognitive abilities and a limited inventory of linguistic forms of expression allow the (re-)production of a potentially unlimited number of content expressions)

– Grammar is a (theory-dependent, and therefore hypothetical) model that is in-tended to describe and explain linguistic processes of understanding in a psycho-logically realistic manner.

Yet this cognitive research agenda already loses its unifying force when it comes to the question of what, then, grammar accounts for as a model “of what a person who knows a language actually knows, and of what a speaker or listener actually does when speak-ing and listening” (Taylor 1995: 4). Is grammar an autonomous human ability that can be investigated independently of other cognitive abilities? Or conversely, due to being highly sophisticated and empirically easily accessible, does it constitute an excellent subject of research that allows the study of universal cognitive processes that equally encompass “basic sensory-motor, emotional, social, and other experiences of a sort available to all human beings” (Lakoff 1988: 120). The remaining part of this section explains both these basic cognitive research positions.

42. Cf. also Schwarz (1992a, b); Keller and Leuninger (2004); Strohner (1995); Sucharowski (1996). Related edited volumes, e.g. Felix, Habel, and Rickheit (1994), Habel, Kanngießer, and Rickheit (1996), also stand under the auspices of generative grammar. An exception can be found in the comprehensive work of Konerding (1993) insofar as he also discusses selected holistic positions.


Whereas historically the holistic and modularist positions have the same roots, they could not diverge more in terms of answering the fundamental issue of linguistic theory concerning the relationship in which language and the human mind stand to each other.43 Adherents of generative grammar maintain that the human mind is or-ganized in a rigidly modular manner. They trace the complexity of mental processes back to the functional principle of the distribution of cognitive tasks. In contrast, the holistic thesis states that perception, motor function, and the language faculty strongly interact, their ontogenetic development is independent of one another, and they are governed by universal cognitive principles.

These divergent points of departure lead to a series of methodological conse-quences that fundamentally distinguish modular approaches from holistic ones (cf. Taylor 2002: 7; also Lakoff 1987: 584; Lakoff 1990). Firstly, the methodological inten-tion to be able to generate all the grammatical sentences of a language with as few principles as possible results in a degree of formalism within modularist theories. The maxims of formalizability are accommodated to the extent that the postulated “deep structures” become increasingly abstract, so much so that they no longer have anything in common with the analysed surface structure, the units of utterance, and can only be retraced to them via a set of purely formal operations. In addition, since the focus is on generative principles, the (ostensibly) idiosyncratic phenomena slip into the pe-riphery of the research focus, insofar as the rules brought to the approach are unable to explain them. As a result, a core area of grammatical phenomena emerges that does not include, for instance, historical or epistemological aspects.

The “core” area of a modular approach to semantics includes exclusively language-inherent factors. In various semantic-level models, these are systematically distinguished from external linguistic factors of meaning construction (Bierwisch and Schreuder 1992: 30). Three broad variations of semantic-level models can be distinguished:

i. Two-level semantics. This posits a fundamental distinction between semantic and conceptual knowledge. Should the meaning of a linguistic expression exhibit an internal, linguistically bound component structure, general world knowledge only becomes relevant at the level of the conceptual system. Proponents of this model type are, in particular, Manfred Bierwisch (1983a) and Manfred Bierwisch and Ewald Lang (1987). Bierwisch repeatedly states that knowledge is structured and organized by frames on the conceptual level, however, without elaborating on this (cf. Bierwisch 1979: 78–79; Bierwisch and Schreuder 1992: 32–33).

ii. Three-level semantics. Monika Schwarz (1992a, 2000) expanded on the two-level model. She, too, distinguishes between the semantic components of an expression

43. Cf. Newmeyer (1986); Huck and Goldsmith (1998). The research profile of holistic approaches emerged out of the conflict with different generatively oriented theoretical approaches. Thus, in the 1970s holists like Charles Fillmore, George Lakoff, and Ronald Langacker increasingly turned away from the foundations of a generative theory of grammar. Lakoff (1987) and Langacker (1987) laid down their own research programme, the basic principles of which, however, were already evident in Lakoff and Johnson (1980) and Langacker (1979); cf. the overview in Lakoff (1991).


that are determined by linguistic principles and the conceptual content of linguistic expressions. However, in her conception, the communicative purpose of a linguis-tic expression does not emerge until the level of utterance meaning, which is when contextual data become relevant. Although Schwarz (1994: 15) expressly distances her three-level model from the two-level model, the two models ultimately posit identical linguistic premises. In addition, Bierwisch himself sometimes seems to favour a three-level model. Thus, he distinguishes at one point between the levels of linguistic knowledge, everyday knowledge, and knowledge of social interaction (Bierwisch 1979: 64ff.). Schwarz’s conception does not vary significantly from this, even concerning the possible incorporation of schemata (or frames) to explain conceptual processes (cf. Schwarz 2000: 111–112).

iii. Multi-level model of meaning representation. The influence of the conceptual sys-tem on the constitution of meaning is only inadequately explained in the works of Schwarz, Bierwisch, and others. For that reason, Johannes Dölling, proceeding from similar linguistic premises, takes greater account of conceptual phenomena like metaphor, metonymy, systematic polysemy, and vagueness (Dölling 2001b). Dölling comes to the conclusion that compositionality and context sensitivity work closely together in the process of understanding the meaning of linguistic expres-sions and that, in the course of contextually specifying the context-independent meaning of an expression, several phases of “information completion” must be passed through (Dölling 2001b: 14). His so-called “multi-level model of meaning representation” allows for interaction between a lexical, a compositional, a concep-tual, and a contextual component. Although, like the two- and three-level models, Dölling’s multi-level model makes a rigid distinction between purely semantic and conceptual knowledge, the innovation lies in the implementation of schemata that provide specifications for contextual enrichment.

These three approaches do not so much pose independent theories as build upon each other. Two-level semantics must be seen as the foundation. Its methodological premises serve as orientation for both Schwarz’s three-level model and Dölling’s multi-level model.

Such a modular conception of the language faculty is countered by a number of holistic approaches, which I subsume under the term “cognitive linguistics”. The primary object of enquiry of Cognitive Linguistics is the cognitive reality of language use, without already dividing it into relevant and irrelevant (component) aspects on the basis of preliminary theoretical decisions. Speaking and understanding individuals have specific needs, are steered by social interests, and take recourse to intersubjectively shared symbolic resources to express these needs and interests. Further, they depend on a very specific psycho-physiological endowment that allows a certain repertoire of cognitive abilities to form (and suppress the development of others).44

44. Cf. Evans and Green (2006: 27–50). A simple example is colour perception. The number and type of photoreceptors determine the possible colour spectrum in which organisms perceive their environment. For instance, since rattlesnakes have one photoreceptor more than humans, they are able to see infrared light.


Holistically oriented linguists hold the conviction that a cognitive treatment of language should take these factors into account as primary phenomena. Even the most primitive of organisms (like unicellular organisms) maintain relations with their en-vironment; however, there are considerable differences in the way these relations are constituted and processed. In this context, language establishes particularly complex relations with the social environment:

Language is only the tip of a spectacular cognitive iceberg, and when we engage in any language activity, be it mundane or artistically creative, we draw uncon-sciously on vast cognitive resources, call up innumerable models and frames, set up multiple connections, coordinate large arrays of information, and engage in creative mappings, transfers, and elaborations. (Fauconnier 1999: 96)

Language constitutes a “window” to cognition, which means a window through which can be viewed the facets of linguistic conceptualization competence. The modularist assumption that there are innate principles (such as those of a “universal grammar”) blocks the view through this “window”.

The radical alternative to the nativist postulate is provided by the so-called “usage-based model” of Cognitive Linguistics.45 According to this, linguistic structures are emergent phenomena that are solely motivated through language use (and not by universals). Since psychological process always play a decisive role in language use, grammar is constantly changing and as a consequence does not form a predetermined system but rather a dynamic one. Two principle assumptions form the basis of this perspective of language:

[C]ognitive linguistics is defined by two primary commitments, what I will call the Generalization Commitment and the Cognitive Commitment. The generalization commitment is a commitment to characterize the general principles governing all aspects of human language. … The cognitive commitment is a commitment to make one’s account of human language accord with other disciplines as well as our own. (Lakoff 1990: 40)

The first assumption places Cognitive Linguistics in a research complex that is affiliated strongly with cognitive (developmental) psychology and anthropology. A cognitively realistic description of human language stands in focus. The second assumption states that such cognitive analyses must always take into account the research findings of neighbouring disciplines.

Within cognitive linguistics a variety of approaches to meaning theory have emerged that depart from the common assumption that linguistic meanings can be equated with conceptual structures. Further, linguistic meanings are “embodied” in human experience. Language is understood as a system of signs, i.e. form-meaning

45. Cf. the comprehensive outlines in Bybee (1995), (2013); Elman et al. (1996); Langacker (1988c), (2000); Barlow and Kemmer (2000); Tomasello (2003).


pairs,46 both on the lexical level and in terms of grammatical constructions. Accordingly, syntax is not an autonomous system, since syntactic categories and syntactic construc-tions fulfil primarily meaning-relevant functions.

This so-called “semantic turn”, i.e. the turn to semantics and the challenge to the primacy of syntax, characterizes many studies from the late 1970s, especially those of Wallace Chafe, Charles Fillmore, George Lakoff, Ronald Langacker, and Leonard Talmy. However, the holistic model of cognition was only established through Langacker’s comprehensive outline of “Cognitive Grammar” (1987), Lakoff ’s (1987) study on prototype theory, and the efforts of Fillmore and Kay to include grammatical constructions in the analysis. Four approaches can be distinguished:47

i. Construction grammar. In a number of studies (such as Fillmore, Kay, and O’Connor 1988; Kay 1997; Kay and Fillmore 1999), Fillmore and Kay propound a model of grammar that they name “construction grammar” for the first time (Fillmore 2013). This is a variation of “head-driven phrase structure grammar”, de-vised by Carl Pollard and Ivan Sag in Stanford (Pollard and Sag 1994). Fillmore and Kay explicitly anchor their model in frame semantics (Kay and Fillmore 1999: 21).

ii. Radical Construction Grammar. William Croft’s so-called “radical construction grammar” incorporates many aspects of Fillmore and Kay’s construction grammar (cf. Croft 2001: 14–29; Croft 2013) and at the same time orients itself methodologi-cally towards Langacker’s work (Croft and Cruse 2004: 278–279). Croft’s approach is “radical” in that, among other things, in contrast to Fillmore and Kay (1999: 1), he postulates not only that all syntactic categories are derived from constructions – that is, from form-meaning pairs – but also that they are also specific to individual languages, i.e. not of a universal nature. Croft also rejects the existence of purely linguistic categories. He argues further that all categories – including syntactic ones – are derived from (the use of) constructions. Croft arrives at these findings on the basis of contrastive linguistic typological studies.48

46. Construction Grammar also speaks of “constructions” and Cognitive Grammar of “symbolic units”. For a more precise distinction cf. Section 4.1.2.

47. An overview is provided by Croft and Cruse (2004: 257–290), Fischer and Stefanowitsch (2006), and Schlobinski (2003: 159–192; 224–226). Apart from the four streams named here, in more recent research there has been a whole array of empirical studies, in particular corpus-based, which stand in the paradigm of Cognitive Grammar and Construction Grammar (Boas 2003; Diessel 2004; Tomasello 2003). I do not summarize these as a fifth approach as they draw on different methods to investigate their respective specific research interests (cf. Ziem and Lasch 2013: 48–67).

48. Frames are not addressed in detail in Croft’s Radical Construction Grammar because Croft concentrates exclusively on the analysis of syntactic representations. This is where “radical frame semantics” (Croft 2001: 62), as Croft in allusion to Fillmore calls it, would find its place.


iii. The Lakoff/Goldberg model. In Women, Fire and Dangerous Things: What Categories Reveal about the Mind, a widely known study on cognitive prototype theory, George Lakoff (1987: 462–584) also makes use of the concept of constuction. This analytical approach sometimes also called “Cognitive Construction Grammar” (Boas 2013), which was presented using a case study on there-constructions, was expanded and systematized by Adele Goldberg (1995). More so than the other ap-proaches, Goldberg and Lakoff ’s analysis concentrates on the correlation between “constructions” (so-called “categorization links”). Both expressly refer to frame semantics.

iv. Cognitive Grammar. There are points for and against considering Langacker’s pro-gramme of a Cognitive Grammar (Langacker 1987, 1991a; also Taylor 2002, 2012; Ziem and Lasch 2013: 41–44) as an independent school of thought. Against this view is the fact that to a large extent departs from the same premises as the three construction grammar approaches already mentioned (Langacker 2005: 102), sim-ply using in part other terminology. On the other hand, Langacker places empha-sis on the circumstance that all grammatical units have an intrinsically symbolic character.49 In contrast to Croft (2001) and Goldberg (1995) Langacker thus rejects placing grammatical information (such as morphological or syntactic) on the form side of “constructions”; rather, they result, in his view, from form-meaning pairing as a whole (Langacker 2005: 104).

Consensus prevails in these approaches, firstly, that grammatical constructions are independent only in their symbolic function, secondly, that constructions are cogni-tively uniform in representation, and thirdly, that they are organized taxonomically in a grammar (Ziem 2014). However, the way in which constructions and grammatical information are precisely defined varies depending on the approach. Accordingly, there is a large degree of terminological variety.

In sum, it is clear that holistically oriented linguistic approaches, including in par-ticular construction grammar and Cognitive Grammar, differ from modular models in many fundamental aspects. In view of the fact that these differences affect the develop-ment of the linguistic research area as much as the methodological issues surrounding its investigation, it becomes clear that a certain dualism runs through the entire field of cognitive research. The research focus and the method of frame semantic analysis may change according to the premises under which frames have been addressed.

49. Cf. Langacker (1987: 56–86). Croft and Cruse follow a similar line of argumentation (2004: 279): “The distinguishing feature of Cognitive Grammar as a construction grammar is its emphasis on symbolic and semantic definitions of theoretical constructs traditionally analyzed as purely syntactic”.


1.3.3 Frames in modularist and holistic approaches

Frames are addressed within both modularistic and holistic approaches to (lexical) meaning. However, by contrast, other linguists with a strong interest in conceptual meaning structures have directly associated frames with modular models. I deal with some examples of this in this final section of first chapter. In doing so, I am not so much concerned with the particular research objectives of the individual studies as with the fact that there is clearly little agreement concerning the cognitive theoretical founda-tion of frame semantics. In the next chapter, I take this deficit as the opportunity to examine the holistic and modularist positions from the viewpoint of frame semantics.

In his introduction to a collection of articles he edited on “semantic conflicts”, Ekkehard Felder (2006) outlines the methodological instruments and the basic mean-ing theoretical concepts that form the basis of the volume. He assigns frames a position of central significance for meaning theory. Although not overtly stated, Felder directly follows Busse (1991a: 37) when he conceives of “knowledge frames” as the general term for all the various forms of knowledge agglomerations relevant to understanding that text linguistics has identified to date (Felder 2006: 19). Frames, he goes on to say, are composed of concepts that constitute a partial meaning of the particular frame insofar as they define aspects of the frame. In turn, Felder expressly does not conceive of these partial meanings as semantic features, since meaning is an “interpretive hypothesis constructed from textual and situational meaning” (translated from Felder 2006: 19). Since he concludes that the assumptions of the semantic component theory are false, Felder conversely departs from a “concept of holistic meaning (holism)” (Felder 2006: 18) according to which semantics and pragmatics cannot be strictly separated. Despite having specified this, a few pages later Felder observes, “not distinguishing between the linguistic and cognitive levels will certainly cause significant theoretical difficulties” (translated from Felder 2006: 26).50 And a few lines later, he states:

Without entering into a detailed discussion of the programmatic assumptions, two directions in cognitive semantics can be diagnosed:The semantic structure is identified with the cognitive structure, which in my view is a false assumption, since this excludes, for example, linguistically bound components;Two-level model: this departs from a semantic form that is thought to be linguisti-cally bound and from a conceptual structure that is thought to be linguistically independent (interlingual) …

50. Felder bases this on Fritz (1998), who in turn after a very brief explanation of cognitive semantics concludes, “These programmatic assumptions cannot be discussed more closely here. I will only suggest that the lack of distinction between the linguistic and cognitive levels… and the resulting problematic equation of idiosyncratic single-language rules of use with cognitive categories create considerable theoretical difficulties” (translated from Fritz 1998: 99). Apart from the fact that I know of no studies that expressly propagate such an equation, it is indicative that subjective description and reasoned discussion give way to sweeping judgements.


However, what are the “linguistically bound components” that cognitive semantics is allegedly not able to account for? Are these “non-contextualized prerequisites to un-derstanding” (Scherner 1979) in the broadest sense? It should be noted that cognitive semantics even makes these prerequisites the point of departure for its investigations (cf. for instance Croft 2009). Conversely, if the two-level model presumes a linguisti-cally bound semantic form, as Felder correctly notes, then the question remains as to the status of these components. What is nonetheless clear is that these do not cor-respond to “interpretive hypotheses” (cf. Sections 2.2 and 3.1), but rather to semantic features whose theoretical value Felder questioned previously with reference to a ho-listic model of meaning.

The fact that neither the holistic nor the two-level model are discussed, yet are eclectically conflated with the aim of providing a linguistic theoretical basis for frames, is symptomatic. Instead of providing a critical review, problematic and often false arguments are presented. Thus, Felder claims, for example, that a conceptualist ap-proach to semantics must face the accusation that “the concept of language that forms its basis comes dangerously close to being an untenable epistemological realism of representation theory” (translated from Felder 2006: 26). Yet, just the opposite is the case: Jackendoff even specially introduced the term “projected world” to counter naive realism.51 Furthermore, Felder states that in linguistics there is a general consensus “that a theory of concepts cannot be identical with a theory of linguistic semantics” (Felder 2006: 26) (whatever the latter may be). At this point it should be noted that, in the Anglo-American world, after 25 years of research history there are certainly as many supporters of cognitive linguistics as there are of the modularist model. Felder continues, “by contrast, in ‘cognitive semantics’, which is at times dominated by psy-chology, it is not rare that concepts are equated with linguistic categories such as ‘mean-ing’ without any explanation” (translated from Felder 2006: 26). Yet, on the contrary, it can be observed that the extent to which semantic (and grammatical) structures are of a conceptual nature is the core theme of Cognitive Linguistics. Both the contribu-tions of the journal Cognitive Linguistics and the more than 50 volumes that have been published to date in the Mouton de Gruyter series “Cognitive Linguistics” are dedicated to this topic.

The fact remains that Felder on the one hand – following a holistic concept of meaning – sees frames as a central category of analysis, but on the other hand propa-gates a distinction between a semantic and a conceptual level of meaning – and thus at the same time advocates a modular model. How can these diametrically opposed approaches be united? What methodological specifications does the application of

51. Cf. the extensive explanation in the sixth chapter. However, I quote the central passage here: “We must take issue with the naive position that the information conveyed by language is about the real world. We have conscious access only to the projected world – the world as unconsciously organized by the mind; and we can talk about things only insofar as they have achieved mental representation through these processes of organisation. Hence the information conveyed by language must be about the projected world.” (Jackendoff 1983: 29)


frames follow, modularist or holistic? What consequences does this decision entail? Can the modularist and holistic theorems be blended at all? These questions not only remain unanswered, but also are not posed at all.

Similar problems arise in a series of other studies, for instance in the corpus-based study by Fraas (1996a).52 Fraas employs frames as a tool to find out how the concepts “identity” and “German” have changed in the course of German reunification. With reference to Konerding (1993), she understands frames as conceptual knowledge units “that allow rational access to stereotypical knowledge that is bound to lexis” (translated from Fraas 1996a: 16). In the same context, she makes it clear that her approach is in the tradition of the frame theory of Minsky (1975) and Fillmore (1977a, 1982a, 1985) as well as the prototype theory of Lakoff (1987) and Rosch (1977). Thus, it only seems consistent that Fraas explicitly distances herself from componential semantics:

Meanings are not to be described as feature aggregates exhibiting sufficient and necessary features, but rather as mental representational units with mandatory and optional components that are mentally delimited by default values yet allow options and therefore function as instantiatable variables, whereby the context plays a significant role. (Translated from Fraas 1996a: 14–15)

However it is all the more surprising that Fraas at the same time takes recourse to Schwarz and Bierwisch’s multi-level semantic model. From here she adopts the assump-tion that linguistic meanings have three representational levels, a lexical and a concep-tual level and a level that concerns utterance meaning (Fraas 1996a: 12). The first level contains purely linguistic information, which refers to intensional meaning aspects:

The lexicon meanings Semsyn, which are stored permanently in the long-term memory, can be contextually enhanced in concrete usage situations…, which is expressed in Bierwisch’s notation as sem (ct) = m. The utterance meaning m is conceived of as a function of the intension of a lexicon unit sem relative to a context ct. (Translated from Fraas 1996a: 13)

What does the “intension of a lexicon unit sem” consist of? The first contradiction in Fraas’ model arises – as it did with Felder – from her recourse to the multi-level model, as Bierwisch leaves no doubt that the lexicon meaning “Semsyn” stored in the long-term memory corresponds to a bundle of invariant semantic components.53 When Fraas borrows from modularly aligned theories, she thus accepts that she is adapting a component theory that she actually rejects.

What is more, Fraas seems to see no problem in assigning frames the function of governing the contextual differentiation of the information units prescribed by the

52. I return to Fraas’ study in the course of this book.

53. In the article Fraas quotes, Bierwisch writes: “Since I have already made the assumption that word meanings are not unanalysed units, this initially implies that SEM must be a configuration of semantic components” (Bierwisch 1983a: 70).


contextually abstract lexicon meaning Semsyn. Since the conceptual level at which frames should be located relies on information provided by Semsyn at the semantic level, the influence of frames would be accordingly significant. Can the manifold fac-ets of knowledge that Fraas identifies when using frames to analyse the concepts of “identity” and “German” really be compatible with these specifications?

In general, the same conclusion applies to Fraas: what the consequences would be of blending distinctly holistic positions (such as those of Fillmore, Lakoff, Minsky, and Rosch) with theories using modular approaches to frames remains unaddressed. This methodological eclecticism, however, can hardly be said to contribute to clarifying the cognitive theoretical premises of frame semantics.

I conclude with two more examples to briefly demonstrate that the positions of Fraas and Felder I describe here are by no means negligible isolated cases. Rather, they demonstrate quite aptly what the current status of the discussion particularly but not solely in German linguistics is as to what constitutes the cognitive theoretical basis of frame semantics

An essay by Christina Gansel (2002) is devoted to the relationship between con-ceptual and semantic structure using the example of the verbal field of “verba dicendi” (verbs of speaking and uttering). In the introduction, Gansel notes that engaging with semantics from a cognitive viewpoint requires making a decision for a modular or a holistic approach.

Modularity forms the basis of my considerations, and I consider the semantic and the conceptual systems as separable systems that can interrelate through and integrate with a variety of procedures and operations. (Translated from Gansel 2002: 277)

Her thesis states that verba dicendi feature a shared “semantic core” or an “archilexeme” (Gansel 2002: 280) that in German corresponds to the verb sagen (say, tell). This lexeme is not only stylistically neutral, but it also has a “barely decomposable general meaning” and is a “candidate for semantic universality”. The correctness or plausibility of this assumption is a moot point. More important in this context is the following: Since a “semantic core” or “archilexeme” is clearly an element belonging to the semantic level54 that is assigned to all verba dicendi, then these verbs distinguish themselves from each other in that they are differently enhanced on the conceptual level. According to Gansel, frames perform this differentiation, a process that the author, building on Barsalou (1992a), understands as being a general mechanism to describe the organiza-tion of human knowledge.

54. Cf. Bierwisch and Kiefer (1970), Schwarz (2000), and Section 3.1.1 on the use of the “core” metaphor in level-semantic models.


However, when we take into consideration that Barsalou – no differently than Fillmore – understands frames as dynamic and purely conceptual units,55 the same contradiction becomes apparent that made the conceptions of Felder and Fraas prob-lematic. Whereas the semantic core is required to be a contextually independent and interlingual variable of potentially universal value, every lexico-semantic item only gains its individual value in relation to a frame that underlies the lexico-semantic field (Gansel 2002: 279). Yet, how do frames structure the meaning of verba dicendi when the verbs are characterized by a “lexico-semantic base structure”? What exactly do frames structure? And again: How can an inherently holistic conception of frames be reconciled with a modular approach?

As in the three cases already mentioned, Andreas Blank (1999) addresses frames in close connection with a modular semantic model without entering into the diffi-culties and inconsistencies arising from this. Blank is interested in the cognitive fac-tors that contribute to change in linguistic meaning and draws upon, among others, Fillmore’s frame theory to do this. Blank understands a “frame” as being a network of concepts “that are structured according to the contiguity principle” (translated from Blank 1999: 126). At the same time he adheres to a structurally motivated distinc-tion between linguistic and conceptual knowledge and adopts a weakened component theory according to which semes are abstracted facets of encyclopaedic knowledge and “thus substantially distinct from other facets of knowledge”.56 In direct conformity with Bierwisch (1983a), Blank bases his observations concerning meaning change on a semantic model that distinguishes between the levels of “single-language sememic knowledge”, “single-language lexical knowledge” and “extralinguistic knowledge” (cf. Blank 1999: 129). Although in the course of his further argumentation the first level does not play any role, the question remains as to how these three levels interact – and how in particular frames are to be integrated into the proposed semantic level model. No explanation is to be found.

There are many more examples that could be added,57 but the ones mentioned are sufficient to illustrate my point: Although frames are apparently becoming increas-ingly important, the linguistic and cognitive theoretical premises under which they are addressed remain oftentimes unaccounted for. Above all, it remains unresolved

55. Cf. Barsalou (1992a: 21): “Furthermore, frame theorists generally assume that frames are rigid configurations of independent attributes, whereas I propose that frames are dynamic re-lational structures whose form is flexible and context dependent”.

56. Translated from Blank (1999: 129); italics in bold in the original.

57. A particularly interesting example is the comprehensive frame theory of Klaus-Peter Konerding, which I return to in Section 6.3. Konerding (1993: 223–233) applies the frame ap-proach that he develops to eliminate what he calls a “desideratum” in Bierwisch’s two-level semantic model that affects the conceptual level. It is not clear whether or not Konerding shares the linguistic and cognitive theoretical premises of the model.


what role conceptual knowledge plays in the constitution of meaning and what influ-ence the selected model of cognition has on determining the function of frames in understanding language.

chapter 2

Cognitive theory and semantic issues

The incorporation of frames into entirely different approaches suggests that frames are being assigned a kind of gap-fill function. When a theory runs up against its limits and in particular cannot explain how conceptual knowledge flows into the process of meaning constitution, frames enter into the fray. In the previous chapter, however, I indicated that the cognitive theoretical framework in which frames are addressed exerts considerable influence on what frames actually are in semantic terms and what semantic knowledge they describe or are able to describe. In the following, I propose that the cognitive theoretical premises under which frames are addressed entail far-reaching consequences regarding the following issues:

– how frames should be modelled, i.e. what their structural elements are, what status should be assigned to them, and what area of semantic knowledge they describe;

– how semantic knowledge is cognitively organized and represented in order to be relevant to understanding at all, and what cognitive operations are involved in the construction of frame structures;

– what frames reveal about specific human abilities, establish epistemic references to the world by means of language, or more generally, what constitutes the language faculty and cognition overall;58

– what the specific problems are that a frame theory must confront.

Appropriate answers to these questions can only be expected once recourse is taken to an entire model of cognition. Since the choice of the cognition model determines, among other things, how far the semantic explanatory power of frames extends and what theoretical-methodological specifications a frame theory should follow, the cog-nition model provides much more than mere theoretical orientation. If it is holistic in nature, then so is the semantic theory. Conversely, if it is modularistically motivated, this will at the very least affect the relationship that the semantic structures maintain to other modules (in particular to the phonological and syntactic ones).59 In the case

58. In this context it is revealing to read Bierwisch (1987a) and Lakoff (1988) synoptically. The two articles, which appeared almost simultaneously, have a trailblazing character for cogni-tive linguistics yet both could not be more different, particularly concerning the basic theory-induced estimation of what constitutes the human mind.

59. Jackendoff, for instance, strictly opposes making a division between a semantic and a con-ceptual meaning level (i.e. the premise of two-level semantics) (Jackendoff 1983: 110). Instead, he argues for a threefold modular structure of the human language faculty in which the conceptual


of two-level semantics, modularity even affects the internal organization of semantic knowledge, as we will see.

Against the background of holism/modularism duality that completely traverses cognition research, there are two possible scenarios. Either frames structure “the con-ceptual level and merely provide a framework for the potential of textualizing knowl-edge” (Fraas 1996a: 27), so that case frames determine what Bierwisch (1979: 77) calls the “actualization context”, from which he wishes to distinguish the purely “linguistic meaning” of an expression; or frames function as a superordinate representational format for “all the different forms of knowledge agglomerations that have been identi-fied in text linguistics to date” (Busse 1991a: 27).

In the former case, frames take on the semantic function of contextually enriching what the linguistically determined so-called “semantic form” prescribes (cf. Dölling 2005). Conversely, in the latter case, there is no semantic level of representation or determination preceding frames; meaning actualization is accordingly always a form of conceptualization (Langacker 1987: 156).

In the following, I introduce a variety of meaning theory models that address frames. The focus is on the conditions under which each of the represented meaning theories becomes reductionist, that is, is unable to further explicate meaning-relevant knowledge. My argumentation will show that frames are only able to unfold their semantic explanatory potential within a holistic approach.

2.1 Holism vs. Modularism: an example

Before entering into the individual positions, I illustrate the conflict between holism and modularism using the semantics of prepositions as an example. Within a holis-tic perspective, language is not assigned a special psycho-developmental status. The guiding premise is that the ability to speak and understand language evolves in social practice under the conditions of our physical and biological disposition, just as other human abilities do. Accordingly, it is precisely within this matrix that linguistic mean-ings should be addressed. Instead of seeing the meanings of prepositions like in, on, before, under, etc. as being predetermined through experientially independent seman-tic forms, a holistic view sees them as being very closely related to our multifaceted perceptual and sensorimotor experience of the environment. Through the ingestion of food, for instance, infants learn early on to perceive their bodies as “containers”. Thus, schematized concepts (image schemata) evolve already in this developmental stage. Lakoff and Johnson argue that these later function as conceptualization templates,

system has a direct interface to the syntactic system. This is the essential difference to the two-level model, which additionally postulates a purely linguistic level that mediates between the syntactic and semantic systems. Cf. Section 3.2.

Chapter 2. Cognitive theory and semantic issues 51

so to speak, for a whole range of linguistic expressions.60 How the meaning potential manifests itself in an ontogenetic perspective – here in the preposition in – therefore essentially depends on body-based experience.

However, experiential correlations not only enter indirectly into the semantic in-terpretation of a linguistic expression, but are also directly involved in the process of meaning actualization. Thus, the concrete utterance meaning of the preposition in is inferred from the conceptual structure in which the word is embedded. The activa-tion of meaning-relevant knowledge in this process runs via a frame that contains the relevant standard information (in the form of default values). In this example, the word in activates the container frame.61 Various informational units can potentially be retrieved via numerous frame elements (slots):

– Container type: What type of container is involved (thermos flask, garbage bin, trouser pocket, etc.)? What distinguishes this container type from others?

– Use: What is the container typically used for (cooking, drinking, etc.)? How often is it used?

– Properties: What material is the container typically made of (porcelain, steel, glass, etc.)? What attributes does the material have?

– Size: What measurements does the container typically have? What shape is it? Is it closed (cupboard, suitcase, etc.) or open (glass, vase, cabriolet, etc.)?

– Content: What does the container typically contain (clothing, drinks, garbage, etc.)? How long/when does it contain it (cf. basket, lift)?

– Parts: What parts of the container can be identified (bottom, side walls, bottle neck)?– User: Who uses the container (a particular person, a particular occupational group,

etc.)?– Location: Where is the container typically found (in living rooms, kitchens, gar-

dens, at the airport, etc.)?

Information can be variably retrieved from this set of data.62 In the complex expres-sion a crack in the vase (an example to which I will return later), the semantic focus is

60. Lakoff (1987), (1990); Johnson (1987), (1999); Lakoff and Johnson (1980), (1999); cf. from the psycho-developmental perspective Mandler (2004), (2005). The container concept is, how-ever, only one example of many. It is omnipresent as a conceptual metaphor (to bottle sth. up, to bellow sth. out, to blurt sth. out, someone explodes, etc.).

61. Cf. also the details on the container frame at:http://www.cogling.com/index.php?title=F:Containers [last accessed: 10 April 2014].

62. Fillmore provides an illustrative example of how the preposition in itself is specified in the container type. Compare the difference between I saw it in the news and I saw it on the news. Fillmore notes (1984: 128): “The difference is in no way directly explained by knowledge of any independently discoverable difference in meaning between in and on, but is rather the result of a complicated set of inferences involving the verb see, the recognition of contexts in which some segments of a larger domain can be referred to as the news …”


directed towards the properties of the vase material while ignoring other knowledge facets (like container type, use, content, etc.). A process of conceptual integration leads to this: A crack is a non-material entity and refers to the material properties of an object, so only this meaning aspect of the expression vase is profiled.63

However, in a modular semantic model such conceptual structures and cognitive processes only play a secondary role. Modularists argue that alongside the conceptual knowledge system there is a linguistic knowledge system that precedes the former during semantic interpretation. Its primary research focus is thus the underlying, purely linguistic meaning representation of an expression. In a modularist reading, the case of the preposition in involves an argument structure with two variables, of which one stands for the localized entity, whereas the other indicates the region in which the object is contained. Apart from these syntactically motivated components, formal-semantic representations also contain at least one semantic feature. Schwarz (2000: 32) explains the meaning of in in this manner: the object to be situated x is contained in a region of y.64 Whereas X and Y are the variables in the argument structure, contained in represents a semantic constant.65 The thesis then is that this abstract representational format predetermines all meaning variants of the preposi-tion in. Different readings arise from the fact that different conceptual values occupy the argument structure. Compare the following examples that Herweg (1988: 15ff.) explains within his modular approach.

(1) the crack in the vase (2) the flower in the vase (3) the water in the vase

63. In a detailed semantic analysis numerous other aspects beyond these two would of course have to be given. The effect of the crack on the functional value of the object depends on the material, cf. the crack in the vase versus the crack in the record.

64. A remark on the notation: Capital letters are usually used to designate that semantic com-ponents are being referred to; I follow this convention here. This does not indicate whether it is a primitive component (semantic constant) or a complex component (conceptual element).

65. The semantic constant thus deviates from the usual information in terms of degree of ab-straction (Lang 1991: 128; Herweg 1988: 74): Herweg provides the following semantic form for the preposition in: λx λy [loc [x, place [y]]]. We will encounter other such “translations” of natural linguistic expressions into the language of formal logic when we examine two-level semantics. The lambda operator (λ) is a functional expression that provides the characteristic function of the sum of individuals that the predicate applies to. More precisely, in the given example in, the localized entity x is situated in a region that describes the place that the entity y occupies. This roughly corresponds to Schwarz’s component contained in (enthalten). In Examples (1) to (3) the variable x is replaced by the values ‘crack’, ‘flower”, ‘water’ and the variable y by the value ‘water’. The key thing to note here is that in the formal representation no information about the properties of the place are given, for instance whether the interior of the vessel is meant as in (2) and (3) or the material of the vessel as in (1).


Whether the conceptual components crack, flower or water occupy the first vari-able in the argument structure of in determines in which manner the semantic content of the expression is specified. How radically semantic representations are underspeci-fied is shown in the fact that they must be sufficiently abstract in order to equally sub-sume all conceptual interpretations. It is solely inferences drawing upon encyclopaedic knowledge that distinguish the three interpretational variants (1) to (3). These include, for example, the knowledge that a vase which contains a flower or water is normally in an upright position, which need not necessarily apply to a vase with a crack (cf. Taylor 1995: 5ff.), but also that a crack is not a material entity, but rather describes the state of the material, while water is a fluid that fills the hollow of the vase, whereas flowers have a solid substance and prototypically protrude from the vase.

Formal-semantic meaning representations do not include such facets of knowl-edge. What manner of semantic feature is required so that it covers differing forms of being-contained without itself exhibiting conceptual structures? Which meaning-relevant function fulfils a formal representation of in that abstracts from all those facets of knowledge constituting the semantic nuances of (1), (2), and (3)? I will take up questions of this nature in Section 2.2.1. The holistic counterthesis is that there is no such purely linguistic level of meaning representation and that semantic interpre-tations are essentially encyclopaedically motivated and thus arise principally from conceptualizing capacity.

This topic touches on a multitude of traditional issues of conflict. The question of cognitive organization is controversial not least due to its epistemological implications. Whether the mind is an indivisible whole and can be considered independently from the human body is known as one of the most enduring issues of Western philosophy, which dates from early idealistic positions such as those of Descartes. The mind-body dichotomy is also the cause of the conflict between modularism and holism in cogni-tive science. The proposition that the language module is autonomous stems from the fact that the linguistic rule system is not seen as being affected by the conceptual, sensorimotor system. Applied to the submodule of semantics, this assumption entails a rationalist objectivism66 that consists of inserting a predetermined and purely lan-guage-motivated meaning level. This exists entirely independently of experience-based cognition and, with the help of predicate logic, can be determined algorithmically as a manipulation of arbitrary abstract symbols (Bierwisch 1987). In this sense, modularity theory conceptions belong to the Cartesian rationalist tradition.67

The concept of experience-based cognition, following Lakoff and other adher-ents of Cognitive Linguistics, relates a priori to a very broad and non-specific area

66. Putnam (1975) speaks in this context of objectivist metaphysics; cf. also Lakoff (1988).

67. In generative grammar, modularity also includes neurophysiological aspects: It is a descrip-tion of how our brain is neurologically equipped and structured. For a criticism of this type of dualist ontology from a philosophical perspective cf. Krämer (2001: 37–54), and from a linguistic perspective, see the contributions in Linguistic Review 22/2005.


of experience that equally encompasses sensorimotor, social, emotional, etc. facets.68 In the conception of two-level semantics, the principle of modularity is responsible for this area of experience remaining completely irrelevant for the foundations of a semantic theory. That which is specified on the first level of semantic representation is merely epistemically “enriched” on the second, conceptual level (Dölling 2005: 159ff.).

Lakoff summarizes the holistic counterposition under three points:

– Where objectivist cognition views human thought as fundamentally disem-bodied, experientialist cognition sees human thought as essentially involving the kind of structured experience that comes from having human bodies, especially from innate human sensory-motor capacities.

– Where objectivist cognition sees meaning in terms of a “correspondence the-ory”, as the association of symbols with external objects, experientialist cog-nition sees meaning as essentially involving an imaginative projection, using mechanisms of schematization, categorization, metaphor and metonymy to move from what we experience in a structured way with our bodies to abstract cognitive models.

– Where objectivist cognition sees thought processes as the manipulation of abstract symbols by a great many highly-structured algorithms, experien-tialist cognition posits a small number of general cognitive processes whose application to abstract highly-structured cognitive models constitutes reason. (Lakoff 1988: 121)

In the meantime this agenda of a holistic theory has been frequently discussed, refined, and supported by empirical studies. However, not all the points are equally important for a frame-semantic conception such as that posited here. The last two points are of central importance, since we are dealing with cognitive procedures that arise through knowledge representations – for instance in the form of frames. The construction of conceptual structures occurs as a cognitive construction process, and in this context the question arises as to which different operations are individually involved in this.

The first point precedes a frame theory and concerns the mentioned problem area of the mind-body dichotomy. With the fundamental epistemological conviction that has entered into the literature under the catchword of “embodiment” (cf. Johnson 1987; Lakoff 1987; Zlatev 1997; Lakoff and Johnson 1999; Wilson 2002; Ziemke, Zlatev, and Frank 2007) cognitive linguists hold the view that the human mind, and consequently language, cannot be considered in isolation from the human body or from specific human cognitive structures and their organization.69 Isolating language ability from

68. Lakoff himself (1988: 120) speaks of “experientialist cognition”.

69. According to Margaret Wilson (2002: 625) six main theses form the basis of the embodiment theorem: “(1) cognition is situated; (2) cognition is time-pressured; (3) we off-load cognitive work onto the environment; (4) the environment is part of the cognitive system; (5) cognition is for action; (6) off-line cognition is body-based”. She considers the fifth thesis to be problematic.


our physical, social, and cultural experience would thus mean investigating a language that had never existed (and probably never will). Conversely, the central task is

…to ask how much of the structure of language is determined by the fact that people have bodies with perceptual mechanisms and memory and processing and limitations, by the fact that people have to try to make sense of the world using limited resources, and by the fact that people live in social groups and have to try to communicate with each other. (Lakoff 1983: 155)

Linguistic categories are therefore not abstract, quasi disembodied, i.e. objective cat-egories, but rather constructed products that are firmly grounded in our experiential world. The gap between a modular and a holistic model is therefore enormous.

The latter focuses its linguistic considerations on those aspects that have been lost sight of in the former (cf. Langacker 1999a: 15–17):70

i. environmental factors like gravitation and interaction with the material environment,ii. biological factors of an anatomical, physiological, neurological, and perceptual

nature,iii. (developmental-)psychological factors, in particular concerning the ontogenesis

of cognitive abilities,iv. historical factors with a view to the limitations of cognitively available knowledge

resources on the one hand and the dynamics of grammaticalization processes on the other,

v. socio-cultural factors that account for the indissoluble fact that language is first and foremost a tool for communication and as such provides a dynamic repertoire of conventional signs.

We will see in the following that the differences are also enormous when it is a matter of meaning representation in the narrower sense – and thus also of frames.

2.2 Modularism

There are a number of variations within the modularist paradigm. In Section 1.3.2, I distinguished between various semantic-level models with respect to meaning theory enquiry. Bierwisch, for instance, describes linguistic meaning as an interaction between two levels, a purely linguistic one and a conceptual one. Schwarz replies with a model that provides for three representational levels. With his model, which is affiliated with

70. In an ironic inversion of Chomsky’s programme (1965: § 1) one could state: Linguistic theory is concerned primarily with an everyday speaker -listener, in a heterogeneous speech-community, and who is always also affected by such grammati cally relevant conditions as memory limita-tions, distractions, shifts of attention and interest, and errors (random or character istic) in ap-plying his knowledge of the language in actual per formance. Also see, for instance, Langacker (1988a).


observations from gestalt psychology, Jackendoff takes up a position within semantic holism in which linguistic meaning is considered as a conceptual entity. At the same time Jackendoff (similar to Bierwisch and Schwarz) denies that syntactic and phono-logical structures have the same conceptual status. This is the point in which the holistic approaches of Langacker, Fillmore, and others differ.

The following section outlines two modular variations from a frame-semantic perspective. It focuses on the extent to which the represented approaches entail se-mantic reductionism that consists in only partially explaining or being able to explain meaning-relevant knowledge.

2.2.1 Two-level semantics (M. Bierwisch)

The modular approach is very closely linked to generative grammar (Bierwisch 1987). From a generative perspective, language as a cognitive subsystem is distinguished from other subsystems (Chomsky 1980: 3; 1986: 5). Every (sub-)system functions autono-mously, i.e. it is constituted through an independent system of rules that is subdivided into further subsystems. The principle of modularity applies for every level of abstrac-tion, specifically for

– the general organization of human cognition (perceptual, sensorimotor, linguistic system),

– the more specific organization of the linguistic subsystem (phonological, morpho-logical, syntactic, semantic system),

– the even more specific organization of the semantic sub-subsystem (semantic and conceptual system).

In this context, the third point is naturally the most relevant. However, it is impor-tant to keep the overall context in mind: Language is understood as an autonomous mental module of moderate degree of abstraction that is composed of independent cognitive systems that interact with each other in language processing, i.e. in language production and reception.71 Bierwisch’s well-known formula to describe the compo-sitional structure of an utterance meaning should be seen against this background (cf. Bierwisch 1983b: 33ff.; and 1979):

(((ins (phon, syn, sem)) ct, m) ius, cs).

71. Fodor (1983) presented the most sophisticated theoretical outline for this and argued for research into modularity with respect to not only the structural but also the procedural features of cognitive systems. He distinguishes between three cognitive mechanisms (transducers, input and output systems, and central processes), which are, however, not relevant for my further argumentation. Therefore, I do not address Fodor’s approach any further and limit myself to the division between a semantic and a conceptual knowledge system.


This states that a linguistic utterance has a grammatical structure that is composed of the three elements phonological structure phon, syntactic structure syn, and semantic content sem. The utterance is a physically describable unit ins that differentiates itself within a context ct to become the actualized meaning m. This in turn fulfils a particular communicative function cs in a particular utterance situation ius. In this formula, the grammatically determined linguistic system (i.e. phon, syn, sem) can be retrieved in the innermost brackets. All other entities belong to the conceptual system. Although in this case Bierwisch distinguishes more precisely between the conceptual system C, constituted through ct and m, and the system of communication rules, constituted through ius and cs, both of them, however, fall back on conceptual knowledge. In contrast to the grammatical system, they are thus conceptual values.

Insofar as generative grammar, in conformity with Chomsky’s extended standard theory, departs from the premise that language and conceptualization competence constitute two autonomous cognitive domains, a sub-area of semantic competence also enjoys a cognitively autonomous status (cf. Figure 1).72 This dimension of semantic knowledge forms a grammatical module, like syntax and phonology. In contrast to these, however, it has the distinction of functioning as an interface with the conceptual system. The elaboration of initially purely syntax-oriented theoretical considerations (Chomsky 1965) to other levels means that the innateness of grammatical structures, generative grammar’s main thesis, also affects a sub-area of semantic knowledge. Since this precedes the human conceptualizing capacity, it not only operates in a modular fashion in processes of understanding language, i.e. detached from experientially based cognition, but in this function is also equally able to be reconstructed as an autono-mous rule system.73

According to this view, semantic knowledge is structured in several levels. For instance, Bierwisch (1979: 64ff.) distinguishes at one point between three semanti-cally relevant knowledge types: linguistic knowledge, everyday knowledge, and knowl-edge of social interaction. All three observe autonomous rules of formation. Whereas knowledge of social interaction and everyday knowledge belong to the conceptual system, linguistic knowledge – by which Bierwisch understands the purely linguisti-cally determined meaning (or logical form) of an expression – constitutes the semantic system. Figure 1 illustrates the relationship between the three levels. Figure 1 also shows that linguistic knowledge, everyday knowledge, and knowledge of social in-teraction stand in a unidirectional determination relationship. The levels of everyday

72. Bierwisch (1982: 13) states: “[S]emantic representations of lexical items in general have a certain amount of internal structure, i.e. they are configurations of semantic primitives deter-mining the context dependent interpretation of linguistic expressions.” And he continues with the observation “that lexical items are in a sense self-contained structure units, whose categori-zation determines their possible function within larger units.”

73. Early critiques of the modular model can be found, for instance, in McCawley (1968), Lakoff (1971), Seuren (1972), and also summarized in Schnelle (1970) and Olson (1974).


and knowledge of social interaction merely further refine what is inherent at the level of language knowledge. This results in something that resembles the conception of a “container of meaning”.74

The scope and limitations of empirically determining the meaning components inherent to semantic processing are established by those methodological principles that for their part arise out of the primacy syntax is given in generative grammar. Noteworthy in particular is the principle of compositionality, which is bound to syn-tactic structure (cf. for instance Bierwisch 1971; Dölling 2001a), according to which the meaning of complex expressions consists of the meaning of their elementary signs as subject to syntactic structure. Further, there is the central methodological premise adopted from structuralism that states “one form – one meaning” (Jakobson 1936; Saussure 2001), which motivates a layered treatment of linguistic meaning, that is, a categorical separation of expression meaning and utterance meaning (cf. Dölling 2001b: 7–8; Dölling 2005: 163–164). “Form” relates to the meaning dimension that is exclusively determined by the language system and is algorithmically ascertainable, whereas “meaning” indicates the contextually dependent dimension of a particular unit of expression.

As already mentioned, the basic notion of such a methodological procedure dates from Chomsky 1965. Here Chomsky lays out the basis for a treatment of semantics based on syntactic theory. In conformity with this, in a description of semantic knowl-edge, modularists are initially concerned with the level of semantic form. Linguistic knowledge is knowledge of linguistic forms, and these provide a

74. I counter this model with the conception of “meaning potential”, particularly in Section 4.2.4.

B (A)

M (t)

CS (t)

Level of linguistic knowledge:

Linguistically determined meaning (logical form)

Level of everyday knowledge:

Utterance meaning of the utterance t

Level of social interaction knowledge:

Communicative meaningof the utterance t

Semantic system

Conceptual system

Dire

ctio

nof

det

erm

inat

ion

Figure 1. The concept of the ‘container of meaning’ within level-semantic models, reconstructed here on the basis of Bierwisch (1979: 64–72)


…context-dependent matrix of conditions for the meaning of a linguistic expres-sion, which in interaction with other cognitive systems is mapped onto a context-specific meaning – the utterance meaning. (Translated from Meyer 1994a: 62–63)

In his language model Chomsky substantiates this matrix of conditions by adopting a syntactically motivated “base component” (Chomsky 1965: 88–90, 113–121), which is considered to deliver input to the semantic component by specifying the syntactic selection restrictions. Only then does the semantic component generate a semantic representation from this deep structure. Thus, already in the base component are those semantic features of a lexical category identified that determine the selection of another lexical category that may be combined with it. For instance, as the verb sleep imposes the selection restriction [– abstract] on the subject, ideas are not able to sleep.

Methodologically, a modularist approach to language implies that encyclopaedic knowledge is excluded from semantic theory. That is to say, the potential selection relationships (for instance, between a noun and a verb) determine purely grammatical relations, and these enjoy an autonomous, i.e. meaning-independent, status (Chomsky 1965: 147–155). As such, they are categorically separate from conceptual knowledge. Ascertaining the selection restrictions ensures a checklist of semantic features. If the features of two lexical categories (e.g. [+ abstract], [– abstract]) contradict each other, they cannot be combined.

However, for a long time, semantics only played a subordinate role in modular approaches. The focus of the studies was largely on the syntactic component of the language faculty. In Chomsky’s conception, this had no interface to extralinguistic cognitive systems. As this is clearly not the case for the semantic system, this is assigned a special function in the modular perspective. The semantic linguistic component me-diates between the linguistic and extralinguistic conceptual systems. Numerous studies, in particular by Bierwisch (1983a, 1983b) and Lang (1985, 1988), have attempted to demonstrate this interface function. On the one hand they take account of the con-ceptual dimension of relevant world knowledge in meaning actualization, but on the other, they adhere to the notion that the core meaning of every linguistic expression is determined by grammar. They thus postulate a two-level representation of linguistic meaning.75 This model has entered into the literature as “two-level semantics”.

The basic modularist premises are now expanded to the internal structure of the semantic system. On the one hand, semantic features are understood as predicate con-stants in the sense of predicate calculus, but on the other hand the so-called lambda operators are applied as additional determinants of the linguistic system that specify

75. Dölling (1995: 128) states: “This thus presupposes that the semantic representation of a lexeme contains no elements at all that according to their origin would have to be classified as knowledge of the extralinguistic world. A particular consequence of this form of modularity is that those conditions that are traditionally understood as restricting semantic selection and therefore containing knowledge of the ontological categorization of entities are no longer avail-able as lexical requirements”.


the syntactic slots. Together they determine “which concepts are expressed linguisti-cally in what way” (translated from Wiese 1999: 94).

Studies on the meaning representation of dimensional adjectives (Bierwisch 1986; Bierwisch and Lang 1987) show how the interaction between the semantic and conceptual systems can be grasped more precisely. The first level, that of semantic representation sem, is of a purely linguistic nature, lexicon-based (i.e. in form and substance bound to the principles of universal grammar and rigidly componentially structured), and thus structurally decomposable into semantic components (cf. also Lang 1994). The second, conceptual level cs is, by contrast, language independent and based on human experiential knowledge. It “determines the mental representation of that which is conveyed through linguistic utterances” (translated from Bierwisch 1987: 6). According to this, semantic representations adopt a specific interface func-tion between the conceptual system and the syntactic system syn in that they, on the one hand, indicate via lambda operators those slots that can be filled by constituents on the syntactic level and, on the other, mark an area of possible conceptual world knowledge elements that come into question for syntactic combination. For this rea-son, Wiese designates this dual function of semantic representation the determination of the syntactic “combinatory potential” (Fügungspotential) on the one hand and the specification of the conceptual “reference potential” (Referenzpotential) on the other (Wiese 1999: 93–94). In the process of meaning actualization, this reference and com-binatory potential is specified by conceptual knowledge. Conceptual knowledge forms the set of possible conceptual elements that can be enlisted to interpret the semantic units, whereas linguistic knowledge is the set of those purely linguistic elements (also called “components”, “primitives”, or “semantic constants”) that restrict the possible interpretations of a linguistic expression to a particular reference corpus (cf. Figure 2).

Adherents of a two-level semantic model combine this separation of linguistic and world knowledge with a theory of the mental lexicon (Bierwisch and Schreuder 1992: 41ff.). The mental lexicon stores only the contextually invariant core meaning of a word, which includes exclusively semantic constants and variables that specify the number and the type of the arguments that require a lexical unit.76 Differing contextual meaning variants of the same linguistic expression are derived from this (such as in the example of the preposition in in Section 2.1). The same semantic representation underlies them; however, on the basis of differing contextual data they are conceptu-alized differently. This is based on the conviction that semantic representations are radically underspecified and are only stored in the mental lexicon in this form. This does not apply to conceptual elements.

In the next two sections I wish to illustrate this division using some example analy-ses. This is particularly in order to show that semantic representations predetermine

76. Bierwisch and Schreuder (1992: 28) state: “[W]hile constants have a fixed, albeit context-independent interpretation in terms of conceptual, perceptual, motoric, and possibly other mental structures, variables are open slots that are to be filled in one of the two ways: they are either specified by other linguistic expressions, or they are to be fixed by appropriate conceptual values”.


the possible conceptualization of a linguistic expression to such an extent that the rich meaning potential of frames, which characterizes them as a semantic form, is seriously inhibited.

2.2.2 Frame semantics vs. two-level semantics: some issues

How do frames fit into a two-level model such as that illustrated in Figure 2? Let us assume, along with Bierwisch (Bierwisch 1979: 78–79; Bierwisch and Schreuder 1992: 32–33, 47), that frames do indeed fit into the modular conception of a two-level semantic model. What immediately becomes apparent is that as a result of their conceptual nature frames would be secondary phenomena. Ignoring all ontological, functional, methodological, and epistemological specifications of frames that derive from the holistic paradigm, the following general picture emerges: Frames structure the reference potential of linguistic expressions and contribute to ensuring the effective step-by-step transition from underspecified semantic representations to contextually enriched concepts. However, this conceptualization process would remain bound to the specifications of the underlying semantic representation format sem. Only variables in the argument structure could be filled with conceptual values, and only semantic constants (components) could be framed and thus become semantically “richer”. The latter would be provided by the default values of a frame. Yet, it is important here to note that these would have no lexicon entry, in contrast to the variables and the seman-tic constants. The cognitive status of both would be significantly different.77

This becomes clearer using example analyses. For words such as church, theatre, prison, school, university (Kirche, Theater, Gefängnis, Schule, Universität), etc. that is,

77. Bierwisch and Schreuder (1992: 28): “Both constants and variables are assigned to specific semantic categories determining on the one hand the type of conceptual interpretation that can be associated with them, and on the other hand the combinatorial structure of SF [semantic form] based on these assumptions”.

sem = semanticrepresentation

[visit (peter, x)]

syn

λx

Combinatorypotential

cs

Referencepotential

Linguistic knowledge Conceptual knowledge

Figure 2. Separation of linguistic and world knowledge in a two-level semantic model


for words that indicate different types of institutions, Bierwisch suggests the following uniform semantic representation sem (Bierwisch 1983a: 86):

(1) λx [purpose [x w]]

The representation must be this abstract so that the formal structure remains common to all expressions denoting institutions without at the same time containing conceptual elements. The component institution itself already describes “an involved complex of conditions” and is consequently – in contrast to the semantic constant purpose – part of a “theory of everyday life” (Bierwisch 1983a: 86). The same applies for the component teaching_and_learning_processes with which (1) can be specified so as to capture the (still very underspecified) meaning of the word school.

(2) λx [institution [x] & purpose [x w] & teaching_and_learning_processes [w]]

Where, then, does the dividing line between linguistic and conceptual knowledge run? In the quoted example it clearly runs between (1) and (2). The semantic representation cannot extend beyond the informational units contained in (1), since the component institution presupposes a complex knowledge context, which for instance provides answers to the following questions: Who creates institutions? What relevance does an institution have for humans? Where do we find institutions? What sort of institutions are there? To what extent does an individual have anything to do with institutions? To distinguish the meaning of the word school from the meaning of the words university or church, it is necessary to have semantic specifications that require more detailed knowledge. Thus, teaching_and_learning_processes are part of the concepts of both school and university, and one may argue whether priests teach the teachings of God and whether the congregation is thus a group of learners, i.e. whether church also contains the conceptual element teaching_and_learning_processes (or may do so at least in certain contexts).

However, the extent to which the institutional forms that are designated by school, church, and university are distinct from each other can be shown in an initial ap-proximation by the answers to the following questions: Who goes to church, who to university, who to school? What is taught there? Why does teaching take place there? Who teaches at a university, who at a school? Is there an obligation to go to church or university are there is to go to school? Why (not)? And so on. Without doubt we have extremely rich conceptual knowledge about institutions that allows us to distinguish institutional types from each other. It is clear that this knowledge is solely encyclopae-dically motivated. Linguistic knowledge cannot be used to differentiate between vari-ous institutional forms, nor can it determine whether a linguistic expression designates an institution at all or, more generally, what an institution actually is.

Bierwisch also admits this, since the knowledge addressed is knowledge of the world, and it is indispensible to understanding and semantically distinguishing be-tween expressions like school, university, and church. Bierwisch is thus concerned with isolating an area of purely linguistic knowledge from this encyclopaedic knowl-edge. Example (1) presents just such a purely semantic representation of the expres-sion school. However, let us note: All that this semantic representation contributes to


understanding the expression school is the knowledge that this contains a component purpose. It does not even contain information as to what purpose this is, as this would already take recourse to conceptual knowledge; therefore both pertinent argument slots remain empty.78 Thus there is a semantic constant purpose contained in the se-mantic representations of the expressions handkerchief, trousers, condom and window pane, insofar as these expressions designate human products that are manufactured to fulfil particular human needs. And from the perspective of a farmer, the semantic representation of the expression rain would probably contain a purpose component (although modularists would naturally counter with the argument that this does not depend on the perspective of particular individuals, but rather on universals, which would be why a purpose component would not be relevant – but what is “universal” and how can “universality” be empirically proven?).

What Bierwisch and all supporters of a two-level semantic model wish to avoid can be summarized as what Herweg (1988: 55) calls “polysemy inflation”. According to this, the different readings of the expression school distinguish themselves merely in conceptual content; their semantic representation always remains the same. For this reason, the expression school is not considered polysemous. It is merely enriched in a conceptually different manner in different contexts. By occupying the variables in the semantic representation in (1) with different conceptual elements, the expression school can be interpreted as an institution in one situation, a building in another, and in yet another as a process. Compare this to the semantic decomposition of the expression school as it occurs within a two-level model in the following examples.79

(3) The school needs to be reformed. λx [institution [x] & purpose [x w] & teaching_and_learning_processes [w]] (4) The school lies on the outskirts of the city. λx [building [x] & purpose [x w] & teaching_and_learning_processes [w]] (5) School is boring. λx [process [x] & purpose [x w] & teaching_and_learning_processes [w]]

To what extent then does the semantic representation sem have an autonomous status? There can only be an autonomous level of semantic representation as in (1) if conceptual processes (metaphorical and metonymical shifts, conceptual integration and blend-ing, etc.) do not intervene on this level. Consequently, from a modular perspective,

78. The argument structure (theta grid) of an expression contains information about the num-ber, placement, and type of obligatory arguments for an expression; the argument structure for purpose is presumed to have two positions, thus x has the purpose w.

79. At this point, for illustrative reasons, I do not begin by discussing problematic presupposi-tions such as whether it is at all meaningful to isolate an institutional reading from a building reading: To what extent do institutions exist without buildings? And what is meant by the com-ponent process in (5)? The contentious nature of the latter question arises from the problem of the extent to which the obviously privileged building reading of It is boring at school can be held up against the process reading in (5) despite its semi-synonymous content.


Example (5) is not a process of metonymic shift. This would then presuppose that the semantic interpretation of the expression school derives from the institutional reading in (3). Instead of accepting such a conceptualization process, Bierwisch insists that all pos-sible readings of an expression are principally equivalent and not to be viewed as being derived from each other. In each case, a conceptual reading is exclusively motivated by the fact that a conceptual element (institution, building, process) occupies the first argument position of the semantic constant purpose. This is not a peculiarity of the given example. Rather this applies analogously for the semantics of any word at all.

At this point two problems emerge. First, this type of approach implies a rather strict division between representational and procedural aspects of language un-derstanding. By definition, semantic representations form the primary level of un-derstanding, whereas conceptual interpretations form the secondary level. Strictly speaking, it is already problematic to speak of an act of understanding with respect to the first level, since no cognitive construction capacities are involved in building up semantic representations. If, however, as Dölling writes “several levels are passed through when understanding an utterance” (translated from Dölling 2005: 163), this must be verified from a procedural perspective. Purely theoretical postulates are not sufficient. Empirical evidence must be provided, for instance using psycholinguistic studies, that purely semantic processing principally precedes conceptual processing. However, to date, there is no conclusive proof of this. Not without reason does Dölling (2005) argue on a purely methodological level.

In the same way, empirical studies have not been able to support the proposition that semantically more complex verbs – i.e. those that require one or more argument positions than others – take more processing time in the process of understanding (Kintsch 1980; Fodor et al. 1980). That would, however, be a necessary consequence of the modular premises. Text linguistic studies clearly indicate the opposite, that in-ferences already have an elementary function at the beginning of every act of un-derstanding (cf. Fillmore 1981, 1984). Further, priming experiments suggest that the transformation should not be modelled on a process of passing through “levels” but rather as an activation of segments of semantic networks (cf. Aitchison 1994: 82ff.; Rickheit and Strohner 1993: 117ff.; Rickheit, Sichelschmidt, and Strohner 2004).

A second problem also results from the modular division between a semantic and a conceptual system. According to this division, different readings of a word, such as the noun school in the example, occur from fundamentally equivalent conceptual shifts. The Examples (3), (4), and (5) – and many other conceptual variants that could be found – are solely distinguished by the fact that different conceptual elements are slotted into the formal representation in (1). Yet, these elements certainly differ in their degree of salience. From an empirical perspective, it is not unlikely that metonymi-cally motivated shifts tend to be favoured in semantic interpretation (more on this will follow). Contextual data provides additional motivations. In the sentence The school has 13 classes, both the institutional and the building readings are possible, but the first is certainly preferred, because this activates the standard assumption that schools generally have very many more classrooms. Two-level semantic models cannot ac-count for such phenomena. According to Meyer (1994b: 37) “default logic would more


adequately reflect usage” than a lambda conversion, the latter being how a conceptual interpretation is achieved within two-level semantics. No contextual conditions can be given that lead to a preference for a particular semantic interpretation, since under the auspices of modularism the process of meaning interpretation is approached from the perspective of the underlying formal representations. The case at hand, for instance, does not capture the fact that the preferred reading of Building is crucially linked to the meaning of the complex expression 13 classes.

Let us return to the question of what it would mean to address frames within a modular cognitive theory. Frames would be a tool for differentiating the content of se-mantic forms. However, since lexical units are never polysemous by dint of their stand-ardized semantic representation, all potential frames would have to be contained in this format. As we have already seen, they would be able to occupy only those conceptual slots prescribed by the argument structure of an expression that is prestructured by its semantic constant(s). The aim of modular semantics is to provide evidence that the in-terpretive variation of a linguistic expression is predetermined “by the system of available SF-constants and their combinatorial properties” (Bierwisch 1986: 771). Consequently, the choice of possible frames is already limited in two respects before conceptualization even occurs, and that means, before the process of semantic understanding begins.

In the next section I use example analyses in an attempt to highlight different defi-ciencies of such a modular theory of meaning. The focus is on the empirical inadequacy of modular semantic descriptions.

2.2.3 Example analyses

Apart from the key cognitive theoretical issues mentioned, it is ultimately not a ques-tion of empiricity or theory whether it does indeed make sense on the basis of assump-tion to so rigidly restrict processes of meaning actualization by a level of semantic representation. In our case, for example, all the possible utterance meanings of school should be able to be derived from two argument positions and the semantic constant purpose. In particular, every meaning-relevant frame element (in this case, default values like ‘institution’, ‘teacher’, ‘lesson’, ‘class’, etc.) should be indirectly, that is, medi-ated via the concept school, traceable to one of the two argument positions. The fact that there are no other slots apart from these follows directly from the basic assump-tions of two-level semantics I outlined.80

80. To further underline this point, an additional quote by Bierwisch: “This view is very much in line with the traditional distinction between lexical and encyclopedic knowledge … Encyclopedic knowledge as a component of C [the conceptual system] is a memory based system of concepts and beliefs that provides context dependent interpretations for SF-representations, which are not necessarily part of individual lexical items. These interpretations are channeled, so to speak, by SF-constants, but depend crucially on the conceptual context that fixes the parameters contained in SF-representations” (Bierwisch 1986: 779; underscoring by the author).


The following usage variants of the word school provide a small selection of the possible aspects of meaning that the word school might evoke.81

(6) The school has bright, spaciously designed classrooms. (7) Lena Kurz is the new principle of the school. (8) (The) school has to prepare children for the competitiveness of working life. (9) The school is organizing an open-door evening. (10) The school offers the course participants sound professional training. (11) Parents can also become involved in shaping the school. (12) The influence of (the) school ends at the children’s doorstep.82

(13) The school Studio ImPuls is a member of the VdG. (14) The 12-year-old took part in a competition during the spring vacation. (15) The domain “schule.de” was reserved worldwide by the German Unix User

Group school task force. (16) Girl-friendly comics … are tearing down old school attitudes.83

The first three sentences are more or less prototypical examples of the three readings introduced above in (3), (4), and (5). Using additional examples, let us go through the points that make the assumption of a contextually invariant semantic representation structure seem problematic.

Firstly, due to the premise that “every (well-formed) utterance model t has a lit-eral meaning” (Bierwisch 1979: 70)84 – meaning that the semantic representation (1) underlies all the example sentences – metaphors and metonyms are deemed to be peripheral phenomena that are not the primary object of semantic analysis. However,

81. The following instances are randomly chosen from over 140,000 instances in the IDS corpus Cosmas II [and subsequently translated for this publication] cf. http://www.ids-mannheim.de/cosmas2/ [last accessed: 10 April 2014].

82. Original: Der Einfluss der Schule endet vor der Haustür der Kinder. In contrast to English, the German original of Example (12) allows two variant readings of the nominal phrase. Einfluss der Schule can refer to both the school as an institution and a particular school, whereas in English the zero article suggests the institutional reading, while the definite article indicates a particular school.

83. This example is sourced from an article found at http://vitaminw.co/culture-society/making-better-comics-changing-sexism-industry [last accessed: 5 February 2014].

84. The well-formedness criterion does not allow any conclusions to be drawn on an autono-mous semantic level – quite the contrary, cf. Fillmore (1973: 276): “[C]onversations, or discourse samples in general, are not well-formed or ill-formed as such, but only on particular interpre-tations. These interpretations have to be based on assumptions about what is going on in the conversation … There is no way, in short, of talking about grammaticality or well-formedness without getting in many ways involved in the details of social interaction by means of language”.


is it not the case here that metonymic processes form an intrinsic phenomenon of the meaning actualization? In order to understand the word school in (9) and (10), it is not necessary to initially activate the semantic component purpose. Instead, school evokes a frame whose elements ‘teacher’ or ‘principle’ are immediately available as the default values of the slot ‘part-whole’. This process of interpretation via an activated cognitive domain remains closed to two-level semantics because the inferred default values cannot occupy any variable in the semantic representation.

Therefore a two-level model is forced to take the following detour: (i) activate the institutional reading, (ii) individualize to a particular institution, (iii) specify the relevant purpose of this institution through conceptual components,85 (iv) derive the meaning-relevant conceptual element school_employee from this component. This is, in fact, an abbreviated version, since presuppositions of the conceptual element have not been explicated in the final step. However the decisive factor is something else. The conceptual element school_employee cannot be readily instantiated in semantic representation (1), because the first argument position of the constant purpose is occu-pied by the concept institution, the second by the concept do_house_advertising or teaching_and_learning_processes respectively. Strictly speaking therefore, the element school_employee does not belong to the predetermined reference potential of school, and interpreting it would require engaging in an additional inference process that lay beyond the specifications of semantic representation (1).

The fact that such fundamental difficulties arise for such an apparently simple phe-nomenon as the understanding of tropes like metonymy is due to conceptual processes being categorized as secondary phenomena. What applies to sentences (9) and (10) applies to the other examples as well.86 This failure to capture primary U-relevant87 data is grounded in the theoretical tenet:

The exclusion of metaphor from the scope of the approach is possibly a reflection of the theory-internal requirement for semantic form to constrain conceptual interpretation. Metaphor is a creative and highly unconstrained phenomenon. … Nevertheless, metaphor is endemic in natural languages. A semantic theory that excludes metaphor is inherently impoverished. (Taylor 1994: 7)

Ignoring the constitutive role of trope shifts in the case of the lexical unit school means failing to recognize not only an obviously highly dominant usage form, but also those

85. Proponents of two-level semantics would insert the component do_house_advertising in the case of (9) and teaching_and_learning_processes in the case of (10). Yet, these are conceptual components that could also break down differently depending on co- and context.

86. Perhaps with the exception of (6) and (15), since interpreting school as a building and an institution respectively is sufficient without having to draw on individual frame elements.

87. Following Fillmore (1985), “U-relevant” stands for “Understanding-relevant”, for more details see Section 1.2.1.


cognitive routines that allow segments of meaning potential to be deduced.88 Further deficits follow from this point.

Secondly, against the background of what has just been said, we must once again ask what meaning-relevant function cognitive processes like conceptual shifts, integra-tion, and blending have. It is highly implausible to assume that semantic constants like purpose are the objects of a “straightforward interpretation in terms of conceptual pattern or operations” (Bierwisch 1986: 777). Quite apart from the fact that interpret-ing without conceptualizing is a contradiction in itself, school would then have to be interpreted first and foremost in terms of the component purpose in sentences (14) and (15) (as, of course, in all the others as well). However, (14) is only concerned with the fundamental feature public institutions have of temporarily closing, which is far removed from the postulated purpose component of an institution, whereas in (15) school has, among other things, the semantic properties of a name and relates indi-rectly to an area of life. Specific cognitive domains are evoked, and individual facets of knowledge (default values) from these domains are profiled while others are shifted into the background, resulting in a specific “semantic perspective” (Croft and Cruise 2004: 58ff.).89 With this in mind, in (14) and (15) only some aspects of the cognitive domain school shift into the foreground. However, these have nothing to do with the semantic constant purpose, but certainly a great deal to do with processes of cognitive meaning construction. The semantic correlation between school and spring vacation does not arise from “an initial structure” of the expression school, from “anchor points by means of which L [linguistic knowledge] gets tied to C [the conceptual system]” (Bierwisch 1986: 781). The conceptual profiling and perspectivation, i.e. the contextual specification of the denotatum, proceeds much more flexibly; in particular it cannot be reduced to a meaning potential limited by linguistic knowledge that takes no account of the social and cultural horizon of experience of language users (cf. Clark 1997).

Thirdly: as an example we should recall to mind at this point what area of meaning-relevant knowledge the first representational level actually comprises. To do so, let us take an example sentence in which the postulated constant purpose is also actively involved in the process of meaning construction. In (12) (The influence of school ends at the children’s doorstep), this is clearly the case. Depending on whether the conceptual component institution or process occupies the first argument position, there are two alternative readings of the noun school, even when the second argument position in both cases is occupied by the component teaching_and_learning_processes. In the first case, the meaning of sentence (12) receives a certain sociological touch; it relates to the limited sphere of influence institutions of schooling have, insofar as they do not (or hardly) penetrate to the private area of life. In the second case, this sociological dimension does remain to some extent, but concrete teaching and learning

88. On this theme, see the analogous phenomenon of the meaning constitution of compounds through cognitive routines that are based solely on contextual data (Clark 1983).

89. Langacker also calls these cognitive processes “focal adjustment” (1987: 117), which he in turn subdivides into three sub-processes: selection, abstraction, and perspectivation.


processes or school teaching shift more into the foreground, with their extramural ef-fectiveness being questioned. Thus, the first reading generalizes more than the second.

What role does the semantic representation sem play here? It merely states that an x fulfils a purpose y. Possible fillers (institution, process, teaching_and_learn-ing_processes) are of a conceptual nature, are motivated by contextual data, and are inferred by calling on experiential knowledge. Therefore, componentially there is little to deduce what is decisive in this example sentence: doorstep meronymically represents the parental home and metonymically what the child experiences and does in this home.90 Speaking of a doorstep only makes sense if it is clear that the parental home is one centre of a child’s life and the school building another. This deals with complex concepts, which relate to each other as follows. In our culture, school attend-ance is compulsory. It is common that children live with their parents until they have completed their secondary schooling. During this period, daily life is structured and determined above all by the ritual of going to school each morning and returning home in the afternoon. The child therefore commutes as it were between the school building and the parental home. The statement that the influence of school ends at the children’s doorstep evokes precisely this knowledge complex.

Fillmore (1977a: 75) refers therefore to “prototypical scenes”, alluding to our pre-dominantly visually shaped experiential horizon.91 The evoked cognitive scenes are in themselves highly sophisticated and decisive for deducing the meaning of (12). They contain information about the features of two central areas of a child’s life. They pro-vide information about the fact that in both areas of life, school and home, different educational principles reign that may even be contradictory at times. In (12), ‘parental home’ is conceptualized as a closed space, ‘school’ – by dint of its genetivus subjecti-vus – as a spatially situated, quasi-personified entity, and ‘influence’ as a path, or more precisely as an entity with spatial-temporal extension that can “end” at a particular point.92 ‘Influence’ therefore demonstrates physical properties that set limits to its ef-fectiveness: It cannot penetrate through the house wall into the parental home. Once the child that moves between ‘school’ and ‘parental home’ enters the latter (doorstep), it is no longer in the sphere of the school’s influence. This does not end already at the gate of the school, but lasts until the doorstep, because ‘influence’ is conceived of as a

90. Taking into account that doorstep also metaphorically represents an entrance into a sphere of influence, we are even dealing with a “metaphtonymy” (Goossens 1995), i.e. a “metonymy-based metaphor”, in the sense of Radden (2000: 93).

91. For meaning construction in (12) this also means that: “The all important role of the notion of prototypical scenes in this process consists in the fact that much of this linking and filling-in activity depends, not on information that gets explicitly coded in the linguistic signal, but on what the interpreter knows about the larger scenes that is material activates or creates. Such knowledge depends on experiences and memories that the interpreter associates with the scenes that the text has introduced into his consciousness.” (Fillmore 1977a: 75)

92. According to Lakoff and Johnson (1980), this is a so-called conceptual metaphor, similar to the case of the container metaphor mentioned above.


unidirectional extension. This masks the fact that the way to school can be a third area of life in which all sorts of things can, of course, happen.93

The necessary integration of such complex background knowledge in order to understand the sentence is based on our experiences, which we perspectivize by means of the described “prototypical scenes”. The sentence is about the possible influence school can have on private areas of life; it is about the particular scope and limitations of that influence; and it is not least about knowledge that distinguish private and pub-lic spheres. To understand (12), it is therefore necessary to have a cultural model in which relevant information (default values) is bundled. It is this model that allows this information to be cognitively accessible in the process of understanding. However, this wealth of meaning-relevant data lies beyond the reach of two-level semantic models.94 Apart from the fact that necessary processes of conceptual integration are by no means compositionally motivated, from the perspective of language-processing theory the input of semantic representation exhausts itself in marginalities, with prespecification thus preventing adequate explanations. This is because:

[T]he processes that accomplish this integration make use of any information they can find, and do not necessarily pass through a phase in which they deliberately restrict themselves to information of any particular kind. (Fillmore 1984: 133)

For collocations like old school in (16), meaning interpretation is, for example, so routine that no phase of lexical decomposition for school needs to be passed through. In individual cases, it is perhaps possible to envision ex post what meaning aspects of a lexeme (here: school) have entered into a phraseologism. In any case, isolated word meaning does not play a role in the relevant process of understanding because the collocation old school has been lexicalized as a whole (cf. also Cuyckens, Dirven, and Taylor 2003: 10–17).

Fourthly: In the previous point I neglected an aspect that crucially motivates al-ternative conceptual interpretations in German of the definite noun phrase der Schule (‘(the) school’) in (12).95 Nouns are always already topically grounded in particular usage contexts by co- and/or contextual embedding. Thus, the interpretation of the collective singular school varies if (12) appears, for example, in an educational science textbook on socialization research. Consider now that the definite article is included yielding ‘The influence of the school ends at the children’s doorstep.‘ If this sentence came from a teacher who is expressing her concern about one of her students, the

93. It is equally irrelevant that school without doubt also exerts an influence beyond the door-step. However, it is clear that another “influence” is intended than for instance the influence homework exerts on the child.

94. Börkel (1995) comes to similar results in his analysis of other examples such as opera or newspaper: “If there should be found one abstract SF [semantic form] for all these readings this SF would be a very abstract one with hardly any useful information.” (Börkel 1995: 71)

95. Cf. footnote 25.


definite article would immediately ground the noun school in the respective commu-nicative situation without requiring a selection of reading type.96 Different readings for school therefore only arise under the counterfactual assumption of a zero context. This must first be empirically validated as a meaningful methodological construct. In any case, in the example just discussed, for instance, there is no need to assign any meaning-relevant function to a zero context with respect to the nominal unit school. And whether this is any different in other cases remains questionable. Epstein observes,

Nominals are not used for evoking objective, pre-existing, homogenous entities. Rather, speakers must construct discourse referents before they can call the hearer’s attention to those referents. … Speakers employ various grounding predications – including definite articles – to facilitate the construction of discourse and, at the same time, to induce hearers to accept each referent into the discourse under the desired guise. (Epstein 2002: 41)

There are, therefore, good reasons to suppose that grounding generally has a primary function in processes of understanding. Relating this to our case, this means that in-stead of underlaying the expression school with an objective and homogenous “level” (distinguished, for instance, by the component purpose), construal operations play a decisive role at the outset already in that, for example, available grounding predications like definite articles are the departure point of inferences.

Next to definite articles, which establish a contextually unique reference, ground-ing predications include grammatical morphemes, which ground utterances in the speech situation with respect to persons and time, and names, which singularize ref-erents. (7) and (13) are examples of the latter. The names “Lena” and “St Gregory’s College” restrict the interpretational variance of school without a detour via semantic representations being necessary. Here are the example sentences once again:

(7) Lena Kurz is the new principle of the school. (13) The school Studio ImPuls is a member of the VdG.

What speaks for such a construction of the utterance meaning and against a level model is also the fact that – contrary to (7) and (13) – the interpretational variance is often not restricted to one reading.97 As we have seen, this applies, for example, to the sentences (10) and (11). I will now take a closer look at the latter.

96. This phenomenon is discussed in Cognitive Linguistics as “grounding”. Brisard (2002: xi) says on this: “Grounding is proposed as a technical term in Cognitive Grammar to characterize grammatical predications that indicate the relationship of a designated entity to the ground or situation of speech, including the speech event itself, its participants, and their respective spheres of knowledge.”

97. In (7) and (13), this is the institutional reading, which arises out of the conceptual integra-tion of other sentence elements; only an institution can belong to an association, and only an institution can have a principal running it (a maintenance supervisor does not run a school).


(11) Parents can also become involved in shaping the school.

Whether the institution or the building is meant in (11) is decided by the default values that are activated or on the basis of further contextual data. If adjacent to (11), sentences (11a) and (11b) would contribute to disambiguation.

(11a) Many join the parents’ association. (11b) Some parents, for example, sign up for voluntary gardening work.

Proponents of two-level semantics would now argue that subsequent conceptual specifications of precisely this kind are motivated by semantic representations in that different conceptual components (in this case, institution, building) replace the variables. However, contextual data usually generate much richer conceptual specifica-tions. They can even effect radical meaning shifts, as in alternative following sentences like (11c) and (11d).

(11c) They could perhaps teach their children to behave properly. (11d) They could perhaps stop always wanting to be involved.

The focus in understanding shifts to meaning aspects of school that are not in the foreground in (11). What is shaped in the school when parents teach their children to behave respectfully? To what extent are parents involved in shaping the school if they stop always wanting to be involved in shaping it? Many background assumptions must be activated to grasp the meaning of school in these two contexts. However, what these are in particular is determined by contextual data and not by zero-context knowledge. It is not possible to resolve the logical paradox in (11d) (‘being involved in shaping by not being involved’) by following the compositionality principle. On a conceptual level, by contrast, this contradiction does not emerge at all. Coulson (2001) extensively addresses this problem complex from a frame-semantic perspective. Using numerous examples, she shows that such processes cannot be grasped compositionally, but rather have an irreducible constructive character.98

The insufficiency of semantic descriptions within a level model becomes obvious once again in connection with so-called grounding predications. Grounding predica-tions allow a text recipient “to (re)construct the meaning of a communicative utterance by referring to all the knowledge he possesses that is relevant for understanding this utterance” (Busse 1997a: 23), and that means in this case, on the one hand, grounding

98. Coulson (2001: 70) states: “Compositional accounts of natural language processing do not provide a mechanism for explaining how people are able to create novel meanings for words, phrases, and idioms. Nor do they provide a natural way of accounting for how frame-shifting can change the evaluative and socio-cultural significance of actions.” Meaning shifts such as those motivated by (11d) that affect the reinterpretation of the nominal unit school in (11) Coulson designates as “frame-shifting”. These are particularly striking processes of conceptual integration. There is reason to believe that such processes are a core phenomenon in the theory of understand-ing language; (11a), (11b), and (11c) also motivate retrospective meaning shifts in (11).


the communicated utterance content in the “text world” to establish semantic coher-ence, as well as grounding this content (for instance in the case of deictic expressions) in the perceptually accessible world of things.

The fifth and final point takes up a fundamental question that in recent years has led to a far-reaching revision of the two-level model. If, on the first level, the semantic interpretation of linguistic expressions is predetermined by semantic constants, then what cognitive operations do they determine? Since no conceptual processes may be involved on the level of semantic representation, a vacuum remains, which also in-cludes the unsolved problem of how semantic representations are determined in their turn. To avoid falling back on an imagination theory of word meaning (for instance in the sense of Plato’s ideai) (cf. Busse 1991a: 25–30), it suggests itself to assume that semantic constants are abstractions produced from conceptual units. However, these assumptions would topple one of the two pillars of two-level semantics. That is to say, a strict division between conceptual structure and semantic knowledge would then be possible only with respect to the syntactic combinatory potential of linguistic ex-pressions. Semantic constants, that is, the reference potential, would by contrast then be subject to the influence of conceptual processes. Different advocates of modular cognitive theories like Ray Jackendoff (1983, 1997, 2002), Monika Schwarz (1992a, 2000), and Johannes Dölling (2001a, 2001b, 2005) have drawn this conclusion and thus conceded more autonomy to semantics compared to syntax.

Finally, using some examples, let us examine what led to this revision. To what extent is the component purpose a semantic constant? Bierwisch posits that there is a limited, but universal inventory of semantic primitives that overall constitutes the in-variable part of the semantic knowledge of a language user. More specifically, Bierwisch (1986: 780–781) differentiates between “created primes” on the one hand and “fixed primes” on the other. Whereas the former have no a priori status, the latter are “part of the universal repertoire available to the language learner as his or her initial equipment on the basis of which linguistic knowledge can be acquired” (Bierwisch 1986: 780). Referring to our example, this means that the component purpose has the status of a “fixed prime”, whereas a component like teaching_and_learning_processes is a “created prime” (Bierwisch 1983a: 86).

This distinction is decisively motivated by two assumptions: firstly, by the idea of structural minimalism, which states that semantic primitives do not have a complex internal structure, thus not being able to be decomposed into further components, and differ in this characteristic from conceptual units (cf. Bierwisch 1971: 410–413); and secondly by the invariance premise, as a result of which Bierwisch assumes that these “fixed primes”, the smallest semantic units, precede conceptual, perceptual, motor, and other experientially based cognitive structures and are therefore – in contrast to them – subject to very little change (Bierwisch and Schreuder 1992: 28).99 Together these two assumptions justify setting content restrictions, besides the purely formal variables of the argument structure, which determine the combinational properties of a linguistic

99. Cf. also Meyer (1994b: 34).


expression. This process is a case of specification by the grammatical system. The prin-ciple of modularity therefore applies here as well: as far as components have the status of semantic primitives, they do not fall under the sphere of influence of conceptual operations and belong to the linguistic system; by the same token, as far as semantic components are structural entities, they form part of the conceptual system.100

Once again: To what extent is the component purpose a semantic constant? As such, it should not be able to decompose into further components. On the other hand – because it belongs to the grammatical system – it should be significantly more rigid than conceptual units. Even if one conceded the latter,101 it would still remain more than dubious whether speaking of purpose as a semantic primitive can be justified. What speaks against this is that for the different readings of the expression school, there is no uniform representation of what a purpose is (cf. Blutner 1995: 45; Börkel 1995: 70). Whereas the purpose of a building, as in Example (6) for instance, consists in it functioning as a location for teaching and learning processes, the purpose of school in the sense of “teaching lessons”, as in (10), lies in an abstract educational goal. What a purpose is in a particular case can therefore only be determined on the conceptual level: “This means that the semantic representation in the lexical entry cannot be a complete representation of purpose” (translated from Meyer 1994b: 37).

Now one could argue that the question concerning the semantic component pur-pose should be distinguished from the question concerning the specific purpose of a particular entity, for the reason that every specific purpose is already a conceptual variant of the semantic constant purpose. However, a number of points speak for the semantic constant purpose being a unique conceptual unit. A purpose is a particular type of causality with the feature that it is intentionally brought about (which is why no component purpose can be identified when decomposing nouns such as storm or deluge).102 In addition to Purpose, Talmy identifies fourteen semantic causation types with differing degrees of abstraction and conceptual (Talmy 2000: 471–548). Using the concept of ‘purpose’ as expressed in (17) and as applies in understanding (18) and (19), the following at least touches on the sophistication of the underlying conceptual structure.

100. Bierwisch thus follows up in a more sophisticated form on a structuralist position that he already advocated at the end of the 1960s and beginning of the 1970s (Bierwisch 1969, 1971).

101. In fact it suggests itself to distinguish between different degrees of rigidity. However, it is not necessary to separate the linguistic and conceptual systems for this. A default logic, such as I propose for frame semantics, fulfils similar aims. Degrees of rigidity correlate to degrees of cognitive salience or prototypicality; cf. Section 4.5.

102. Componentialists would argue in this manner: A storm does not have the purpose of de-stroying houses, etc., but “only” causes damage; a school building, by contrast, has the purpose of providing a location for teaching and learning processes because this building is built specifi-cally to do this (or was converted into such a building). Nevertheless, this analysis depends upon the worldview underlying it. Within a theocratic worldview, a storm or a deluge would be the punishment of God. They both would most certainly then serve a purpose.


(17) School buildings have the purpose of functioning as (institutionally anchored) locations for teaching and learning processes.

(18) and (19) are examples of the building reading:

(18) The school lies quite near the municipal woods. (19) The school has 48 classes, a gymnasium, and numerous facilities.

What genuinely conceptual interpretation with respect to purpose is there in (18) and (19)?103

i. The concept ‘purpose’ (as formulated in (17) and exemplified in (18) and (19)) describes a relational connection. Overall it presents a threefold structural unit, in that two entities are connected with each other by means of a relationship. Teaching and learning processes (= the result) take place at an institutionally an-chored location because (= the relationship) school buildings exist.104

ii. The precondition is itself of a conceptual nature. Certain preconceptions (default values) are required in order to understand what purpose they fill.

iii. Precondition and result stand in a conceptually organized foreground/background relationship. In (18) the concept of ‘teaching and learning processes’ forms the back-ground to understanding, whereas the ‘school building’ shifts into the foreground (thus forming the “figure” or the “profile”). In (19) exactly the converse is true.

iv. The relationship between the precondition and the result must remain intact dur-ing the process of understanding the word school even if one of the two conceptual entities shifts into the foreground. If it is interrupted, then a semantic reinterpre-tation occurs and the purpose context is revised. This occurs, for instance, if the sentence following on negates the respective background to understanding, thus if (18), for instance, is followed by the sentence Now it only serves as a barracks or (19) by It is about to be torn down.

v. The result (institutionalized teaching and learning processes) can occur if the pre-condition (existence of a school building) is fulfilled. If it is not fulfilled, then no institutionalized teaching and learning processes occur. This clearly shows what type of relationship it is. The precondition is not a sufficient but only a necessary condi-tion. (In other cultures, school classes take place even without a school building.)

vi. Precondition and result stand in a specific temporal relationship. In Examples (18) and (19), we have the (rather rare) case that precisely this temporal relationship does not play a role – at least in a standard reading. The school does not first have to be built so that classes can take place there; rather the presupposition of its exist-ence feeds into the process of understanding school. (Another example for clarity: So that a medication can fulfil its purpose of healing, it must take effect; therefore a certain period of time must ensue.)

103. Cf. Talmy (2000: 494–495).

104. While learning processes in turn are the result of teaching processes.


purpose is therefore anything but a primitive component. It is conceptually even more complex than the putative semantic constant caus (“causality”) that Bierwisch constantly uses as his prime example (cf. for instance Bierwisch 1979: 75; 1983a: 90; 1986: 769). The six conceptual properties explained above apply analogously to this. Similar analyses could be carried out with any of the other components.105

To illustrate the division between conceptual and semantic content, Bierwisch and Schreuder go so far to claim at one point that the distinction equally applies to proper names.

[L]inguistically, we know that John is a proper name by means of which we may refer to male persons identified by this name. This is the information represented in the corresponding lexical entry in the format discussed above. Conceptually, we have a varying amount of more specific knowledge about each John we happen to know. This knowledge is mentally organized in systematic ways, although it has nothing to do with the knowledge of English. (Bierwisch and Schreuder 1992: 31)

However, it is obvious that whether a proper name refers to a male or a female crucially depends on what a language user knows about the world. Thus, a native speaker of German or English travelling in Italy will discover that Andrea refers to a male there, and she learns this – as is the case with any other proper name – not because of any linguistic-grammatical properties of the word, but by drawing on referential knowl-edge (cf. Busse 1997a: 19–20). She learns which linguistic or extralinguistic segment of knowledge an expression can refer to, and does so without having to distinguish beforehand between linguistic and encyclopaedic information.106 There is no sense in making such a distinction, because this dichotomy is subverted in every act of estab-lishing reference. Against this background, it is not at all surprising that Bierwisch at no point more closely defines the extraconceptual status of semantic constants, but instead assumes that they probably involve “the same substance”.107 That means, in plain language:

105. Cf. Börkel (1995), Meyer (1994b), and from a cognitive perspective also Taylor (1994), (1996). Because conceptual properties, in the sense of (i)–(vi), are intrinsic to linguistic mean-ings, Fillmore (1971: 384) also speaks of presuppositions: “Many of the features of spatial orienta-tion treated by Bierwisch will take their place, in other words, in the presuppositional compo-nents of the semantic descriptions of words usable as predicates.” “Features of spatial orientation” are purported components like vert (verticality) and max (maximal dimension).

106. Directly contradicting Bierwisch, Fillmore remarks: “The question a lexicographer must face is whether these matters have to do with what one knows, as a speaker of a language, about the words in that language, or what one knows, as a member of a culture, about the objects, beliefs and practices of that culture.” And he answers the question as follows: “These are serious questions, but we can of course avoid facing them by making … the decision not to insist on a strict separation between a dictionary and an encyclopedia” (Fillmore 1971: 383).

107. Cf. Schwarz (2000: 32). Schwarz refers here to Bierwisch (1997) and verbal communication with Bierwisch.


If constants may be grounded in conceptual knowledge, and if they are subject to interpretation in terms of conceptual knowledge, why not simply say that they are conceptual entities? (Taylor 1994: 13)

This consequence is inevitable in view of the preceding considerations, even if gen-erativists repeatedly claim that it is essential to assume semantic primitives in order to avoid the learnability paradox. However, in her study on semantic acquisition, Brigitte Löbach (2000) showed that within a holistic approach, the acquisition of conceptual structures can indeed be made plausible without having to base this on a strong de-composition hypothesis.108

At this point I break off the analysis, although from a frame-semantic perspective there is certainly much more to be said on the individual example sentences. At this point I also dispense with illustrating how frame semantics deals with the problem of polysemy in these cases109 as this requires some preliminary explanation. Above all, the five last points outlined here clearly show how markedly the premise of a modular cognitive theory interferes with the object of study of frame semantics.

2.2.4 Three-level semantics (M. Schwarz)

The extensive analysis in the previous section revealed that not only theoretical reasons speak for not addressing frames within a two-level semantic model. The main reason is that the empirical explanatory power of frames is significantly restricted by the meth-odological premises of two-level semantics. Innumerable facets of knowledge relevant for understanding a linguistic expression lie outside the reach of this approach. What is more, a two-level model even prevents such aspects from being registered.

Thus, if semantics is restricted to what is algorithmically computable from lin-guistic units, the resulting semantic representations will be so limited and impov-erished relative to how expressions are actually understood that we would hardly recognize them as reasonable approximations to their meanings. (Langacker 1988a: 17)

In this sense, a reductionist concept of meaning underlies two-level semantics. The meaning potentials of linguistic expressions are reduced to the interaction between a predetermined construction and reference potential that specifies semantic represen-tations ex hypothesi.

Monika Schwarz takes this criticism into account to some extent. Similar to Dölling (2001b, 2005), she advocates a model

108. Cf. in particular Löbach (2000: 174–182). Löbach also argues here against Jackendoff ’s as-sumption of semantic primitives (I turn to Jackendoff ’s conception of semantics in Section 2.3.2).

109. Cf. Martin (1997). For the modelling of meaning variants of school within a network model, cf. Pörings and Schmitz (2003: 34–38).


…that does not postulate a strict division between semantic and conceptual repre-sentational levels, but rather takes into account representational interdependence processual interaction of these two knowledge systems. (Translated from Schwarz 2000: 37)

Thus the close connection between the semantic and the conceptual systems, between linguistic knowledge and conceptual world knowledge, is no longer denied. For similar reasons, therefore, both Schwarz and Dölling abandon the two-level semantic postulate that particular semantic components, that is to say constants or primitives, are able to be identified with elements in the purely linguistic system.110 However, they both continue to adhere to the autonomy of syntax. Accordingly, syntactic principles de-termine the selection of possible conceptual elements (i.e. the combinatory potential of expressions), before any understanding of meaning occurs at all. In the following, I use examples in an attempt to show that this model unnecessarily restricts frames in their explanatory potential. More precisely, I will address Schwarz’s position, leaving aside the nonetheless similar argumentation used by Dölling.

Schwarz argues for an integrative model that appears to circumvent the holistic/modularist dichotomy. She herself speaks of a three-level semantic model (Schwarz 1992a: 91–106; 2000: 33ff.). By using this designation, however, she already indicates that central generative grammar premises remain unchanged. What is new in compari-son to the two-level model is her conception that the semantic system is not independ-ent but rather a semi-module that interacts with other cognitive systems.

The amodal conceptual level is the base level and contains the mental content for the semantics of a language. Lexical meanings are the second stage: conceptual contents that are coupled to phonological forms and syntactic grids. The third level is the level of actual meanings. (Translated from Schwarz 1992a: 101)

Semantic and conceptual elements are no longer distinguished with respect to their sta-tus. Informational units from both the first and second level are stored in the long-term memory. In contrast to two-level semantics, Schwarz stresses that semantic content structures, which Wiese terms “reference potential”, do not belong to the linguistic sys-tem but to the conceptual system (Schwarz 1992a: 89). Thus, it is questionable whether word meanings should be described as decomposable quantities of necessary or suffi-cient features (sem). However, herein lies the key argument for distinguishing between semantic and encyclopaedic knowledge that supporters of two-level semantics use. The following observations give Schwarz grounds for seeing this distinction as problematic:

110. Dölling (2001b: 11) draws attention to these central deficits of two-level semantics in the following questions: “By what means are the constants of a semantic form that have access to an interpretation in conceptual units determined? Are these constants units of an essentially different nature, or are they rather merely certain linguistically related abstractions about par-ticular conceptual units? To what extent, then, does a strict separation between conceptual structure and semantic form, between conceptual and semantic knowledge, make any sense at all?” (Translated from 2001b: 11)


– To explain possible semantic interpretations of linguistic expressions, it is not plausible to posit that semantic representations are carriers of diverse information on the one hand while being inflexible activation processes on the other.

– The hypothesis of the decomposability of linguistic expressions is accompanied by the complexity hypothesis, according to which the processing of more sophis-ticated meanings is cognitively more complex and therefore must require more time. To date, no empirical evidence has been found for this. Rather, pertinent studies indicate that there is no significant difference in processing time.

– Furthermore, there is a lack of theoretical and empirical evidence for making a distinction between a level of semantic representation and one of conceptual rep-resentation. In particular, more recent neurophysiological findings indicate that there are no biological grounds for upholding the principle of modularity.111

– Language understanding is ultimately a cognitive-constructive act that takes re-course to inferential strategies to establish coherence. Activating lexico-semantic knowledge only plays a subordinate role in this process; the central function is in fact held by conceptual world knowledge.

Schwarz draws conclusions from this that amount to her taking an intermediate posi-tion combining holistic and modularistic assumptions:

However, I hold the view that rejecting the complexity hypothesis does not neces-sarily require rejecting the decomposition hypothesis. Structural and processual assumptions could, or rather must, be distinguished from each other. The structural organization of the long-term memory does determine (in a way yet to be re-searched) the sequence of cognitive processes, but other factors such as processing depth and cognitive complexity play a significant role in the individual processing situations. Decomposition in semantic features is thus not an obligatory but rather an optional operation in processing meaning that is undertaken according to situ-ative and cognitive requirements… (Translated from Schwarz 1992a: 89)

Her intermediate position therefore consists in Schwarz on the one hand denying that meanings are composed from a set quantity of nuclear components, thus contradict-ing the postulate of strict compositionality that characterizes two-level semantics.112 However, what still remains unclear is the extent to which decomposition is an “op-tional operation” and what in particular the cognitive reality of this operation should consist of. If it were purely optional, it would be cognitively irrelevant, because then

111. For this reason, in the previously mentioned special issue of Linguistic Review (22/2005) on the current status of cognitive linguistics, a number of cognitive linguists recently distanced themselves from the modularity hypothesis, including the weaker version as proposed by Schwarz.

112. The helpful distinction between “strict” and “loose compositionality” goes back to Taylor (2002: 97–116). Dölling diverges from Schwarz on this point; he continues with the assumption of strict compositionality (Dölling 2001b).


solely other cognitive principles would determine the process of understanding. On the other hand, it is clear that Schwarz does not doubt the psychological reality of contex-tually invariant features per se. She only questions whether semantic decomposition constitutes an independent, processually separate level of language understanding. Her uncertainty in determining what status such a “contextually invariant, underspecified core meaning” (Schwarz 2000: 38) then has becomes clear in her attempt to detach the core meanings of linguistic signs from the so-called “conceptual scope”. “Conceptual scope” refers to the encyclopaedic cognitive domain (or frame) in which a core mean-ing is embedded. According to Schwarz (2000: 38), the core meaning of house (German: Haus) is residential_building and of to shoot someone dead (‘erschießen’) is agent x kills patient y with instrument z (firearm z). Are these “fixed primes” or “cre-ated primes”, using Bierwisch’s terms, or conceptual units? Schwarz does not provide a clear answer to this. However, since she rejects Jackendoff ’s view that semantic and conceptual structures are on an equal level, and since by speaking of contextually invariant core meanings she conversely adopts Bierwisch’s invariance premise, core meanings should probably be understood in the sense of “fixed primes”.

However, this means that Schwarz’s position in fact comes down to being a level-semantics model that needs to clarify the correlation between semantic constants and conceptual knowledge elements. A model such as this would also need to clarify to what extent “cognitive processes” (Schwarz 1992a: 89) directly access the level of se-mantic representation. What is certain is that due to a lack of empirical evidence, Schwarz rejects the complexity hypothesis; depending on context, semantic compo-nents can also be selectively activated. At the same time, she does not abandon the principle of decomposition and thus inherits all the difficulties of two-level semantics that were outlined in the previous two sections.

In addition, Schwarz upholds the generative grammar hypothesis that semantic units are coupled with syntactic grids and thus simulate an argument structure that has nothing to do with the conceptual system. In doing so she takes up Chomsky’s prem-ise that assigns syntax priority over semantics. Although the theta grid of a linguistic expression does not provide an inventory of features that affects the understanding of meaning in terms of content, it does function as an interface between syntax and con-ceptual structure, and in this function it only opens quite specific slots, which are then able to be conceptually specified. This affects the formal understanding of semantic structures, not the content. In fact, the theta grid is a fixed matrix for possible syntactic realizations, which thus results in a third level of consideration – in addition to the distinction between semantic and conceptual elements I have just outlined – that this time concerns the set of conceptual elements. Either a conceptual element belongs to the “conceptual scope” (Schwarz 2000: 37) of a linguistic expression that prescribes its theta grid, or it is removed from this scope and is thus not relevant to understand-ing. The previous section showed that this is accompanied by a severely reductionist description of semantic knowledge.

Although Schwarz tries to avoid the meaning-reductionist dilemma of two-level semantics by positing that core meanings are directly linked to conceptual units (like


frames) via “procedural routes”(2000: 37–38), in reality in doing so she at best di-minishes the gap between semantic and conceptual representations. The meaning-reductionist core remains for the simple reason that conceptual cognitive domains can only fulfil a meaning-relevant function if they refer to an argument position in the expression in question. In fact according to this, argument positions function as it were as departure points of the “procedural routes”; they are anchor points of lexi-cal knowledge that show the way to potentially connectable conceptual knowledge. According to Schwarz, the contextually invariant lexicon entry for to shoot someone dead contains three variables: “x” for the ‘agent’, “y” for the ‘patient’ and “z” for the ‘instrument’ (Schwarz 2000: 38). Cognitive domains can now be variably linked into (cf. Figure 3).113

shoot_dead (x, y, z) X (agent) kills Y (patient) with Z(instrument) is firearm Z

motive attendant circumstancesconsequences

Figure 3. Part of the cognitive domain of SHOOT_DEAD according to Schwarz (2000: 38)

The large, lower box in Figure 3 contains the semantic information of the lexicon entry, whereas the three small boxes represent three possible conceptual domains. Both semantic and conceptual information are connected to each other by activated procedural routes (whose direction is indicated by the arrows), they are substantial but equal, and they are stored in the memory in close connection with each other.

However, neither the sketch nor its explanation clarifies how such a cognitive connection should exactly occur. If the substance is the same, what then justifies sepa-rating semantic information from conceptual information? And presuming that both can be distinguished from each other, what would motivate a “procedural route”, i.e. the activation of a conceptual domain? These questions remain unanswered. Figure 3, however, yields two possible answers.

Option one: The arrows are not assigned to individual argument positions but to a lexicon entry as a whole. In this case, the meaning-reductionist problem does not present itself to the same degree as outlined above for a two-level semantic model. Nevertheless, every domain would still have to be somehow derivable from this con-text-invariant entry. But how? Speaking very generally of “activity states” in which a domain is replaced by the lexicon entry does very little to help. Considered in terms of a language-processing theory, adhering to underspecified core meanings means that

113. I have attempted to adopt Figure 3 as exactly as possible from Schwarz (2000: 38). The graphical illustration presents some central problems – not only for our purposes – which Schwarz does not clarify in her explanation of the diagram.


these serve as fixed inference bases. But as information postulated on a semantic basis, in what connection then do these fixed inference bases stand with conceptual domains? Under the condition that both have the same substance, it would be crucial to explain the intrinsic connection, as the danger is otherwise great that they both represent two ontologically different cognitive systems – in the sense of two-level semantics. This first option, the one Schwarz seems to have in her sights, raises many important questions that on the one hand demonstrate the independence of Schwarz’s model but on the other hand remain unanswered.

Table 1. Possible correlation of argument positions with cognitive domains

Argument position(s) Cognitive domain

Agent → ConsequencesPatient → Attendant circumstancesShoot_dead (x, y, z) → Motives

Option two: Every conceptual domain is directly attached to an argument position. This would partially accommodate the problem in the first option. In the example given, the domains motive, consequences, and attendant circumstances would have to be retraceable via procedural routes to one of the argument positions x, y, or z. This corresponds to the conception of two-level semantics as discussed on the basis of Bierwisch. Each arrow in Figure 3 would be firmly correlated with one or more argument positions. Clearly it makes little sense to interpret the arrows in Figure 3 as if they had the fixed correlations shown in Table 1.

In fact, it would be highly questionable to correlate the cognitive domain motive with the entire lexicon entry; it would then seem that this undoubtedly important domain could not be correlated with the variable x (‘AGENT’), because this variable is already attached to the domain consequences. At first sight, however, it would be plausible to attach the domain motive to the argument position ‘AGENT’ and to postulate two possible attachment capabilities for the domain consequences, that is the two argument positions ‘AGENT’ (for instance, psychological and/or legal con-sequences) and ‘PATIENT’ (for instance, death and/or consequences for those left behind, etc.). But what argument position would the domain attendant circum-stances then relate to? This knowledge clearly relates to the context of an action (in this case: the fatal shooting), i.e. the situative context in which an action occurs. It is obvious that such facets of knowledge do not have to exhibit a direct relationship to the putatively invariable lexicon entry. Thus, adhering to the (perhaps abductive) inferentiality of conceptual knowledge on the basis of grammatically determined core meanings just to get a grip on the problem of polysemy causes the same meaning reductionism to emerge that is a feature of two-level semantics.

Apart from these two precarious interpretations of Schwarz’s model of meaning, some overarching questions arise. How are the argument positions determined, and how is it guaranteed that these all have the same status? In contrast to the variables


x (= ‘AGENT’) and y (= ‘PATIENT’), z (= ‘INSTRUMENT’) is already specified by firearm z in the lexicon entry for to shoot someone dead. In comparison to the other argument positions, this is much more concrete, and its (potential) conceptual scope would be as a result restricted to relatively few informational units, for instance the details of precisely what type of firearm it was, what material it was made of, etc. The determination of the argument positions appears to be guided by a type of interpretive heuristic rather than by a comprehensible system. Otherwise the core meaning would have to allow less marginal conceptual information to be deduced that obviously is not attached to any of the three argument positions, for instance details about the location of the event (e.g. the chancellery in Berlin vs. a shopping street in Bagdad), details about the action frame in terms of Goffman 1974 (e.g. fictional shooting in a children’s game or a stage play vs. non-fictional shooting in war, a robbery, etc.), details about the larger action frame or the motivational framing (e.g. world war vs. assassination of the Pope), etc. In Schwarz’s model, these would consequently have to count as irrelevant to understanding.114

However, an additional point seems to me to be even more crucial. In the ab-sence of any reference to superordinate cognitive domains, the putatively abstract and grammatically determined lexicon entry for to shoot someone dead remains empty. (i) Shooting someone is an action, and only as such does it require an agent and a patient. However, how can an action be distinguished from mere behaviour without drawing on conceptual world knowledge? (ii) The meaning of to shoot someone dead implies deontic knowledge. Shooting someone dead can be an intentionally focused act, in which case it is murder, otherwise it is manslaughter or self-defence. However, what is intentionality in the one case without relation to the world? And what is mur-der or manslaughter without having a concept of “law”? (iii) Shooting someone dead means that agent x is causally responsible for the death of a patient y. Yet, the previ-ous section showed that a multitude of specific conceptual properties feature causal relationships. To name a few examples: How can the temporal connection between shooting and being shot be grasped without empirical knowledge of this process? Why doesn’t the process of shooting (in contrast to strangling) last several minutes? (iv) It

114. It would be easy to give examples of contextual frames that show the central significance of such data that is relevant to understanding. See the substitution of the following information for instance: Today a tramp/the Chancellor/ the Pope was shot dead by an unknown assailant in the stage

play X/in the chancellery/ in Bagdad/ in the civil war.Follow-on sentences could now refer to an array of knowledge facets concerning shoot dead. – In the stage play X the bullet didn’t reach him, but he fell down dead anyway. – In the chancellery, as in all of Germany, this assassination caused great distress. – In Bagdad such public executions have apparently become commonplace.And so on. As it is clearly evident that such facets of knowledge are relevant to understanding, I dispense at this stage with any further analysis of examples.


is only possible to shoot dead living creatures and then usually only particular living creatures. A horse, a dog, or a man can be shot, but not an alarm clock or a fly. How can this be distinguished without possessing knowledge of the domain living_creature?

These examples illustrate the “intrinsicness of conceptualisation to lexical mean-ing” (Taylor 1994: 15). The three-level model as suggested by Schwarz also ignores large sections of meaning-relevant knowledge because Schwarz understands conceptualiza-tion as a cognitively subordinated operation.

Adhering to the distinction between a semantic and a conceptual representational level has very particular consequences for Schwarz’s theory of indirect anaphors. For instance, to explain the process of understanding indirect anaphors, she has to posit different activation types (Schwarz 2000: 98–117). One of these is designated “ground-ing on verb-semantic roles”. To illustrate this she gives the following example (trans-lated from 2000: 99):115

(20) I was about to unlock the door when Moretti [sic] leapt out of the bushes. I dropped my key in fright.

She argues that the theta grid for the verb unlock is responsible for the success of the indirect anaphoric reference to the word key in the following sentence. Schwarz makes a distinction between cases like this and the following (2000: 111):

(21) In those last August days … the girl called Rita Seidel awoke in a small hospital room…. The nurse steps up to her bed.116

This time she makes use of frames (however, calling them “schemata”) for her explana-tion. An activated frame establishes the relationship between nurse and hospital room. The frame contains the information that a hospital is a building with rooms and wards in which ill people are treated, as well as the knowledge that this treatment is carried out by nurses. In the previously introduced terminology, hospital room and nurse are two instantiations (or fillers) in two different slots of the hospital frame. However, why should this process of schema instantiation differ significantly from the first example of reference establishment? The verb unlock also activates a frame, and the word key also functions as an instantiated filler. Resolving the two indirect anaphors leaves us with the following sentences:

(20′) I was about to unlock the door with the key… (21′) … The nurse steps up to her bed in the hospital room.

115. To what extent it makes sense to use the term “anaphor” to describe these phenomena is a moot point here.

116. Christa Wolf (2003: 4–7), They Divided the Sky. Translation by Luise von Flotow of Der geteilte Himmel. University of Ottawa Press. Original text: “In jenen letzten Augusttagen erwacht in einem kleinen Krankenhauszimmer das Mädchen Rita Seidel… Die Krankenschwester tritt an das Bett”.


Simply taking the primary data as the basis for the process of language understanding,117 it is difficult to find a plausible explanation as to why the semantic relationship between the lexical units key and unlock on the one hand should be significantly different to nurse and hospital room on the other. Knowing that nurses work in hospitals is just as much a part of the prerequisite conceptual world knowledge as knowing that a key serves to unlock a door. In the former case, an examination of the German original may be instructive:

(21′) In jenen letzten Augusttagen erwacht in einem kleinen Krankenhauszimmer das Mädchen Rita Seidel… Die Krankenschwester tritt an das Bett.

In this case, Krankenhauszimmer (lit: hospital room) and Krankenschwester (lit: hos-pital sister = nurse) are both determinative compounds. The first element determines precisely which entity is intended and shows that both expressions belong to the same frame. Yet, this is a cognitive-interpretive act that integrates a set of assumptions into a superordinate frame. A Krankenhaus is a “house” for the care of ill people, an insti-tutional facility with hospital wards and rooms in which nurses who are responsible for this care work.

The second case is no different.

(20′) Ich wollte gerade die Tür aufschließen, als Moretti [sic] aus dem Gebüsch sprang. Vor Schreck ließ ich den Schlüssel fallen.

The semantic nucleus of the German particle verb aufschließen (lit: ‘open up’ = ‘un-lock’) and of the derivative Schlüssel (‘key’) allocates both to the same frame. The relationship in which the two expressions stand to each other within this frame is then inferred on the basis of world knowledge. Not all doors can be unlocked, only particular doors must be unlocked at all, and a key isn’t always necessary to do this (for instance, a door between rooms inside a house is normally not locked). Thus, a conceptual integration of relevant assumptions about the world also already occurs here, and in this case this makes it unnecessary to occupy a syntactic slot specifying what is used to unlock a door.118

Clearly the distinction made between the different types of indirect anaphor are extremely theory induced (cf. Coulson 2001: 43ff.; Lakoff 1991). It is not the result of typification of primary data relevant to understanding, but the result of a prior,

117. I draw on a distinction made by Fillmore (1984: 127; 1985: 235). Data that are relevant to the understanding of an expression are primary; data that are highly theory induced are secondary.

118. In Example (20) the information in the sentence is sufficient to clearly resolve the “indirect anaphor”; it is not necessary to make use of additional contextual data (which are, of course, always present in natural processes of understanding language – here, for instance, we are deal-ing with a quote from the novel One for the Money by Janet Evanovich). It states that “a door” is to be unlocked and that in its direct vicinity there are “bushes”. Thus: The door is the entry door to a building (since bushes are prototypically not found inside a room).


syntax-theory oriented distinction, i.e. of “theory-defined data, which scholars of par-ticular persuasions find manageable and regard as proper for their theories”: “data that are associated with single sentences that are used assertively and in separation from any context of use” (Fillmore 1984: 128). Schwarz’s own objective of positing a cogni-tively economical model would be much better met if all anaphoric referentialization or activation processes were based on a uniform descriptive format (cf. Ziem 2013, 2014). Frames are an obvious candidate for this.119 The other two types, the so-called inference-based and meronymy-based activation types (Schwarz 2000: 104–111, 114–117), can also be effortlessly explained using a frame model (Ziem 2014, submitted). Apart from the examples taken up here as (20) and (21), Schwarz discusses a range of other case examples and is forced to finally admit that the four postulated types cannot be clearly distinguished from one another. However, instead of proposing a universal explanatory model, she posits the existence of “mixed types” that lead to an “anaphoric continuum” (Schwarz 2000: 117). It is solely Schwarz’s syntax-oriented language model that prevents uniform modelling.

Once again, the particular effects of a modularist position can be explicitly seen here. Without completely abandoning the decomposition theory, Schwarz shifts the content structures of linguistic meanings onto the side of the conceptual system, and formal structures without reservation onto the side of the linguistic system (cf. Figure 2). Thus, in key points she remains committed to a modularist cognitive model. Almost all the problem areas of two-level semantics that I outlined in the previous section therefore remain for the three-level model. The reason for this should be a little clearer in the meantime. Determination of the semantic research focus and the repertoire of methods available is extremely dependent on whether a modular or a holistic cognitive model is given precedence – no matter in what form or variation. This choice also establishes the scope and limitations of frame semantics.

There is one position for which the holism/modularism dichotomy does not apply to the same extent: the so-called “conceptual semantics” of Ray Jackendoff. As in the holistic model, the departure point in this case is the hypothesis of the structural identity between semantics and cognition. Nonetheless Jackendoff embeds this under-standing of semantics in a modularist model of cognition. Although this only indirectly affects (frame-)semantic descriptions,120 the impact does become tangible with respect to the larger issues surrounding the model of cognition (cf. Deane 1996).

119. In Section 6.5, I briefly return to indirect anaphors and demonstrate that frames can indeed serve as a uniform descriptive format to explain all four of Schwarz’s postulated activation types.

120. Thus similar to what I do here, Jackendoff advocates a default logic that uses considerations from prototype theory to explain semantic structures (Jackendoff 1983; 1990: 283ff.). However, in Jackendoff ’s conception, semantic units bear remarkable similarity to the “fixed primes” of Bierwisch (1986). In fact, Jackendoff ’s conceptual structures, like semantic forms, consist of a fixed set of semantic primitives that determine the syntactic behaviour of a linguistic expression. On the repercussions for semantic descriptions cf. Taylor (1996).


In the following, I dispense with a detailed outline of Jackendoff ’s Conceptual Semantics. However, since the holistic position can be easily illustrated in contrast to this, in Section 2.3.2 Jackendoff ’s approach serves as the foil, as it were, for the holistic approach. In contrast to Jackendoff, cognitive linguists proceed from the assumption that the language faculty closely correlates with general mental abilities. Accordingly, phonological, morphological, and syntactic processes are specific manifestations of conceptualizing capacities.

Section 3.1 initially addresses the holistic position only with respect to semantics. It is directed against level-semantic models; Jackendoff also supports it in this form.

2.3 Holism

The designation holism is only to be found in the German literature on this subject. It was popularized by Monika Schwarz (1992a: 16–19; 1992b: 49–51), who was clearly at-tempting to establish a counter term to “modularism”.121 By contrast, Anglo-American literature generally speaks of an “integral”, “unitary” or “uniform” model. This uni-tary aspect should perhaps be particularly highlighted here. “Unitary” means that, in contrast to what I just illustrated for the level-semantic models, the descriptive and representational formats do not follow different rule systems. The cognitive descrip-tion of anaphoric references would only have been uniform if one and the same format underlie all reference types and they were described by the one and the same cognitive operations.

The unitary principle can, on the one hand, refer to the entire cognitive system. In this sense, the cognitive psychologist John R. Anderson stresses

…the unity of human cognition, that is, that all the higher cognitive processes, such as memory, language, problem solving, imagery, deduction and induction, are different manifestations of the same underlying system. (Anderson 1983: 1)

The holistic notion that the ontological development of the cognitive system, includ-ing language, can only occur in interaction with the environment had already been advocated by Jean Piaget (in 1983). He rejects Chomsky’s nativist position, which states that universal and specifically linguistic parameters are innate. Inversely this means that a holistic language theory must illustrate the operative effect of general cognitive processes on all levels of the organization of linguistic signs. In particular it must be able to uniformly describe different cognitive phenomena (semantic, syntactic, phono-logical, etc.). This approach constitutes the methodological departure point of cognitive linguistics and in particular of Cognitive Grammar as first animated in Langacker 1987.

121. Although Lutzeier (1985: 105–06) already speaks of “holistic” semantic conceptions, he is not setting his sights on cognitive theoretical positions, but rather making general reference to strong sociolinguistic approaches (such as Putnam 1975).


However, the unitary principle also states that cognitive data within one and the same subcategory do not vary in terms of status from one another. Of course, level-semantic models also infringe this principle, since they posit two formats for the repre-sentation of linguistic meaning, a logical-formal and a conceptual-encyclopaedic one. Both are manifestations of different cognitive systems. Jackendoff doubts the plausi-bility of semantic levels, this in this respect maintaining a holistic position, yet at the same time he challenges the uniformity of overarching semantic cognitive processes.

In the following, I adopt Schwarz’s proposed terminology and use the term holism exclusively for the cognitive model of cognitive linguistics. I sympathetically take into account the fact that this entails an underlying reference to a philosophical school. From the time of Quine (1979), the term holism has served in particular to designate an epistemological position of “antifundamentalism”, which states (in contrast to Frege and Carnap) that the meaning of linguistic expressions cannot be confirmed or refuted by perceptual experience. What sentences mean depends in fact on an entire network of epistemological convictions, and these are never questioned in their entirety.122 Schwarz most certainly did not imply this in her use of the term holism. However, in large sections there is an astonishing similarity (which has until now received little attention) in terms of fundamental epistemological convictions between the ency-clopaedic semantic conception of cognitive linguistics and the semantic holism of analytical epistemology.

2.3.1 Meaning as conceptualization

The fact that the holistic approach concerns not only the larger architecture of human cognition, but also to the same extent subcomponents of this means in our context: If level-semantic models posit that contextually dependent usage variants of an expres-sion share a grammatically determined pattern restricting the meaning potential of linguistic expressions to a particular reading, then this form of “disambiguation” must be performed solely by conceptual processes in a holistic model (cf. Langacker 1997; 1999b: 327–332). In the modular model, the question remained as to what cognitive reality a purely linguistic representational level should pass through if no conceptual processes such as categorization participate in this. In the holistic perspective, this problem does not even arise. Since semantics without cognition is not possible but language understanding ultimately concurs with meaning understanding (and not with the “understanding” of formal-syntactic structures), the processes involved in this are seen as being fundamentally conceptually motivated.

122. Sellars (1999) adds that experiences can never confirm or refute statements, because they themselves have no propositional content. Only other statements can confirm or refute a state-ment. Thus, a cognitive “supererogation” always supervenes.


So if one does not accept a causal link between cognition and semantics, one has to accept the bizarre claim of semantic arbitrariness. If, however, one is willing to accept cognition as the motivation and the input for semantic categorisation, one may be able to avoid the deadlock. (Dirven 1994: 144)

Of course, a holistic approach thus faces the task of explaining how language under-standing is possible at all against the background of the irreducible polysemy of lin-guistic expressions. However, it is important to recognize that the problem of polysemy as it occurs in level-semantic models is one of their own making, i.e., it is a strongly theory-induced problem. The danger of “polysemy inflation”, as Herweg (1988) puts it, only arises at all in a modular semantic conception because of the precedence given to those data that are ascribed to an expression outside of any context and outside of any communicative function.123 However, even compounds or simple phrases (like school principal or fantastic school) would no longer be ambiguous to the same extent, because they would already be contextualized by a cognitive model that consists of a multitude of background assumptions (default values). The activation of relevant back-ground assumptions leads to a hierarchical categorization of potential interpretational variances. Contextual knowledge, understood here as the “base of our understanding of a word” (Fillmore 1976b: 23) is thus ineluctable.124 In other words, tabula rasa and understanding are mutually exclusive. Thus, within a holistic approach the question shifts from how language understanding can occur without semantic representations (in the sense of modular conceptions) to one of which uniform cognitive operations (apart from other cognitive capacities) control language understanding.

What is remarkable is that, by and large, subjects engage in quite similar construc-tions on the basis of similar prompts, and thereby achieve a high degree of effective communication. The reason seems to be that cultural, contextual, and cognitive substrate on which the language forms operate is sufficiently uniform across in-terlocutors to allow for a reasonable degree of consistency in the unfolding of the prompted meaning constructions. (Fauconnier 1999: 98)

123. Cf. for instance Busse and Bickes (1984); Busse (1988) provides a critique of such word-semantic theories from a communications theory perspective.

124. In this context Busse (2007) suggests integrating the term “contextualization” into a theory of language understanding as a comprehensive category of analysis. Langacker (1997) makes a very similar suggestion from a cognitive perspective. According to this suggestion, any form of (language-based) knowledge actualization is a form of contextualization. In my view, the key question in this perspective is how contextually relevant knowledge is cognitively constructed. Specifically, it needs to be clarified which conceptual contextualization format and which cognitive contextualization operations are to be distinguished. In contrast to level-semantic models, frame semantics understands itself in this sense as “contextualization semantics” and U-semantics.


Fauconnier mentions the following cognitive operations, which I elaborate on later to some extent in connection with frames:

…viewpoints and reference points, figure-ground/profile-base/landmark-trajector organization, metaphorical, analogical other mappings, idealized models, framing, construal, mental spaces, counterpart connections, roles, prototypes, metonymy, polysemy, conceptual, fictive motion, force dynamics. (Fauconnier 1999: 96)

Utterance units constitute the departure point of the usage-based approach of cognitive linguists. This therefore raises the number of those potentially relevant assumptions that can be called upon for the semantic interpretation of an expression. Even when a maximally “decontextualized” communication unit is postulated (for instance, a note lying on the street on which the words fantastic school are written), we have a plethora of context-relevant assumptions with which we can construe alternative communica-tive contexts.125 The cognitive operations named above contribute to the success of this construction, and at the same time they restrict the interpretational variance.

What this means for studies within lexical semantics is that, in order to analyse word meanings on a usage basis it is necessary to call on the knowledge contexts in which they occur. Under the premise that words, in effect, never occur in isolation, but rather function as a component of larger units, it should be evident that any model that abstracts from this knowledge context and sets an invariant component struc-ture as the basis of meaning-theoretical considerations is theorizing at cross purposes to linguistic reality. This results in purely theoretical constructs like the problem of “polysemy inflation”. Quite the contrary, language users do not have to distinguish beforehand between relevant and non-relevant or between centrally and marginally relevant knowledge but have an almost inexhaustible potential of meaning-relevant data at their disposal. Which data actually feed into the process of understanding in particular is another question. The semantic problem of polysemy shifts for holistic approaches such that they have to specify the contextual restrictions governing the contextualization processes.

In the course of this book, I have already stressed a number of times that it makes sense to distinguish those restrictions that involve the construction of conceptual structures from those that refer to the conceptual representational format itself. This corresponds to the usual distinction between operational and representational aspects of language processing:126

125. Scollon and Scollon (2003) introduce the term “transgressive emplacement” in this context. In this context, potentially context-relevant questions could be: Is the utterance of any relevance to me? Why is the note lying here? Is it near a school that could be being referred to? Who could have written the note? Does the punctuation indicate an illocution? And so on.

126. Cf. the summary remarks in Fillmore (1984: 125ff.) and Schwarz (1994).


i. Cognitive operations: These concern the fundamental cognitive processes (like pattern recognition, schematizing, categorizing, figure/ground distinctions, etc.) that participate in processes of language understanding. Since from a holistic per-spective the linguistic knowledge system is not autonomous, these principles just as much govern non-linguistic cognitive processes.

ii. Conceptual structures: These concern the semantic structure respectively “semantic units” (in Langacker’s terms, cf. Section 4.1). Conceptual structures exist, on the one hand, as schematized units in the long-term memory and, on the other, as knowledge-enriched semantic representations of a linguistic utterance unit in the short-term memory (that is, as a representation of what expressions mean in an actualized communicative context).

A certain predominance of operational relations can be detected within holistic theory construction. The particular effect of this is that the focus is not so much on the attempt to unify the different representational formats (“mental spaces”, “idealized cognitive models”, “frames”, “domains”, “schemas”), but on the aim to describe them using a uni-form set of cognitive operations (cf. Croft and Cruse 2004: 40–73; Taylor 2002: 8–16). The fact that a range of representational formats compete with each other (or simply exist alongside each other) and that the interaction between actualized representational and procedural aspects is neglected are issues that I see as a deficit in current research (cf. also Section 1.2.2). All in all, there seems to be a prevailing view that different formats explain different cognitive phenomena and that structural aspects should be ignored until cognitive operations have been exhaustively explained.

2.3.2 Language as conceptualization (R. Langacker vs. R. Jackendoff)

To begin with, it is necessary to sketch out the shared cognitive-theoretical premises. Section 2.3 then elaborates on individual points that are strictly relevant to frame semantics. The following only focuses on the most important aspects of a holistic approach. These are most clearly understood by comparing Jackendoff ’s conception of Conceptual Semantics with the holistic-encyclopaedic conception as coined by Langacker.

The most important common feature in Jackendoff ’s and Langacker’s conceptions is that semantic and conceptual structures are on equal footing, while the most impor-tant difference is that Jackendoff (similar to Bierwisch and Schwarz) ascribes syntactic structures a quasi-autonomous status, whereas Langacker understands syntactic con-structions primarily as semantic functions.127 Thus in the holistic conception, syntactic

127. On this common feature cf. Jackendoff (1996: 96). He explicates the reference to syntax as follows: “It [Jackendoff ’s “conceptual semantics”] is committed to the existence of an autono-mous syntax, though one that interacts richly with meaning in a way that is potentially congenial to the findings of cognitive linguistics.” (Jackendoff 1996: 93)


structures are not the result of an autonomous – i.e. independent from semantic influ-ences in particular – rule system, but rather forms of expression for meaning-relevant construction processes. Grammatical categories and syntactic constructions indicate how a linguistic expression (or sequence of linguistic expressions) is to be semantically interpreted.128

Therefore Jackendoff ’s approach has the character of semantic holism while main-taining cognitive modularism. When engaging with Langacker’s position, Jackendoff formulates his premises as follows:

I diverge from Chomsky’s practice in not treating syntax as the principle of genera-tive component of grammar, from which phonology and meaning are interpreted. Rather, I treat phonology, syntax, and conceptual structure as parallel generative systems, each with its own properties, the three kept in registration with one another through ‘interface rules’. (Jackendoff 1996: 96)

His rejection of a non-conceptual, semantic representational level is motivated by his scepticism towards a meaning theory that, in conformity with Chomsky, is primarily based on notions from syntax theory. Jackendoff criticizes this type of “syntactocen-trism” (Jackendoff 1997: 15–19) and counters with a conceptual semantic model that has a direct interface to the syntactic module without – unlike in the semantic level models – having to take recourse to semantic representation (sem). Nonetheless, he adheres to the principle of the modular organization of human mental faculties. In his model, the language faculty features three modules that interact with one another via correspondence rules (interface modules) and contribute to the construction of representations of linguistic structures. These modules are the phonological, syntactic, and conceptual systems. What Jackendoff refers to as “parallel generative systems” in the quote above he later calls a “tripartite parallel architecture” (or TPA) (Jackendoff 1997). Within the TPA, only the conceptual system possesses interfaces to extralin-guistic systems such as emotion and the perceptual and motor systems. The functional autonomy of all systems still remains.129

This is illustrated in Figure 4. Whereas the holistic model of cognition posits inten-sive mutual relationships between language, perception, and motor function, which all participate in the construction of conceptual representations (irrespective of type),130

128. For an overview cf. Croft and Cruse (2004: 308–326); for a particular view of syntax the-ory within “Construction Grammar” see Goldberg (1995), and within “Radical Construction Grammar” see Croft (2001).

129. Concerning this cf. Figure 10.5. in Jackendoff (2002: 305).

130. Lakoff and Johnson (1980, 1999), for instance, have illustrated how “image schemata” (equivalent to the already mentioned sensorimotor-motivated container concept) evolve on-togenetically and then become relevant in the form of conceptual metaphors for the process of language understanding. Further, some studies indicate that even off-line cognition (that is, cognition without direct perceptual input like daydreams, hallucinations, etc.) operates on a


TPA maintains the principle of cognitive division of labour (or autonomy) for the con-ceptual, phonological, and syntactic systems. According to Jackendoff, since syntactic patterns and phonological units are not conceptual units, cognitive capabilities such as categorization play no role at all in this process.131 Instead inherent rule systems of a purely formal nature determine these mental modules.

The characteristics of Jackendoff ’s intermediate position that cannot be ascribed to either modularism or holism can be derived from his tripartite model of cognition. According to TPA, every lexicon entry contains phonological, syntactic, and concep-tual information. Since linguistic meanings are represented in the conceptual system – and not in an autonomous semantic one – Jackendoff ’s approach differs in this respect significantly from the two-level model (cf. Wiese 1999) and tends to bear similarities to Langacker’s holistic position.

body base because it draws on sensorimotor resources. In this case, sensorimotor simulations of online cognition occur (Talmy 2000; for an overview: Wilson 2002).

131. Conversely, Taylor (2002: 243–261), (2003: 247–165) and Kumashiro (2000) have pro-vided evidence that phonological processes are also essentially guided by categorizations. Cf. Goldberg (1995) and Fillmore, Kay, and O’Connor (1988) on the conceptualization of syntactic information.

Concepts

Perception Motor functions

Inferences

Language = Repertoiresymbol. unit

COGNITION

WORLD

Inte

grat

ion

Categorization Figure/Grounddistinction

Conceptualblending

etc.

Restrictions through cognitive capacity

Objects Objects

Sign-Token

MIND

Figure 4. The holistic model of cognition


Following the gestalt psychologists … as well as more modern perceptual psy-chology …, I maintain that a theory of meaning must pertain to the “behavioural reality” of the organism, or more properly, the world as it is constructed by the organism’s mind. (Jackendoff 1996: 96)

When Jackendoff sets linguistic meanings on a par with gestalt entities, thus shift-ing them in the proximity of frames (cf. also Jackendoff 1983: 23–29, 140–143), two convergences become immediately apparent. On the one hand, categorizations play a central role in the construction processes mentioned. Meaning forms emerge from the conceptual integration of linguistic and perceptual (such as visual or auditory) infor-mation and information stored in the long-term memory.132 Furthermore, Jackendoff posits that conceptual representations, i.e. the results of these categorization processes, (a) fundamentally exhibit fuzzy boundaries, (b) are graded on the principle of family resemblance, and (c) are prestructured according to default values. Thus, he shares all the essential assumptions of the conceptual theory presented in this book.

In TPA, however, linguistic meanings are not grounded in human experience to the same extent as in Langacker’s encyclopaedic semantics (Langacker 1994). A key difference is, in fact, that Jackendoff concedes relative autonomy to syntax, which stands in opposition to the semiotic premise of cognitive linguistics, which states that form and meaning are conventionally linked to each other not only on the lexi-cal level but also with respect to grammatical constructions (Langacker 1987: 56ff, for an overview: Ziem and Lasch 2013: 18–26, 90–95; cf. also Section 6.1.1 and 1.2). Consequently, according to Jackendoff visual information has no effect whatever on syntactic patterns. On the other hand, cognitive linguistics equally posits an (also ontogenetic) intrinsic relationship here (Lakoff 1987: 462–585).133 In Figure 4, this is symbolized by the double arrow connecting language, perception, and motor func-tion. Thus, in TPA the modularist distinction between linguistic and non-linguistic knowledge with respect to syntactic and phonological knowledge representations stands in contrast to an understanding of language as an “integral part of human cognition” (Langacker 1987: 12). In this respect, Jackendoff ’s Conceptual Semantics corresponds to the level-semantic model of Schwarz and Bierwisch. In contrast to them, however, Jackendoff rejects syntactic restrictions on semantics. As semantic structures are conceptual structures in TPA, they are subject solely to the restrictions of cognitive capabilities.134 Consequently, as in the holistic model, semantic concepts

132. In analogy to the prototype theory considerations of cognitive linguistics, Jackendoff (1983) developed a theory of so-called “preference rules”, which is intended to explain the activation of semantic default values – in the same sense as frame theory.

133. Langacker (1987) argues that lexical units, idioms, and syntactic constructions are analogue phenomena that only vary according to their specificity in level of abstraction. This book ad-dresses the questions following from this only in passing.

134. On the one hand, Deane claims that Jackendoff ’s assumption of modularity leads to a two-level model of sorts. He writes (Deane 1996: 38): “Jackendoff essentially splits meaning into two


(understood here as knowledge-enriched semantic representations of an utterance unit) emerge through inferential processes of conceptual integration. In Figure 4, this is symbolized by unidirectional arrows.

Against this background, it is misleading that Schwarz denotes Jackendoff ’s conception overall as “holistic” (Schwarz 1992b: 51). More precise differentiation is required. Also according to Schwarz, a cognitive theory is holistic if the cognitive principle of modularity is rejected and the linguistic knowledge system is explained by general cognitive principles, “which equally form the basis of all mental capabili-ties” (translated from Schwarz 1992b: 49). However, precisely this does not apply to TPA. A holistic model of language is distinct from TPA in five points, as illustrated in Figure 4:135

i. Language, perception, and motor function interfere and interact with each other. They do not constitute autonomous cognitive systems with their own process-ing principles, but rather are part of a cognitive processing cycle that is open to the external world, i.e. the material environment. Cognitive linguistics adheres in this sense to an integrative model (to “experiential cognition”, cf. Lakoff 1988), which states that language processing is essentially codetermined by perceptual and sensorimotor processes – such as in the form of “image schemas” (Johnson 1987; Lakoff 1987; Hampe 2005) or “conceptual metaphors” (Lakoff and Johnson 1980; Lakoff and Turner 1989; Johnson, Mark 1999).

ii. The mind and the world do not belong to different ontological spheres, but rather the boundary between them is permeable (Lakoff and Johnson 1999; Sinha 1999, 2004) yet cannot be discarded (Wilson 2002). Language, perceptual and motor functions only exist insofar as they have extramental correlates.136 In contrast to the TPA model, linguistic cognitive processes are grounded or embodied directly – and not indirectly via interface modules – in human experience.

parts: a linguistic portion instantiated in conceptual structure and an extralinguistic portion encoded in the correspondences between conceptual structure and other mental modules”. On the other hand, linguistic knowledge does not restrict the semantic process of conceptualiza-tion, but rather only mediates between conceptual units and the other modularly organized information units.

135. Compare this to Jackendoff ’s own sketch of the model (2002: 305).

136. In Figure 4, this correlation between cognitive and extramental entities is shown by two-directional arrows. The reason for this is that a unidirectional effect does not do justice to certain phenomena. Language, understood as a cognitive repertoire of symbolic units, changes to some extent only in language use, that is, mediated via extramental units, which in turn are the result of cognitive processes. This applies similarly to perceptual units such as visual or auditory impres-sions. Although they are the result of extramental units, they are simultaneously the products of specific cognitive functions. Thus, our perception of colour, for instance, depends significantly on the number of colour receptors we have.


iii. Accordingly, language is not to be understood as an interaction between different cognitive systems, but rather as a repertoire of symbolic units that language users correlate to variably (re-)con struct communications. In particular, syntax does not have an autonomous status. In fact, even syntactic constructions fulfil symbolic functions; they are expressions of conceptualizing capacity (Croft 2001; Kay and Fillmore 1999; Langacker 1999b; Goldberg 1995).

iv. Inferential processes occur on all levels of cognitive organization. They involve linguistic conceptualizations just as much as the construction of specific sensori-motor and perceptual concept units (Anderson 1983; Minsky 1975). The proces-sual operations superimpose themselves, as for instance in the case of conceptual metaphors (Lakoff 1987) or the conceptualization of deictic expressions and prep-ositions (Brugman 1988; Schulze 1991). Within TPA, however, these processes operate separately.

v. In a modular model, module-specific rule systems ultimately restrict the con-ceptual interpretational variance of a linguistic expression. Since Jackendoff sets semantic and conceptual structures on a par, this particularly affects syntactic and phonological rules, whereas in level-semantic models this additionally affects the specific semantic restrictions that determine the reference potential of a linguistic sign (Bierwisch and Schreuder 1992: 28). However, in the holistic conception, the interpretation of an expression coincides with the process of conceptual inte-gration, which is governed by universal cognitive principles like categorization, schematization, figure/ground distinctions, etc. (Langacker 1987: 116–137; Talmy 1988: 165–205).

These are fundamental features of a holistic model of cognition. In summary: Language is conceptualization on all levels of its organization. What this specifically means for a frame-semantic conception can already be inferred at this stage. Under the auspices of holism, a distinction between encyclopaedic and linguistic knowledge representations is just as untenable as a distinction between conceptual and non-conceptual cognitive processes.

As early as the mid-1980s, Lutzeier (1985) and Busse (1987) developed models that amount to a similar encyclopaedic semantic conception, although both approaches argued not so much on the basis of cognitive as communication theory. In the context of our cognitive theory, it was largely the explanatory inadequacy of the semantic descriptions found in modular approaches that led to this position.

In sum: In two-level semantic models, predetermined component and argument structures restrict the meaning potential of linguistic expressions to such an extent that large proportions of U-relevant knowledge (“Understanding-relevant knowledge” in the sense of Fillmore 1985) remain unexplained and inexplainable. Numerous exam-ple analyses have demonstrated this. Schwarz recognized the problematic of a strong decomposition theory, yet in her own three-level model failed to correct the “syn-tactocentrism” (Jackendoff 2002) that led to the semantic explanatory inadequacy of the two-level model. Jackendoff tackles this correction by consistently understanding


semantic structures as conceptual structures. If, however, these are prototypically or-ganized and such structuring is the result of cognitive operations, why, then, is prefer-ence not given a priori to the methodologically more obvious variant that describes language understanding only under the psychologically realistic condition that the same principles control all processes of cognition?

Jackendoff ’s cognitive model, too, is a far cry from being holistic. However, what does holism mean for frame semantics? Interestingly, Schwarz clearly outlines the consequences at one point, although these consequences are by no means evident to this radical degree in her three-level model:137

Word meanings can no longer be represented by means of feature aggregates that are defined by sufficient and necessary features. Individual meanings are structures as mental prototypes. They are mental representational units with obligative and facultative components that are mentally restricted by defaults, yet allow options and thus function as instantiatable variables. The individual components demon-strate (depending on their degree of typicality) different activation threshold levels (threshold values for activation). Typical units have a lower threshold value than non-typical units. (Translated from Schwarz 1992a: 91)

And she continues:

The entire semantic cognitive system in the long-term memory should be mod-elled as a network in which meanings are connected with each other through specific relations. When lexically activated, they absorb action potential [“Aktivierungspotential”, AZ] and pass it on to other meanings. (Translated from Schwarz 1992a: 91)

Indeed, those are the key consequences: Meanings should be examined solely from the perspective of prototypicality, where meanings are to be understood not as feature aggregates (as in structuralist feature semantics or the standard version of prototype theory) but rather as potential. This potential is prestructured in mental models by default values and thus grounded in the experiential world of language users. Default assumptions can be flexibly activated and replaced by other assumptions. Ultimately, neither the default values themselves nor the model in which they are embedded are autonomous entities, but rather they constitute sections of a comprehensive semantic network.

137. What is particularly precarious is that she designates the consequences as premises that “[should] form the basis of every semantic lexicon model given the current state of research” (Schwarz 1992a: 91; italics not in original). Yet, Schwarz herself posits a weak decomposition theory that nonetheless departs from the premise that expressions possess contextually invariant features (cf. the lexicon entries for to shoot someone dead and house, Schwarz 2000: 38). Further, her three-level model is diametrically opposed, both methodologically and empirically, to a network model.


However: How can these “semantic cognitive systems in the long-term memory” be described? What “specific relations” connect linguistic meanings in a network? How can “default values” or “mental prototypes” be determined? What determines the “threshold value of activation”? To what extent do default values function as “instantiat-able variables”? How do linguistic meanings emerge through the absorption of “action potential”? How can action potential be passed on “to other meanings”? And above all: In what mental representational format does this all occur?

Schwarz provides no answers to these questions, perhaps for the reason that she is not able to pose them in this form within her quasi-modular model. In the follow-ing, I will attempt to answer some of these questions by regarding frames as a uniform representational format within a holistic model of cognition.

chapter 3

The holistic paradigm

The most notable finding in Chapter 2 is that not only do modular semantic models proceed in a reductionist manner, i.e. large sections of knowledge pertinent to un-derstanding tend to be excluded from and avoided in linguistic analysis (as noted by Busse 2005a: 55), but also – and even more crucially from the perspective of theories of meaning – the methodical assumptions underlying them in fact hinder adequate se-mantic description. Even when the postulated underlying logical and formal structure of linguistic signs is considered as only a meaning-based departure point for semantic description (for instance, as in the works of Dölling) and as a result the onus of expla-nation is delivered over to the lower levels of semantic interpretation (as with Schwarz in her use of schemata, or Dölling, also using abductive inferences), no exhaustive explication of meaning can take place. Certainly the established but unquestioned metaphors “semantic level” and “semantic core” imply that linguistic and encyclopaedic knowledge may be clearly differentiated in analytical terms; they further imply that a non-inferential area of meaning-relevant knowledge exists that is a human biological endowment without being itself the result of cognitive capacities.138 However, in the previous chapters we saw that the key conceptual metaphors (entirely in the spirit of Lakoff and Johnson 1980) of level and core remain in many respects inadequate for modelling the research area. This is not related to the fact that they are “merely” meta-phors. Metaphors may certainly be secondary cognitive phenomena according to level theorists, but here they nevertheless clearly motivate axioms of a theory of meaning, such as that of drawing a sharp distinction between linguistic and world knowledge and thus linguistically setting off purportedly primary data from secondary data. The decisive reason for the emergence of this explanatory deficit lies rather in the funda-mental circumstance that semantic knowledge is based on experience, i.e. acquired in particular communicative contexts and on the basis of already existing previous

138. To what extent such metaphors are a force that influences generative approaches has been clearly demonstrated by Langacker (2006: 108): “There was first the conception of language as a distinct mental ‘organ’ or ‘faculty’. … A language was represented as a box labled G, which im-posed a partition on the set of possible sentences (strings of discrete symbols), dividing them into two discrete subsets: those that were (categorically) grammatical, and those that were (categori-cally) ungrammatical. The grammar of a language (G) was thought of as a machine, or assembly line, in which well-formed sentences were constructed step by step and given as ‘output’. Of course, the objects being constructed took the form of inverted trees, from the bottom branches of which hung discrete lexical items (attached by asserting them into slots). These lexical items were thought of as containers …”


knowledge, and is thus employed holistically when understanding language.139 All facets of meaning-relevant knowledge, including the most fundamental, are emergent phenomena. They are a result of language usage, and they perpetuate themselves or modify themselves through language usage. This applies to all linguistic units on all levels of the organization of signs.

This assessment leads to a specific basic premise for a semantic model, which I examine more closely from the point of view of frame theory: the thesis that it is impos-sible to specify meaning abstracted from context. If it makes sense at all to speak of a “zero context” (or “neutral context”), then at most this can only mean a context with (variable) standard conditions.140 What is sometimes called “literal meaning” is thus itself a collection of communicatively stabilized knowledge elements:

“Neutral-context” interpretations are simply those which are so accessible from our entrenched conceptual structures that their construction requires very little help from a specific context in the outside world. (Sweetser 1999: 137)

Such solid conceptual structures constitute the so-called “default values”, the com-municatively stabilized frame instances within the frame-semantic model proposed in this book. I attempt to outline how such default values evolve in Section 6.5. The argumentation shows that communicative stabilization is to be understood in a cogni-tive sense as the routinization of individual categorization operations.

To begin with, it is decisive that taking the level-semantic model as the meth-odological anchor has wide-ranging consequences for any variety of semantic frame theory, as for instance in the case of Fraas’ approach (Fraas 1996a). In Fraas’ concept, the “contextualization potential” (Fraas 1996a: 5) of linguistic expressions would have to be far lower and by far more rigidly prestructured than the empirical results suggest. As has been shown, in the modular model this contextualization potential is severely limited by the semantic representation underlying each utterance. Ultimately this is identical with the so-called construction and referential potential described by Wiese (cf. Figure 2 in Section 2.2.2), but only very narrowly overlaps with the “potential of the communicatively meaningful contextualization of a concept” aspired to by Fraas

139. “What is commonly designated as ‘terminology’ seems to be a conglomerate of attribu-tions of features, abstractions, and oblique and situational references which are the result of a multitude of communicative acts in which individual linguistic signs could be the trigger of or reference point of crystallisations of aspects of meaning (constituted through the contexts in which they were used), but which can never represent the ‘object’ (or the ‘term’) as a unity of its whole.” (Translated from Busse 1991b: 3)

140. This understanding of zero context is fundamentally distinct from zero context as it is ad-dressed within a multi-level approach to semantics. In the former, communicatively stabilized and acquired background assumptions constitute elements of the zero context, not “knowledge” that precedes language use. In this context, Busse (2007) drew attention to the ineluctability of cognitive contextualization, without which it would not be possible to understand meaning; cf. especially Section 2.

Chapter 3. The holistic paradigm 101

(1996a: 5). The reason for this is clear: The capacity for “communicative meaning” is not anchored in linguistic knowledge, but rather is contingent on the constructive-integrative interaction of acquired knowledge on the one hand and current contextual knowledge on the other.141 Cultural knowledge forms the basis of both lexicon and grammar, as stated by Langacker (1994: 33).

Drawing on the results of recent cognitive linguistics, in particular construction grammar and Cognitive Grammar, in the following sections I outline a approach that takes into account the creativity of the linguistic construction of meaning. This theory questions the distinction between linguistic and conceptual knowledge (Section 3.1). Linguistic understanding does not oscillate between the poles of linguistic and con-ceptual knowledge, but rather occurs in a holistic “mental space of understanding” (Section 3.2). Drawing on terminology proposed by Fillmore (1985), I speak in this context of a postulate of “U-relevance”. This forms the basis of both semantic-episte-mological and inference-theoretical approaches (Section 3.3).

3.1 Are linguistic and conceptual knowledge distinct entities?

The frame theory presented here aims to provide a non-reductionist description of linguistic meaning. “Non-reductionist” means that meaning-relevant facets of knowl-edge are not excluded in advance from the scope of the semantic study on the basis of theory-methodological assumptions, such as on the grounds that these are aspects of non-linguistic knowledge that can be assumed to be given.142

This extension of the area of research requires a reorientation of key semantic questions. Fillmore (1971a: 384–385; 1971b: 274) already formulated this in his early articles to the effect that the question is no longer what linguistic signs mean, but rather what knowledge is required to be able to appropriately use or understand a linguistic form.143 Some years later in his article Scenes-and-frames semantics he states:

I think it is misleading to separate a word from its context just for the sake of capturing in one formulation the common features of these two kinds of scenes.

141. An overview of linguistic approaches that focus on this interaction is given by Aschenberg (1999: 44–177); however, aspects of cognitively oriented contextual theories such as that of Langacker (1997) are not considered.

142. The semantic level model elaborated on here can be seen as a modern variant of structur-alist reductionism. It is no coincidence that the basic ideas of feature-based semantics live on in generative approaches, although there embedded in a comprehensive linguistic theory that even proposes linking semantic components with the hypothesis of genetic predetermination.

143. Yet without recognizing that he thus proposes a position that Ludwig Wittgenstein already propounded – for instance with recourse to “customs” and “life forms” as language-relevant reference quantities (Wittgenstein 1989: 199, 241) – for which no less than Karl Bühler’s field theory (cf. Bühler 1934: 154f.) is of central importance.


It is misleading, that is, if we are trying to capture by the semantic description of a word what it is that a speaker of the language needs to know in order to use the word appropriately. (Fillmore 1977a: 68)

Standing in distinction to the reductionist approaches that were oriented towards a system paradigm, Fillmore thus formulated the premise of a holistic theory of mean-ing. Following the postulate of U-relevance, the relevance of understanding, “what one knows, as a member of a culture, about the objects, beliefs and practices of that culture” (Fillmore 1971a: 383) thus shifts into the semantic field of vision. Instead of asking about the structural properties of signs as such, the users of these signs them-selves become the focus, or more precisely, the knowledge that they make use of in the act of understanding and the specific abilities they use to interpret linguistic signs by incorporating the complex of communicatively relevant factors making up the situ-ational context (Schmidt 1973: 104–105). In brief: An ontological perspective of signs gives way to a cognitive perspective of sign users.144

A linguistic theory of meaning is non-reductionist if it equally includes all U-relevant aspects of meaning in the analysis and in addition makes it clear in what way the necessary conceptual integration in processes of understanding occurs. Accordingly, the postulate of U-relevance and the attempt to avoid semantic reduc-tionism are two facets of the same methodological endeavour to achieve the most comprehensive explication of the meaning of linguistic units without at the same time treating meanings as atomized or reified objects.145 This endeavour inevitably leads to a semantic epistemology, because the forms of knowledge we are dealing with here are integral components of communicative, cultural practices and thus extend far beyond the limits of language knowledge in its narrowest sense. Meaning-relevant knowledge is social knowledge.146

144. Bransford, Barclay, and Franks, who may be said to be early proponents of a holistic con-ception, wrote over 35 years ago already: “In a broader sense the constructive approach argues against the tacit assumption that sentences ‘carry’ meaning. People carry meaning.” (Bransford, Barclay, and Franks 1972: 207) I return to their conception of a maximalist inference theory later on.

145. For a critique of this, see Zlatev (2003) and Section 2.2.3 and 2.2.4.

146. In German, the term Erkenntnistheorie is largely synonymous with the English term epis-temology. In the Anglo-Saxon context, however, the focus is rather on the subject-philosophical conditions of the constitution of knowledge. Certainly every conception of semantics – whether implicit or explicit – holds an epistemological position (on the relationship between epistemol-ogy and cognitive semantics cf. Geeraerts 1993). A semantic epistemology in the sense intended here, however, aims more towards the historical formation conditions of, for instance, cognitive patterns that control semantic interpretations. Geeraerts (1993: 73) notes: “Now, while the dy-namic flexibility of prototypical categories does indeed demonstrate the interpretative potential of linguistic categories, it also constrains the set of possible interpretations, as these are required to link up with existing concepts. Interpretation does not go wild… because speaker and hearer


Chapter 2 showed that one key question about the cognitive theoretical basis of a semantic epistemology concerns the structuring of the cognitive system. In contrast to the modularist models, the holistic conception subscribed to here disputes that it makes sense to distinguish between an autonomous semantic representational level and a conceptual representational level. Thus, the distinction between linguistic and conceptual knowledge also becomes obsolete. To avoid such distinctions, I speak of “U-relevant knowledge”, knowledge that is relevant to the process of understanding. I wish to elaborate on the pre-empted hypothesis that U-relevant knowledge is of a purely encyclopaedic nature and thus conceptually organized. It lays the foundation stone for a frame theory in the holistic paradigm.

First of all it is important to recognize that it is only possible to distinguish more or less meaningfully between linguistic and conceptual knowledge within a theoreti-cal context. It is by no means obvious what language consists of, not even if this sub-sumes – as is usually the case – elements of phonetic (or graphical), phonological (or graphematic), morphological, and specifically semantic knowledge. In fact, such an understanding of linguistic knowledge presupposes in turn prior specifications of the linguistic subdisciplines mentioned and thus the explication of an entire linguistic theory. Thus, the question arises, among others, of the extent to which, for instance, syntactic and morphological descriptions can be meaningful when disregarding uni-versal linguistic factors. Modular approaches, like the two- and three-level models, and holistic approaches, like Cognitive Grammar and construction grammar, arrive at completely opposing assessments in this point. Thus, we can conclude that criteria that are called upon to determine linguistic and conceptual knowledge pertain to the methodological premises of the approach being followed (cf. Dirven and Taylor 1988). Due to this interrelation with fundamental theoretical issues, this issue of determining what belongs to semantic knowledge is considered to be “one of the most difficult theo-retical (and – in the case of, for instance, lexicography or language teaching – practical) problems of linguistics” (Busse 1997a: 16).147

Although many semantic approaches take recourse to a heuristic division be-tween linguistic and conceptual knowledge, the methodological justification for doing so can only be found in language theory preconceptions.148 Such preconcep-

share a system of preferential interpretations that also constrains the form deviant readings may take.” From a historical epistemological perspective this depends on analytically determining this shared knowledge – for instance, in the form of default values.

147. Cf. also Baker (1999: 4–5).

148. A critical reflection on this is, however, all too often lacking, regardless of whether the research is oriented towards meaning theory or textual linguistics; cf. for example Coseriu (1980: 41–47); Gansel and Jürgens (2002: 102–116); Heinemann and Viehweger (1991: 95); Reischer (2002: 97). Even Scherner, who subscribes to an anti-reductionist position in his work Sprache als Text (Language as Text – Scherner 1984) and elaborates on – most certainly following the postulate of the relevance of understanding – cognitive constructivity as a basic function of


tions are intensively reflected in level semantics because it is only through them that a division between levels can be legitimized in the first place (cf. Bierwisch and Kiefer 1970; Bierwisch 1979: 64ff.; Bierwisch 1983a: 62–63; Bierwisch 1986; Bierwisch and Schreuder 1992: 30; Lang 1994: 26–29). As a general principle: As the legitimation of distinguishing between linguistic and conceptual knowledge proves to be shaky, so too does the distinctivity between linguistic-semantic and conceptual-encyclopaedic structures lose its plausibility. In contrast to the case in lexicography, the distinction between lexical and conceptual knowledge is not a result of practical considerations, but rather part of a linguistic programme.149

The question should therefore be: If a distinction between linguistic and concep-tual knowledge is not valid outside of a language theory, what reasons might then justify it within one? To support his conception of a Cognitive Grammar as an ency-clopaedic one, Langacker begins by discussing three reasons that do not derive from a specific language theory, but rather lay claim to validity across disciplines. Firstly, he states that this outlines a genuinely linguistic area characterized by “the privileged status of a restricted class of semantic properties” (Langacker 1987: 156), whether they are lexemes arranged in the lexicon according to structural principles or semantic features as the basic constituents of linguistic expressions, as in the case of two-level semantics. According to him, this leads to the fact that meanings can be identified as secondary phenomena of an underlying autonomous, formal system, which itself con-sists of a clearly analysable set of elements and relations. The second reason Langacker gives is of an epistemological nature: “[H]uman conceptualization is not amenable to empirical inquiry and precise description”. In this statement Langacker is alluding to the original Saussurean motivation of separating langue from parole, that is, the aim of

linguistic understanding, asks in this context on what theoretical basis his distinction between linguistic and world knowledge makes any sense at all. In an article published in 1994 he writes: “‘Linguistic knowledge’ allows the recipient to construct a … not yet saturated preliminary read-ing level, which is then enhanced to become a reading that is meaningful for the recipient by his consulting additional cognitive knowledge. The cohesion generated by linguistic knowledge functions metaphorically speaking as a ‘clue’. The recipient proves its cognitive ‘coherence’ to himself by following the clue and completing its interpretation.” (Translated from Scherner 1994: 328). But what does this “preliminary reading level” consist of? Can it be empirically verified? In what connection does the postulated lexical system stand with the larger semantic system and thus with syntactic, morphological, and phonological processes? To what extent does world knowledge retrospectively enhance or complete – as suggested by Scherner – the process of understanding? All the questions that I posed with respect to the two-level model also repeat themselves here. They can only be answered in the context of a theory of grammar.

149. From a lexicographical viewpoint, the question of the demarcation of linguistic knowledge is above all a question of the practical feasibility of drawing up dictionaries; cf. Wiegand (1988); Bergenholtz and Kaufmann (1996). In the meantime, the use of new media also opens up the possibility of non-reductionist descriptions of meaning; cf. Atkins, Fillmore, and Johnson (2003); Fillmore and Atkins (1994); Konerding (1993).


achieving a more objective form of language research. His third point is that any such objective form could only be achieved by formal logic in the context of a theory of meaning. Formal logic offers “a rigorous account of semantic structure based on truth conditions”. However, this assumption goes beyond positing an autonomous linguistic system, as the process of meaning construction is reduced even further to a particular, purely formal type of inferential processes.

Langacker doubts whether it is in fact meaningful to depart from linguistic knowl-edge in interpreting the cognitive phenomena of semantics. This is because: (i) There is no foundation for describing language as an autonomous system; (ii) meanings are precisely what they are said not to be: conceptualizations;150 (iii) certain aspects of meaning cannot be grasped in formal logic terms. This is why Langacker argues for an “encyclopaedic conception of linguistic semantics” within his programme of a Cognitive Grammar (Langacker 1987: 156). He thus follows the postulate of the U-relevance by proposing a double hypothesis: with respect to representational as-pects, the hypothesis that an exhaustive description of U-relevant knowledge can only succeed – at least in approximation – by recurring to a comprehensive explication of human cognition; and with respect to procedural aspects, the hypothesis that the area of research focus can only be adequately described in close connection with such cog-nitive processes that can also be equally found in extralinguistic cognitive processing operations.

In my view, Langacker leaves many questions unanswered in all three points. With respect to (i): From a semantic perspective, what exactly speaks against positing a complementary relationship between linguistic and conceptual knowledge? Why is it not possible for a modular semantic conception to comprehensively deduce linguistic meaning? With respect to (ii): To what extent are meanings solely the result of concep-tualizations? And why does this speak in principle against a distinction between lin-guistic and conceptual knowledge? With respect to (iii): Why can’t knowledge elements that cannot be grasped in formal logic terms be seen as “completing” or “enriching” the understanding of a linguistic expression? Is it really possible to deduce that formal logic descriptions are useless simply because they cannot deliver a complete explication of meaning? In my view, the fact that certain aspects of meaning cannot be grasped in formal logic terms speaks neither for nor against encyclopaedic semantics. Adherents of formal semantics could instead always argue that their position could be extended and enhanced without, however, having to share the premises of encyclopaedic seman-tics such as that proposed by Langacker.

Against the background of the results from Chapter 2, we can quickly deal with points one and three. For two-level semantics it was shown that the restrictions placed by the linguistic system were so severe that they prevented the inclusion of a multitude of knowledge features relevant to understanding. For methodological reasons it was thus impossible to comprehensively interpret meaning. The inevitable empirical ex-planatory inadequacy that arises as a result within the level-semantic models is in fact a

150. See also Langacker (1988b).


valid argument for lifting the categorical distinction between linguistic and conceptual knowledge. The second point, by contrast, provokes a whole range of new questions, for instance concerning the conceptual processes that participate in meaning interpreta-tion and the knowledge structures used to do this. The answers to these questions can only be found within a comprehensive theory of meaning.

Diverging from Langacker’s approach, the following presents arguments that are less sweeping but nonetheless definite (and, as I will attempt to show, problematic) prerequisites for a making a distinction between linguistic and conceptual knowledge. In an article that has in the meantime gained great influence, which Langacker also refers to, John Haiman (1980) gathered and systematically organized all those argu-ments used in the relevant literature to justify distinguishing between conceptual and linguistic knowledge. He summarizes them in six distinctions:

i. Essence vs. accidenceii. Analytic vs. synthetic truthsiii. Language vs. cultureiv. Semantics vs. pragmaticsv. Subjective vs. objective factsvi. Proper nouns vs. common nouns

Haiman elaborates in detail on each distinction (made with entirely different interests and in entirely different contexts), and ultimately comes to the conclusion that none of them can be justified. At this point I do not wish to outline the individual argu-ments, which ultimately led Haiman to the conclusion that it is not only practically impossible to make a distinction between linguistic and conceptual knowledge but also methodologically misleading (Haiman 1980: 331). Instead, I address selected is-sues concerning the attempt within two-level semantics to designate language as an autonomous cognitive domain.151 Since this attempt fails, the cognitive foundation for a frame theory is to be sought in encyclopaedic (or conceptualist) semantics. Just such an approach was initially proposed by Lakoff (1987) and Johnson (1987), although Fillmore (for instance, 1984, 1985) and Langacker (for instance, 1988b) arrived at simi-lar conclusions. Encyclopaedic semantics is aimed directly against the central premises of generative grammar.152

The first four dichotomies outlined by Haiman (and listed above) can be found in level-semantic models. They will be discussed in turn below.

151. For discussions of other attempts to distinguish linguistic and world knowledge from each other cf. apart from Haiman (1980) also Dirven and Taylor (1988) and the overviews in Scherner (1975) and Peeters (2000).

152. For instance against the idea of cognitive modularity, grammar as a generative system, and the possibility of maximal formalizability of linguistic expressions. See also the discussion in Langacker (1988c).


3.1.1 Essence vs. accidence?

One potential argument for distinguishing between linguistic and conceptual knowl-edge is revealed by the metaphors “meaning core” and “meaning periphery”, which were introduced by Bierwisch. Bierwisch and Kiefer speak of “core” and “periphery” for the first time in an early article from 1970. Bierwisch returns to this distinction in later articles, and Schwarz (2000: 38) also uses the concept of “core meanings” to illustrate the first level of her three-level model. To what extent these metaphors are relevant to our context becomes particularly clear at one point in Bierwisch and Kiefer 1970:

The core of a lexical reading comprises all and only those semantic specifications that determine, roughly speaking, its place in the system of dictionary entries, i.e. delimit it from other (non-synonymous) entries. The periphery consists of those semantic specifications that contribute to the meaning of a lexical entry without distinguishing it from other dictionary entries, i.e. of specifications which could be removed from the reading without changing its relation to other lexical readings within the same grammar. (Bierwisch and Kiefer 1970: 70)

Two pages later they explain this in the following way:

It seems reasonable to propose now that the distinction between C (E) [the seman-tic core C of a lexical item] and P (E) [the semantic periphery P of a lexical item] corresponds essentially to that between linguistic an encyclopaedic knowledge. In other words, the periphery of an entry E of the language L contains what might be called non-linguistic knowledge about the objects, facts, or properties of the speaker of L, conceptualized by the reading of E in the dictionary of L. The core of E, on the other hand, contains the linguistic knowledge associated with E, in the sense that the core includes all and only those specifications that specify the delimitation of E, within the dictionary of L.

What belongs to the “meaning core” of an expression is thus part of linguistic knowl-edge, whereas what counts as its “meaning periphery” has the status of conceptual world knowledge. If one and the same linguistic expression is specified by different semantic “cores”, it is assigned another value in the linguistic system, completely in-dependently of which elements of conceptual knowledge determine it. According to Bierwisch and Kiefer, polysemous expressions thus behave as follows:153 Polysemous expressions are distinguished solely by dint of their “semantic periphery” whereas the “semantic core” remains identical.

The decisive question is now: How is the boundary between core and periphery, between linguistic and conceptual knowledge, drawn in individual cases? However, Bierwisch and Kiefer recognize that a certain degree of exchange may occur between elements that belong to the core and the periphery, i.e. that the boundary between the two cannot be fixed for all time. Thus, they provide the example of the word spoon,

153. Cf. for instance the examples like school discussed in Section 2.2.3.


which has, on the one hand, the semantic core components ‘physical object’, ‘artefact’, ‘used to eat fluid food’ and, on the other hand, peripheral elements of conceptual knowledge such as ‘consisting of solid material’, ‘of a particular size’, and ‘not used in Asian cultures’ (Bierwisch and Kiefer 1970: 70). Although a “peripheral” knowledge element can certainly become a “core-semantic” one, Bierwisch and Kiefer must nev-ertheless adhere to the fundamental distinguishability of “core” and “periphery” if the difference between these is not to be eroded. This boundary must be of an objective and very stable nature. If it should transpire that the “semantic core” of language users varies from context to context, the distinctivity of “core” and “periphery” would be jeopardized and ultimately superfluous.

However, precisely this semantic invariance is not guaranteed. In the individual case, it is not possible either to clearly distinguish specifications of a “core semantic” nature from those of a “periphery semantic” one or to prove that the split between the two is objectively, i.e. intersubjectively, valid. In fact, if as Haiman (1980: 340) explains the “core” is characterized by only essential and the “periphery” by only accidental semantic specifications – which is what the specifications of Bierwisch and Kiefer amount to154 – then the floodgates are open to subjective, i.e. not generally applicable, semantic attributions:

This distinction [between essence and accidence] is vitiated by the fact that what we consider essential to a category seems to vary from person to person, and even more from culture to culture. (Haiman 1980: 340)

What are the essential semantic specifications of the expression spoon? The degree of curvature at the top end? The length of the spoon handle? The fact that a spoon is ide-ally suited to removing boiled eggs from boiling water (perhaps because I have only ever encountered a spoon in this function)? Or the fact that at the end of the spoon there is an engraved image of Empress Maria Theresa (because I have used such a spoon to remove eggs from boiling water my entire life and assume that everyone also does so)? What semantically “essential” specifications are and what not clearly and crucially depends on recurrent experience, and only on the basis of this is it possible to prove that one or the other semantic specification is more or less plausible. In this respect, the status of any “semantic core element” can be doubted, because historical or cultural changes give cause to or because a language user classifies this element as accidental or purely and simply because the given context demands it. This also applies to apparently simpler and clearer examples. If we define man – as Haiman does with critical allusion to Aristotle – as essentially rational and accidentally bipedal, then a non-rational man must be denied his humanity. Haiman adds sarcastically that he knows of no culture

154. This shows the structuralist influence on Bierwisch’s component-based approach to lin-guistic meaning. Bierwisch and Kiefer (1970) postulate their distinction between “core” and “periphery” in analogy to the feature-semantic conception based on essential vs. accidental, on given vs. non-given semantic specifications.


to date in which, as it were, savants were treated like animals or even slaughtered and eaten due to their irrationality.

The dichotomy of semantic “core” vs. semantic “periphery” and the contingent contrast between essential and accidental semantic specifications are thus useless as justification for the categorical distinction between linguistic and world knowledge. In fact, this is a gradual phenomenon,155 which leads us to two more explanatory options: to the dichotomy between synthetic and analytic truths and that between culture and language. Both of these are also used in two-level models to justify the autonomy of linguistic knowledge.

3.1.2 Synthetic vs. analytic truths?

Bierwisch and Kiefer (1970) only mention in passing the relationship of “core se-mantic” specifications to analytic truths and of “peripheral semantic” specifications to synthetic truths. They briefly state (Bierwisch and Kiefer 1970: 72): “We are touching here on the much disputed problem of a priori, analytic, synthetic, and empirical truth.” The fact that this is certainly intended as a serious attempt to justify the autonomy of linguistic knowledge is demonstrated by later remarks. For instance, in an article pub-lished in 1983, Bierwisch characterizes semantic representations – semantic “cores” in the old terminology – as “prerequisites that specify the conditions of possible contexts” (Bierwisch 1983a: 98). Three years later – the terminology had changed yet again, with “semantic forms” consisting of “fixed” and “created primes” – Bierwisch writes (1986: 780): “Let me call the two types fixed and created (or a priori and a posteriori) primes of SF [semantic forms].“ And he adds: “Fixed primes are part of the universal repertoire available to the language learner as his or her initial equipment on the basis of which linguistic knowledge can be acquired.”

To begin with, the reference back to Immanuel Kant’s transcendental philosophical terminology is highly questionable in terms of content. For one thing, Kant’s teachings on the conditions of possibility of human understanding “only” had the intention of specifying formal conditions, not material ones. However, “fixed primes”, of which Bierwisch gives the examples do, caus, become, and know, are obviously specified by semantic content, such that their semantic power of restriction extends far beyond

155. Cruse (1988: 79) also comes to this conclusion: “It seems undeniable that some aspects of lexical knowledge belong more intimately to the words themselves, while other aspects belong to words in a less direct fashion, being more intimately related to the things that the words refer to (or to our conception of these things). Some facts concerning word-meaning can be ascertained by observing things in the world; for instance that dogs bark, whereas cats do not … But there are some aspects of lexical know ledge that can only be studied by making some sort of observation on words: the information is not be found in the referents, nor in the concepts which mediate our dealings with referents.” Knowledge of word class is an instance of these, but not knowledge of basic meanings as in Bierwisch’s sense when he speaks of semantic “core”.


formal categorical specifications.156 In addition, Kant by no means understood condi-tions of possibility as innate and thus biologically endowed. In fact, they are conditions that result solely from epistemological reflections. Reifying them would be subscribing to a nativist fallacy. Precisely this occurs when Bierwisch attributes “fixed primes” to the innate biological endowment and argues that for this reason they cannot be learnt.

Besides these methodological discrepancies, the distinction between analytic and synthetic does not provide the necessary means for distinguishing linguistic knowl-edge from encyclopaedic knowledge. This is due to a fundamental weakness in the philosophical concept of analyticity, as noted by Quine (1979). Kant maintained that analytic truths, in contrast to synthetic truths, do not amplify understanding. However, according to Quine this can only be proven by appealing to synonymy. Thus, if an analytic truth (in contrast to a synthetic truth) is characterized by the predicate being already contained in the subject, this begs the question of the degree to which subject and predicate are synonymous. Quine (1979: 25, 29) illustrates this using, among oth-ers, the example of the expression bachelor.

(1) All bachelors are unmarried men.

When are the expressions bachelors and unmarried men synonymous? Quine’s answer is thus: When they both have the same definition. But at this point the problem arises that a definition that attempts to demonstrate the equality of meaning of definiens and definiendum must already contain antecedent synonyms, which for their part are presupposed thus remain unchallenged. Otherwise it would remain unclear how the definiens could correspond to the definiendum. This dilemma is not fundamentally altered if we assume that meaning only arises in context, since even here the problem of synonymy arises, but this time with respect to the context. Although the premise is no longer that the definiens be synonymous with the definiendum in its antecedent usage, now the entire context in which the two are used must be the same in meaning. Consequently, even here there are always antecedent, unexplained synonymy relation-ships. With each explication of a synonymy, new synonyms are required that in turn require explication. This leads to an infinite regress. Ultimately, according to Quine, the problem of the explication of analyticity cannot be resolved even if it is assumed that the synonymy of two expressions consists in the statement remaining true when they can be exchanged. In our case this means that unmarried men and bachelors are two expressions that can be arbitrarily exchanged in any context without changing their meaning. This cannot explain analyticity since a prior understanding of it is required. How can it otherwise be possible to judge whether both expressions can be exchanged while maintaining the (sentential) semantics? The explanation is circular.

Quine’s inevitable conclusion is that a distinction cannot be made between ana-lytic and synthetic truths: “That there is such a distinction to be drawn at all is an

156. The use of small capitals indicates that these are not normal linguistic expressions, but rather semantic components. According to Bierwisch, do, cause, become, know are examples of this.


unempirical dogma of empiricists, a metaphysical article of faith” (Quine 1979: 42), an article of faith that is also useless for legitimizing the distinction between linguistic and world knowledge.

3.1.3 Culture vs. language?

In the context of the first dichotomy, between essential or accidental semantic speci-fications, drawing a rigid line between linguistic and world knowledge has proven to be impossible. However, Bierwisch’s proposition of “fixed primes”, i.e. innate semantic constants, now appears to have opened up a possibility of making this distinction. Haiman (1980: 331–336) subsumes explanatory attempts such as this under the di-chotomy of culture vs. language.

The distinction between created primes and fixed primes was outlined in Section 2.2.3. Bierwisch (1986: 780–781) had already introduced this to more specifi-cally define the status of semantic specifications. Thus, the criterion that now could be used to distinguish whether semantic specifications belong to cultural or linguistic knowledge lies in their status. This is a perspective that has been of central significance to generative grammar from the beginning. In order to resolve the learnability paradox, Chomsky argues that there must be a universal grammar, an innate ability to learn language. Bierwisch now formulates this perspective for semantics.157

Without entering into the discussion in detail, the fact remains that recent research in the area of first language acquisition has increasingly found empirical evidence for the fact that linguistic structures can only be learnt inductively on the basis of lin-guistic input. This proves the learnability paradox to be figmentary. Apart from this, Bierwisch’s approach begs the question as to where, then, the information given by semantic primitives should come from if not from experience. At this point, I return to the arguments already outlined in Section 2.2.3.They are taken up again here, but not further elaborated. To what extent can, say, caus (Bierwisch 1986: 780), that is knowledge of causality, be a primitive, innate semantic component? With reference to Talmy (2000: 471–548), I already demonstrated how complex and, indeed, heterogene-ous concepts of causality are. What is more, knowing which entities are causally linked in what manner in the world is without doubt contingent on individual experiential and learning processes. (For instance we learn this painfully when we touch a hot stovetop.) What should know – another semantic component raised by Bierwisch to the status of fixed prime – “mean” without making reference to a prior understand-ing of knowledge modes?158 What should become “mean” without taking recourse to sensually perceived phenomena? Awareness of a broad variety of processes of change

157. Wierzbicka makes an alternative attempt to tag linguistic primitives as innate (for the first time in Wierzbicka 1972). Haiman (1980: 336) comments on her approach with these words: “Without experience, there is no thought, and certainly no language”.

158. On the distinction between knowledge modes, cf. Busse (1991a: 187).


arises from the fact that we perceive our bodies as the experiential centre and place it in relation to “non-selfs” in our empirical environment. Thus, become is not innate, i.e. not pre-experiential.

To illustrate the problem with another example, the verb teach cannot be under-stood because a process of lexical decomposition occurs that breaks it down into the elements do, caus, become, know (cf. Bierwisch 1986: 769), but because classifica-tion schemata based on experience are activated. The meaning of teach is deduced by means of multiple categorization processes of the type “x belongs to the category y” and “x is a type of y” (Fillmore 1984: 141). For instance that a person does something (do) belongs to the category of ‘human action’. I know that an action can cause a result (caus) because I can bring it in relation to my everyday, concrete experience; it is, so to speak, in the nature of experience type y.159 I also know that knowledge is acquired, that I was not always a knowing being, that I learn daily, that many of my parents’ ac-tions served to teach me what they had already mastered or knew, and so on. An array of categorizations participate in these reasoning processes. If we follow Bierwisch and assume that ‘become’ is relevant to the understanding of teach, this can only indicate a very abstract concept, that is, teaching understood as the transition from a state of not-knowing to one of knowing. Yet, how can ‘become’ precede my experience?

It cannot be denied that the components named by Bierwisch are relevant to un-derstanding in some form or other. However, what can be denied is that they are independent of experience, have the status of semantic primitives, and do not stand in any intrinsic connection with practical life. Fillmore formulates the departure point of the countermodel in this way:

Whenever we use a word that has a classificatory function, we interpret its use by being aware of the classificatory schema within which it has a role. (Fillmore 1984: 142)

This interpretation process occurs against the background (or in the space – cf. Section 3.2) of our experience. This does not require a set of “fixed” and “created primes”. Fillmore (1975) rightly criticizes this for being a semantic “checklist”. Such a checklist requires conditions that state when a component is part of a meaning, when it belongs to the “meaning core” or the “meaning periphery”, when it is an analytic or synthetic part of a concept, or when it is part of our biological endowment or culturally contingent. To avoid having such an infinitely long checklist (which precisely cannot explain how meaning is understood), criteria would also need to be given for the ex-clusive and exhaustive specification of components. In addition, within all of this, no access to encyclopaedic knowledge would be permitted to take place. As we have just seen, this undertaking does not even succeed with the attempt to reason that linguistic knowledge is part of our biological endowment.

159. This also results in the ability to distinguish between perlocutionary verbs, such as to con-vince sb., and illocutionary verbs, such as to hit sb.


3.1.4 Semantics vs. pragmatics?

Finally, there is one last option to illustrate the extent to which linguistic knowledge enjoys – at least relative – autonomy. This is based on the idea of a dichotomy between semantics and pragmatics based not only, as is usually the case, on the practical pur-poses of research, i.e. on heuristic reasons, but also on systematic reasons. Certainly, in important points this argumentation draws on aspects that proved to be problematic in the previous section. This concerns, above all, the proposition of a strong composition-ality principle (that is, Frege’s original hypothesis that the meaning of a complex expres-sion is determined by both the meaning of its parts and the manner in which they are composed). Nonetheless, the dichotomy between semantics and pragmatics presents a rather self-contained argumentation because, in contrast to the others, it does not attempt to legitimize the autonomy of semantics by dint of its nature, but rather does so by highlighting the independence of other factors, namely pragmatic ones. The extent to which such a viewpoint can indeed be justified has yet to be discussed.

According to Bierwisch (and thus all adherents of level-semantic models), lin-guistic meanings do not “emerge” within communicative practice; rather they are the result of pre-communicative, autonomous rule processes:

And in this sense, my hypothesis is that linguistic meaning enters into linguistic action as its own determination framework and cannot be reduced to this by sur-rendering an abstraction. (Translated from Bierwisch 1979: 64)

Thus one must conclude that the communicative purpose of such “linguistic meanings” constituted each time a linguistic action is performed are subject to autonomous rule processes, namely such that occur subsequent to those rule processes that constitute “linguistic meaning”. Bierwisch summarizes this as follows:

Language behaviour as the linguistic expression of thoughts and as the compre-hension of linguistically formulated thoughts is determined by the interaction between language knowledge and everyday knowledge. Communicative language behaviour is determined by the interaction of both systems with the awareness of social interaction structures. (Translated from Bierwisch 1979: 65)

However, instead of justifying the autonomy of the linguistic determination frame-work, i.e. of so-called “linguistic knowledge”, via the indirect route of its status (as with the postulate of “fixed primes”) or via the compositional character of linguistic mean-ings, here the argumentation takes another turn. The claim is that, apart from a purely linguistic meaning dimension that determines “utterance meaning”, there are prag-matic functions that specify the “utterance meaning” (Bierwisch 1979: 66). The crucial step is then a small one. According to Bierwisch (1979: 65–66), the process of inferring the communicative purpose of an expression sequence runs through three levels, each with their own knowledge types and determination frameworks: firstly, the level of “linguistic knowledge”, which fixes the “linguistic meaning”; then the level of “utter-ance meaning”, which contributes “everyday knowledge” for the semantic specification


of linguistic meaning; and finally the level of “communicative purpose”, which feeds “knowledge of interaction conditions” into the act of understanding, thus enriching the utterance meaning to become a unit with communicative content. Accordingly, in this conception linguistic knowledge and the conceptual system obey different rule systems, which would equally necessitate different theories.160 The borderline between “linguistic knowledge” on the one hand and “everyday” and “interaction knowledge” on the other, collates roughly with that drawn between semantics and pragmatics. Thus, as Bierwisch makes clear on many occasions, metaphorical constructions do not belong to the narrower area of semantics, but are rather secondary phenomena in that they are based on conceptual processes that are in turn predetermined by linguistic knowledge. In later articles Bierwisch reinforces the path of modularist argumentation he has taken, but dispenses with a more detailed definition of the pragmatic compo-nents, limiting himself largely to the analysis of the first level.161

In my view, a whole array of reasons speak against making this type of generalized distinction between semantics and pragmatics, and I will attempt to summarize three of these lines of argumentation.

The first reason is based on an observation that Haiman makes in his critique of non-encyclopaedic semantic conceptions. When Bierwisch propagates a methodologi-cal distinction between semantics and pragmatics, this occurs, as Haiman formulates it, in the sense

…that the distinction between them is not one of mere administrative conveni-ence (like the difference between syntax and morphology, or between morphology and phonology), but one which is justified in principle. I believe that the mainte-nance of this point leads to a view of semantics which is indistinguishable from Chomsky’s view of syntax and hence entirely vacuous. (Haiman 1980: 344)

Haiman is by no means referring to level-semantic models, and yet he hits a central point of Bierwisch’s approach. The autonomous status of semantics in question is, in Bierwisch’s model, definitively motivated by Chomsky’s syntax theory. That is to say, which selection restrictions underlie a linguistic expression is already determined within syntax. However, Haiman rightly points out that it is extremely dubious whether

160. That is to say, firstly, a “theory of language” that investigates the “phonological, mor-pho-syntactic, and logical structure of language; secondly, a “theory of everyday knowledge” (“Alltagskenntnis”) that covers the “construction of the conceptual system”; and thirdly a “theory of social interaction” that analyses “the structures of interindividual actions” (translated from Bierwisch 1979: 64). Accordingly, a theory of conceptual world knowledge would investigate the “conceptual system” and “action structures”.

161. Most clearly in the previously mentioned formula “(((ins (phon, syn, sem)) ct, m) ias, cs)” (cf. Bierwisch 1983b: 33ff.), according to which a speech act is the result of a linearly conceived determination chain in which the context (“ct”) and the interaction relation (“ias”) only plays a role at the end.


selection restriction can be syntactically motivated at all.162 To illustrate this, Haiman used one of Makkai’s example sentences (1971: 487):

(2) The rock is pregnant.

Certainly, the sentence subject breaches the permissible selection restrictions of preg-nant, since it contains the component [– animate]. However, on what basis can it be determined which components belong to the expression? And what leads to the as-sumption that the sentence subject in (2) is being used metaphorically? There is no need to add to Haiman’s commentary:

But the categorization of rocks as inanimate and hence, a fortiori, barren, is a belief about the world, and one which is not necessarily shared by everyone. (I happen to know at least one myth, the Hittite story or monster Uli Kummi, in which a rock does get pregnant.) (Haiman 1980: 347)

If the semantic decomposition of an expression can only be carried out with an implicit reference to its metaphorical mode of usage, then this particularly demonstrates its dependence on conceptual knowledge.

Makkai argues in a very similar way (1971). He draws attention to the fact that there is no principle of semantic deviance, that semantic selection restrictions and as-sumptions about the world are not to be separated, and that as a result it is solely the latter that determines what is semantically acceptable and what not. Makkai speaks in this context of the “contextual adaptability principle” under which the apparently most semantically anomalous expressions can make sense163 if – as Olson formulates it (in translation) – an “intellectual rearrangement of our perceptual experience” occurs, i.e. if an appropriate context can be constructed, “in which this sentence specifies an intended reference object” (Olson 1974: 185).

What is decisive is what is considered to be normal, and this is a culturally vari-able quantity that only makes sense within a system of presumed convictions and as-sumptions. Contrary to the much advocated hypothesis, Chomsky’s famous example sentence Colorless green ideas sleep furiously need not by any means remain without meaning; it merely appears absurd without any contextual embedding. For instance

162. The analysis of numerous examples in Section 2.3 already showed that this is not possible. The scepticism about whether selection restrictions are syntactically motivated stems from an early study by Fillmore (1970) on the grammar of the verbs hitting and breaking. Fillmore (1970: 131) arrives at the conclusion: “It looks very much as if for a considerable portion of the vocabulary of a language, the conditions determining the appropriate use of a word involve statements about the properties of real world objects rather than statements about the semantic feature of words.”

163. Makkai demonstrates the effectiveness of the “contextual adaptability principle” using a sentence that can hardly be outdone in terms of alleged semantic deviance, yet, under certain – in this case, cultural-spiritual – contextual assumptions is semantically well formed. The sentence is: “It was five years before he was born and seven years after he died that the baby divorced his grandmother.”


the referential expression ideas could easily specify a reference object in the context of a political satire. One could imagine that the word ideas were being metonymically applied to refer to people who had ideas, while colorless meant something like without contours and the attribute green referred to a political ethos. The fact that these ideas, also metaphorically speaking, sleep furiously then means that the large coalition part-ner – for instance in the German context the Social Democrats – has severely sanc-tioned the ideas of green politicians for the time being, while the latter are anything but satisfied with the situation and are only waiting for a politically opportune moment to “reawaken” their ideas.164

Olson appears to have such examples in mind when he writes that “the meaning of an expression depends on the context of the alternatives” (Olson 1974: 192–193).165 Consequently he argues that it is precisely perceptual knowledge that forms the basis of meanings (Olson 1974: 186).

Let us address the second argument that speaks against separating semantics and pragmatics. This, too, is of a methodological nature and is directed at the same cir-cumstances from another perspective. In brief it states: Knowing which semantic com-ponent belongs to a linguistic expression means knowing which facets of knowledge this elicits in specific knowledge contexts. This objection can be illustrated using the expression bachelor, which has acquired a certain fame. Fillmore (1977a: 67) lists a number of different meaning variants of the same word (in American English), among which are the following:

(3) A bachelor is an unmarried adult male human being. (4) A bachelor is a knight bearing the banner of another knight. (5) A bachelor is a male fur seal without a mate during the mating season.

In contrast to (4) and (5), (3) is also a common usage variant of the expression Junggeselle in German. In addition the following meaning spectrum can be identified:

(6) A Junggeselle is a young journeyman. (7) A Junggeselle is a man who conducts a restless lifestyle that diverges from the

norm.

Bierwisch also mentions the example Junggeselle in several places (e.g. in Bierwisch 1979: 78) and notes that its semantic decomposition reveals three components: [+ un-married], [+ adult], [+ male human being]. But how do proponents of level semantics arrive at such results? They would have to argue that the word Junggeselle in (3), (6), and (7) is not a case of systematic polysemy (in the sense used by Dölling 2001), but

164. In this context, see also the frame-theoretical considerations in Minsky (1975: 231).

165. One paragraph later he adds: “Speaking generally, an expression can apparently never exhaust all the potential of a perceived reference object. New properties can be found for any event as long as the event is confronted with a different set of alternatives.” (Translated from Olson 1974: 193)


simply a case of homonymy, because the “linguistic meaning” of this expression dif-fers in all three cases – and the same argument would apply to the American variant bachelor with its corresponding Examples (3), (4), and (5). However, this means know-ing which facets of knowledge are central in which knowledge context. Otherwise it would not be possible to identify why (6) and (7) are not proof of the polysemy of Junggeselle. Yet, the knowledge necessary to do this can only result from pragmatic knowledge of the various usage forms of the expression. Thus, in (6) – the common meaning of Junggeselle until into the 16th century – the profession of a person shifts into the foreground, while his marital status is of little significance. Understanding what the expression Junggeselle in (7) refers to presupposes knowledge of certain ways of life, but not necessarily assumptions about the marital status or age of the person in question. In both cases the context in which the expression Junggeselle occurs governs what is meant by it, and knowledge of life circumstances (and not “purely linguistic” knowledge) is crucially involved in this process.

If we ignore the case of polysemy and only concentrate on Example (3), a very similar difficulty also appears with contextually abstract analyses of meaning. Both Fillmore (1977: 67) and Lakoff (1987: 70) draw attention to numerous background assumptions that must be fulfilled apart from the three components in order to label a person as a bachelor. It would be most peculiar to use the word bachelor to describe the Pope or a man who has lived with a woman for ten years despite not being married, and this although both cases deal with unmarried, adult male human beings. Fillmore (1982b) provides another example in another place: it would make just as little sense to use bachelor to describe a boy who was abandoned in the jungle and grew to maturity there far removed from our social norms. As absurd as such examples may appear, they nonetheless demonstrate how far background assumptions extend for constituting understanding. They encompass assumptions about social, institutionalized practices (marriage, divorce, celibacy) just as much as perceptions of socially relevant norms (heterosexuality vs. homosexuality, monogamy vs. polygamy). And that means that purportedly purely semantic components are ultimately pragmatically founded and in themselves completely arbitrary. Background assumptions that are relevant to under-standing should therefore already be accounted for on the level of “linguistic meaning”. However, then the distinction between linguistic and conceptual knowledge would be superfluous. Rudi Keller summarizes this dilemma as follows when he discusses level semantics from a semiotic viewpoint:

But how do speakers and hearers know which “logical form” “specified” by these features is “denoted” by the expression Junggeselle? And through what features is the expression in a position to “denote” precisely this form? The answer to the second question is the secret of all representational approaches. The answer to the first question appears to me to be: speakers and hearers know that Junggeselle represents this set of features (if they know it) because they have learnt to use the word Junggeselle to refer to unmarried, adult male human beings or to state that adult men are unmarried. (Translated from Keller 1995: 191)


The knowledge of the “linguistic meaning” of an expression is itself founded in com-municative practice.166

The third argument is already inherent in the quote by Keller. Semantics and prag-matics can only be meaningfully separated from each other if it is possible to exclude references to reference objects from the research area of semantics and qualify them as purely pragmatic. After all, the reference objects of linguistic signs are not extra-linguistic entities (objects, circumstances, persons, actions, events, etc.), but rather cognitive constructs of these, and such constructions, which are indeed always already structured or profiled with respect to relevance areas, are based on conceptual world knowledge.167 For this reason, the level of linguistic meaning in modular models must remain free “of any referential ‘impurity’” (Kleiber 1998: 13). However, this does not succeed, as Kleiber explains, even if an attempt is made to disregard the referential in-terpretation of semantic features and exclusively work them out by juxtaposing lexemes and specifying them in terms of distinctive features. The reason is that the meaning must already be known before the lexemes can be confronted with each other. A theory of linguistic reference is not outwardly a meaning theory, but conversely an integral and indispensible component of one.

Since a semantic theory is not possible without taking referential knowledge into account, but a linguistic theory without substantiated semantics remains vacu-ous, it is thus not possible to conceive a meaningful linguistic theory departing from the axiom of a prior distinguishability between linguistic and extralinguistic knowledge … (Translated from Busse 1997a: 20)

166. However, Keller does not differentiate between modularist and holistic models. At one point, therefore, his criticism of “representationalistic” approaches, as he calls them, evidently subsuming cognitively oriented approaches per se (Keller 1995: 82–86), well overshoots the mark. It is a fundamental error (and a false interpretation of Langacker) to claim that holistic approaches assume “the only way to avoid an objectivist, truth-functional semantic conception [consists] in shifting meanings into the head of the speaker” (translated from Keller 1995: 84). On the contrary, it is widely known and frequently discussed that in contrast to generative ap-proaches, the departure point of cognitive and construction grammar is the so-called “usage-based model” (cf. the overview in Barlow and Kemmer 2000 and Bybee 2013). Its premise is based on precisely what Keller claims for his part: that meanings (as well as grammatical phe-nomena) are “result[s] of the conventionalization of previous communication practice” (trans-lated from Keller 1995: 193). In a wide variety of areas, studies have been carried out under the basic cognitive-holistic premise that linguistic knowledge is solely a result of language usage, for instance studies on language acquisition (Behrens 2005; Lieven et al. 2003), morphology (Bybee and Slobin 1982; Bybee 1985), semantics (Langacker 1988c; Tuggy 1993), and phonology (Bybee 2001; Taylor 2002: 143–160). Also see my observations on this theme in Section 6.1 and 6.5.

167. Consider, for instance, the simple example of adjective-noun constructions. If we speak of a red light, the attribute is ascribed only on the basis of our foreknowledge: the (traffic) light as a whole is not ascribed the colour red, but only the light of one lamp; this presupposes knowledge about the composition and function of traffic lights. Compare such expressions as red button, red alert, red car, etc.


Of course, the distinctive features of expressions could be determined, for instance, with the help of a commutation test,168 but not without including the extralinguistic conceptual reference area of these features.

In addition, not only must the standard semantic interpretation of an expression already be available to semantically specify an expression, but, in individual cases, conceptually specific background assumptions must also be activated to first specify the relevant area. What, for instance, does the expression “attentive waitress” mean? Depending on the context, the meaning may vary so starkly that it has nothing more to do with the standard interpretation. Imagine, for example, a bar that is known for having a male transvestite waiting on tables. The bar has the distinction that the guests pay an entry fee and then serve themselves with drinks. The waitstaff ’s job is to pay attention that no full glasses stand unobserved, since there has been a recent spate of guests’ drinks being spiked with drugs. What does it then mean if a guest calls the staff member an “attentive waitress”? Most likely, above all, that a man in a bar is collecting drinks rather than serving them.169 Presuming that no metaphors are involved, the meaning actualization occurs solely against the background of an intersubjectively shared life-world. The fact that the waitress is a man and is attentive by dint of the fact that she does not serve her clients raises no problems. The “principle of contextual adaptability” mentioned previously can, as in this example, even lead to linguistic expressions establishing reference to cognitive models (or frames) that completely contradict the standard interpretations. Of course, there needs to be a minimum of conventionalization of the referential scope of linguistic signs. However, it is by no means necessary to have pre-fixed meaning components that demand validity in every arbitrary context. This would not do justice to the creativity of linguistic meaning construction and in individual cases would even inhibit a contextually appropriate reading (like the component [+ female] for the lexeme “waitress” in our example). For every component, a context that annuls it could be given.

This also shows that semantic knowledge is not simply to be understood as a modular determination framework that can be uncoupled from pragmatic factors. The meaning of every expression depends on the conceptual structure that it refers to.

A word’s meaning can be understood only with reference to a structured back-ground of experience, beliefs, or practices, constituting a kind of conceptual prerequisite for understanding the meaning. Speakers can be said to know the meaning of the word only by first understanding the background frames that motivate the concept that the word encodes. Within such an approach, words or word senses are not related to each other directly, word to word, but only by way

168. Although I would tend to speak of “distinctive background assumptions” in this context, on the one hand to avoid conceptual-ontological implications and on the other to clarify that the prescribed methodological framework does not subscribe to the premises of a componential theory.

169. Cf. also Searle’s familiar discussion of the sentence The cat is on the mat in Searle (1998) and Coulson’s frame-semantic reinterpretation of the same example in Coulson (2001: 4–10).


of their links to common background frames and indications of the manner in which their meanings highlight particular elements of such frames. (Fillmore and Atkins 1992: 76–77)

Words do not carry meaning themselves (not even in the most abstract sense), nor do they carry meaning by maintaining relations to other words, at least not in the structuralist sense of purely langue-based meaning relations. They become meaningful units when they are set in relation to a (background) frame. Rather than a metaphor of “level”, Fillmore thus argues for an “interpenetration model” in which encyclopae-dic knowledge stands at the beginning of the interpretation process and from which purportedly “purely” grammatical knowledge cannot be isolated. Interpenetration occurs in such a way

…that linguistic knowledge, including lexical and grammatical knowledge, en-ters at each of these levels, not only at the top, and that knowledge about worlds, contexts, and speakers’ intentions enter at every level, not only at the bottom. (Fillmore 1984: 132)

In the sense of a holistic language theory, this could also be formulated in this way: All linguistic means that are available to us deliver a specific contribution to the constitu-tion of meaning and ultimately serve to facilitate the transition from what is said to what is meant.

Finally, there is only one conclusion that can be drawn from the four attempts to strictly separate linguistic and conceptual knowledge that have been discussed in Sections 3.1.1 to 3.1.4: Since all demarcation attempts have failed for the most varied of reasons, a cognitive theory of language must take into account that humans are born into a specifically structured life-world (in the broadest of socio-cognitive senses) and only learn to use a language as a social instrument in interaction with this. Under the influence of the innateness dogma, level-semantics proponents have lost sight of this banal fact. Suggesting that frames can be integrated into a level-semantics model, as Bierwisch does on several occasions, is therefore entirely misleading and supposes the compatibility of the incompatible. Bierwisch suggests, for instance, positing frames as representational formats of social interaction knowledge, ignoring the fact that Fillmore’s approach diverges fundamentally in all of the previously described aspects from his modular determination model. He claims that Fillmore gives

…a range of examples of a similar nature, for the analysis of which he uses the concept “frame”, which originates from AI research, in a similar, if somewhat more specific, sense to that of CA (t) [= representation of interaction knowledge]. (Translated from Bierwisch 1979: 78)

However, according to Bierwisch, before an interpretation based on interaction knowl-edge is arrived at, two modularly organized determination levels have already been passed through.170 Apart from the fact that frames would, as a result, only have a

170. Concerning this cf. Figure 1 Section 2.2.1.


marginal gap-fill function, they would not be able to be modularistically integrated on any level for general methodological reasons.

Fillmore had already emphatically subscribed to the counterposition of an ency-clopaedic, holistic approach semantics at a time when the basic idea of frame semantics had not yet been developed. The (almost cynical) concluding remarks in his article A grammarian looks to socio-linguists state:

My proposal for determining the boundary between linguistics and other disci-plines connected with a speaker’s control of language use is to start a project for writing an instruction manual for an Immigrant of the sort I have in mind, to have on this project a large and capable research team with workers from a great many academic disciplines, and to determine empirically which tasks the linguists can carry out without any help from the others. (Fillmore 1973: 285)

The remaining part of this book establishes just such a holistic, encyclopaedic approach to semantics that is not able to lay claim to disciplinary autonomy.

3.2 The “space of understanding” (C. Demmerling)

Linguistic or conceptual knowledge? Obviously this is the wrong question. A approach can only be non-reductionist if it questions the possibility of distinguishing between linguistic and conceptual knowledge. Differentiation and typification of knowledge relevant to understanding must follow a different guideline. However, before coming to this, I first outline the departure points of the holistic countermodel to level semantics. To do so, I draw on the metaphor of “space of understanding” (‘Raum des Verstehens’) coined by the philosopher Demmerling (2002).

The previous argumentation has initially shown two epistemological conse-quences. No area of semantic knowledge can be shown to exist prior to understand-ing, and equally no fundament of understanding can be identified that is not itself both the result and object of understanding. Neither of these points is given validity in generative and purely formal-semantic approaches. In a multitude of forms this reflects the assumption that linguistic structures exist prior to understanding. Syntactic deep structures consist, for instance, of a repertoire of abstract symbols and recursive rules that determine the combinatory potential of these symbols. However, neither the rules nor the symbols must be understood, in fact they cannot be understood, because by dint of their modular nature, they enjoy an autonomous status and precisely for this reason are not subject to semantic processes.171 The situation is no different for the semantic primitives postulated by level semanticists; they are primitive due to their property that they themselves need not be analysed.

171. The same criticism of formal and generative approaches prompted later adherents of cog-nitive and constructive grammar to classify syntactic constructions not as autonomous but as meaning-relevant and thus conceptual phenomena. The line of determination between syntax and semantics is reversed; cf. the agenda-setting remarks in Langacker (1987).


The postulate of U-relevance questions altogether the existence of such basic struc-tures claimed to precede understanding. Instead of assuming that innate language-understanding principles, that is, principles that pre-exist communicative practice, guide the process of understanding, this task is performed by cognitive schemata (or frames).172 Such patterns are emergent structures that are acquired by means of induc-tive abstraction processes from authentic language material. Such processes can be ver-ified in a multitude of ways in first-language acquisition. The rationale of usage-based approaches in language acquisition research is that linguistic categories can be acquired inductively on the basis of the information distribution and the token frequency in the language input (Tomasello 2003; 2006: 27–32). The basic psycholinguistic unit with which small children learn a language is the utterance. Accordingly, communicative units constitute the database for empirical analysis. Taken representatively for usage-based approaches overall, Michael Tomasello concludes:

Following the general strictures of cognitive linguistics, to identify the fundamen-tal units of language use we must begin with basic processes of human cognition and communication. Following the general lead of many functionally oriented theorists, my candidate for the most fundamental psycholinguistic unit is the ut-terance … An utterance is a linguistic act in which one person expresses towards another, within a single intonation contour, a relatively coherent communicative intention in a communicative context. (Tomasello 2000: 63)

What Tomasello states here using oral communication as an example is naturally analogous for the understanding of written utterances; however, from an ontogenetic perspective these are secondary for language or meaning acquisition.

Level-semantic models are under criticism not only due to the lack of attention to communicative units that this implies. It has already become evident that their em-pirical explanatory inadequacy renders a radical revision necessary. This was shown in the discussion of Bierwisch’s and Schwarz’s approaches in Section 2.2. A theory of meaning can only have explanatory adequacy, in the sense that it approaches an exhaustive description of knowledge relevant to understanding, if it proceeds holisti-cally. Syntactic constructions are not subject to their own rule system, but rather fulfil meaning-relevant functions because they are form-meaning pairs.173 This constitutes

172. Cf. Behrens (2006: 3): “In constructivist and emergentist approaches, no specifically lin-guistic innate representations are assumed. Instead, it is argued that children are very efficient pattern and intention recognisers so that they can induce linguistic structure based on the language they hear.”

173. Cf., for instance, Goldberg (1995: 4). Lutzeier (1991: 83) writes in reference to his so-called CRMS theory, which is very similar to Langacker’s conception of Cognitive Grammar: “We engage in syntax not for the sake of syntax, we engage in syntax for the sake of semantics”. The hypothesis concerning form-meaning pairs as minimal units for syntax as expressed in this is outlined further in Section 4.1.1 and 4.1.2.


the semantic turn of more recent cognitive approaches such as construction grammar and Cognitive Grammar.

But what does it mean to understand an utterance? How do we grasp meanings conveyed by language if only a fraction of the knowledge relevant to understanding can be determined by means of the principles of compositionality? Using the metaphor of “space of understanding”, the following will clearly show what epistemological conse-quences the postulate of U-relevance has. The core hypothesis is that every linguistic utterance is embedded in a “space of understanding” and can only be sufficiently in-vestigated by including those conditions that constitute this space.

To begin with, from Chapter 2 we can note that semantic interpretations of linguis-tic expressions do not obey the restrictive specifications of purely linguistic knowledge, not even “only” in the first instance or on a first level. Rather, from the very beginning the entire knowledge that a language user has at a particular time is at their disposal. Demmerling attempts to express this multidimensional framework of conditions by using the metaphor of “space of understanding”. He departs from the premise that we, as understanding individuals, have always moved and oriented ourselves in this space. We are, as it were, locked in this space.

It is the space of understanding that carries our usage and our understanding of language, as well as our understanding in general. The space of understanding is just as broad and inexhaustible as the space of life and encompasses everything that we think, say, experience, or do. (Translated from Demmerling 2002: 161)

Demmerling’s use of the space metaphor refers on the one hand to hermeneutic phi-losophy (in particular that in the vein of Heidegger and Gadamer), which speaks in this context of “pre-understanding” or of the “fore-structure of understanding”.174 On the other hand, he refers to philosophers like Richard Rorty, John Searle, and Charles Taylor, who prefer the term “background” to designate the same circumstance. However, by using the metaphor “space of understanding”, Demmerling consciously accentuates a perceptual-sensorimotor factor that is inadequately represented in the concepts of “background” and “pre-understanding”, yet from the perspective of holis-tic-cognitive approaches it deserves particular attention.

The mode of operation of the non-propositional in the space of understanding, be it a situation, be it a linguistic utterance, an action, or a person, is closely bound with our sensual perception: the capacity to perceive situations with all our senses belongs to this; what also belong to this are the multifaceted possibilities of relat-ing affectively to situations, utterances, actions, or persons, or ultimately how our physical being overall constitutes a sounding board for the space of understanding. (Translated from Demmerling 2002: 162)

174. With respect to forming linguistic theory cf. Scherner (1984: 57–71; 162–176). He gives an early overview of “non-textualized prerequisites to understanding” in Scherner (1979).


The fact that the world is always accessible in this manner within our (also linguistic) performance of actions and life, does not at all mean that non-propositional types of givenness (such as affective references to situations or persons, the perceptually ex-perienced, or phenomenal knowledge of oneself) are not transferrable into linguistic expressions. Quite the contrary, it is only through this that they become units in the consciousness, and only in linguistic, i.e. propositional, form can non-propositional elements be discussed. Nonetheless, states of mind also exist even if we cannot linguis-tically objectify them, although it is only through this process that they first gain the status of epistemic facts. Retrospectively making them the object of the consciously experienced in the form of convictions does not mean, however, that they were previ-ously given in the form of propositional structure, but rather merely that they can, in principle, be transferred into convictions or assumptions – that is, into linguistic structures.

Taking recourse to Demmerling, what this can cast into doubt is the possibility of reducing understanding to linguistic understanding, that is, to linguistically medi-ated and thus propositionally structured references to the world. In his view, “space of understanding” extends significantly further. Not only is it impossible to draw a boundary between linguistic and conceptual knowledge, but a categorical distinction between propositional and non-propositional references to the world proves to be equally untenable from the perspective of a theory of understanding.175 Both belong to the “space of understanding”; however, the area of the non-propositional does not belong to the narrower research focus of linguistics (but in fact to perceptual psychol-ogy and, if applicable, to neuropsychology; cf. Lakoff and Johnson 1999). One should only bear in mind that an abstraction of non-propositionals already amounts to small-scale reductionism.

The fact that feelings and non-propositional attitudes can become elements of knowledge relevant to understanding is sufficiently well known from interactional lin-guistics, for instance in the area of doctor-patient communication or institutional com-munication.176 What is more, it is this dimension of understanding-relevant knowledge that “image schemas” refer to (Johnson 1987; Lakoff 1987; Lakoff and Turner 1989). Image schemas already unfold their understanding-relevant function in the area of the preconscious. They are based largely on early childhood bodily experiences that pre-figure neuronal structures and in this way lay the foundation for the emerging ability to understand linguistic expressions (cf. the summary in Mandler 2005; Rohrer 2005).

175. This view is shared by Busse (1994: 222), who speaks of the “non-reducibility of language processing to the interpretation of symbols or signs”.

176. Fillmore has already alluded to this circumstance: “In the first place, I would like to inte-grate a semantic theory with a theory of language understanding… I find, however, that this frequently involves appeals to information of the kind that is not strictly information about language. … In the second place, I am convinced that an enormously large amount of natural language is formulaic, automatic and rehearsed, rather than propositional, creative or freely generated” (Fillmore 1976a: 8f.).


Apart from non-propositional structures of this nature, which only play a subordi-nate role in the further course of this book, the “space of understanding” encompasses the various convictions we have, intentions we bear, behaviours we are accustomed to, knowledge we have acquired, and the abilities and capacities we possess to apply all this information successfully in our everyday lives.177 Overall these are the “resources” available to us to successfully manage the complexity of everyday life. They do not emerge within language, but nonetheless act on this in a multitude of ways, for instance, in the case of oral communication imparted by paraverbal elements such as intonation, prosody, and accent or by nonverbal elements such as eye contact, gesture, and facial expression. To maintain the principle of “perceptual constancy” (“Sinnkonstanz”, as introduced by Hörmann 1994: 179–212) in the interpretation of linguistic expressions, there is no knowledge base that could in principle be considered as irrelevant for in-terpreting meaning. Hörmann himself already noted:

Meaning is not an association, but rather the knowledge of relationships (as ac-tualized in acts of thinking and understanding)…. What is meant is a cognitive availability in the sense that correspondences between experiences, actions, and the consequences of actions can, with more or less conscious reflection, be made the basis of actions and experiences that are meaningfully built on this. (Translated from Hörmann 1977: 121)

I return to the aspect of the cognitive availability of knowledge at a later point in con-nection with issues of inference theory.

Ignoring suprasegmentalia and non-linguistic factors and taking only the simplest case of a contextually abstract word-semantic specification, the result is by no means any less complex. For instance, to know what the word (apple) core means (i.e. to know how to use the word appropriately), it is necessary to have the conviction that apples (or pears, etc.) are edible and digestible and that they have particular properties; it is also necessary to have knowledge of the manner in which apples are usually eaten in our culture; and finally an array of abilities is required in order to eat an apple at all. If the core were also eaten – in the same way as the fruit flesh of an apple – there would pos-sibly be no distinction in everyday language between the expressions apple and core.178

177. The philosopher Oliver Scholz remarks in this context that we cannot simply dissolve “the interrelation of the meaning concept with the practice of understanding”. And he contin-ues: “When the meaning of an expression is explained, the correct usage of this expression is explained. How a word, a turn of phrase, or a sentence is actually or usually used is not all that is explained, but how the relevant expression is to be used and to be understood is explained. A normative concept of correctness is at play here.” (Translated from Scholz 1999: 277–278)

178. Fillmore (1976a: 26) has more to say on this: “[I]f a language has a word, there must be some category of thought, identified by an associated cognitive schema current in the speech com-munity, which this word activates”. And at another point, on the same example of the apple core, Fillmore (1984: 138) notes: “In a sense, knowing what is meant by ‘apple core’ requires knowing something about how people in our culture eat apples. If we ate apples, seeds and all, straight through rather than around the middle, we would probably not have formed such a category”.


When, within the space of understanding, Demmerling distinguishes between convictions on the one hand and skills and abilities on the other, this corresponds to the cognitive-semantic distinction between knowledge representation and knowledge interpretation or knowledge structure and procedure that has already been introduced in connection with the concept of representation. Demmerling (2002: 168) also stresses that the junction between the two is fluid, yet not seamless. We only have convictions or assumptions because we have the ability to situate them in the space of understanding and relate them to other elements there. Conversely, the ability to situate something in the space of understanding presupposes assumptions about the world or about the properties of the space.

Demmerling (2002: 174) uses the example of the apple to illustrate another fact: “An utterance always ‘hangs’, so to speak, in a particular type of situation.” What is precisely meant by “situation” remains unclear at that point; however, against the back-ground of his explanation of the space of understanding, Demmerling appears to mean an orderly structure of elements in the space of understanding, e.g. the assumptions that are relevant in order to understand the word core. The concept of “situation” can thus be seen as similar to Fillmore’s earlier conception of “case frames” and their later frame-semantic extension.179 As Tesnière did before him, Fillmore stresses the central function of the verb within a sentence. Both Fillmore and Tesnière greatly empha-size the fact that the verb evokes a scene or situation by requiring particular actors (with Fillmore reformulating this as deep case). Tesnière compares this with a stage play: There is a circumstance that stands in the foreground and a set that forms the background to the circumstance, which although not itself part of the circumstances nonetheless cannot simply be ignored in terms of understanding the action or process about to unfold. It is no coincidence that the stage-play metaphor also plays on the interaction between propositionality and non-propositionality.180

As the analyses in Chapter 2.2.3 have already shown, a considerable degree of se-mantic knowledge is always required, and this cannot be deduced from the context-free

179. Cf. the outline in Section 4.2.3. On case frames Fillmore (1982a: 115) retrospectively writes: “In particular, I thought of each case frame as characterizing a small abstract ‘scene’ or ‘situa-tion’, so that to understand the semantic structure of the verb it was necessary to understand the properties of such schematized scenes”. The concept of “situation” already occurs in rela-tion to frame semantics in Fillmore (1975: 127), however, succeeded by the concept of “scene”. According to Fillmore, if a word is associated with a cognitive scene, a wide range of conceptual experience units is meant, from visual scenes and bodily experiences to convictions about the world (Fillmore 1977a: 63) although one can speculate to what extent a person can have convic-tions or “human beliefs” that are not only mediated by language.

180. The example apple core clearly shows the parallel between his frame conception and Demmerling’s theory of space of understanding. Overall, Fillmore notes (1977a: 75): “Such knowledge [that is relevant for understanding an expression] depends on experiences and memories that the interpreter associates with the scenes that the text has introduced into his consciousness”.


analysis of the corresponding lexical unit, even though this is crucial for understand-ing the expression. As a member of a particular culture, every language user has this knowledge in a standardized form because it is part of the experiential circumstances that recur there. One could almost believe Demmerling is referring to frames when he writes:

Neither the structure of the language, our knowledge of the relationships between sentence and word meaning or corresponding rules for composing sentences from words, nor even “mental” or organic foundations of linguistic competence are capable of providing such clues [skills and abilities]. (Translated from Demmerling 2002: 165)

Actions and activities only gain their meaning by being able to be embedded in larger, always familiar contexts. It is not until actions or linguistic utterances are embedded in a more comprehensive space that they obtain the participant per-spective that is indispensible for appropriately understanding the meaning of an action or utterance. (Translated from Demmerling 2002: 166)

Demmerling concludes from the ineluctability of such “embedding structures” (Scherner 1984: 239) that the space of understanding is inexhaustible and for this rea-son is not able to be fully grasped in the context of a theory, because its epistemic elements constantly reposition themselves in this “space” and build up complex rela-tionships with each other. A further clue to support the hypothesis of inexhaustibility is that again every (semantic) explanation departs from premises that in turn equally require explanation; ultimately explanations themselves are linguistically composed and consequently by no means self-explanatory.181 The inexhaustibility of the represen-tation consequently affects the “space” as a whole to begin with; sub-areas or sections of the “space” certainly do allow an almost exhaustive description, for instance words or sentences used and actions performed.

Understanding a linguistic expression includes actualizing knowledge that is relevant to understanding and presupposes the ability to correlate this knowledge. “Convictions and abilities or skills, propositional and non-propositional as well as absolute and relative prerequisites structure this space” (translated from Demmerling 2002: 181). Convictions (or background assumptions) constantly figure as partial struc-tures within an entire network of convictions. The assumption, for instance, that apples are edible and digestible indicates assumptions about our biological disposition; these presuppose in turn convictions about the way in which human organs like the stomach and intestinal system function and how foodstuffs may influence this function. This by

181. Similarly, Busse states (translated from Busse 1998: 535): “Yet we always explain what we want to explain, with the same instrument that our investigation is focused on. We can only describe meanings – by setting new meanings. Thus, we replace meanings with meanings, com-ment on words with words, sentences with sentences, text with texts, which all cast us back to the questions from which our efforts first departed”.


no means completes the set of potentially relevant assumptions. They could be easily continued until we reach very fundamental and thus rarely ever questioned or even rec-ognized convictions about our image of the world and humans. Medico-physiological assumptions about the function of human organs could perhaps be embedded in and epistemically supported by a completely different network of convictions in, for in-stance, primitive cultures or within a rigidly theocentric worldview. Clemens Knobloch therefore characterizes understanding as a “scalar phenomenon”.182 We only ever more-or-less understand (the meaning of a word, sentence, text).

Quite obviously, in this case, the space metaphor also falls short here, as it implies a certain containment and measurability. Semantic-empirical analyses can only ever grasp a part of the referential structure of linguistic signs. The parts that are excluded in an analysis are anything but measurable. One aim of Chapters 4 to 6 consists in developing an adequate theoretical basis to carry out semantic case analyses under holistic premises. For this some semiotic specifications, among other things, are nec-essary, since we must assume that the general openness of sign-bound interpretation processes correlate with particular properties of linguistic signs that allow the sign user to “charge” the communicatively used token with knowledge. Before I come to this in Section 4.3.2, I introduce two approaches that follow the postulate of U-relevance and that are central for my further line of argumentation.

3.3 The postulate of U-relevance

The following line of thought serves to make the postulate of U-relevance accessible from an inference-theory perspective. If a holistic approach follows the postulate of U-relevance, it departs from the premise that the dichotomy of linguistic and concep-tual knowledge is false. Language understanding always takes place in a comprehensive “space of understanding” in which such a distinction cannot be meaningfully made. When comprehending individuals constitute their social “reality” within this “space” this concretely means that they recur to imputed foreknowledge when interpreting linguistic signs, and this recourse to foreknowledge occurs by constructing numerous inferences. The postulate of U-relevance is thus very closely tied to an epistemological and a psychological research perspective: with the study of U-relevant knowledge (and its representational properties) on the one hand, and with the study of the cognitive processes that participate in the relevant processes of understanding on the other.

182. Knobloch 1994: 182 (in translation): “Understanding is a gradual phenomenon. Language, with its very general schematizations in the lexicon, creates in us if nothing else but the impres-sion of having understood even when just below the surface this has come to a standstill: We all know what the ‘ozone hole’ is, ‘understand’ that it is something threatening for our skin, and that it is caused by the CFCs that cool our beer and clean military equipment. But the actual circumstances indicated there can only be explained by an expert”.


What Demmerling formulates solely with epistemological but not linguistic in-terest (and thus without any form of semiotic foundation) is of direct relevance for linguistic theories of meaning. This relevance can be summarized succinctly using Fillmore’s observation “that linguistic semantics cannot be properly separated from an examination of people’s experiences with language in context” (Fillmore 1977a: 72). In the German setting, a semantic-holistic perspective of language understanding goes back to two research strands that for a long time took little notice of each other, despite having similar linguistic theoretical premises – historical semantics and psycholinguis-tic research on language processing.

I now elaborate on the two fields with respect to their orientation towards the postulate of U-relevance. To begin with, the aim is to bring together two research per-spectives that are relevant to frame-theory. My working hypothesis is that they both determine a specific dimension of the “space of understanding” under what initially appear to be disparate premises. More precisely, these are, on the one hand, an epis-temological specification of what is “found” in a “space of understanding” and, on the other, an inference-theoretical specification of the manner in which these epistemic “realities” are achieved. I begin with Busse’s conception of a historical (or, as designated in his later work, “explicative”) approach to semantics and then address the psycholin-guistic study of inference processing.

3.3.1 Busse’s explicative semantics

The concept of U-relevance, which I have established as a key concept in frame-seman-tic research, was first introduced by Dietrich Busse in his communication-theoretical conception of explicative semantics.183 Of central importance here, as well as for a frame theory, is a semiotic model that does not even allow meaning reductionism such as in level-semantic models. The expressive and appellative function (in Bühler’s sense, 1934: 24ff.) are placed equally with the representational function of linguistic signs.

With respect to individual linguistic signs, this means that, by employing all meaning-relevant factors and through reference to their experiences of percep-tion as linked in previous acts of communication to the linguistic sign (being used in the current utterance), the communication participants attempt to make the sequence of signs meaningful (which can only ever occur in an utterance as an uttered sequence of signs). (Translated from Busse 1988: 258–259)

183. Busse (1987) initially only spoke of “meaning-relevant” and “language-relevant” knowledge in his conception of “historical semantics”. In his later works – which, however, follow other aims and also go under the title of “explicative semantics” (Busse 1991a) or “deep semantics” (Busse 2000a, 2000b) – he uses the term verstehensrelevantes Wissen (i.e. knowledge relevant to under-standing, which I designate here as “U-relevant knowledge”). Key references are, among others: Busse (1991a: 138–140; 1994: 224–230; 1997a: 19–21; 2000a: 42; 2003a: 25–26; 2003b: 182–183; 2005a: 51ff.; 2005b: 38ff.; 2007: 95ff.).


Busse laid the foundation for a communication-theoretical perspective in his 1987 book Historische Semantik (‘Historical Semantics’), however, without using the expres-sion relevance to understanding. However, in many respects the agenda of “historical semantics” already defines the horizon of a holistic theory of meaning and in so doing it demonstrates the important facets of the postulate of U-relevance.

The departure point of Busse’s considerations is a usage-based concept of meaning, which advocates that linguistic meanings only constitute themselves in the individual communicative act, but at the same time that they draw on a multitude of epistemo-logical premises that are not only evolved from each individual communicative act. To illustrate this, Busse takes recourse to different theoretical approaches from the semantic-pragmatic periphery (for instance to Wittgenstein’s usage-based approach, to Grice’s theory of speaker meaning, Hörmann’s psycholinguistic considerations, and Lewis’s conception of convention). He concludes that because linguistic meanings are social facts, they can only be adequately described in the context of the social condi-tions of their emergence, which “as the background and indispensible prerequisite of the performance of individual action have a meaning-constituent function” (Busse 1987: 272). In other words, semantic analyses only achieve a minimum of complete-ness and explicitness if they integrate situative and communicative factors and do not exclude them (for instance, as mere connotations). The smallest unit of analysis is, thus, a linguistic-communicative unit, from whose multifactorial grid of prerequisites meanings arise as emergent products.

Therefore, successful communication requires a shared knowledge base, which functions as the historical condition that makes intersubjective (language) understand-ing possible. Taking up Demmerling’s metaphor, the “space of understanding” must have similar “coordinates” and “props”; therefore, individuals communicating with each other must also dispose over similar cognitive abilities for interpreting linguistic meaning, as well as similar background assumptions, even though the “space of under-standing” of the communication participants can never be fully congruent. This can only ever be approximative simply by dint of their respective individual experiential and communication history. Nonetheless, maximal congruence is required particularly in the most fundamental requisites if understanding is to be possible at all.

Busse shows how these fundamental conditions of possibility can be more pre-cisely described by recurring to Foucault’s discourse theory. Discourses mark the cul-turally shaped, historically variable framework within which it first becomes possible for language users to convey meaning (Busse 1987: 248–251).184 Thus, discourses tran-scend individual linguistic actions and texts, which, however, does not diminish their epistemic effectivity. This lays the foundation stone for a meaning theory that extends its analytical focus to the “entire complex of knowledge necessary to understanding” (Busse 1987: 251), that is, that equally takes account of all factors and can determine

184. See Busse (1987: 222–226) and Busse and Teubert (1994) (regarding empirical analysis) for more on the discourse concept (from a theoretical perspective). See also from a frame-semantic perspective in Ziem (2008a).


only in particular empirical analyses which factor gains the greatest weight in the respective meaning constitution.185 Accordingly, linguistic signs themselves do not function as carriers of meaning; their function is rather to “allude” to knowledge that language understanders have to reconstruct on the basis of their foreknowledge (Busse 1991a: 90). What is, above all, semantically relevant is the manner in which those sub-areas of conceptual world knowledge (frames, cognitive domains) to which linguistic signs construct relationships are made available.

The in fluence of school end s at the child ren ’s door step.

Core focus

Field of focus

Area of relevance

Discourse knowledge

World knowledge

Pre-history Present time Post-history

Levels of abstractionof knowledge relevant

to understanding

Time axis

Focu

s of a

tten

tion

Figure 1. Distinction between forms of U-relevant knowledge according to Busse (1991a: 141ff.) using the example sentence The influence of school ends at the children’s doorstep

This remembered knowledge evoked by linguistic signs must be structurally formed, and the structural forming must in turn feature supra-individual similarities and a certain stability. This alone guarantees that the text recipient will understand what the text producer means by an expression. Understanding means that an uttered expression is connected with approximately the same conceptual content that the text producer intended to convey by it. From this perspective, Busse is not so far removed from a psycholinguistic inference theory such as that addressed in the following section. However, contrary to this, he is concerned with the direct relevance of epistemological considerations for a theory of linguistic meaning.

185. Busse himself sees his work as a comprehensive theoretical programme, which can only partially do justice to empirical analysis due to the range of sub-aspects to be dealt with. In this respect, the selection of the field of analysis alone constitutes setting a focus in the sense that weight is given to certain factors, cf. Busse (1987: 261–267).


An epistemological perspective of language understanding is justified by the in-trinsic relationship between linguistic and world knowledge on the one hand and by the irreducibility of communicative features on the other. In every semantic analy-sis, therefore, there is a multitude of U-relevant factors to consider. Busse speaks in later works of an analysis of manifold factors “which allow the decomposition of all the elements that contribute to the communicative constitution of meaning” (Busse 1997a: 19). It is in this factual connection that he applies the term verstehensrelevantes Wissen, i.e. knowledge that is relevant to understanding, which I designate here as U-relevant knowledge. This knowledge can be phenomenally differentiated with respect to three coordinates (Busse 1991a: 139–159):

i. Levels of knowledge. These concern the phenomenal aspects of the process of un-derstanding. Busse distinguishes here between a horizontal, temporal axis and a vertical, paradigmatic axis (cf. Figure 1). In the temporal perspective, each process of understanding arranges itself in (a) a communicative or interactive pre-history (which is referred to retrospectively in the language, either directly or indirectly), (b) the current moment of understanding, and (c) the post-history (what is rel-evant to the later course of communication and possibly prospectively referred to). This tripartite division arises from the principle of linearity (in the sense used by Saussure) and thus concerns the cognitive process of syntagmatically connecting semiotic units. In contrast, the vertical axis describes the distribution of attention (focusing), which means the knowledge dimensions, graded according to different degrees of abstraction, that can feed into the meaning constitution of a linguistic expression. In the example sentence The influence of school ends at the children’s doorstep, the semantic “core focus” is directed on the social relevance of a public institution;186 “relevance knowledge” is equally required here (for instance, about what status school has in our educational system) as well as, where relevant, dis-course knowledge (for instance, the debate about education in the wake of the Pisa study results).

ii. Types of knowledge. At the same time, U-relevant knowledge can be differentiated according to type, independently of the “level” of its occurrence. Since it is not pos-sible to deductively determine such a typology, Busse understands it as a heuristic approximation. It is contingent on empirical analyses and is thus not exhaustive. Busse names thirteen types of knowledge (Busse 1991a: 149–150):

186. This is independent of the syntagmatic embedding of the expression; “core focus” may be considered as a synonymous expression for the term semantic “profile”, which is convention-ally used in cognitive linguistics. In contrast to Busse (1994: 227), Busse makes an earlier ad-ditional distinction between “core focus” (‘Kernfokus’) and “focus field” (‘Fokusumfeld’) (Busse 1991a: 146–147). The term “focus field” corresponds for its part with the cognitive term “base”. In this sense, “core focus” and “focus field” (or “profile” and “base”) form a functional unit. With respect to the expression school in the quoted example, the “core focus” is on the social relevance of school as a public institution. This relevance can only be interpreted by engaging the “focus field”, which includes background assumptions about the role and function of schools.


a. knowledge of the spatial/temporal coordinates of deictic expressions, b. non-textual knowledge about the communication partner that is supported

by current perception, c. knowledge of the rules of usage and structuring of text elements, d. knowledge of the already constituted textual world, e. knowledge of social forms of action and interaction, f. knowledge of forms of text arrangement and structuring, g. the experiential knowledge of the text producer, h. knowledge of forms of everyday action and life, i. knowledge of the perceptually experienced world, j. discourse-related abstract knowledge, k. knowledge of the emotions of the communication participants, l. knowledge of evaluations and attitudes, m. knowledge of intentions, goals, motives.

iii. Modes of knowledge. Finally, the status of the epistemic content varies from the lin-guistically conveyed. This also belongs to U-relevant knowledge. Busse (1991a: 187) distinguishes between eight modes, which can be placed on a scale between the two extreme points of “considered as true” and “considered as false”. Additional modes are: taken-for-granted, presumed, considered as probable or improbable, and considered as possible or impossible. Modal or attitudinal particles and para- or non-verbal signals can indicate these modes; but sometimes the “mode reading” is also inherent in presuppositions.187

This inner differentiation of U-relevant knowledge provides an initial specification of those elements that structure the space of understanding. However, this tends to be rather a phenomenology of U-relevance that provides neither concrete instructions for empirical research nor information about how aspects of the concrete process of cognitive meaning construction interact.188 Nonetheless the value of this phenomenol-ogy against the background of the outlined holistic semantic model cannot be stressed highly enough. Busse deliberately undermines the distinction between linguistic and world knowledge. Knowledge of the usage and structuring rules of text elements, that is, strictly linguistic knowledge, is only one type of U-relevant knowledge among many and serves in particular to govern the focus of attention. Morphological and syn-tactic constructions (one need only think of examples like phrasal compounds and passive constructions) expand or restrict the linguistically evoked area of U-relevant

187. Such presuppositions indicating mode are central to Fauconnier’s theory of “mental spaces”. An example is the sentence Oedipus wants to marry his mother (Fauconnier 1985: 49); due to presuppositions, only the reader (and not Oedipus himself) considers the circumstances ex-pressed here to be true.

188. Busse gives instructions for the empirical application of this programme elsewhere (Busse 1987: 264–266).


knowledge (that is, the extent of the core focus, the field of focus, the area of relevance, etc.). Thus linguistic and world knowledge ultimately prove to be an inseparable unit.

Of course, whether a comprehensive empirical analysis in the sense aspired to by Busse can be realized remains doubtful, alone due to time and resource constraints (cf. Dieckmann 1989). However, in my view, this presents no general cause of objection to the approach, but may at best draw to the attention of those conducting empirical analysis its restrictions and inexhaustibility (in Demmerling’s sense). Conversely, a justified point of criticism seems to me to be that the fundamental cognitive issue has been ignored, although Busse increasingly addresses this, especially in more recent works (Busse 2005a; 2006, b, c). The cognitive dimension appears to play a large role in his approach already in his early studies (such as Busse 1991a),189 because when he characterizes “levels of knowledge”, in terms of a phenomenology of understanding, as being an interaction between two axes, this corresponds to the usual differentiation made in cognitive linguistics between cognitive processes and their participant rep-resentational structures. Thus, the understanding of a sentence is accomplished over time as a constructive process of conceptual integration. Representational structures always participate in this, as shown in the example of school. Busse addresses pre-cisely this representational dimension by differentiating between various embedding structures (“core focus”, “focus field”, “relevance area”, etc.) with respect to the current “attention focus” or “focus of interest”. Thus, overall his usage-based approach shifts a significant portion of the onus of meaning-theoretical explanation onto the question of how certain segments of U-relevant knowledge can become cognitively relevant. How are conceptual units, i.e. segments of U-relevant knowledge, organized in the memory? Which cognitive abilities participate in connecting conceptual units? More concretely, how can, for instance, segments of “discourse knowledge” become mean-ing-constituent parts of the expression school? And which factors contribute to fixing conceptual structures and therefore belong to the available repertoire of U-relevant knowledge? Besides this, it remains to be seen what possibilities there are of empiri-cally studying these features.

Without answering the questions in detail, Busse stresses that the concept of “frame” is “a foundational concept of the first order within an epistemologically ori-ented semantic theory” (Busse 2007: 95). At another point he states:

Knowledge frames are effective for the usage and understanding of linguistic signs in such elementary positions and functions that, without them, mutual linguistic comprehensibility and thus the expressibility of knowledge would not be imagi-nable. (Translated from Busse 2005a: 47)

189. The inclusion of cognitive aspects is, however, only limited in a usage-based theoretical conception of meaning that is strongly oriented towards Wittgenstein; for more on this, see Busse (1995). Despite Wittgenstein’s reservation against speaking in any form of the mental or cognitive, I am nevertheless firmly convinced that the divide between the usage-based cognitive linguistics programme and Wittgenstein’s theory of meaning is not insurmountable; for further information, see the studies linking the two in Garnham (1980), Sinha (1999), and Zlatev (2003).


Thus, Busse holds out the prospect of answers to these questions, which are all based on a frame approach embedded in cognitive research. A large part of this book pursues the basis for just such a approach. This also includes inference-theoretical considera-tions, which I now address.

3.3.2 Approaches in psycholinguistic research on language-processing

At first sight, psycholinguistic research on language processing is in many fundamental points distinct from the conception of explicative semantics. Cognitive psychologists model language understanding as an information assimilation process in which lin-guistic input is transferred into a cognitive structure and at the same time via diverse procedures becomes a meaning-saturated conceptual structural unit (cf. Hoppe-Graff 1984: 15). In contrast to Busse’s conception, this appears to entail a significant reduc-tion in factors. It would seem that both the individual communication situation and a differentiated consideration of different knowledge types are not taken into account. In addition the distinction between linguistic input and conceptual content gives rise to the problematic assumption of two “languages”, “namely, the real, external one, with which we communicate, and another ‘inner language’, so to speak” (Busse 1994: 217). The impression of reductionism is additionally reinforced by the fact that repeatedly a distinction is made between linguistically determined meaning, i.e. explicit (word, sentence, text) meaning, and linguistically unexpressed meaning, i.e. implicit mean-ing, and thus (word, sentence, text) meaning that is to be inferred.190 This leads to the basic deficit of a lack of semantic-theoretical reflection on the inference theories being subscribed to. To what extent, however, can there be meaning actualization without inferencing? And what semiotic (and not cognitive-theoretical) prerequisites must be fulfilled for this?

I only partially share the criticism and scepticism expressed here. Some points of criticism can be refuted within the theory itself or against the historical background; others resolve themselves if they are addressed within a larger context (such as that of frame semantics). Conversely, I believe it would be a deficit not to pay systematic attention to the role of inferences in language understanding. Inference-theoretical observations reveal facets of the postulate of U-relevance that until now have remained unheeded. What is more, particularly a meaning-theoretical perspective provides junc-tions to Busse’s proposed semantic epistemology. The concept of inference is of crucial importance to this.

190. The semantically unqualified conception of linguistic signs as codes, that is, as entrenched form-meaning pairs, can also certainly be found in more recent psycholinguistic research, for instance when speaking of the “concept[s] explicitly addressed in the text” (translated from Strohner 2006: 197) or it is stated that “understanding a text involves both the information con-tained in the text and the knowledge the recipient already has about the circumstances being addressed in the text” (ibid: 189).


Biere (1989: 101–124) elaborates on where such junctions are to be found: in the inferential structure of linguistic signs. In fact, with an orientation towards a pragmatic semiotic concept, the idea of inference construction becomes the departure point of the process of understanding. In the previous section, I indicated that the bridge to a cognitive viewpoint is already inherent in a non-reductionist semantic conception. Signs do not “possess” meaning; it is comprehending individuals who “give” signs meaning, albeit mostly without reflecting on or controlling this process themselves. Meanings force themselves, so to speak, on the comprehending individual and arise automatically. However, this should not detract from the fact that even the most mun-dane of tasks of categorization (such as assigning a proper name to a person) remain individual functions of understanding and require access to structured knowledge units. This leads us to deduce a semantic dimension with genuine cognitive features. Every linguistic expression functions as a departure point for deductive processes that assist in specifying the content of this expression as dependent upon the given context. Inferences therefore serve to construct meaning. Busse himself observes:

Communicatively uttered signs are… orientation points that are always situated in a predefined, usually already existent epistemic space. They are inducements for a comprehending individual to construct the meaning of the utterance by establish-ing relationships to his own knowledge. For this reason, inference construction is not the criterion for a distinction between linguistic and extralinguistic under-standing, but is rather constitutive for any semiotic understanding and thus for linguistic understanding per se. Inference is nothing more than a circumscription for interpretation, and thus for the most elementary of abilities that humans need to orient themselves in the world. (Translated from Busse 1994: 232)

However, if linguistic signs serve merely as “orientation points” in an “epistemic space” (in the sense of the “space of understanding”), and thus sign forms first need to be placed in relation to knowledge, then inferences must have a central place in a theory of meaning, because any non-inferential semiotic theory would regress into a problematic understanding of signs as “containers” of meaning, as is the case in the intelligence-communication code metaphor or a feature- or component-reductionist model of meaning. It is important to note, however, that the concept of inference is a technical term from psycholinguistic research on language processing. No independent infer-ence theories have been developed within linguistic semantics as yet; in this context, inferences merely have a semantic compensation function.191 At this point the ques-tion arises as to the systematic connection between linguistic meaning constitution and inference construction.

191. For instance, in recourse to abductive reasoning in the previously mentioned semantic level model by Dölling. In this case, inferences fulfil a pragmatic function of meaning enrichment; the area of semantics is a sovereign domain of formal logic. Conversely, Sperber and Wilson (1986) have developed a more fundamental understanding of inferences in their relevance theory.


Psycholinguistic research on language processing can be credited with using two elementary processes of meaning constitution to identify the ubiquity of inferences: referentialization and coherence establishment. Reference concerns relating a linguis-tic sign to relevant knowledge. Since linguistically performed referentialization is a cognitive act as such and can only occur by recurring to foreknowledge, the reference objects, i.e. the referents, are not extralinguistic objects or circumstances, but cogni-tive models. And these, in turn, result from potential inference processes. Coherence extends beyond the construction of a cognitive model and concerns connecting one constructed conceptual structure to another. Recent research empirically confirms that semantic inferences occur on all levels of language processing (of words, sentences, texts) and that these levels influence each other in different ways.192

At this point, let us simply take a rather general concept of inference that clearly belongs to the basic inventory of explicative semantics and turn to its relevance for semantics in terms of cognition theory. Because inferences are realized in a knowl-edge base, i.e. signs can only be placed in relationship with the knowledge of a sign user, the question arises as to how knowledge is stored in the long-term memory so that it can operate efficiently in processes of language understanding. In the course of the last three decades, cognitive psychology has devised a range of experimental procedures that allow empirically supported criteria of plausibility to be developed for different cognitive representational formats.193 In this, an important role is given to schema-oriented models. Schemata serve to motivate certain inferences and block others. From this perspective, the “cognitive turn” consists in the assumption that there is an intermediate level on which communicatively uttered tokens become coherent, conceptually well-structured units in our conscious experience. With reference to this intermediate level, Jackendoff (1983) speaks of a “projected reality”, a mental repre-sentation of “reality”, which we construct mediated by our idiosyncratic sensorimo-tor disposition. Signs function, in the sense used by Busse in the previous quote, as prompts to construct representations of what is meant. These are projections (and not representations) because the mediation process between “reality” and the cognitive representation of this “reality” is conducted by schemas. Sign interpreters only have typified knowledge at their disposal to interpret signs (which is partially due to reasons of memory efficiency). Among these are frames, which form the semantic inference base of language understanding.

Arguing that this is a case of a problematic assumption of two “languages”, an ex-ternal and an internal one, seems to me to bear little sense, as the “external language”

192. On inferences at the word level cf. Tabossi (1988), McKoon and Ratcliff (1992), at the sen-tence level McKoon and Ratcliff (1986), and at the discourse level Revlin and Hegarty (1999). On the relationship between these levels Hess, Foss, and Carroll (1994: 80) conclude, “[W]e do not need a separate account for word-level, sentence-level, and discourse-level phenomena. The language processing system is geared to discovering relations among the input items at all levels; coherence drives the system”.

193. A brief overview is provided by Coulson (1995: 128–131).


hardly deserves to be called “language”; all linguistically relevant processes are cogni-tive phenomena (in the sense of Cognitive Grammar). If linguistic meanings were not conceptual constructs, one would have to assume that signs themselves carried meaning. In reality, however, a form of translation process occurs, in the sense that language users interpret signs on the basis of all the knowledge resources available to them. This is a cognitive process par excellence.

In recent years, within psycholinguistics there have been discussions on three in-ference theory approaches in particular, which are distinct in terms of which semantic functions are assigned to inference and which factors trigger them. After a critical review, a great deal speaks for positing a fundamental role for the function of infer-ences in language processing and also seeking the inference triggers in non-textualized resp. non-linguistic parameters, which makes inferences particularly interesting for frame theory and places them in line with explicative semantics. The following three approaches to inferences describe tendencies in psycholinguistic theory formation in over thirty years of research history. They should not be misunderstood as distinct research areas or as developmental phases. In their subdivision I have followed the classification proposed by Rickheit and Strohner (1999, 2003) and Strohner (2006).

i. Minimal-inference theories. Advocates of minimal-inference theories such as Kintsch (1988), Kintsch and van Dijk (1978), and McKoon and Ratcliff (1992) see the function of inferences to consist merely in bridging local gaps in textual un-derstanding and correcting already constructed text representations. Accordingly, inferences only create local coherence, for instance by linking directly sequen-tial propositions, as in the approach of Kintsch and van Dijk. Such inferences can involve, for instance, causal or co-referential connections, as Schwarz (2000) showed using indirect anaphors as an example. McKoon and Ratcliff (1992: 440) also recognize alongside this all inferences that are “quickly and easily” available. Therefore, concept modulation is potentially also an inferential function. For in-stance, the context of moving house could suggest that the concept piano will be specified with respect to particular properties such as weight and size (Barclay, Bransford, Franks, McCarrell, and Nitsch 1974).

In all, however, it is important to see that, from a minimalist perspective, in-ferences are not a prerequisite for understanding text, because the prior construc-tion of mental propositions or conceptual units is not contingent on inferences. Inferences only contribute to a more comprehensive textual understanding. This type of approach doubly deserves to be called minimalist. On the one hand, it does not allow inferences that concern global textual coherence, for instance, the con-nection that the propositional content of a sentence has with the text topic (or the macroproposition, in the sense used by van Dijk 1980) or with larger action goals; on the other, it excludes right from the start many types of world knowledge that could possibly serve as a knowledge base for inferences. Most of the thirteen types of knowledge Busse names cannot be included because they are non-textualized prerequisites for understanding. Yet, they are not represented in propositions at all.


ii. Maximal-inference theories. This approach posits a far more fundamental function and semantic reach for inferences in language processing. The hypothesis is that inferences are also constructed with respect to objectives, intentions, and topical and causal relationships. Bransford, Barclay, and Franks (1972) already provided empirical evidence for this in their early studies, even if they did not attend to the question of when the identified inferences are constructed. For this reason they were heavily criticized. For psycholinguistic (but scarcely for semantic) studies, it makes a difference whether inferences are drawn during the reception pro-cess (“online”) or only afterwards (“offline”). The fact that Bransford, Barclay, and Franks ignored this distinction does not diminish their insight into the abundance and constructivity of inference processes.

In the meantime, Graesser, Singer, and Trabasso (1994) have substantiated the maximalist position on the basis of new research results and formulated three premises that overcome the implicit division between cognition and communi-cation in the minimalist theory. Firstly, the motive of the text reception has an influence on the inference activity (“reader goal assumption”, Graesser, Singer, and Trabasso 1994: 371). Secondly, inferences arise in the attempt to establish coherence, and this in both a local and global range (“coherence assumption”, Graesser, Singer, and Trabasso 1994: 372). And thirdly, questioning the reasons for actions and events that are brought to linguistic expression (explanation as-sumption”, Graesser, Singer, and Trabasso 1994: 372) leads to further inferences. Graesser, Singer und Trabasso (1994: 385–391) thus directly criticize the explana-tory inadequacy of the minimalist model and introduce a number of empirical studies that validate their own position. According to their results, inferences adopt a much more fundamental function in language understanding than the minimalist approach allows for. The most significant distinction consists in the fact that inferences occur not only where coherence between textual elements is to be established, but also where such gaps in coherence do not exist, yet where other inference constructions would certainly be possible. Whereas the minimalist ap-proach posits three inference types – namely, those that involve the interpretation of co-referential relations, those of case roles, and those of causal relations between directly sequential propositions – Graesser, Singer, and Trabasso (1994: 375) name ten additional inference types alongside these. They concern not only the semantic relationship between textual elements, but also the relationship of textual elements with extratextual knowledge. Inferences occur with respect to

a. superordinate action goals (with what goal does the action expressed in the text occur?),

b. the emotions of persons in the text (what is the emotional state of a person being spoken of, what feelings does this person have at a particular point of time against the background of what is presented?),

c. the causal results of an action or an event (what additional consequences for particular persons, events, etc. can be deduced from what is presented?),


d. part-whole relationships (what knowledge can be derived on the basis of meronymic relations that an expression maintains to others?),194

e. instruments (what instruments or aids are necessary in order to carry out an action or bring about an event?),

f. subordinate action goals (what subactions do complex actions contain?), g. characteristics of persons and objects (what relevant properties can be mean-

ingfully assigned to an entity?), h. the emotions of the text recipient (what emotions does the text elicit?), i. the intentions of the author (what does the text producer aim to achieve?), j. thematic relations (what is the connection in content between the entities

presented and the larger theme or macroproposition in the sense used by van Dijk 1980?).

Graesser, Singer, and Trabasso were able to verify these inference types in reading studies. For this reason, it can be assumed that, for verbal communication, ad-ditional inference types can be added that are accounted for by knowledge based on current perception and current spatial/temporal coordinates (in the sense of the first two knowledge types in Busse’s typology). With respect to the distinction between inferences constructed “offline” and “online”, the authors have provided empirical evidence that the knowledge types (c) to (g) are inferred offline, unless the test recipient is following particular goals or interests during the reception process or draws inferences based heavily on previous inferences. However, it could not be definitively clarified whether inference types (h) to (j) are drawn offline or online.

iii. Situated inference theories. Departing from the “search-after-meaning principle”,195 a maximalist perspective allows that recipients can recur to their total world knowledge for the semantic interpretation of linguistic expressions. The function of inferences is thus constitutive to understanding and not merely supportive of it. In recent times, an extension of the maximalist model to include pragmatic issues has asserted itself within a so-called “situated inference theory” (Rickheit and Wachsmuth 2006). In the fundamentally revised version of their propositional model, van Dijk and Kintsch (1983) already understand coherence establishment more strongly as a communicative process. Subsequently, inferences are largely drawn according to strategic calculation. The introduction of the strategy concept serves to take account of contextual assumptions alongside cognitive ones. These

194. Graesser, Singer, and Trabasso (1994: 375) name this inference type a little inappropriately “instantiation of noun category”. The quoted example clearly demonstrates that meronymical re-lations are meant by this. Thus, the test subjects infer the instances bacon and egg from breakfast.

195. Graesser, Singer, and Trabasso (1994: 372). This “search-after-meaning principle”, as they call it, dates back to Bartlett (1932). Because different experimental methods are used to test how subjects semantically interpret particular text passages and thus how they establish the sense of the passages, I render this term in German as “Prinzip der Suche nach Sinn”.


include, among others, the situative embedding of a text and social parameters such as motive and norms, as well as the speech act function of a text (Rickheit and Strohner 1999: 285), whereby in my opinion it remains questionable whether the latter can be analytically specified – if at all.

In the maximalist approach, two additional issues can be identified that are hard to find in van Dijk and Kintsch 1983. In his inference concept, Clark (1978) addressed more strongly the almost forgotten role of the text producer in infer-ence processes. He thus succeeds in integrating the text author’s intention more strongly into inference theory. Above all, as Clark argues (1978: 295), such infer-ences were intended to be constitutive to understanding by the text producer himself. Although this appears to carry the risk of being an intentionalistically truncated perspective of inferences that ignores the background assumptions involved in the process of understanding,196 Clark deflects this risk in his later studies (as in Clark 1996) by recognizing the shared knowledge of the communi-cation participants as the basis for drawing inferences.197 Conveying background knowledge (for instance, discourse-related abstract knowledge) does not usually belong to the communicative intention of a text producer, and yet it may function as an inference basis.

This last point is made clear by the second central aspect of a situated in-ference theory. Inference constructions occur not because the text base induces them, but because a situative model motivates them. Graesser, Millis, and Zwaan (1997), and prior to this Johnson-Laird (1983), showed the relevance of infer-ence constructions that arise out of the communicative situation, that is, that are motivated situatively and in relation to linguistic action. This concerns precisely the two knowledge types that Busse identified, which have escaped the lens of the maximalist model: person/time/place deixis on the one hand and knowledge supported by current perception, that is, knowledge obtained extratextually on the other. The role of the non-linguistic context as an inference basis is barely consid-ered in the maximalist approach, since this orients itself in the paradigm of the written communication situation, namely the reading comprehension of written language texts. Here, communication is reduced to the fundamental elements of the communication situation.198

196. Cf. also Searle’s criticism of Grice (1979: 70–73).

197. On the controversy between a constructivist inference theory following Bransford, Barclay, and Franks (1972) and Crothers (1978) on the one hand and an intentionalist approach such as Clarks on the other cf. Biere (1989: 92–100). Biere was, of course, unable to take into account Clark’s later work e.g. (1996).

198. Fundamental elements are elements of the text formula itself; extratextual factors occur at most in a highly constrained manner (cf. Busse 1991a: 137f.). In compensation, an extrapolation occurs onto the “ideal general-type text producer” (idealtypisch generalisierten Textproduzenten), as Busse formulates it following George H. Mead.


Two types of situative model have been discussed in the literature to date: the theory of cognitive scripts (Schank and Abelson 1977) and the scenario theory (Sanford and Garrod 1981). Scripts and scenarios represent global knowledge structures that are activated by linguistic expressions and can therefore become U-relevant. Thus, the expression tip evokes the restaurant script and the expression defendant a courtroom scenario.199 At this point, what is relevant above all is that, in the sense of the situated inference theory, both induce action and/or situational inferences that in individual cases are not only textually motivated, but also due to particular non-verbal situational factors (cf. also Minsky 1975). This does not contradict the basic idea of the maximal-inference approach, but simply extends its list of inference-inducing factors by adding text-external factors. Or, as Rickheit and Strohner formulate it when criticizing the metaphor of the “semantic core”:

In contrast, the situated inference theory rejects the assumption of a situatively independent semantic core and posits that inference construction is permeable and situatively dependent. (Translated from Rickheit and Strohner 1999: 290)

If we once again ask in what relationship the postulate of U-relevance stands with the psycholinguistic study of inference processing, the answer is clear. Inferences are not a support to, but constitutive of understanding. Language understanding without inference construction is impossible, because inferences are effective on all levels of language processing and not only, as the minimalist view holds, restricted to the se-mantic correlation of text elements. Without doubt, inferences that connect a proposi-tion or text element with a previous proposition or text element are certainly essential. But over and above this, there is strong empirical evidence that so-called “extratextual inferences” are significantly involved in the process of understanding.200 Such infer-ences derive from generic and specific knowledge structures that the text recipient introduces via cognitive schemata (such as cognitive models, frames, scripts, scenarios) to interpret individual expressions and expression sequences (Graesser, Singer, and Trabasso 1994: 376).

199. Rickheit and Strohner seem uncertain about whether frames, scripts, and scenarios belong rather to a maximalist or a situated-inference theory. They are considered to be types of situ-ative models in Rickheit and Strohner (2003: 573f.), but conversely, in Rickheit and Strohner (1999: 277f.), as mental models that do not do justice to the communication situation.

200. On the distinction between “extratextual inferences” and “text-connecting inferences” cf. Graesser and Bower (1990); Singer and Farreira (1983); Trabasso and Suh (1993). In the literature, extratextual inferences are often considered under the aspect of coherence, and text-connecting inferences under the aspect of cohesion.


3.3.3 Comparison of knowledge types

The gulf between the conception of explicative semantics of the type Busse proposes and psycholinguistic research on language processing can, in my opinion, be bridged within a situated inference theory. Each focuses on the same problem from a differ-ent perspective and they thus tend to stand in a complementary relationship to each other. What takes the form of epistemological analysis in Busse’s approach, due to its focus on the semantic specification of content, corresponds to analysis of the cognitive representation and processing of semantic knowledge in the psychologically orientated inference theory. However, because knowledge is of a conceptual nature and the result of inferential processes, the research focus is ultimately identical. The semiotic concep-tion of explicative semantics stands face-to-face with cognitive-psychological, empiri-cally supported studies on inference construction that offer a variety of opportunities to empirically test Busse’s typology of U-relevant knowledge.

This demonstrates the added value of the proposal to connect the theory of ex-plicative semantics with the empirical study of inference processing. Busse’s typology of different forms of knowledge emerged introspectively and is of heuristic value. At the same time, however, there is the question of how one arrives at precisely these results, how the posited forms of knowledge actually become U-relevant, and, when this is the case, how they can be empirically tested. I believe that the maximalist and situative inference theories can provide and have provided answers to these questions. However, the study of inference processing has, for its part, never recognized the direct relevance of its approach for a semantic theory because it has been far too preoccupied with psychological enquiry. Conversely, this relevance is what concerns Busse. When the role of inferences in the framework of explicative semantics becomes evident as a result of a fundamental scepticism towards any form of substantialist semiotic concep-tion, then it does so without recurring to psychologically supported empirical findings.

Besides this “complementarity”, there is a second added value in the fact that they each highlight a particular feature that the other tends to neglect: the knowledge forms participating in language understanding in inference theory are stressed in terms of the aspect of processuality, i.e. the knowledge interpretation process, whereas Busse em-phasizes their representational aspect. However, both aspects belong closely together. As such, seeing them as entwined is a more accurate approximation of the “reality” of language processing.

Interestingly, all of the inference types empirically identified by Graesser, Singer, and Trabasso (1994) find their correlation in Busse’s knowledge types. And those situ-ative inference types that Graesser, Millis, and Zwaan (1997) additionally identify in their extended study also find their counterpart in the two pragmatic knowledge types (“knowledge of spatial/temporal coordinates” and “knowledge supported by current perception”) in Busse’s typology. At the beginning of this section I drew attention to the objection of factor reductionism that has been raised against inference theory. This objection thus proves to be unfounded (cf. Rickheit and Wachsmuth 2006).


Table 1. Correspondence between the inference types identified in Graesser, Singer, and Trabasso 1994 (GST) and Graesser, Millis, and Zwaan 1997 (GMZ) and the knowledge types outlined by Busse201

Interference types in GMZ and GST201 Knowledge types in Busse 1991a

– Knowledge about the current communication situation [GMZ]

→ Knowledge of the spatial/temporal coordinates of deictic expressions

– Knowledge of characteristics of persons/objects [GST]; knowledge about the current communication situation [GMZ]

→ Non-textual knowledge about the communication partner that is supported by current perception

– Knowledge of part-whole relationships (meronymic knowledge) [GST]

→ Knowledge of the rules of usage and structuring of text elements

– Knowledge of thematic relationships between linguistically represented and superordinate themes; knowledge about the current communication situation [GMZ]

→ Knowledge of the already constituted textual world

– Knowledge of the instruments necessary to carry out an action or bring about an event [GST]

→ Knowledge of social forms of action and interaction

– Knowledge of thematic relationships between linguistically represented and superordinate themes [GST]

→ Knowledge of forms of text arrangement and structuring

– Knowledge of the author’s intention [GST] → The experiential knowledge of the text producer

– Knowledge of subordinate action goals or subactions of more complex actions [GST]; knowledge of superordinate actions [GST]

→ Knowledge of forms of everyday action and life

– Knowledge of the causal consequences of an action or event [GST]

→ Knowledge of the perceptually experienced world

– [No correspondence] → Discourse related abstract knowledge

– Knowledge of the emotions of persons in the text [GST]; knowledge of the emotions of the text recipients [GST ]

→ Knowledge of the emotions of the communication participants

– Knowledge about the current communication situation [GST]

→ Knowledge of evaluations, attitudes

– Knowledge of superordinate action goals [GST]; knowledge of subordinate action goals or

subactions of more complex actions [GST]; knowledge of the author’s intention [GST]

→ Knowledge of intentions, goals, motives

201. The abbreviations in square brackets indicate which publication the knowledge type is drawn from. “GST” stands for “Graesser, Singer, and Trabasso (1994)”, “GMZ” for “Graesser, Millis, and Zwaan (1997)”.


Table 1 demonstrates the correspondence between the inference types and the knowl-edge types identified in Busse’s explicative semantics. It shows clearly that there are no one-to-one relationships. For instance, the inference type ‘knowledge of the current communication situation’ corresponds to three different knowledge types in Busse’s typology (namely ‘knowledge of spatial/temporal coordinates’, ‘knowledge supported by current perception’, and ‘knowledge of the already constituted textual world’). Apart from this inference type, Busse’s typology deals mostly with generic forms of knowl-edge in that the corresponding inference types tend to concern more specific facets of knowledge. In this context, it comes as no surprise that one knowledge type from explicative semantics, namely ‘discourse-related abstract knowledge’, finds no coun-terpart. This is because there have been no comprehensive studies within inference theory on the U-relevance of abstract discourse knowledge.

However, one cannot justifiably speak of the complementarity of explicative and inference-based semantic conceptions until both take the same methodological prem-ises as the departure point of their studies. This seems to be mostly the case. Firstly, the conception of an explicative semantic model and psycholinguistic inference theories both circumvent the distinction between linguistic and world knowledge. Inferences only arise on the basis of foreknowledge, which is itself the result of linguistically medi-ated experiences in the broadest sense. In addition, from the perspective of explicative semantics, any attempt to determine an autonomous area of purely linguistic knowl-edge is doomed to fail. Secondly, both explicative semantics and inference theory over-come what Clark called the “dogma of understanding”, which highlights the key role of the sentence in language understanding.202 Since investigations of isolated words or sentence units only skim the surface of the knowledge required for the semantic interpretation of a linguistic expression, the approach must account for foreknowledge on the part of the recipient as well as the contextual embedding structure. Thirdly, this makes a rigid distinction between the semantic levels of word, sentence, and text superfluous.203 A communicative action constitutes the smallest analytical unit – and with it the entire complex of factors that codetermine it. Finally, both approaches fa-vour a process-oriented, anti-substantialist view of meaning constitution. Linguistic meanings stand at the end of a dynamic-cognitive construction process initiated by linguistic signs. Instead of a constant semantic “core”, cognitive routines ensure the

202. In this context, Clark (1997) speaks of “dogmas of understanding” and calls for an exten-sion of the perspective of enquiry to do justice to the communicative complexity of processes of language understanding. He particularly criticizes the assumption that sentence meanings exist independently from utterance meaning. Cf. also the brief historical overview in Rickheit, Sichelschmidt, and Strohner (2004: 25–28).

203. Referring back to Hörmann, Knoblauch concludes: “It would be completely naive to model language processing as if it began with word meanings that were linked by means of grammati-cal rules to become sentence meanings, which were then perhaps related to the situation. The meaning orientation of our entire waking lives is part of the essential foundation of language and speech processing.” (Translated from Knobloch 1994: 182)


effective interpretation of linguistic meanings Inference construction becomes routine if it is motivated by recurrent patterns of knowledge and action (that is, schemata of varying types, cf. Section 5.2). In the further course of this book we see that frames have a particular role in this process.

First of all we can note that under the postulate of U-relevance the “space of un-derstanding” can be described more precisely from two perspectives. The first focuses on the result of meaning constitution and is interested in the structural and “material” properties of the semantic knowledge involved. Because, in the broadest sense, this is a case of analysing knowledge in the context of particular socio-culturally shaped spaces of understanding, this research perspective assumes the form of a semantic epistemology. It inspects, so to speak, the epistemic “realities” that fill an actualized space of understanding. The metaphors “realities” and “space of understanding” are, of course, not intended to promote reification and should thus not be misunderstood to the effect that realities are persistent and exist independently of the sign interpreter. Rather, by “realities” I mean aspects of a linguistically constituted reality that particular communication participants share with each other at a particular point of time within the framework of a particular situative context.

The second perspective focuses more on the cognitive procedures that lead to the interpretation of these realities. This is a psychological perspective of the pro-cess of meaning constitution, precisely the type of inferences that I have addressed. Without the interaction of procedural and representational processes of understand-ing, the space of understanding could not exist. It exists only because inferential pro-cesses occur on the basis of linguistically previously constituted knowledge. In other words, inference constructions operate on the basis of knowledge in that they recur to standard representations of relevant knowledge (i.e. to frames with default values), and inference constructions produce knowledge representations themselves (i.e. in frames with default values and fillers). Both representational and procedural aspects contribute to the “structuring” of the “space of understanding”, as Demmerling puts it.204 Neither of these structuring principles may be absent. On the one hand, we must dispose over background assumptions and convictions in order to be able to under-stand an expression, and on the other, we must possess abilities to inferentially place a linguistic expression, regardless of its complexity, in relation to these assumptions. Representational and procedural aspects are mutually dependent.

204. However, as previously mentioned, Demmerling dispenses with identifying the structur-ing principles on the grounds that this would lie outside his philosophical interest in language processing.

chapter 4

Semiotic issues

Chapter 3 showed that language processing can only be poorly explained if the space of understanding is not included. Inference constructions, in which a wide variety of knowledge types participate, are constitutive for understanding. Busse’s typology of what I call U-relevant knowledge conveys an impression of how diverse the knowl-edge resources that participate in the understanding of linguistic expressions can be. Although Demmerling, with these knowledge resources in mind, speaks of the space of understanding being epistemically “inexhaustible”, he introduces as evidence the “open-endedness” of meaning interpretations and attributions without, however, jus-tifying this more closely in semiotic terms:

A theory of meaning that aims to explain at the same time how it is possible that speakers mean something and hearers understand something has to consider the space of understanding as its open end… (Translated from Demmerling 2002: 183)

Meaning descriptions are therefore ultimately interminable simply because every lin-guistic meaning is embedded in a network of epistemic assumptions about the world. In other words, every explicans requires a new explicandum and can for this reason only ever be provisionally but never definitively specified.

The departure point of the following considerations is the hypothesis that the semantic inexhaustibility of the space of understanding must be reflected in the struc-tural properties of linguistic signs, because the fact that the content side of linguistic signs is epistemically inexhaustible means nothing more to begin with than that it is possible to place the expression side in relation to infinitely many assumptions relevant to understanding.

The distinction between a linguistically fixed (or “explicit”) meaning dimension on the one hand and a linguistically non-expressed (or “implicit”) meaning dimension on the other hand does not exist in this conception. This would imply a distinction be-tween linguistic and world knowledge, which cannot be theoretically justified and em-pirically leads to inadequate explanations (in the sense of the postulate of U-relevance). Nonetheless, the dichotomy of explicit/implicit meaning dimensions has found its way into psycholinguistic inference theories: inferences are said to occur to interpret the U-relevant implicit meaning dimension whereas explicitly expressed semantic content is obvious, as it were. Certainly some advocates of maximalist inference theories also depart from this widespread myth of the explicit givenness of U-relevant data – for instance Graesser, Singer, and Trabasso (1994), and Schnotz (1994), who serve here as representative:


The textbase provides a shallow representation of the explicit text but does not go the distance in capturing the deeper meaning of the text. Deeper meaning is achieved by computing a referential specification of each noun. … Deeper com-prehension is achieved when the reader constructs causes and motives that explain why events and actions occurred. (Graesser, Singer, and Trabasso 1994: 373)

Text understanding requires the activation of not only linguistic but also content-based foreknowledge. As a result, the mental representations construed in the process of understanding always extend beyond the explicit information conveyed, and understanding is dependent upon the respective context. (Translated from Schnotz 1994: 50)

The authors clearly share the assumption of explicit, as it were, self-explanatory tex-tual meaning, even if this does not encompass all of the U-relevant data. Very similar statements can be found among some supporters of situated inference theories. As already highlighted, Rickheit and Strohner (1999), as well as others (cf. Rickheit and Wachsmuth 2006), did indeed directly criticize the implicit meaning theory assump-tion of a semantic core in conceiving a situated inference theory. We already encoun-tered the core metaphor in connection with Bierwisch’s two-level semantic model and in this context it led in both methodological and empirical respects to very strong and ultimately untenable premises. Although the metaphorical terminology of “explicit” and “implicit meaning” vs. “semantic core” and “semantic periphery” certainly dem-onstrate similarities, Rickheit and Strohner also insist that there is a more explicitly given meaning dimension. The same applies to Clark, as the following passage shows:

In the study of comprehension, it is important to discover how we draw author-ized inferences [that is, inferences the speaker intended the listener to draw] as we listen to people talk. … We solve this problem … using three main ingredients:1. The explicit content of the sentence.2. The circumstances surrounding the utterance.3. A tacit contract the speaker and listener have agreed upon as to how sentences are to be used. (Clark 1977: 244)

Rickheit and Strohner formulate something similar:

One aspect of inferences in the understanding of texts is the cognitive representa-tion of information that is conveyed by the text. The explicit information in the text constructs certain semantic knowledge structures that are, if necessary, expanded and completed by inferences. As a result, new semantic structures emerge that are in general perceived by the text recipient as the meaning of the text. (Translated from Rickheit and Strohner 1990: 534)

Both reference and coherence are largely established with the help of inference process. As a result, inferences are distinguished by the fact that they draw on world knowledge that is not explicitly addressed in the text. (Translated from Strohner 2006: 197)

Chapter 4. Semiotic issues 149

What does the “explicit” meaning, the “explicit” information, that a text conveys consist of? Clearly the distinction between “explicit” and “implicit” meaning is motivated by a particular perception of cognitive schemata. Schemata govern inference constructions and once any schema becomes U-relevant it already contains a set of relevant back-ground assumptions that – according to this dubious hypothesis – do not themselves require interpretation but are rather directly present for the comprehending individual. Garrod (1985) speaks in this context of “pseudo-inferences”. Other assumptions, which are certainly also U-relevant but must first be derived from the schema in connection with further textual information, would by contrast have to be specifically interpreted. According to Garrod, only these are inferences in the stricter sense. The distinction between “inferences” and “pseudo-inferences” can be seen in the following examples: If on the basis of textual foreknowledge the reader of a text interprets why a person was not able to leave a tip after having been to a restaurant (for instance, because the text previously addresses the fact that the person concerned spent beyond his means), this is a case of inference; conversely, it is a pseudo-inference to make other assumptions from the tip schema, such as the amount of tip, and when or how it is given. Pseudo-inferences are thus not motivated on the basis of the given text, but arise out of the information given be the schema (that is, those that I refer to as default assumptions or default values).

In my view, the distinctions between “inferences” and “pseudo-inferences” and between “explicit” and “implicit” meaning are misleading and ultimately not tenable. They are relicts of a more or less substantialist understanding of meaning that posits a semantic dimension that does not first have to be interpreted because is, after all, completely explicit. If pseudo-inferences are “immediately present” to the reader, as Schnotz (1994: 176) writes, are they not then propositionally organized structures? Or does “immediately present” mean perhaps pre-predicative? The step towards a concep-tual ontology of Platonic vein would then just be a small one.

Meanwhile, this relict of a quasi-substantialist conception of meaning as is ex-pressed in the dimension of “explicit” meaning proves not to be as harmless as it initially seems, as other dichotomies are clearly closely linked with it. For instance, in the quote above Schnotz places “linguistic knowledge” in contrast with “content-based foreknowledge”, and Rickheit and Strohner suggest distinguishing between semantic knowledge structures that are constructed without inferences and those that are inferentially expanded. This appears to be a new attempt to separate lin-guistic and world knowledge. Are there dimensions of linguistic knowledge (such as, perhaps, knowledge of how to correctly connect signs on the morphological and syntactic level to make more complex expressions) that can be described in complete independence of their semantic relevance? Further, is there a sub-area of actual-ized meanings that is not the result of inferential processes? And why (or why not)? In my view, the literature on inference theory scarcely reflects on these questions, above all because the empirical findings lack a semiotic and semantic foundation (cf. also Kindt 2001: 113). The results of the research – which, as we have seen, are very comprehensive – are insufficiently related to the structure of the linguistic material


itself.205 On the basis of what semiotic premises do the inference constructions that are found become possible?

In the following sections, I engage with questions of this nature by pulling back to the key language-theory focus of frame semantics: to the holistic semantic conception of construction grammar and Cognitive Grammar. More precisely, I address, on the one hand, the so-called “symbolic principle”, which also concerns the semiotic premises of frame semantics (Section 3.1). I further argue that a distinction between conven-tional and contextual aspects of meaning is useful and necessary for semantic analysis, and that it is not “situations” and “backgrounds” that structure the conventional mean-ing dimension, but rather cognitive “scenes” or “frames” in the sense used by Fillmore (Section 3.2). On this basis, it only makes sense to speak of semantic (“evocation” or “invocation”) potentials (Section 3.3) rather than “explicit” meanings.

4.1 Linguistic signs as constructions

Construction grammar and Cognitive Grammar are two cognitive linguistic theo-ries that follow holistic premises. Adherents of both approaches posit that U-relevant knowledge is not modularly constructed.206 Within construction and Cognitive Grammar, form and meaning constitute a “symbolic unit” that can vary in complex-ity. Subsequently, syntactic and morphological categories are also linguistic forms that are (in the Saussurean sense) conventionally associated with (particular aspects of) meaning. The theorem underlying such a perspective of language can be summarized as follows:

The structure of human language can be exhaustively described on all levels of se-miotic organization (that is, also with respect to morphological and syntactical aspects) as an inventory of variable, conventionally linked form-meaning pairs that are called “constructions” (as in construction grammar) or “symbolic units” (as in Cognitive

205. At this point, the extent to which this lack is also due to the paradigm of system-theory that Rickheit, Strohner, and also Schnotz locate themselves in is a moot point. However, in studies like those of Rickheit and Strohner (1993) and Strohner (1995) it is obvious that system-theoretical descriptions of communicative processes are thought to be able to manage without genuine linguistic foundation; they contain no sign of semiotic reflection.

206. It is misleading when Schlobinski concludes without any qualification that Fillmore and Kay’s construction grammar is a “competence-based” grammar approach that posits “mental representations of linguistic knowledge” (translated from Schlobinski 2003: 213), thus distin-guishing itself in this respect from Langacker’s paradigm of a Cognitive Grammar. As we have seen, Fillmore also rejects such distinctions as linguistic vs. world knowledge and, in the genera-tive version, competence vs. performance; Fillmore and Kay’s construction grammar certainly does differ from Langacker’s approach, as well as from other construction grammars, in that is not a strictly usage-based model.


Grammar). The elucidation of this theorem, which I refer to hereafter as “the symbolic principle”, is the focal point of the following three sections.207

4.1.1 The symbolic principle in construction grammar and Cognitive Grammar

Although neither Cognitive Grammar nor construction grammar constitutes as yet a uniform, homogenous theory, both approaches – as well as the different streams within these approaches – nonetheless share a range of linguistic and cognitive-theoretical assumptions, of which I will only discuss those that are relevant to semiotics (for an overiew cf. Ziem and Lasch 2013: 9–30, 90–110; Hoffmann and Trousdale 2013). In particular, all of the approaches share the view that the symbolic principle in the sense outlined above constitutes the basis of a theory of grammar.

I shall at least briefly mention additional fundamental premises. One common methodological conviction is that linguistic descriptions should be consistently usage based (e.g. Barlow and Kemmer 2000; Langacker 1987, 1988a; Tomasello 2003). Because it is agreed that making a theoretically based demarcation between semantics and pragmatics is not possible, any prior distinction between competence and perfor-mance also proves to be invalid. Bifurcations of this nature are common in generative approaches, taking there the form of representational levels of semantic knowledge with respect to meaning theory. Adherents of construction and Cognitive Grammar counter this with an analysis perspective that orients itself towards actual language usage. The path this takes leads from language use to theory and not vice versa, from theory to language use. Thus, the problem of retrospectively mediating between dif-ferent levels or modules does not even arise.

When authentic language material is investigated, this often occurs on the basis of large text corpora. Introspection – that is, using individual, usually self-constructed examples to establish plausibility – is being increasingly rejected as a sole method of justification (cf. Gries and Stefanowitsch 2006). Corpus linguistics can provide reliable information about the configuration of linguistic constructions. All those who advo-cate cognitive and construction grammar follow a conceptual position that posits there is no knowledge specific only to language (Kay 1997: 124–126). Linguistic knowledge is in fact an epiphenomenon that is subject to conceptual processes just as every other form of knowledge is as well. Under these specifications, a division between linguistic and world knowledge is not only problematic but also meaningless in terms of the approach. Instead of presuming innate linguistic knowledge, the departure point is provisionally “that human infants are endowed innately with a special ability to induce

207. The term “symbolic principle” that I have chosen follows from Taylor’s (2002: 38ff.) term “symbolic thesis”, which is increasingly asserting itself in the Anglo-American literature. However, I consider this term misleading. It is not the thesis that is “symbolic”, but the relation-ship that according to construction and cognitive grammar prevails between the form and content sides of linguistic signs.


linguistic structures from linguistic data, that is, to acquire linguistic constructions” (Kay 1997: 126). That linguistic structures can be inductively acquired whereas this capacity for inductive learning itself is not acquired is not a methodological dogma, but simply a provisional working hypothesis, the validity of which must first be empiri-cally proven. In the meantime, numerous studies demonstrate reliable evidence for the accuracy of this assumption (cf. Elman et al. 1996; Lieven et al. 2003; MacWhinney 1999; Tomasello 2003).

In my explanation of the above-named hypothesis of the symbolic principle in the following, I do not exclude any stream of construction or Cognitive Grammar.208 Methodologically Langacker’s Cognitive Grammar differs from construction grammar approaches above all in that he does not place grammatical units (such as morphologi-cal and syntactic categories) on the form side of symbolic units. Rather, they result from form-meaning pairs as a whole (Langacker 2005: 104); I briefly return to this issue in the next section. As a result, Langacker’s theory focuses strongly on semantic categories and relations. Thus, the theory of “symbolic units” also becomes the methodological foundation in John Taylor’s comprehensive presentation of Cognitive Grammar – as mentioned, Taylor speaks of the “symbolic thesis” (cf. Taylor 2002: 38–120).

Langacker and Taylor discussion of the semantic-holistic premises relevant to frame theory is the most extensive and concise. As far as the key issues of semiotics and grammatical theory that are most significant to me are concerned, there is a minimum consensus in the cognitive and construction grammar approaches that is decisive: they directly oppose the central premises of “mainstream linguistic theory” (Langacker 2005: 101), specifically of generative grammar. (These include all those issues that I extensively criticized in my discussion of level semantic models.) And they all take up a construction-grammar position by following the symbolic principle.

“Construction grammar”, “radical construction grammar”, or “Cognitive Grammar”? To provide a semiotic basis for the conception of a frame-semantic model as I develop it here, it is not necessary to decide on one particular cognitive approach. Although there is great terminological confusion, this dissolves to some extent with frames in mind, because all the approaches largely share the same relevant semiotic premises. Langacker describes this as follows:

Cognitive Grammar is both construction grammar and radical, but is called nei-ther. Construction grammar is not limited to Construction Grammar, but it also includes Cognitive Grammar and Radical Construction Grammar. (Langacker 2005: 102)

In the next section, I address what a Cognitive Grammar shares not only with a con-struction grammar as proposed by Fillmore and Kay or Lakoff and Goldberg, but also with Croft’s radical construction grammar, thus making it to an overarching theory.

Distinctions between these different streams exist above all with respect to the specific (but not fundamental) specification of what characterizes “constructions”

208. Cf. the brief overview in Section 1.3.2.


as grammatical units, that is, the question of how the inventory of constructions is structured, and with respect to some theoretical-methodological issues, such as the extent to which the model subscribed to is expected to provide rather a psychological or a formal description (cf. Fischer and Stefanowitsch 2006: 8–14). Overall it can be noted that on the one hand there is no agreement in the specification of the form side of constructions or symbolic units: What status do syntactic categories have? What syntactic relations are to be posited? On the other hand, questions that are directed at the nature of the relationship between constructions and the nature of the taxonomy of constructions are considered controversial (Croft and Cruse 2004: 257–290; Ziem and Lasch 2013: 95–102).

Apart from these differences, it is important to recognize that all the approaches depart from the same basic premises that are also relevant for a frame-semantic model.

All of the theories conform to three essential principles of construction gram-mar …: the independent existence of constructions as symbolic units, the uni-form representation of grammatical structures, and the taxonomic organization of constructions in a grammar. (Croft and Cruse 2004: 265)

That “constructions” (that is, the smallest form-meaning pairs) or “symbolic units” con-stitute the building blocks of a grammar and are uniformly represented and taxonomi-cally organized opens up the possibility of integrating frames directly into a theory of grammar. However, apart from the following considerations I will not be entering any further into this issue (cf. for instance Goldberg 1995 for further discussion). Instead, I focus exclusively on the principle of the independent existence of constructions or symbolic units. The term “symbolic unit” used to designate form-meaning units of dif-ferent complexity originates in Langacker’s Cognitive Grammar but is largely identical to the term “construction” in construction grammar. Hereafter I ignore the differences between the approaches, at least as far as possible and meaningful, and concentrate on specifying the concept of construction insofar as it is relevant for the understanding of frames.

4.1.2 What are constructions and symbolic units?

Every linguistic form-meaning unit that cannot be reduced to other form-meaning units has the status of a construction. Constructions are considered the only and thus the fundamental units of a language and its grammar. This therefore negates the exist-ence of simple, nuclear, i.e. non-assembled or non-constructed U-relevant elements (such as semantic primitives or semantic forms), because these are not constructions but at most elements of constructions.

Every construction is composed analytically from three elements or, in Langacker’s terminology, three “basic types of units”:


Only three basic types of units are posited: semantic, phonological, and symbolic. A symbolic unit is said to be “bipolar”, consisting of a semantic unit defining one pole and a phonological unit defining the other: [[sem]/[phon]]. (Langacker 1991b: 16)

Phonological and semantic units never occur alone, but only combined as constructed complex units. They constitute what Langacker calls a “symbolic unit”. The term “sym-bolic unit” is a technical term from Langacker’s Cognitive Grammar, just as are the terms “phonological structure” and “semantic structure” (or alternatively “phonologi-cal pole” and “semantic pole”). In the simple case of lexical morphemes, symbolic units have the [[sem]/[phon]] structure as given above. The word dog, for example, is a “symbolic unit” in that it combines a phonological utterance structure with a seman-tic content structure. Accordingly, the complex expression dogs already contains two constructions. In addition the form side of the grammatical morpheme [-s] shows that this is a plural form. The content structure linked to this accordingly modifies the con-struction of the lexical morpheme dog. Or more precisely, conceptually linked together, both constructions together form the complex symbolic unit [[dogs]/[dogs]].209 This corresponds to a tripartite form-meaning correlation as expressed in Figure 1.

A “symbolic unit” (as a technical term in Cognitive Grammar) is distinguished from a “construction” (as a technical term from construction grammar) in that it may consist of several symbolic units, whereas a construction cannot be further decom-posed. A word that is composed of two morphemes, i.e. two form-meaning pairs, thus does not have the status of a construction, but rather a complex symbolic unit.

What is symbolic about a symbolic unit? Strictly speaking, the unit itself is not symbolic, but rather the relationship that exists between the phonological and the semantic structures (for which reason Langacker’s terminology is imprecise here). Whereas Langacker understands the term “phonological structure” very broadly, thus meaning the cognitive representation of material, sensorily perceivable (i.e. both graphical and acoustic) aspects of a linguistic expression in the broadest sense (cf. Langacker 1987: 60–61), he uses the term “semantic structure” to refer to the meaning correlate that a sign user links with the phonological structure. The semantic structure constitutes the content side, the meaning of a symbolic unit.210 A form-meaning pair constitutes a “symbolic unit” when the link between form and content (i.e. between

209. Here, I follow Langacker’s notation proposal of indicating the content side in capital letters and the form side in lower case. Both symbolic units and their sub-units are indicated by square brackets. The slash “/” indicates that the phonological expression structure forms a symbolic unit with the semantic content structure.

210. Langacker (e.g. 1987: 77) additionally distinguishes between “semantic structure” and “conceptualization” of the same. This distinction is roughly identical with the delineation be-tween literal (in the sense of context-abstract) meaning and utterance meaning as a so-called “usage event”. To be sure, these do not form separate levels, but rather a transitional continuum. However, the term “conceptualization” is confusing because “semantic structures” are also


the phonological and semantic structures) has become conventionalized, that is, it has become so fixed through frequent usage that form-meaning allocations occur automatically, as it were.211 Therefore, relationships between form and content are “symbolic” in that they are arbitrary and must be learnt: every construction is based on convention. In this point, construction grammar follows Saussure’s sign concept. In other points, however, it diverges from it, as we will see.

Langacker speaks of “cognitive routines” in reference to the cognitive processes that lead to symbolic structures being fixed (Langacker 1987: 100). Native speakers of English have no difficulty, for instance, perceiving the phonological structure of the expression dog as a distinctive unit. And it causes no difficulty either interpreting the conceptual meaning evoked by that same word or establishing the connection between the expression and content sides. On a higher level of complexity such cognitive rou-tines play a crucial role, for instance when frames are involved in the conceptualization processes.

At this point we arrive at the first answer to the question of what role frames play in constructions (and symbolic units). My hypothesis is that a frame describes the conceptual structural organization of the meaning side of a construction (and a symbolic unit).212 Or more precisely, the form side of a linguistic sign or sequence of linguistic signs evokes a frame, i.e. a language user places a linguistic form in relation-ship with a particular conceptual content. This constitutes the moment of evocation in the form side of constructions. The evoked meaning can be of an extremely complex nature, composed of a network of numerous concepts. A frame is just such a network of concepts. For example, in the case of the term dog, a language user disposes over conceptual knowledge comprising physical features (colour, size, etc.), abilities (bark-ing, biting, running, etc.), functional range (hunting, guarding, as a pet), and various other convictions about dogs. These assumptions form overall an integral, gestalt-like frame structure that is evoked in processes of understanding.

It seems reasonable to raise the objection that within such a construction model there is no longer any room for morphology and syntax, that it is not possible to suf-ficiently distinguish between morphological, syntactic, and phonological phenomena. However, such is not the case. Lakoff and Croft make it clear in their specification of the construction concept that both the form and the meaning sides encompass far many more factors than appears to be the case with Langacker’s use of the terms “phonologi-cal” and “semantic units”. Certainly, concerning the “phonological unit” Langacker

intrinsically conceptual elements. I return to this in Section 4.3.2 and argue, in conformity with Allwood (2003), for a more precise specification of “semantic structure” as meaning potential.

211. “As a matter of fact, ‘unit’… refers to a structure which has been entrenched and automated, through frequency of successful use. A structure with unit status can be accessed as an integrated whole, without a person having to pay attention to its internal composition.” (Taylor 2002: 26)

212. For this reason, grammatical phenomena in the sense of complex form structures and their correlation with the meaning side are not pursued any further.


(2005: 104) remarks, “I would of course generalize this to include other symbolizing media, notably gesture and writing.” However, at the same time, in contrast to all the other construction grammar approaches, he makes it clear that in his view the form side of constructions does not include grammatical forms.213 Rather, he sees grammar as being reducible to combinations of symbolic units as wholes.214 Thus in this respect, Langacker’s conception does not concur with the illustration of the form side presented in Figure 1. Morphological and syntactic categories are not recognized as integral com-ponents of the form side in Cognitive Grammar; rather, they are externally symbolic units and represent, as it were, effects of the combination of different symbolic units.

Complex symbolic unit

Complex form structure (“phonological structure”)

Complex meaning structure (“semantic structure”)

Form[dog]

Meaning[DOG]

sym

b. re

l.

sym

b. re

l.

sym

b. re

l.

Form[-s]

Meaning[PLURAL]

Figure 1. Connection between two constructions resp. form-meaning pairs, illustrated using the example dog

213. Croft (2001: 62) presumes that constructions are symbolic, namely “a pairing of a morpho-syntactic structure with a semantic structure”. Similarly, Goldberg (1995: 51) speaks of “a pairing between a semantic level and a syntactic level of grammatical functions.”

214. “Grammar is symbolic in nature, where symbolic structures reside in the pairings of se-mantic and phonological structures. On this view, grammar (or grammatical forms) does not symbolize semantic structure, but rather incorporates it as one of its two poles.” (Langacker 2005: 105)


The alternative to Langacker’s conception is formulated in the approaches of Cruse, Fillmore and Kay, and Lakoff and Goldberg. Lakoff ’s definition of a “construction” is thus:

Each construction will be a form-meaning pair (F, M), where F is a set of condi-tions on syntactic and phonological form and M is a set of conditions on meaning and use. (Lakoff 1987: 467)

According to this, syntactic properties belong just as much to the form side as prag-matic properties do to the meaning side. Croft (2001: 19) adds another factor:

The term ‘meaning’ is intended to represent all of the conventionalized aspects of a construction’s function, which may include not only properties of the situation described by the utterance but also properties of the discourse in which the utter-ance is found … and of the pragmatic situation of the interlocutors …

Besides semantic and pragmatic elements, discourse-functional elements belong to the meaning side, and besides phonological and syntactic elements, morphological elements belong to the form side. Although a symbolic unit can thus continue to be described as a form-meaning pair, in addition, on both the form and content sides one can distinguish between different information units, as I choose to call them. For more on this, see Figure 2.

Even the smallest symbolic units, namely grammatical and lexical morphemes, can refer to different information units on the form and meaning sides. For example, whereas the morpheme -s forms the information unit [-s] on the form side and on the meaning side indicates a plural form (that is, it takes on a semantic function in the form [plural]), by contrast a definite article can be used in a discourse functional manner, for instance insofar as it deictically refers to a reference object shared by the communi-cation participants, whether in the form of person or place deixis or what Karl Bühler (1934) calls “Deixis am phantasma” (that is, referring to absent referents).215 Finally, more complex symbolic units such as the greeting formula Good morning contain on the form side (besides morphological and phonological information units) a syntac-tic information unit (which is in turn composed of simpler symbolic units). On the meaning side, these fulfil a pragmatic function, for instance insofar as the phrase can be more closely specified with respect to the speech act and includes social deixis – one can assume that the communication partners use the formal form of address. At the

215. That certain similarities with Bühler’s semiotic model resonate here cannot be rejected out of hand. Because the expressive, appellative, and representational functions of linguistic signs are each to the same degree sign-constitutive factors, semantic and pragmatic aspects of mean-ing interfere with each other. However, Bühler’s organon model seems to take more account of the role of a concrete sign sender and receiver than is done in the construction concept (Reisigl 1999 argues that Bühler focuses more on the role of the sign-sender in his model than that of the sign-receiver). However, on the other hand, this complicates a semiotic reflection on gram-matical phenomena.


same time, the greeting formula Good morning may serve in a discourse-functional sense to open a conversation.216

Croft and Cruse (2004: 257–258) completely overlook these differences when they expressly correlate the illustration of the form side in Figure 2 with Langacker’s term “phonological unit”. However, I do not wish to elaborate on this discrepancy any further. To be sure, it also concerns the key issue of the degree to which syntax and morphology enjoy relative autonomy;217 Croft, Goldberg, and Fillmore and Kay grant grammatical forms greater independence in that they posit that these can inde-pendently organize themselves without direct reference to the meaning side of con-structions. However, since this is concerned exclusively with the issue of the nature of grammatical phenomena and consequently the content side of constructions is not affected, this problem complex plays no further role in my further line of argumenta-tion. The following devotes its attention entirely to the meaning side of constructions (and symbolic units). Frames are introduced as specific schemata that structure the meaning side of constructions and symbolic units, thus making conceptual informa-tion units available to language users.

Construction

Form

Information of a semantic natureInformation of a pragmatic natureInformation of a discourse-functionalnature

Information of a syntactic natureInformation of a morphological natureInformation of a phonological nature

Meaning

symbolic

relationship

Figure 2. Symbolic unit as form-meaning pair (following Croft 2001: 18)

216. Whether the pragmatic and discourse-functional information units can be separated from each other, as Croft supposes, may remain open here. Croft himself provides not criteria for such a distinction. Other approaches, such as functional pragmatics, would raise their doubts about whether such a distinction is meaningful.

217. Note that we are speaking here of “relative autonomy”. No proponent of construction gram-mar would accept “autonomy” in the sense of modular organization. This would contradict the “symbolic principle”.


To begin with, one can certainly note one thing as a generalization in our context: U-relevant knowledge always occurs in the form of constructions and symbolic units, even if it is a matter of explaining phonological, morphological, or syntactic categories of a language. There is uniform consensus on this. Consequently, lexicon and gram-mar are equally subject to the symbolic principle. The division between the two is not categorical but rather gradual. That there is a continuum between lexicon and gram-mar is a core hypothesis of all construction grammar approaches (Boas 2010; Ziem and Lasch 2013: 90–95).

A pivotal claim of cognitive grammar is that grammatical units are also intrinsi-cally symbolic. I maintain, in other words, that grammatical morphemes, catego-ries, and constructions all take the form of symbolic units, and that nothing else is required for the description of grammatical structure. (Langacker 1991b: 16)

Grammatical forms fulfil symbolic functions. They have a meaning-side correlate, and they are never of an abstract nature (in the sense of a self-sufficient, purely formal unit). This constitutes the most fundamental difference between a construction grammar and a generatively inspired linguistic theory.

As I have already established, Langacker departs from a very broadly understood meaning concept when he maintains that grammatical units (like syntactic categories) are symbolic units and thus fulfil meaning-relevant functions. This also encompasses, for instance, semantic profile constructions on the sentence level. Alternative syntactic constructions such as the varying clause-initial positioning in example sentences (1) and (2) differ, for instance, as a result of their semantic foreground and background.

(1) Birgit geht nach Hause. Birgit go-prs-sg towards home ‘Birgit is going home’ (2) Nach Hause geht Birgit. towards home go-prs-sg Birgit ‘Home is where Birgit is going’

(2) is a case of marked clause-initial positioning, which shifts the specification of lo-cation into the foreground. “Marked” simply means that this type of clause-initial positioning occurs less frequently in language usage, occurring more usually as a sub-ject-predicate-object construction, and as a result of this it is not as fixed a construction as the latter.218 That the information structure and thus the semantics of (1) and (2) are certainly different is shown by the fact that (2) may give rise to more specific infer-ence constructions. In specific contexts it can be inferred from (2) that Birgit is going home and not somewhere else; on the basis of (1), such inferences, which naturally

218. Of course, this need not be, and it would be feasible for the frequency relationship to be reversed. Within a usage-based model of grammar the relationship between deviation and the norm is always dynamic.


assume concrete form in situative contexts, are barely possible (i.e. at least they are less intended by the speaker). Therefore, what we have here is the interesting case that the form side of extremely complex constructions such as the construction [subject predicate object], which includes (1), and [Object predicate subject], which includes (2), has meaning-side correlates. Goldberg concludes from this that “constructions themselves carry meaning, independently of the words in the sentence” (Goldberg 1995: 1). Since (1) is an instance of the construction [subject predicate object], whereas (2) is an instance of the construction [object predicate subject], the meaning content of the two differ. Kay and Fillmore (1999) come to very similar results in their study of the construction “what’s X doing Y?”.219 The following are examples of this construction:

(3) What’s that book doing on my desk? (4) What’s that fly doing in my soup? (5) What’s that dustcloth doing in my bed?

The form side of the idiomatic expression “what’s X doing Y?” as a whole contains syntactic and semantic information units that cannot be derived either from other con-structions or from the sub-units of this construction. This also applies to the semantic information unit evoked by the form side, which is, for instance, inappropriate in each circumstance expressed in (3), (4), and (5). The book does not belong on the desk, nor the fly in the soup, nor the dustcloth in the bed. Part of the pragmatic information content is that such circumstances are not to be tolerated, so the construction “what’s X doing Y?” thus has an appellative character. When a guest asks the waiter what a fly is doing in his soup, the waiter cannot answer with the observation that it is swimming. Kay and Fillmore argue that the reason lies in the nature of the complex construction “what’s X doing Y”, or more precisely in its meaning-side information content, which cannot be explained by conversational implicatures (Kay and Fillmore 1999: 4).

(3), (4), and (5) are instances of the construction “what’s X doing Y?” The variables X and Y can be flexibly filled. I now return once again to this relationship between schema and instance in connection with another issue. Which prerequisites have to be fulfilled so that conventionalized linguistic expressions (morphemes, words, idi-oms, etc.) have the status of “constructions” or complex “symbolic units”? Fischer und Stefanowitsch (2006: 6) name three conditions that are closely related to the three mentioned sub-units of constructions – form, meaning, symbolic connection.

(i) [The expressions’] form is directly paired with a particular meaning or function, (ii) their form cannot be (completely) derived from other forms of the language, and (iii) their semantics is not (entirely) compositional. (Translated from Fischer and Stefanowitsch 2006: 6)

219. Cf. also Fillmore, Kay, and O’Connor (1988) in their discussion of the construction “let alone”.


The first condition addresses the previously explained symbolic relationship between form and meaning. Also aspects of the linear (phonological, morphological, syntac-tic) organisation of signs are treated under this symbolic principle, because linguistic signs can only exist in the form of a symbolic unit of form and meaning. This has far-reaching consequences:

From the symbolic nature of language follows the centrality of meaning to vir-tually all linguistic concerns. Meaning is what language is all about; the analyst who ignores it to concentrate solely on matters of form severely impoverishes the natural and necessary subject matter of the discipline and ultimately distorts the character of the phenomena described. (Langacker 1987: 12)

That form and meaning constitute a symbolic unit is reminiscent of Saussure’s bilateral linguistic sign concept. However, in Cognitive Grammar, Saussure’s linguistic sign theory undergoes a correction in that both the form and meaning sides are posited as certainly obeying their own – albeit not autonomous – principles of organization. Accordingly, form and meaning do not stand in a one-to-one relationship with each other, but each display internal structures, which are symbolized in Figure 1 by the two horizontal dashed lines.220 Here, frames are meaning-side organization structures. The holistic premise that neither phonological nor semantic structures constitute modular representational levels is not affected by this; “the claim is, quite simply, that these levels may be organized in ways that are independent of the role in symbolic relations” (Taylor 2002: 58).

The second and third conditions named by Fischer and Stefanowitsch, namely on the one hand that the form side of language constructions, i.e. phonological, morpho-logical, and syntactic features, may not be derived from other forms and on the other that the meaning of a construction may not be compositional, is linked to a particular conception of cognitive economy. This makes the direct reference to the holistic para-digm particularly clear.

The point of departure is that the set of constructions in a language constitute overall a structured inventory of what language users know about the conventions of a language (Croft and Cruse 2004: 262). Constructions, it is further assumed, are organ-ized in a taxonomic network that lays claim to psychological reality.221 Constructions need not occur repeatedly nor be reducible to each other in order for a network like this to be systematically acquired. Every node in this network represents a specific construction that is connected by categorization links to other constructions on a higher or lower level of abstraction. In other words, a construction and its sub-units, i.e. its form and meaning sides, function as schemata in that less abstract constructions

220. In Langacker’s theory, these relations are characterized more precisely as encoding relationships.

221. This is the case in all four approaches named here, although Kay relativizes this claim at one point (Kay 1997: 129).


or their sub-units are instantiated in them; and conversely, a construction and its sub-units function at the same time as instances in that they themselves are instantiated in more abstract constructions. Schema instantiations arise in this process through specific categorization links that relate constructions with each other. The holistic paradigm provides the basis for positing that grammatical and semantic knowledge is organized in a unitary format, namely in constructions. All the aspects of U-relevant knowledge specified by Busse (1991a: 139–159; cf. Section 2.3.1) would therefore have to be cognitively represented in the form of constructions and categorization links between constructions (schema instantiations). What makes a construction and the formation of more complex constructions possible is our capacity to categorize, i.e. to automatically establish schema instantiations of manifold types, as it were.222 A language constructivist approach understood in this manner functions without rules (in contrast to generative grammar). It is solely built on the categorization competence of humans, which also proves to be just as constitutive in other domains – like that of the cognitive processing of visual or auditory data.

Here are two examples to illustrate this. As already mentioned, the lexical mor-pheme dog constitutes a form-meaning unit of the nature [[dog]/[dog]]. To which nodes are the form and content sides of this symbolic unit connected? To begin with, the form side [dog] is instantiated in the schema [noun], and this forms, on an even higher level of abstraction, an instance in the schema [word].223 Syntactic properties of the lexical morpheme dog are specified via these categorization links (cf. Figure 3, example (b)). On the meaning side I only name the two most important schema in-stantiations, namely hyperonym and hyponym relationships (cf. Figure 3, example (a)). We will encounter these again, among others, in closer connection with frames. The semantic unit [dog] functions, on the one hand, as a schema for subordinate desig-nations such as mastiff, Golden Retriever, lapdog; on the other hand, it itself forms an instance in the schema [animal] (which is in turn an instance in the schema [living creature], etc.). The more abstract the meaning-side schemata are, the more nonspe-cific the semantic specifications are. In Figure 3, the schema instantiations described here are illustrated on the meaning side under column (a) and on the form side under column (b). The arrows symbolize categorization links and show the increasing degree of abstraction of an instance to a schema.

In addition to the word-specific constructions illustrated under (a) and (b), Figure 3 illustrates four nodes of an idiom containing a transitive verb, like to kick the bucket (example (c)). Firstly, the idiomatic construction c1 to kick the bucket as a whole forms its own node, because it is semantically idiosyncratic, i.e. its meaning cannot be compositionally construed from knowledge of the symbolic units bucket and kick. A second node is of a more abstract nature and concerns the relevant verb-specific

222. On the relevance of categorizations on all levels of the organization of signs, cf. Taylor (2003).

223. Children inductively learn such abstract symbolic units as [noun] and [word] on the basis of language input. Once learnt, these units form schemata that adopt concrete morphosyntactic functions. On the acquisition of word classes, cf. Behrens (2005).


construction c2 [subject kick object]. Even more abstract is a third construction, c3, which reflects the valency of the verb kick, which requires a subject and an object: [subject verb object]. Finally, this is in turn an instance of the highest node in the hier-archy, c4 [sentence]. None of the four constructions can be retraced to another.224 They are linked with each other by schema instantiations. Thus, an idiom forms a highly complex symbolic unit: on the lowest level of abstraction is c1, whose form and content sides form instances of the construction schema c2, which in turn is an instance of c3, and this functions on its part as an instance of c4, the most abstract and “highest” node of the network. These categorization links are also illustrated by unidirectional arrows in Figure 3 under (c). In contrast to (a) and (b), the schema instantiations concern the construction as a whole.

It is clear from the three examples in Figure 3 that the complexity of constructions varies considerably in natural languages. Whereas morphemes represent the smallest constructions, which can, for example in the case of inflection, indicate meaning-relevant features with regard to time, person, modality, etc., idioms or entire sentences are of a highly complex nature (Langlotz 2006). And even abstract syntactic categories, such as word classes, are no less constructions, so that one can concur with Croft (2001: 17) in saying, “everything from words to the most general syntactic and semantic rules can be represented as constructions”. Uniformly representing a morpheme, a complex word, an idiom, or a sentence in the form of a construction means converting them into symbolic units that are composed of the three elements I have just explained, namely of a form side, a meaning side, and a symbolic relationship through which these two sides are associated with each other.

224. Conversely, however, all verb phrases with a transitive verb could be reduced to k3 and k4.

[subj. kicks the bucket]

[subj. kick obj.]

[subj. verb obj.]

[sentence]

[[DOG] / [dog]]

[noun]

[word][living creature]

[animal]

[[GREAT DANE] / [Great Dane]]

(a) (b) (c)

Figure 3. Schema-instance relationships between (a) the meaning sides of constructions, (b) the form sides of constructions, and (c) constructions as a whole, illustrated using the example of the word dog and the idiom sb. kicks the bucket


4.1.3 Constructions in the “space of understanding”

Before speaking about the connection between frames and constructions and sym-bolic units, it makes sense to briefly recapitulate the holistic context in which the “space of understanding”, a construction-grammar sign conception, and a semantic-epistemological and inference-theoretical perspective of language processing is to be seen. In the previous sections, I addressed each of these issues in detail and yet was barely able to address their shared connection with the holistic paradigm. This will be made up for here.

If the space of understanding of which Demmerling spoke is characterized by not being able to be phenomenologically subdivided into independently existing subfea-tures, but in fact is always present as a unity in diversity – as a unity of cognitive abili-ties, assumptions about the world, non-propositionally organized body perceptions, etc. – this also has consequences for the theory of language and cognition. Perhaps the most significant consequence is to abandon the dichotomy between linguistic and world knowledge to pave the way for a conceptual perspective on language. The more specific consequences for a semantic theory express themselves in a semantic-episte-mological conception on the one hand and an inferentialist understanding of meaning construction on the other.

An epistemological perspective is concerned with specifying and internally dif-ferentiating those segments of world knowledge that are actual in the construction of a semantic representation (semantic unit or semantic structure), and thus cognitively evoke a linguistic expression in a language user. What is holistic here is the basic prem-ise that this “evocation potential” of the form side of linguistic signs is not restricted by preconceptual factors.225

An inference theory addresses that way in which such a process of the cognitive activation of knowledge proceeds. Rejecting semantic primitives (irrespective of their form) forces us to understand representational structures as being constructed solely as a result of inferential processes. The holistic position is also expressed in the rejection of a non-conceptual meaning level; linguistic signs are constructed by means of infer-ences and thus obtain the form of symbolic units. Even understanding such a simple symbolic unit as [[dog]/[dog]]226 already requires five interwoven inference processes that aim at different sub-aspects of the sign constitution:

– Phonological units. Inferences are necessary to identify as discrete entities parts of the form side (whether of a phonological, graphical, or other nature) of linguistic signs, as is here the case with the units [dog] and [-s].

225. The term “evocation potential” is a technical term I introduce in Section 4.3.2. It merges Fillmore’s concept of frame evocation with Allwood’s proposal to assign a “meaning potential” to linguistic expressions.

226. As previously discussed, this unit consists of two constructions, that is, of two symbolic units integrated into each other. Expressed in formal terms: [[[DOG][-S]]/[[dog][s]]].


– Semantic units. Analogously, the content side of linguistic signs forms itself by means of inferences against a background of relevant background knowledge (cog-nitive domains). This concerns the units [dog] and [-s].

– Symbolic relationships. Form and content form a conventionally linked unit, which is not endowed as such until symbolically connected. This applies just as much to the symbolic relationship between [dog] and [dog] as it does to the symbolic relationship between [-s] and [-s].

– Syntagmatic relationships. Complex semantic (and phonological) units arise if the conceptual content of a semantic (or phonological) unit is combined with the conceptual content of another semantic (or phonological) unit and is inferentially integrated into a “composition structure”, in this case [dog] and [-s] to become [dogs] on the one hand and [dog] and [-s] to become [dogs] on the other.227

– Schema instantiations. Every form-meaning pair (as well as the form and content sides in themselves) is specified by figuring as the instance or schema of another form-meaning pair (or its form or content sides). Schema instantiations cause phonological and semantic units to emerge. In this respect, such categorization relationships are indistinguishable from those inference processes that construct phonological and semantic units.

From a semiotic point of view, this confirms the hypothesis formulated in Section 3.3 in conclusion to Busse’s explicative semantic model and psycholinguistic inference theory: Constructing inferences is the most fundamental ability forming the basis of language processing. Every form of linguistic categorization is of an inferential nature. It depends on knowledge and can at the same time create new knowledge.

At this point, I wish to return to the distinction between explicit and implicit meaning, which is often to be found in the literature on inference theory and which I critically examined at the beginning of this chapter. On the basis of a cognitive-constructivist semiotic model as outlined here in line with construction and Cognitive Grammar, it is no longer possible to justify the quasi-substantialist semiotic concep-tion that finds its expression when speaking of the “explicit meaning dimension”. Distinguishing between the explicit and only implicit givenness of (parts of) textual meaning in fact suggests that there is a dimension of sign constitution that is not of an inferential nature. If, however, constructions – that is, the smallest form-meaning pairs – are already products of inferences, where then should an explicit, as it were self-given, meaning dimension originate? According to maximalist inference theories, the departure point of inferential processes is referentialization of action goals, intentions, and the thematic and causal connections between text elements, and if one follows situative inference theories, also the referentialization of situation models. Although

227. “Syntagmatic combination is the integration of two or more component structures in se-mantic, phonological, or symbolic space to form a composite structure of greater size in the same domain. The integration of symbolic structures involves their integration at both the semantic and the phonological poles.” (Langacker 1987: 94; italics in bold in the original)


the focus here is on abstract and partially extralinguistic U-relevant factors, this should not hide the fact that linguistic signs themselves – completely irrespective of their complexity – already represent the language users’ own active cognitive operations.228

Although it is obvious that the basic inferences participating in the construction of every symbolic unit cannot be easily empirically proven, this does not by any means justify drawing the conclusion that they are explicitly given text information. As soon as we posit that constructions are the basic units of the grammar and lexicon, the concept of inference becomes a fundamental concept of the first order. Specifying inferences on the level of constructions seems to be primarily an empirical problem of operation-alization; this does not affect the issue of the existence or non-existence of inferences.

However, it is possible to distinguish differing degrees of cognitive effort. The cog-nitive effort required to establish a symbolic connection between linguistic form and content sides increases to the degree that the inner complexity of form-meaning pairs grows. Although the content side of a construction emerges from inferences and there-fore has a conceptual status, this construction of a semantic unit proceeds in a highly automated manner and is at first barely recognizable as an inferential performance. This is also the case when connecting two constructions to become a symbolic unit, as in the case of the quoted complex symbolic unit [[dogs]/[dogs]]. On the sentence level, however, the participating cognitive routines are already less pronounced. The low cog-nitive effort required to form a construction has apparently led psycholinguistic infer-ence theory to overlook the inferential operations involved in forming constructions.

Against the background that research on inferences in language processing (irre-spective of whether it is of a minimalist, maximalist, or situative provenance) focuses on extremely complex form-meaning relationships without, however, characterizing them more precisely, the analysis of constructions in cognitive and construction grammar tends more to have a fundamental theoretical character. I subscribe to the view that psycholinguistic inference theory and the cognitive theory of symbolic units could be shifted into a complementary relationship and that this would contribute to a more dif-ferentiated understanding of semantic conceptualization processes. On the one hand, the inference concept in construction and Cognitive Grammar receives too little attention, although it is inherent in a constructivist semiotic model. Conversely, inference theory does not sufficiently reflect on the semiotic, cognitive, and grammar-theoretical bases. The theory of constructions or symbolic units compensates for this deficit.

228. Due to its relevance to further argumentation I have used the concept of construction to demonstrate that signs are the result of constructive, inferential operations made by sign users, a fact that has been repeatedly stressed at least from the time of Humboldt. Wegener notes that there is initially nothing lying in linguistic words and signs that expresses substance by nature” (translated from Wegener 1885: 88f.); he sees each word as being, in fact, an imperative that evokes a conceptual image in our consciousness. Approximately seventy years later Hartmann writes: “Language serves only to rouse our knowledge, to remind us – we would say it serves only to give a mental directive… to execute thought” (translated from Hartmann 1958: 158–159). On this theme, cf. Scherner (1994), who emphasizes in a similar vein that significat and significant issue from the “dynamics of complex inferential text processing” (translated from Scherner 1994: 321).


In addition, an important conclusion can be drawn from the previous considera-tions concerning the status of frame semantics. Frame semantics is not merely a useful method for specifying the semantic content of linguistic expressions. It is also embed-ded systematically in a comprehensive theory of grammar and language (although this cannot be treated in detail here). Considering frames on the one hand as integral components of “constructions” and “symbolic units and on the other as the results of inferences based on world knowledge prevents semantic reductionism. After all, autonomizing semantic principles and processes (as in a level-semantic model) and stipulating the modular independence of syntactic, morphological, phonological, etc. principles and processes means in the broad sense denying them a “constructive” – and thus U-relevant – character. In a model such as this, conceptual processes always remain merely secondary phenomena and only come into play when on a descriptive level recursive rules no longer apply. Rules – or at least what adherents of generative grammar consider to be rules – form the pivotal point of the reductionist procedure. For this reason construction-grammar approaches dispense with a rule-based explana-tory approach. Instead, they posit categorization links that exist between schemata and instances. Both schemata and instances have the status of constructions, which stand as the basic units of our linguistic competence.229 Linguistic knowledge is thus the knowledge of constructions, i.e. the knowledge of conventionalized form-meaning units of varying degrees of complexity.

Of course, in view of this multidimensionality, U-relevant knowledge cannot be investigated in one fell swoop. For practical reasons alone it is necessary to subdivide the task. What is decisive, however, is that this practice of subdivision is not confused with a modular organization of the subject area, that is, with cognitive subdivision. In fact holistic principles prevail on the subject area level. Thus, we can note that holism means that

– linguistic knowledge is represented in a unitary structure, namely in constructions;– these constructions are themselves emergent products of both our individual and

our shared social experiences;230

– superordinate cognitive abilities – such as for categorization, schematization, distinguishing foreground and background – are decisively involved in language processing.

229. In mainstream linguistics, the concept “competence” has established itself as part of a frame that contains, for instance, the default value that language competence constitutes a modular area that is subordinated to the area of performance. Here, what I mean by “competence” is simply the specific human capacity to learn, use, and continually modify such a complex symbolic system as that of language. According to the approach I subscribe to here, language competence and performance are inextricably linked with one another.

230. This insight has recently led to a socio-cognitive extension of the construction-grammar approaches discussed here; Bergen and Chang (2005) speak in this context of “embodied con-struction grammar”.


However, such an understanding of holism does not automatically lead to a strictly anti-nativist position. In fact, there are good reasons to suppose that the development of fundamental cognitive abilities is neurophysiologically endowed (cf. Langacker 1987: 148–150; Rohrer 2005). However, it is important to recognise that these abilities are not language specific but rather universal, and what is more they can only emerge through body-based experience of the environment. Further, representational struc-tures have a purely conceptual status and are not innate.231 In both of these points holistic approaches are fundamentally distinct from level-semantic models.

The results of the last section suggest that frames fulfil an important U-relevant function and in particular that they can be addressed in close connection with con-structions (cf. also the close connections between frames and constructions within in FrameNet; Boas 2010, for an overview: Ziem and Lasch 2013: 118–122). I now address the notion that frames structure the meaning(s) of constructions.

4.2 Frames and symbolic units

Until now, cognitive semantics has paid little attention to the distinction between constructions as a structured and “virtual” inventory on the one hand and their actu-alized meaning within concrete communication contexts on the other. One exception is Langacker’s basic model of a semiotic and grammar theory, which is outlined in a slightly modified form in Figure 4 (Langacker 1987: 77). This may be considered as the most detailed conception to date.

As already noted, Langacker does not use the term “construction” but “symbolic unit”. By definition, a construction cannot be decomposed into additional construc-tions. By contrast, Langacker’s concept of “symbolic unit” equally covers both simple and complex form-meaning pairs. Thus, even a complex linguistic expression (a multi-element word, a phrase, a sentence, a text) composed of numerous constructions forms a symbolic unit. In the following, to include all linguistic expressions regardless of their complexity, I use the term “symbolic unit” in preference to “construction”.

Following from the consideration outlined in the previous sections, i.e. that sym-bolic units represent form-meaning pairs, Langacker distinguishes between symbolic units and so-called “usage events”. Whereas symbolic units involve conventions that constitute in sum the grammar of a language, “usage events” designate those detailed,

231. This also applies to such abstract structures as “image schemas”, because these must emerge ontogenetically first and foremost in ego-world interaction. Representational structures certainly have neurophysiological correlates: From a neurophysiological perspective concept formation cannot be explained without neural pattern formation. “[T]o say that I have formed a concept is merely to note that a particularly pattern of neurological activity has become established, so that functionally equivalent events can be evoked and repeated with relative ease” (Langacker 1987: 100). The basic principle is: The more abstract the representational structure, more firmly the neuronal pattern has been established.


contextualized, and language-conveyed information units that a speaker actually in-tends to convey with the help of symbolic units in a particular communicative situation or that a recipient actually understands on the basis of the symbolic units used in a particular communicative situation. In our context, concentrating exclusively on the content side of form-meaning pairs (that is, on what Langacker calls the “semantic space”) results in an important distinction between “semantic units” on the one hand and “[semantic] conceptualizations” on the other. Instead of Langacker’s term “[se-mantic] conceptualizations” I prefer the term “conceptual meaning”;232 this makes it clearer that not so much the process as the result of a meaning actualization is meant.

The incidental observation made earlier that the content side of more complex symbolic units is by no means amorphous now shifts into the focal point of the fol-lowing remarks. The working hypothesis is that frames have a fundamental function, both in constructing the content side of individual constructions and in integrating several constructions into more complex conceptual units (symbolic units). At least six sub-aspects are affected by this:

232. Langacker’s terminology is confusing at this point in as far as the term “conceptualization” serves to characterize the utterance meaning of a linguistic expression, while implying, however, that a semantic unit for its part is not the result of a conceptualization process. This is certainly not the case, simply because an encyclopaedic semantic theory is the methodological departure point. Semantic units also have a conceptual, encyclopaedic status (cf. Langacker 1987: 63 and 87).

Symbolic space

Semantic unit

Phonologicalunit

Semantic space

Phonological space

Symbolic unit Conceptualmeaning via [semantic]

conceptualization

Phonologicalconceptualization

symb.

symb.

cod.

cod.

Usage event

cod.

Grammar (ling. conventions)

Sanctioned structureTarget structure

Figure 4. The relationship between symbolic units and usage events in Langacker’s Cognitive Grammar (following Langacker 1987: 77)


i. To what extent do symbolic units differ from the communicative purpose that is construed on the basis of performing a speech act?

ii. How can the content side of a symbolic unit be more precisely characterized from a frame-semantic perspective?

iii. How can both the connection between the form and content sides of a symbolic unit and the connection between the content side and the conveyed communica-tive purpose (“conceptual meaning”) be more precisely characterized?

iv. What role do schemata play in the construction of a semantic unit?v. What do schemata consist of?vi. Provided that both the construction of a symbolic unit and the construction of its

communicative purpose are governed by schemata, how then can the components of these schemata be empirically determined?

The next three sections are devoted to the first two questions. In Section 4.3, I then address question (iii). It will be shown that the other questions, (iv) to (vi), are not yet able to be satisfactorily dealt with at this point of time.233

4.2.1 Conventional vs. contextual aspects of meaning (R. Langacker)

To what extent do symbolic units differ from the communicative purpose that is con-strued on the basis of performing a speech act? With this question I begin by turning to the area I first introduced in Section 4, Langacker’s semantic space. A feature of semantic space is its bipolarity. “Semantic units” counterpose “conceptual meanings”, with the former serving as a “code” for the latter (represented by the abbreviation “cod.” in Figure 4). A categorization link connects the two. In Figure 4 this is represented by the unidirectional arrow.

In view of the key position held by the distinction between semantic units and conceptual meanings in Langacker’s work, the illustration of the differentia specifica remains rather imprecise. At one point, Langacker defines “semantic units” as follows:

[S]emantic units are characterized relative to cognitive domains, and … any con-cept or knowledge system can function as a domain for this purpose. Included as possible domains, consequently, are the conceptions of social relationship, of the speech situation, of the existence of various dialects, and so on. Second, linguistic semantics is held to be encyclopedic. … Finally, units are acquired through a process of decontextualization. …If a property (e.g. the relative social status of a speaker and hearer) is constant to the context whenever an expression is used, the property may survive the decontextualization process and remain a semantic specification of the resultant unit. (Langacker 1987: 63; italics in bold in the original)

233. Nevertheless, in the following sections I already offer partial answers (however, without being able to go into all too much detail). Among these is the fact that frames are cognitive schemata with three structural constituents – slots, fillers, and default values. Cf. the detailed discussion in Chapter 6.


The following specification of conceptual meaning, or in his terms “conceptualization”, stands in contrast to this:

The task of finding appropriate linguistic expression for a conceptualization can be referred to as the problem of coding; its solution is a target structure … The target is therefore a usage event, i.e. a symbolic expression assembled by a speaker in a particular set of circumstances for a particular purpose: this symbolic relationship holds between a detailed, context-dependent conceptualization and some type of phonological structure … Conventional units provide an essentially unlimited range of symbolic potential … (Langacker 1987: 65–66; italics in bold in the original)

Both semantic units and conceptualizations have an encyclopaedic, conceptual status; only against the background of at least one cognitive domain can they assert their status as cognitive units. In addition, conceptual meanings exhibit a complex internal structure. In this respect, too, they do not differ significantly from semantic units.234 In fact, differences seem to occur with respect to three features. Firstly, in contrast to con-ceptual meanings, semantic units are conventional units. Secondly, the latter are un-derspecified in contrast to the former. Finally, conceptual meanings are, by dint of their contextual dependence, to the greatest possible extent singular, emergent phenomena. In terms of terminology, these three features can be summarized simply: Whereas the conceptualization of a linguistic expression corresponds to its contextual meaning (Langacker 1987: 157), the semantic unit of an expression corresponds to its conven-tional meaning (Langacker 1987: 66). I now elaborate on this, coming to the conclusion that the quoted definitions remain inadequate in the context of Langacker’s holistic meaning theory, but can, however, be substantiated using frame-semantic means.

Which semantic unit is symbolically associated with a particular language form is arbitrary and can vary from language community to language community. The conven-tional meaning of a linguistic expression is not equivalent to a possible conceptualiza-tion of the same expression in a specific context, because conceptual meanings contain many more and much more detailed semantic specifications (Coulson 2001: 89; Clark and Gerrig 1983).235 The number of possible specifications is ultimately infinite. Thus, there are already a vast number of possible specifications of the semantic unit [dog]

234. See also Langacker (1987: 87), who states: “Despite the effective simplicity of a unit (in the sense of requiring no constructive effort), it is explicitly conceived as an integrated system, possibly very complex internally. Semantic units are defined relative to knowledge structures …”

235. In analogy to the distinction between conventional and conceptual meanings, Clark and Gerrig (1983) contrast two processes of meaning construction with each other: “sense selection” und “sense creation”. The same properties distinguish them from each other as do conven-tional and conceptual meanings (Clark and Gerrig 1983: 606). However, Allwood (2003) quite rightly points out that speaking of “selection” indicates a tendency to reify, in that it posits the pre-existence of semantic components. By contrast, to emphasize the cognitive-constructive contribution of the semantic components themselves, Allwood prefers to speak of “activations”.


solely on the basis of the meaning aspect [size]. Whereas information about how large a dog is on average, i.e. on the basis of recurrent experience, has found its way into the conventional meaning [dog], conceptual meanings contain more concrete informa-tion and are for this reason more complex conceptually than semantic units. That is, for instance, the case when a language user refers to the neighbour’s dog Glamis, who is exactly 50cm tall at the withers.236

What is the nature of the interdependency between semantic units and conceptual meanings? Since semantic units are of a conventional nature, they precede conceptual meanings in temporal and logical terms. Semantic units are of a schematic nature and in this respect abstracted from specific information that issues from an individual co- or context. Further, the meaning, or conceptual content, that semantic units rep-resent is intersubjectively relatively stable. Nonetheless this results exclusively from the social practice of individuals interacting with each other.237 Or more precisely, a facet of knowledge belongs to conventional knowledge once it has recurred in the use of a linguistic expression in a particular contextual framework and as a result language users internalize it. Semantic units thus result from language usage, from the decon-textualization of conceptual meanings.

Thus, it is not that the expression intrinsically holds or conveys the contextual meaning, but rather, that conventional units sanction [the] meaning as falling within the open-ended class of conceptualizations they motivate through judge-ments of full or partial schematicity. … From the encyclopedic nature of contex-tual meaning that of conventional meaning follows fairly directly. The latter is simply contextual meaning that is schematized to some degree and established as conventional through repeated occurrence. (Langacker 1987: 158)

The example dog can be used to demonstrate this conventionalization process as fol-lows. Starting from knowledge from many usage events of the expression dog, we have the ability to inductively gain a schematic meaning for the expression that contains information about typical representatives of the category [dog]. Schematic aspects of meaning such as this “emerge via social interaction and collective reflection on shared experience” (translated from Schlobinski 2003: 165), in that the experiential contexts within a language community are relatively homogenous and repeatedly occur in the same way as the predicative attributions that language users encounter with the expres-sion dog in a particular context.

Langacker emphasizes that conventional aspects of meaning cannot be reduced to specific, temporally persistent structures nor could they be congruent with those

236. As is well known, Saussure’s theory of signs already highlighted the fact that an expression has a conventional meaning within a language community; cf. Pörings and Schmitz (2003: 13); Taylor (2002: 38–42).

237. Langacker (2000: 22). The same applies for Saussure’s theory of meaning; cf. the detailed discussion in Busse (2005b).


facets of knowledge that derive from knowledge of an abstract schema (in this case, for instance, from [animal] or [living creature]). For this reason Langacker (1991b: 3) suggests understanding conventional meaning as a network structure:238 “The conven-tional meaning of a lexical item must be equated with an entire network, not with any single node.” Conventional aspects of meaning thus do not solely result from vertical relationships between semantic units (schema instantiations), but rather just as much from horizontal relationships. Langacker (1991b: 266–267) characterizes the latter as semantic-similarity relationships. Similarity also occurs when two semantic units di-verge from each other in at least one facet of knowledge, but nevertheless remain in the same schema from a vertical perspective. Such relationships determine, for instance, the relationship between the semantic units [dog], [cat], and [monkey]. On the one hand, these semantic units have the status of a schema themselves (so that, for instance, [glamis] may function as an instance of [dog]), but at the same time they serve as instances of the same schema, namely [animal], whose slots they fill with specific val-ues.239 Figure 5 sketches the possible similarity and schema instantiations that [dog] has with other semantic units. Together they form a segment of a semantic network.

The unidirectional arrows in Figure 5 symbolize schema instantiations. They must be unidirectional because they can only be categorized as instances in schemata, but not vice versa, schemata in instances. The relationship between schema and instance cor-responds in this case to the semantic hyperonym-hyponym relationship. The following applies: An instance (hyponym) of a schema (hyperonym) functions at the same time on a lower level of abstraction as a schema (hyperonym) for other instances (hyponyms). At the same time, a schema [a] characterized by more abstract knowledge specifica-tions is an instance of schema [a]. This is a recursive process, since this instance itself forms a schema [b] for other instances. Following this pattern results in the hierarchy of schema instantiations depicted above: [glamis] → [golden_retriever] → [dog] → [quadruped] → [animal] → [living_creature]. Schema instantiations also prevail between [dog] and [bark], as well as [cat] and [meow], although the semantic units correlating to each other belong to different ontological categories, as it were.240

238. Similar considerations are to be found in Tuggy (1988), illustrated using the example of the verb run.

239. It will be shown later in connection with default values that the abstraction level that [dog], [cat], and [monkey] belong to has an important U-relevant function. This is because these semantic units represent so-called “basic level categories” (formulated in terms of the prototype theory of Eleanor Rosch and her team) and therefore exhibit a higher cognitive salience (or “cue validity”) than other categories.

240. What remains unclear is whether the correlation between [dog] and [bark] (and [cat] and [meow]) is a unidirectional schema-instance relation. Thus, [bark] can stand in a schematic relationship to [dog], for instance if certain dogs – in contrast to others – are counted among the total of all barking animals, as in Our dogs don’t bark. But one could equally argue that [bark] is also an instance of the schema [dog], as in the sentence The dog hasn’t forgotten how to bark. Clearly there is a certain degree of flexibility. In general the following applies (cf. Section 6.4.2):


[dog]

[living_thing]

[animal]

[golden retriever]

Similarity relationships

Sche

ma–

inst

ance

rela

tions

hip

[monkey][cat]

[bark]

[glamis]

[quadruped] [biped]

[activity]

[meow]

Figure 5. Network of similarity and schema instantiations departing from the semantic unit [dog]

The dashed arrows in Figure 5 indicate similarity relationships. Their bidirectionality makes it clear that each semantic unit differently specifies the same schema they be-long to. Although [dog] and [monkey], for example, are instances within the schema [animal], thus sharing all the general facets of knowledge that this schema provides, they differ nonetheless with respect to more specific facets of knowledge, which is shown, for instance, by their each belonging to the different schemas [quadruped] and [biped] respectively.241

The subject of a sentence (on the word level: the head of a determinative compound) evokes a frame and thus functions as a schema in which the predicate of a sentence (the determinans of a determinative compound) is instantiated, so that this has the function of being an instance (or filler).

241. Besides similarity and schema instantiations, Langacker (1991b: 266–272); (1988c: 134) also names a third type: extension relationships. Extension “implies some conflict of specification between the basic and the extended values.” Extension occurs, for instance, when the word dog is used metaphorically. If a person is designated as a dog, then a similarity relationship between [dog] and the metaphoric unit [dog’] does exist, but not in the sense explained above. Neither of them constitutes an instance in the schema [animal]. Thus, [dog’] contains specifications (“extended values”) that play no role in the conventional meaning; cf. also Taylor (2002: 464f.). The formation of metaphors is easily explained in the theoretical framework outlined here. More on this later.


Having arrived at this point, several questions arise. If the schematic, conventional meaning of an expression restricts the range of possible conceptual meanings, how exactly does knowledge of conventional meaning “sanction” – as Langacker (1987: 158) writes in the quote above – a potential conceptual meaning? And to what extent does this same knowledge simultaneously “motivate” certain conceptualizations without already prescribing specific concepts with particular values? Put differently: What form must conventional, schematic meanings assume so that they fulfil the condition of being semantically underspecified without constituting their own independent seman-tic level with autonomous principles (in the sense of level-semantic models)? To do justice to the holistic rationale of Langacker’s encyclopaedic meaning theory, conven-tional meanings and conceptual meanings must belong to the same representational format. Yet, what structural elements does this uniform representational format consist of? And what ensures that the representational format is indeed uniform?

Although Langacker does not provide a satisfactory answer to these questions, his remarks provide some indication of where the answers are likely to be found. As I just have just shown, he emphasizes the important role of schemata:

Schematicity can be equated with the relation between a superordinate node and a subordinate node in a taxonomic hierarchy; the concept [tree], for instance, is schematic with respect to the concept [oak]: [[tree] → [oak]]. In such relation-ships I call the superordinate structure a schema, and the subordinate structure an elaboration or instantiation of the schema. (Langacker 1987: 68; italics in bold in the original)

Apart from schemata, Langacker only allows for two additional elements in Cognitive Grammar (Langacker 1987: 53–54): (a) symbolic structures (which are in turn com-posed of semantic and phonological structures) and (b) the categorization relation-ships that prevail between these structures. As was made clear in the discussion on constructions, schemata have overall a central function in holistic theories, because the content side of a construction constitutes the instance of a more abstract content side of a construction. [glamis] may, for example, function as an instance of the schema [golden retriever], which is in turn instantiated in the schema [dog]. However, even if [glamis] furnishes more specific information than [dog], this does not yet say anything about the shared representational structure that forms the basis of the conventional meaning of these expressions and of their possible conceptual meanings.

The background outlined here implies that very similar schema instantiations allow conceptual meanings to be construed through knowledge of the conventional meaning of an expression. Contextual information functions, as it were, as instances within the conventionalized conceptual schema, and this is where frames come into play as a shared representational format. Frames function as a uniform representational format when co- and context-relevant knowledge is to be integrated into an already established frame structure consisting of conventional aspects of meaning. Figure 5 illustrates a segment of just such an established frame structure. In this context let us examine how Langacker himself understands this process leading from the conven-tional meaning of an expression to a potential conceptual meaning.


Coding … takes place across the boundary between convention and usage. It is a matter of finding an appropriate target structure that “fits” a sanctioning unit within some expected range of tolerance. … [T]he conceptualization must be plausibly categorized by the semantic unit to which it corresponds … (Langacker 1987: 77–78)

What he calls here “coding” is the task that every language user faces when under-standing an expression.242 On the one hand, the knowledge of conventional meanings motivates certain conceptualizations and excludes others. The knowledge of the con-ventional meaning of the expression dog motivates, for instance, conceptualizations concerning abstract schemata such as [size], [activity], [colour], etc.243 Conversely, they exclude conceptualizations that draw on schemata like [length of time], [par-tial results], etc. (which would, however, be relevant in the case of symbolic units that form instances of the schema [event], such as [[stroll]/[stroll]]). On the other hand, co- and contextual factors simultaneously motivate very specific conceptualiza-tions. The conceptual meaning then results from the conceptual integration of these facets of knowledge into the already established frame structure.

There are good reasons for frames being a uniform format to represent conven-tional and contextual aspects of meaning. Firstly, a semantic unit not only obtains its unit status relative to at least one other knowledge domain, but also itself represents a conceptual cognitive domain, which in turn forms the background of understand-ing for other semantic units, and this with respect to both schema instantiations and similarity relationships.

Secondly, semantic units themselves function as knowledge domains and in this function exhibit all the key features of a frame (Taylor 2002: 203), yet in terms of frame theory, important distinctions can be made, which are hidden by Langacker’s general use of the term “cognitive domain”. In his discussion it remains unclear how conven-tional aspects of meaning are to be specified and what status they have.244 The frame-semantic answers to this are: Conventional aspects of meaning are default values, and these have a predicative structure.245 The set of default values that are attributed to a frame within a particular language community at a particular time correspond to the conventionalized aspects of meaning that constitute the internal complexity of

242. See also Langacker (1991a: 294–298).

243. As we will see later, abstract schemata such as this have a central function in frame seman-tics; cf. my remarks on slots in Section 6.3.

244. Langacker speaks sweepingly of “entrenchment” and “routinization”; a usage event is then “subject to entrenchment or routinization when it recurs with sufficient frequency” (Langacker 1987: 349). See also Langacker (1987: 59f.); (1991b: 3).

245. Like all predications, default values exhibit a structure of the type a is b (at least they can be converted into this structure), and epistemologically applied they correspond to either assump-tions about the world (cf. Section 6.5) or conceptualizations in Langacker’s sense (1988b: 54).


semantic units. If the same predicate occurs frequently, it becomes entrenched, becom-ing a default value. Without entering into the role of default values in detail, Fillmore expresses it as follows:

The conventional (or ‘literal’ or ‘properly linguistic’) meaning of a sentence is that set of conditions on the interpreter’s understanding of the sentence which figure in all of its contexts; in determining the situated meanings of uses of the sentence, one integrates the sentence’s conventional meaning with its linguistic and extralinguistic context. (Fillmore 1985: 233)

For this reason the third point is: Default values function as condition grids, with whose help language users understand symbolic units of differing degrees of ab-straction – ranging from morphemes through to text. In this process, default values emerge through language usage. There is no need to assign them a special ontological status (“semantic primitives”, “semantic cores”, “purely linguistic components”, etc.). Integrating default values in fact deflects the danger of treating conventional and con-textual aspects of meaning as separate, autonomous entities. There is just as much a continuum between values and default values as there is between lexicon and grammar.

There is also no distinction with respect to the structural disposition of values and default values. Conceptual meanings can be equally understood as a set of predications. Although Fillmore speaks of “situational meanings” in this context, he also emphasizes that information not only from the linguistic but also from the extralinguistic context can become relevant to understanding. If language users integrate knowledge from the linguistic and extralinguistic context into a schematized, conventional meaning of an expression, they also actualize those default values that are not replaced or specified by contextual information. Thus, contextual aspects of meaning (values) are structurally no different from conventional aspects of meaning (default values). Both occur in the form of predications.246 The difference lies alone in their cognitive status. Only default values have a conventional character.

Let us once again consider the content side of the symbolic unit [[dog]/[dog]] to illustrate this. In our linguistic and cultural sphere, a range of default predications have crystallized that concern information about the size, colour, activity, ability, etc. of dogs. That means that a speaker uses the phonological unit [dog] to prompt a con-ceptualization of the content side, in which abstract-schematic semantic units such as [size], [colour], [activity], and [ability] are filled with default values such as [50 cm at the withers], [brown], [bark], and [guard].247 Although in individual cases frequently occurring individual experiences become integrated into a semantic

246. For instance in the form: A dog is x, where x is replaced by predicates such as a quadruped, an animal that can bark, an animal this big, etc.

247. Abstract semantic units such as [size], [activity], etc. correspond to the slots in a frame; cf. Section 6.3.


unit in the form of default values (Busse 2007),248 by far the greater proportion is of a conventional nature and belongs to the shared social knowledge of a language com-munity (which is explained by shared, recurrently occurring experiential connections).

The default values represented in the frame accordingly generate a schematized meaning of the expression dog, which is the functional equivalent of a semantic unit in Langacker’s concept. This in fact functions as a conceptual projection surface for the actualization of a conceptual meaning. For example, if a language user refers to the neighbour’s dog Glamis when using the expression dog within a speech act, the conceptual meaning of this expression – dependent upon the foreknowledge of the communication participants – can be specified in a multitude of ways. Glamis has a particular size, a particular colour, particular abilities, etc., so the default values are accordingly replaced by concrete values in the corresponding abstract semantic units. Against the background of relevant contextual assumptions, language users concep-tualize the meaning of the expression dog more comprehensively. If in comparison to conventional meaning far more specific information becomes relevant, contextually non-substantiated aspects of meaning are retained as default values and are similarly integrated into the conceptual meaning. Since, for example, [glamis] constitutes an instance in the schema [living creature] and we easily produce this categorization relationship, we can, for instance, also deduce default predications about the biological properties of a dog. In the case of the expression dog, this may seem trivial, and since dogs are a very well-known species in our cultural sphere, the detour via the schema [living creature] may be superfluous, as corresponding facets of knowledge are directly accessible via [dog] in the form of default values. However, for language users from other cultures, the relevance of the schema-instance relationship between [dog] and [living creature] can be very important, for instance if the conceptual content has only been formed to a rudimentary degree. (By contrast, many Europeans will certainly have only a very vague concept of what the word Yak means; by means of the categorization relationship [yak] → [living creature] we can at least make some judgements about the entity being addressed and by far more concrete ones with the help of the categorization link [yak] → [ox].249)

As already mentioned, default values, in contrast to co- and contextually induced information, may be considered as collective knowledge intersubjectively shared by a language community. Nevertheless, they are also subject to numerous changes over the course of time, and even in a language community that is to the greatest possible

248. For instance the schema [fear]. Busse speaks in this context of “syncognized cognitive representations”. His example is thus: “If the neighbour’s Great Dane Arko leapt at me in drooling rage whenever I approached his kennel enclosure when I was a child and was only held back from mauling me thanks to the stability of the fence, this may have a far greater influence on my future cognitive representations of the concept dog than can ever be compensated for by all the future harmlessness of the Pug that my present neighbour owns.” (Translated from Busse 2007: 87)

249. Cf. Husserl’s (1980: 38) distinction between “meaning-fulfilling” and “meaning-giving” mental acts; also: Busse (2008a).


extent homogenous, it cannot be assumed that language users dispose over the same system of symbolic resources or conventional knowledge (Taylor 2002: 66). Thus, viewed in this way, semantic units are “phenomena of the third kind”, in the sense used by Keller (2003: 87–94). They consist of fixed segments of semantic knowledge that are not subject to the sphere of influence of individual linguistic action, yet can modify themselves within (part of) a language community on the basis of recurrently performed predications.250

Although the example analyses I have introduced are grounded in introspectively gained plausibilities and thus depend on common sense knowledge, they have – so I hope – fulfilled their intended illustrative purpose. Which default values the frame dog in fact has is ultimately a question that can only be answered on the basis of quantitative corpus linguistic analysis. To demonstrate this, I carry out such an analysis in the last section of this book using the example locust. I later also conduct an extensive frame-semantic specification of fillers and default values. It should be noted at this point that the content side of symbolic units exhibits an internally complex network structure that corresponds to a frame with given default values. With the help of frames, it is possible to explain how co- and context-induced U-relevant facets of knowledge are actualized without breaching the holistic premise that both conventional and contextual aspects of meaning are represented in one and the same cognitive format.

4.2.2 Are “situations” and “backgrounds” elements of semantic units? (J. Zlatev)

My argumentation so far has led to the following conclusion: To grasp the commu-nicative purpose of a linguistic expression it is necessary to engage encyclopaedic foreknowledge that has been acquired in the course of previous communication pro-cesses and that is available in the form of schematized conceptual knowledge. This schematized knowledge becomes U-relevant via the actualization of the content side of a symbolic unit. The content side is structured by frames. What triggers this actual-ization process is a linguistic form that is arbitrarily associated with the content side. Linking a linguistic form with a particular linguistic context means activating a frame with predefined default values.

Langacker leaves no doubt that semantic units do not have a purely linguistic character, but rather are inextricably intertwined with our general world knowledge. As sources of potentially U-relevant knowledge he mentions, among others, knowl-edge of the social relationship between the language producer and recipient as well as

250. Keller illustrates this by describing how a trail is formed. A trail is not intended as a result by individual users, but is rather an emergent phenomenon that only forms when a number of users act in a similar way. Keller’s explanation of this is thus, “that the system of trails is the unintended causal consequence of those (intentional, final) actions that consist in arriving at particular destinations on foot while saving a maximum of energy” (translated from Keller 2003: 100). As I explain in Section 6.5.2, default values are precisely in this sense phenomena of the third kind. Default values are cognitive trails.


knowledge of sociolinguistic status and the register a linguistic expression is accorded (Langacker 1987: 63). In Langacker’s argumentation, however, these examples only ful-fil a representative function. In his encyclopaedic semantic model, literally every form of knowledge fulfils a U-relevant function. On the basis of Busse’s explicative semantic model and psycholinguistic inference research, we can substantiate this in that ten of the thirteen knowledge types named above are potentially reflected in semantic units (cf. Section 3.3.1). Only three knowledge types form the exception:

a. knowledge of the spatial/temporal coordinates of deictic expressions,b. non-textual knowledge about the communication partner that is supported by

current perception,c. knowledge of the rules of usage and structuring of text elements.

Because of their contextual variance, (a) and (b) cannot be the object of conventionali-zation processes. Knowledge type (c) is equally unable to be integrated into semantic units. This encompasses knowledge of a formal nature, for instance morphological, syntactic, or text-structuring knowledge, and thus definitively constitutes the form side of symbolic units.

“Knowledge of the already constituted text world” could pose a point of conten-tion. Is this knowledge able to be integrated into semantic units in a conventionalized form? Or, in Busse’s concept, is this a knowledge type that emerges in an actualized communication context? In my view, this type of knowledge can certainly be of a conventional nature, because it is assumed that every text element – whether explicit or implicit – figures as part of a discursively shaped textual universe (in the sense used by Busse and Teubert 1994), and therefore it should not be difficult to discover forms of weak conventionalization here either. The purposeful use of certain key words or argument patterns in particular discourses is an example of this.251 In each case, con-ventionalized knowledge originates in a previously constituted textual world.

Regardless of whether the previous discussion has made the status of conventional aspects of meaning within a cognitive approach any clearer, the issue of their status against the background of the critical debate with level-semantic models becomes more exacerbated. Since Langacker identifies the conventional meaning of a linguistic expression with a semantic network of recurrent similarity and schema instantiations, he feeds the suspicion that conventional aspects of meaning tend to become reified and atomized as fixed, usage-independent units, since every node in such a network forms its own entity, at least in the graphical illustration (as in Figure 5). Although every node is linked to others by means of categorization links, at the same time, however,

251. Cf. for instance the overview in Wengeler (2005, in particular 272–276 and 278–280). Instead of argument patterns, Wengeler also speaks of “topoi”. These are characterized by a con-nection that, having become fixed in the course of frequent usage and thus weakly conventional-ized, links a form side (formal inference pattern, e.g. reasoning by analogy, causality, etc.) and a content side (the meaning of elements involved in the inference pattern/reasoning). Therefore, this suggests that “topoi” should be understood as complex manifestations of symbolic units (or constructions).


it must exhibit a high degree of distinctively, without which a categorization relation-ship would not first occur. This circumstance has led to a controversy concerning the empirical determinability of nodes with a semantic network, once again raising the issue of the status of nodes and thus of conventional aspects of meaning (for a sum-mary, cf. Langacker 2006).

Sandra and Rice in their critique (Rice 1996; Sandra and Rice 1995) conclude that there are no clear criteria for assuming the existence of certain nodes and relations, which is why different studies on the same subject area inevitably come to different results.252 The same argument obtains a methodological coat of paint with the accu-sation that many results are based less on empirical, corpus-based studies and more on introspectively gained evidence. Concerning the same issue but following another direction, Zlatev (2003) and Allwood (2003) argue for understanding the conventional aspects of meaning of a linguistic expression more strongly as meaning or usage po-tential to avoid the danger of reificatory tendencies.253 The properties of semantic units themselves would then stand in question.

In replying to these objections, Langacker (2006: 142–146) begins by drawing at-tention to the fact that the relationship between nodes is always based on individual language users performing categorizations. And that means that whether an element belongs to a category is based on the judgement of individuals. Categorical affilia-tion is thus a scalar phenomenon and not an objective fact. Every category has fuzzy boundaries because the category itself represents a product of an abstraction process derived from individual experiences. The realization of a schema instantiation crucially depends on the knowledge that a language user invests in performing the categoriza-tion. Some may realize the relationship [yak] → [living creature], whereas a few certainly that of [yak] → [ox]. In this sense, nodes in a semantic network may by no means be regarded as distinct entities.

A lexeme does define a field of potential for meaning and use that for all intents and purposes is continuous. The network model is inappropriate if pushed too far. In particular, it is wrong to “reify” the senses in a network by viewing them as well-delimited islands representing the only linguistic meanings a lexeme can assume. Such atomization of the field of meaning- or use-potential is artificial and leads to pseudo-problems, e.g., the problem of ascertaining which discrete sense a given use instantiates. (Langacker 2006: 145)

252. An example of this is the various studies on the meaning of the word run in American English: Gries (2006), Langacker (1988c), Taylor (1996), Tuggy (1988). Despite having similar cognitive theoretical and methodological premises, the results deviate from each other, at times substantially.

253. Cf. also Peirce’s distinction between quali-, sin-, and legisigns, also called tones, tokens, and types. Peirce defines qualisigns or tones as not yet realized potential. A qualisign is a potential sign, a sign that is potentially able to be received via the five human senses (Peirce 1985: 151; 1993: 123–124). Both the (sensory) quality and the potentiality (in contrast to actuality or real-ness) are definitive elements for Peirce’s concept of qualisign.


Thus Langacker essentially agrees with the quoted points of criticism, but ultimately takes up an intermediate position, which cannot take up Zlatev’s radically usage-based argumentation in every aspect. This is at least the conclusion one must draw if both models of meaning are compared with each other. In fact, in Zlatev’s anti-represen-tationalist approach, which is conceptually oriented towards Wittgenstein’s “language game”, “criteria of appropriateness” (Zlatev 2003: 457) largely replace those aspects of meaning that Langacker ascribes with a conceptual and representational status. According to Zlatev, knowing the conceptual meaning of a linguistic unit means know-ing how to appropriately use an expression. This makes it superfluous to assume that the meaning side of a symbolic unit is of representational nature, that is, that it is itself derived from conceptualization processes and stored in the long-term memory. Subsequently, it is rather the use of a word that is conventionalized and not – as with Langacker – the conventional knowledge that one connects with this word indepen-dently of individual usage. Strictly speaking, default values could thus not be part of a semantic unit. Default values are, in fact, not merely conventional facets of knowledge that despite emerging first in usage nonetheless exist independently of use. Rather, they are also representational units of knowledge in the long-term memory. Langacker’s network model is an attempt to illustrate the connection semantic units have with each other.254 Thus, in contrast to Zlatev’s model, the usage-based representational variant is: although a word must have been used frequently in order to say what it means (only in this way can one know what the predication potential of this word is), one need not reuse it in order to actualize schematized meaning knowledge, because by dint of previous usage the individual already disposes over this in the form of representational knowledge units such as frames.

Zlatev rejects a representational dimension of semantic knowledge such as this. Nonetheless he levels out non-conventional and contextual aspects of meaning, al-though these are only logical.255 Instead he identifies the semantic side of a symbolic unit with a “situation” or a “situation type”:

254. Admittedly, even the metaphor of the “network” used to illustrate semantic knowledge has its limitations and, like every metaphor, is misleading when analogy formations pertain to detailed phenomena. Langacker is fully aware of this. The alternative metaphor of a “con-tinuous field” (cf. Langacker 2006: 145) is, however, inappropriate in terms of approach, since it obscures the third dimension (which is why Demmerling prefers the metaphor of “space” to that of “background”) and additionally implies that there are no temporarily distinct units (phonemes, morphemes, word meanings, etc.). This criticism of the field metaphor is shared by Zlatev (2003: 458) and Allwood (2003: 56).

255. Cf. Zlatev (2003: 454): “[T]he relationship between (elements of) form and (elements of) meaning is a matter of social convention, whereby the latter – rather than ‘truth conditions’ or ‘conceptualizations’ of individual minds – constitute conditions (criteria) of appropriate use of the former”.


However, the semantic pole does not consist of a ‘conceptualization’, but of a situation (type), which can be either real or virtual, and constitutes a humanly significant, in part linguistically construed aspect of reality, not an objective, language-independent state of affairs. Situations can be partially analyzed into semantic categories … These categories are assumed to be mutually dependent, forming aspects of meaningful wholes as in frame semantics … On the utterance level, a similar partial analyzability is assumed. (Zlatev 2003: 455–456)

However, if there is a distinction between the meaning side of a symbolic unit and its potential conceptual meaning and if the former is itself not a conceptualization but rather a “situation” or a “situations type”, the question arises as to what status the meaning side then has. That it is of neither a conceptual nor a representational nature is based on Zlatev’s assumption that linguistic meaning is fixed only relative to a socially shared background. Like Demmerling before him, he bases this on Searle’s concept of “background” (cf. Searle 1992: 179ff.). “Background” in this context means shared practices and thus abilities to carry out practices and perform actions.256 In analogy to Demmerling’s “space of understanding”, Zlatev therefore stresses that usage potential is ultimately infinite and cannot be explained in all its detail.

However, considering the “background”, in Searle’s sense (1992), as a necessary prerequisite for the understanding of semantic units shoots far beyond the mark of Demmerling’s concept. Whereas the “space of understanding” serves Demmerling as a phenomenologically irreducible reference parameter, because every usage and every understanding of language encompasses significantly more than just encoding and decoding signs, Zlatev locates the “background” within the sign itself. The conven-tional meaning of a linguistic sign can therefore not only be understood against the “background” of social practices, etc., but is rather part of the conventional meaning itself. As made clear in the previous quote, the same applies no less for “situations” or “situation types”. These, too, are integrated into the semantic side of a symbolic unit.

256. Although Searle (1992) equates “background” with non-intentional abilities/actions that make intentional abilities possible, this specification is nonetheless complicated by the fact that he introduces the restrictive concept of “network”. What is meant are the “intentional circum-stances” or the background assumptions necessary for understanding. Deviating from the defi-nition of “background” in Searle (1992), at another point Searle (1998) describes “background” expressly in terms of background assumptions that are not realized in the semantic structure of a sentence; accordingly the background would be propositionally structured after all and more or less identical with “network”. In the cognitive paradigm of the frame semantic model presented here, one could probably best deal with this terminological confusion and inconsistency by following Johnson’s suggestion (1987: 178–190). Johnson draws parallels between cognitive “image schemas” and Searle’s “background” and “network”: “Image schemata, as structures emerging in our perceptual interactions and bodily movements, fall clearly into Searle’s notion of Background. But… they are sometimes conceptualized and drawn up within a representational Network, too”. (Johnson 1987: 183)


Zlatev argues that, apart from (a) the “background” and (b) “situations” or “situa-tion types”, ultimately (c) “other words of the utterance” also determine what he calls the “usage potential” of an expression.257 These three knowledge units specify the conventional part of a linguistic meaning, since the usage potential of a linguistic expression corresponds to the possibilities of its conventionalized usage. Clearly the concept “situation” serves here as a catch-all for all the non-linguistic factors that could become relevant for understanding a linguistic expression or even only occur para-linguistically. A “situation” thus exhibits a certain similarity with Fillmore’s concept of scene – to which I return shortly. What Zlatev means by “other words of the utterance” is not elucidated any further. Presumably he is referring to routinized expressions that contribute to semantic units being closely associated with each other or, expressed in terms of Langacker’s network model, being directly connected by a categorization link, i.e. without going through a mediating intermediate node.258

Zlatev adheres without doubt to the principle of recurrence for all three knowledge types. Only under the prerequisite that an expression occurs with a high frequency (a) against the same or a similar “background” (b) within the same or a similar “situation” and (c) in co-occurrence with the same or similar words can corresponding features become components of the usage potential. This resembles Wittgenstein’s “language game” in that the usage potential constituted by means of recurrence is an inextricable tangle of linguistic and extralinguistic elements and originates solely in the practice of linguistically acting humans. Thus, in the sense of the postulate of U-relevance, Zlatev’s position is more radical than Langacker’s because extralinguistic factors are directly reflected in the meaning of linguistic expressions.

Yet, this radicality also entails risks. To what extent can the “background” or a “situation” contribute to an expression “having” a specific usage potential? And what type of knowledge is it that is relevant for this? Quite in contrast to Demmerling, Zlatev makes no distinction between different forms of U-relevant knowledge. Demmerling distinguishes between knowledge that is propositionally given to us in the form of assumptions about the world and knowledge that has a non-propositional character in that it belongs to the performance of primary, spontaneous actions. Regarding this, he writes the following:

At most, individual parts or elements of this [semantic] space are propositionally structured; others can be expressed in principle with the help of propositional sentences or statements, which, however, does not mean that they are as a result propositionally given or would have to be given in this form in different primary routines of action and normal life. The propositional structure of the pertinent

257. Zlatev (2003: 454–459). Zlatev (2003: 458) states: “lexical meaning is not only relative to scenes, but to MDLG [minimally differentiated language games, i.e., language games in Wittgenstein’ sense], consisting of (i) situations, (ii) other words of the utterance, and (iii) backgrounds”.

258. Fraas (2001), for instance, has shown using numerous examples the extent to which co-occurrence can function as an indicator of conceptual knowledge.


circumstances can be deduced without difficulty from the mere possibility of ex-pressing something retrospectively in the form of sentences or convictions that have factual circumstances as their object or are aimed at factual circumstances. (Translated from Demmerling 2002: 161)

Against the background that propositional and non-propositional elements in the space of understanding differ in the nature of their givenness, when we once again pose the question as to what type of knowledge it is when Zlatev speaks of a “back-ground”, a “situation”, and “other words of the utterance”, the answer must be as fol-lows. Neither “situations” nor “backgrounds” are given as such for language users. In fact, what is cognitively effective is rather the assumptions about a “situation” or a shared “background” (in the Searlean sense of shared practices). The same applies to a greater degree for routinized expressions, because they involve connections between symbolic units, and the creation of such connections is the result of cognitive opera-tions. In contrast to routinized expressions, situations and backgrounds in the form of non-propositional elements can certainly also be U-relevant in themselves within – as Demmerling writes – “primary actions and normal life routines”259. However, in this form they do not enter into the usage potential. This is because the constitution of usage potential depends on the ability of a language user to recognize the potential and to reconstruct relevant assumptions on the basis of previous experiences, which in turn requires conceptualizations of experiential relationships. Schematization processes play a crucial role in this, namely in that experiential relationships are not relevant for the usage potential of a linguistic expression in all their detail and particulars but only in terms of typical details and particulars. Thus, when Zlatev speaks of a “situation type”, this is precisely the result of conceptualization processes and is itself nothing other than conceptual knowledge.

Like Demmerling, one can state that only in primary actions and normal life routines can situations and backgrounds be U-relevant as implicit – i.e. non-propo-sitionally given – knowledge. According to Busse’s classification, they then belong to the knowledge type “knowledge of spatial/temporal coordinates” or “knowledge about the communication partner supported by current perception”. However, as soon as they become U-relevant over and above primary actions and normal life routines, we find ourselves in the retrospective mode, and in this context we can recall knowledge only propositionally and judgementally, by means of assumptions or convictions about situations or backgrounds. As a result, conventional aspects of meaning of an expres-sion are always propositional elements in the space of understanding and thus have a conceptual character.

259. This knowledge need not be propositionally given at the moment of performing the linguis-tic action in order to become U-relevant. Consider an example from institutional communication. Of course, a patient who is explaining their concerns to the doctor “knows” about the situative setting, and this knowledge certainly influences his linguistic behaviour. But this does not neces-sarily take the form “that p”. This knowledge can only be retrospectively converted into this form.


This finding is also supported by more recent studies in memory theory, which point towards language and memory being conditional on one another. Memory op-erations are required for knowledge to become U-relevant in the retrospective mode. Every facet of knowledge that is actualized in language processing and that can be referred to in a non-situationally deictic manner must be stored in some form in the human memory. Because such memory operations cannot be explained without language, Roger Schank (1999: 3) comes to the conclusion: “Any theory of language must refer to a theory of memory …” This has the far-reaching consequence that our memory is itself structured homologously, or at least similarly, to language. The sup-position that memory structures have a propositional disposition also explains the inherent dynamics of ongoing learning processes, an aspect that Schank considers to be fundamental for an explanatorily adequate memory model (cf. also Busse 2008a).260 At least contrary to Zlatev’s position, non-propositional elements of the space of un-derstanding cannot develop their effectiveness (not even as usage potential) in the retrospective mode, as this would mean that such elements could be activated without having to recur to any memory function.

At this point, introducing Zlatev’s suggestion of “criteria of appropriateness” (Zlatev 2003: 457) as the only analytical category for the meaning side of symbolic units in order to deny their conceptual status does not help any further, either. The decisive questions are then: How can “criteria of appropriateness” become effective in the process of understanding without taking reference to foreknowledge? How do language users know whether they have used a linguistic expression appropriately or what it means to use an expression appropriately? And what knowledge makes dis-tinguishing different semantic units possible? Zlatev cannot answer these questions because the answers would have to draw on a representational dimension of semantic knowledge that we apply in order to use a word appropriately. Zlatev is without doubt correct in stressing that practical abilities and situative factors are involved in the process of meaning actualization, and there is no doubt that these abilities cannot be resolved into representational structures (cf. Figure 6). But this is not at all necessary. If Zlatev distinguished properly between representational and procedural knowledge and between propositional and non-propositional knowledge, his strong hypothesis that there is no representational dimension of semantic knowledge could be quickly put into perspective. In its place there would be a more differentiated hypothesis that both knowledge types always interact in language processing and that both should be equally taken into account within a holistic theory of meaning.

The question whether “situations” and/or “background” elements are semantic units has therefore become more pointed: Given the case that (a) the content side of symbolic units, i.e. semantic units, have a conceptual-encyclopaedic status, (b) the unit in its inner complexity can be described by frames, and (c) this unit is the result of schematization

260. “No human memory is static; with each new input, with every experience, a memory must readjust itself. Learning means altering memory in response to experience. It thus depends upon the alteration of knowledge structures that reside in memory” (Schank 1999: 3–4).


processes of conceptual meanings, to what extent then can the extralinguistic context be kept out of a frame? In view of the previous considerations, the answer must be thus: “Situations” and/or “backgrounds” cannot form elements of semantic units, because semantic units are defined precisely by their conventional status and are thus distinct from conceptual meanings. However, the result of this last consideration is not just a negative one, since the answer to the question has also led to the insight that semantic units are specific elements in the “space of understanding”. Semantic units exhibit a propositional structure, or at least they can be transferred into one.

Directing our attention to contextual aspects of meaning rather than conventional ones, we can observe that, in Zlatev’s terms, both the “background of practices” and the “situation” contribute significantly to the construction of the conceptual meaning. As illustrated in Figure 6, these organise the currently given “space of understanding”. In the absence of any correlation between semantic units (in this case, [dog] and [b]) and elements in the situative setting (or the “embedding structure”, using Scherner’s termi-nology), no conceptual meanings can occur.261 Referential expressions (as in Figure 6 dog) are only one example of this. At the same time, performing a linguistic action involves the knowledge of (social) practices. We need to know what types of linguistic

261. Taking up an extreme case that I have already mentioned: A situative setting (or an “embed-ding structure”) is also given when someone attempts to understand the words in a note that s/he has found in the street. The situative setting consists precisely in the fact that this note was found in a particular place, at a particular time, by a particular person, and under particular circumstances.

Background ofpractices

b dog… Conceptual meaning

Situation

[[[a] /[B]] [[dog] / [DOG]]… ]symbol. unit

Figure 6. Dimensions of the space of understanding in the process of constructing a conceptual meaning, in this case using the example dog (cf. also Zlatev 2003: 457)


action can be expected in a situation so as to apply practices ourselves. One extremely fundamental practical ability consists in correlating semantic units with extralinguistic entities in the situation or cognitive models of these entities. Even the association of phonological units (in this case, [dog] and [a]) with semantic units (like [dog] or [b]) is the result of a very fundamental cognitive ability. Thus seen, cognitive abilities are part of the “background”, very much in the sense used by Searle (1992). They help to construct conceptual meanings on the basis of available knowledge (semantic units).

4.2.3 Are “scenes” elements of semantic units? (C. Fillmore)

The fact that situations and backgrounds are not themselves elements in semantic units, but rather are only involved in semantic units in the form of linguistically formulated assumptions about situations and backgrounds (propositions) certainly does not mean that they both belong to linguistic and not world knowledge. After all, U-relevant assumptions cannot occur outside of individual experiential processes. However, a further distinction can be made with respect to the conceptual status of U-relevant situational and background knowledge. In fact, in his earlier works, Fillmore (1975, 1976a, 1977a, 1977b) made a terminological distinction between “frames” and “scenes” to differentiate between a linguistic and a non-linguistic conceptual dimension.262 Although Zlatev (2003: 458) also brings his approach in connection with the techni-cal term “scene”, he fails to recognize that in Fillmore’s concept this concept, just like the concept of “frame”, describes conceptual and representational units (cf. Fillmore 1976a: 13, 19; 1977a: 63). In our context, this begs the question whether “scenes” also constitute semantic units in addition to “frames” and, if this is the case, to what extent they differ from each other.

At first sight, in using the concept of “scene” Fillmore appears to be highlighting the relevance to understanding of extralinguistic factors, thus assigning the concept no cognitive or conceptual dimension at all. He draws attention to the close relationship between language and extralinguistic reality in the process of acquiring lexical mean-ings. He points out that the way in which someone learns the meaning of a linguistic expression exerts a great influence on which concrete form a frame assumes, i.e. which default values it takes:

It appears … that in meaning acquisition, first one has labels for whole scenes, then one has labels for parts of particular familiar scenes, and finally one has both a repertory of labels for schematic or abstract scenes and a repertory of labels for entities or actions perceived independently of the scenes in which they were first encountered. (Fillmore 1975: 127)

262. Indeed, the term “scene” already appears in articles prior to 1975, for instance in Fillmore’s well-known study on case grammar (Fillmore 1968; for an overview: Ziem 2014). There is reason to believe that the term “scene” ultimately dates back to Tesnière (1959), although Fillmore only refers to this in passing (cf. Fillmore 1982a: 114).


Here, Fillmore’s concept of scene seems largely to overlap with Zlatev’s concept of scene and situation.263 At least the first part of the quote implies that “scene” is to be inter-preted as “part of the extralinguistic reality”. In addition, Fillmore (1975: 127, 1977a: 61; 1982a: 115) also uses the term “situation” without any recognizable meaning distinction to that of “scene”. Fillmore argues that a child does not learn isolated word meanings, but rather that the association of a meaning to a word form in fact occurs relative to a situative setting. One of his examples is that a child may connect the word pencil with the experience of sitting with its mother in a particular room drawing circles, without being in a position to distinguish between these sub-aspects. It is only later that it ac-quires the words like paper, draw, and circle to describe parts of this “scene”, and only then can it distinguish them and separately name them.264 Accordingly, meaning ac-quisition would thus occur as a gradual process of abstraction. According to Fillmore, this “consist in developing schematic scenes with some of the positions ‘left blank’, so to speak” (Fillmore 1977a: 63). A child thus understands the word pencil against the background of an activated “schematic scene”, which certainly does not always have to be cognitively present in all its details, but in some of them.

At this point it becomes clear, however, that Fillmore’s concept of “scene” is not just referring to extralinguistic, situative factors, but rather to conceptual-schematic cognitive structures. Due to their schematic properties, “scenes” share essential char-acteristics with “frames”. For instance they have slots (“positions”) that can be variably occupied. Where, then, lies the difference between “frames” and “scenes”? Mirjam Petruck suggests interpreting Fillmore so that “scenes” are conceptual entities whereas “frames” are linguistic entities.

In the early papers on Frame Semantics, a distinction is drawn between scene and frame, the former being a cognitive, conceptual, or experiential entity and the latter being a linguistic one … (Petruck 1996: 1)

This interpretation implies that “frames” have a non-cognitive and a non-conceptual status, which raises the same problems that I discussed in the previous sections.

As I see it, Petruck’s definitions miss the point, and not only for this reason. In my view, in his early works Fillmore in no way denies that “frames” fulfil primary cogni-tive functions. On the contrary, the essential difference between “scenes” and “frames” seems to lie in other points. Whereas Fillmore’s “frames” refer initially to linguistic expression formats (in the sense of “case frames”), he reserves the concept of “scene” for the relevant presuppositions necessary for understanding a linguistic expression

263. On the term “scene” Zlatev (2003: 458) remarks: “[L]exical meaning is… relative to scenes.” And later he states (482): “[W]ords are learned in particular linguistic and situational contexts (language games) whose parts co-determine each other’s meaning”.

264. Quine already subscribed to a similar theory of inductive learning. Quine (1960: 56) writes: “We learn ‘bachelor’ by learning appropriate associations of words with words”.


In most natural conversations, the participants have, already ‘activated’, a number of shared, presupposed, scenes that we can speak of as being in their consciousness as they speak. (Fillmore 1975: 126)

Because such presuppositions go well above and beyond the arguments that a linguistic expression requires in its predicative function, on another occasion Fillmore expressly distinguishes “linguistic frames”265 from “scenes”.

[P]eople, in learning a language, come to associate certain scenes with certain linguistic frames. I intend to use the word scene… in a maximally general sense, to include not only visual scenes but familiar kinds of interpersonal transactions, standard scenarios, familiar layouts, institutional structures, enactive experiences, body image; and, in general, any kind of coherent segment, large or small, of human beliefs, actions, experiences, or imaginings. I intend to use the word frame for referring to any system of linguistic choices (the easiest cases being collections of words, but also including choices of grammatical rules or grammatical catego-ries – that can get associated with prototypical instances of scenes). (Fillmore 1977a: 63)

In fact, Fillmore emphasizes the fact that “frames” and “scenes” mutually activate each other and that “frames” are networked among each other just as “scenes” are. This is the basis of the inextricability of linguistic and encyclopaedic knowledge. Nonetheless one gains the impression that “scenes” are shape-like and accordingly visual perceptual impressions of comparable units, whereas “frames” are above all linguistic correlates or, in Fillmore’s words “labels” for linguistically naming and highlighting individual elements within “scenes”.

Although the conceptual status of both remains unquestioned, at the same time it can be noted that Fillmore consciously leaves the concept of “scene” vague. At one point he even contrasts cognitive “scenes” with “real-world scenes”.266 A “scene” could thus just as easily be the flora and fauna in a garden as a film scene or indirectly per-ceivable entities like the internal organs of a human (Fillmore 1977a: 71). As outlined previously, “real-world scenes” themselves can only become U-relevant under certain circumstances, namely, only in the retrospective mode that I spoke of in that context, and then only in the form of assumptions or convictions about “real-world scenes”. Fillmore is fully aware of this, and therefore by “scenes” he generally does not mean real-world facts but rather mental models of them. Nonetheless, his use of the terms “real-world” and “visual scenes” is to some extent deliberate. That is to say, in contrast to “frames”, “scenes” are characterized by a certain excess of epistemic information. They contain manifold details: numerous entities with a variety of external forms and

265. In the original: “language frames” or “linguistic frames”; cf. Fillmore (1975: 123, 126; 1977a: 63, 66).

266. “Communicators operate on these scenes and frames by means of various kinds of proce-dures, cognitive acts…, comparing presented real-world scenes with prototypical scenes, and so on.” (Fillmore 1977a: 66)


shades of colour, entities that distinguish themselves spatially from each other by mul-tiple figure/ground relationships. Our seeing eye encompasses such details of a “visual”, “real-world scene” even though they may in the next moment be withdrawn from our conscious perception. There is essentially little difference in the case of a “schema-tized scene”. A “scene” is schematized precisely because our consciousness can only access certain elements, i.e. retrieve them from memory, in the retrospective mode. Nonetheless, the volume of given information is still substantial.267

It is in this point that “scenes” and “frames” differ fundamentally from each other, because Fillmore uses the latter to refer merely to a “system of linguistic choices” (1977a: 73; my italics) and precisely not the experiential relations associated with “col-lections of words” or “grammatical categories”.

This changes in his later frame-semantic writings. Fillmore’s seminal essay Frame Semantics, published in 1982, marks an interesting change. Fillmore (1982a: 116–117), reflecting retrospectively on his case-frame theory, yet explaining only the concept of “scene” here at length, then introduces a decidedly semantic category under the label “cognitive frames” (117), which he explains as follows:

A ‘frame’, as the notion plays a role in the description of linguistic meanings, is a system of categories structured in accordance with some motivating context. … The motivating context is some body of understandings, some pattern of practices, or some history of social institutions, against which we find intelligible the creation of a particular category in the history of the language community. (Fillmore 1982a: 119)

After this redefinition, it seems only logical that in Fillmore 1984 the terms “scene” and “frame” should be replaced by the terms “schema” and “schematization” respectively.268 Fillmore defines “schemata” as complex cognitive units that evoke words – i.e. pho-nological units, using Langacker’s terminology – in recipients (Fillmore 1984: 143).269 One year later, Fillmore (1985) finally completely abandons the term “scene”. “Frames” now function as the ultimate embodiment of U-relevant knowledge.

Against this background, in the following I would like to propose the hypothesis that the concept of “scene” serves Fillmore as a means of gradually establishing a broad frame concept that is no longer limited to case frames, which is, above all, focused on the valency of verbs (just like frames in the FrameNet project).

267. Take, for example, a typical living-room scene as described in Minsky (1975), (1988: 249f.).

268. At one point the concepts “frame” and “scene” appear again: “The language we use reflects the ways in which we ‘frame’ or ‘schematize’ the world of the text …” (Fillmore 1984: 137), and one page further on he states, “[S]cenes and schematizations introduced quite incidentally in one part of a text can provide major scaffolding for later parts of the text’s envisionment”.

269. According to Fillmore, here the activation proceeds via categorization links (for instance, those which exist between schemata and instances), which is entirely in the sense of the discussed “symbolic principle”.


Case frames describe which deep cases (or “semantic roles”270) characterize the combinatory properties of verbs with respect to semantics (Fillmore 2006: 616). For this, Fillmore already understands case frames as construction schemata “that struc-ture our experience independently of language” (Fillmore 1977c: 102). Therefore, case frames are also closely linked with certain (world) knowledge domains and should not be misunderstood as “purely” syntactic formats. For this reason Fillmore equates case frames with schemata that allow us to linguistically access a domain of our world knowledge:

Our knowledge of [a] domain consists in knowing what the … schema is and in knowing how particular lexical items index particular parts of it. I see no particu-lar advantage in separating out one part of this as strictly semantics and another part as something else. (Fillmore 1977c: 102)

Although, according to this, case-frame theory already represents the postulate of U-relevance, thus rejecting a distinction between linguistic and world knowledge, many facets of knowledge are not taken into account in case frames that are contained in “scenes”. Thus, a case frame only provides information about how many and what semantic roles a linguistic expression, usually a verb, requires. Without laying claim to completeness, Fillmore (1968: 24–25) names six different deep-case types.271 However, on the one hand, it should become evident that these six slots cannot cover all areas of U-relevant knowledge, whereas, on the other, a case frame says nothing at all about the epistemic richness of potential elements that adopt the function of a semantic role. For instance, in this context consider the element seller, which occupies the semantic role of ‘agentive’ in the case frame of the verb flog off. The verb flog off certainly motivates inferences about the potential properties of the seller and the potential reasons why he is “flogging something off ”. I return to similar examples shortly.

By contrast, Fillmore’s later frame concept aims for a comprehensive semantic description that also encompasses the prerequisite knowledge actualized in the process of understanding as well as its structural-cognitive properties:

Such a frame represents the particular organization of knowledge which stands as a prerequisite to our ability to understand the meanings of the associated words. (Fillmore 1985: 224)

In contrast, the degree to which Fillmore’s earlier frame-semantic approaches (for instance Fillmore 1975, 1976a, 1977a) still follow the conception of case frames is

270. Dirven and Radden (1977) speak consistently of “semantic roles” instead of “deep case”, which does more justice to the sentence-semantic orientation of Fillmore’s theory. Semantic roles are fundamentally different to grammatical case. To this extent, the term “deep case” is misleading.

271. Namely, agentive, instrumental, dative, factitive, locative, and objective. On other occasions some of these are augmented by others (cf. Fillmore 1971a; 1971c).


made clear by his reduction of “frames” to formats for potential linguistic-grammatical expressions. Within a “scene” one can gain orientation through “the kinds of frames known as case frames, and a set of procedures or cognitive operations” (Fillmore 1977a: 66). Certainly, in contrast to case grammar, situative roles and not semantic roles form the object of reference for frames.272 Unlike semantic roles, situative roles need not be morphologically marked; they can also constitute elements of a frame independently of valence condition attributes. Nonetheless, often situative roles can be deduced from semantic roles.

This can be illustrated using the example of the commercial transaction frame, an example that Fillmore uses on several occasions.273 This frame contains the elements (a) ‘buyer’, (b) ‘seller’, (c) ‘goods’, and (d) ‘money’, which correspond to verbs such as buy, sell, cost, and spend, respectively (cf. Table 1). Each verb evokes the entire com-mercial transaction frame, but places it in its own specific perspective.274 Compare the following examples:

(1) Peter has bought a car from John for 20,000 dollars. (2) John sold Peter a car for 20,000 dollars. (3) The car cost Peter 20,000 dollars. (4) Peter spent 20,000 dollars on a car.

In (1), the verb buy places the perspective on the buyer, whereas the other frame ele-ments (b), (c), and (d) shift into the background, but still remain relevant to under-standing. Examples (2)–(4) are similar. What is crucial is that what Fillmore attempts to illustrate with a view to semantic frames is not significantly different to case frames. In (3) and (4), the case frames of cost and spend do not have a syntactic slot on the surface level for the frame element ‘buyer’, whereas in the other two cases all four elements are realized. This certainly does not mean that the frame element ‘buyer’ does not play a role in (3) and (4). Precisely because all of the verbs named, including cost and spend, evoke the whole frame, that is, the entire meaning potential of an expression, this ele-ment may not be ignored in the meaning analysis either. What is remarkable is that all four frame elements can be transferred into a deep case (see Table 1).

272. Cf. also Petruck (1996: 1).

273. Fillmore (1976b: 25; 1977a: 58–60; 1977b; 1977c: 99; 101–102; 1982: 117). In his later papers, in which the term “scene” no longer appears, Fillmore interestingly no longer uses this example.

274. Perspectivize means here to fill syntactically relevant slots with values. The slots of a frame that are evoked but not filled have the status of a meaning potential (they are thus slots filled with default values). This makes it possible, for instance, to establish coherence between sentences. Compare the following examples, in which the shared default value “money” is crucial: The car costs 20,000 dollars. He paid cash. The car costs 20,000 dollars. He spent every last cent on it.


Table 1. Relationship between elements in the commercial transaction frame, verbs, and deep cases (following the typification in Fillmore 1968)275

Frame elements Verbs Deep case

buyer → buy → agentiveseller → sell → dativegoods → cost → objectivemoney → spend → instrumental275

The added value of speaking of “schemata” or “frames” instead of case frames is thus extremely low. He largely limits himself to shifting the analytical focus from issues of valency to those of meaning theory, however, without significantly extending the area covered by U-relevant knowledge in doing so.

However, the introduction of the concept of “scene” shifts the area of U-relevant knowledge into view in its entire range. So as not to proceed reductionistically and succumb to a “syntactocentric” fallacy (as I already discussed in connection with the level-semantic models), Fillmore uses the concept of “scene” in his early papers in a regulatory manner, as it were, to fulfil a compensatory function with respect to the relevance of understanding, a postulate he also subscribes to (for instance, in Fillmore 1985).276 A wealth of knowledge factors are required to understand a single frame ele-ment. Of course, it is plausible, and within case-theoretical considerations certainly makes sense, to posit that, for instance, the verb cost not only refers to the frame ele-ment ‘goods’, but also evokes the entire frame. However, U-relevant understanding is not exhaustively covered by this. Costs may be high or low, reasonable or excessive; they may say something about the status of the “goods” (take the cost of a painting at an art auction, for example) and thus distinguish the buyer in multiple respects; and not least, they may be differentiated with respect to the reference object (run-ning costs, consequential costs, overheads, supplementary costs). The list could go on indefinitely.277

275. By “instrumental” Fillmore means, above all, the means necessary to perform an action (and not the purpose of an action). Inasmuch as the noun money does not occur in the sentence subject position and in combination with the verb spend, the deep case “instrumental” differs from the three other deep cases. What is important here is that the verb spend (in contrast to the other verbs) profiles the instrumental deep case or, in other words, shifts it semantically into the foreground.

276. Following Jackendoff ’s criticism of the standard model of generative grammar. “Syntacto-centrism” is a “scientific mistake” (Jackendoff 2003: 654), “just an assumption that itself was partly a product of historical accident” (Jackendoff 1997: 19).

277. In complete contrast to case-frame theory, Fillmore’s frame semantics follows an approach “in which there can be an unlimited number of semantic role systems, associated with individual frames, an approach which even allowed some frames to be unique to single lexical items” (Fillmore 2003: 466).


Fillmore recently very appositely characterized the way in which the frame concept had evolved in his work:

The case frames started out as clusters of participant roles using, initially, names from an assumed universally valid finite inventory of such roles and it was thought that any verbal meaning could be seen as using some collection of these. The frames of current frame semantics, in contrast, are described in terms of charac-teristics of the situation types themselves, including whatever could be said about the background and other associations of such situations. (Fillmore 2006: 616)

“Scenes”, understood as mental models, are “situation types” in the sense that Fillmore explains it here. They contain those background assumptions necessary to understand a linguistic expression within a particular context.278 However, as soon as a genuinely frame-semantic frame concept is introduced, the term “scene” no longer has any place within in the theory. The term has become superfluous because it has cleared the way for a new understanding of frames as a representational format for U-relevant knowledge.

The question of whether “scenes” constitute elements within semantic units there-fore cannot be answered with a clear yes or no. Fillmore’s concept of scene is there-fore too undifferentiated and too ambiguous. If a “scene” is used to refer to a (largely conventionalized) experiential context that language users connect with a particular linguistic expression, and if this can also occur in the retrospective mode in the form of (propositionally structured) background assumption, then “scenes” are indeed in-tegral components of semantic units.279 However, in this case the concept of “scene” is identical with the concept of “frame” in Fillmore’s later publications. In contrast, if a “scene” refers to a “real-world scene”, then it is certainly a perceptually and phenom-enologically relevant factor that may without doubt function as an implicit reference object “in primary actions and normal life routines” (Demmerling 2002: 162), but in this case it is not reflected in semantic units, because it cannot be U-relevant due to its non-propositional disposition in the retrospective mode.

Overall, it is thus not necessary to posit a second organizational format for seman-tic units in addition to “frames”.

278. Contrary to Zlatev, for instance, Fillmore does not use the term “background” in the Searlean sense here. He uses it in its broadest sense of presuppositions, that is, propositionally structured units, which is why I prefer the term “background assumptions”. In contrast, Searle, at least in Searle (1979), is referring to social practices; however, in Searle (1998: 147), he equates “background” with background assumptions.

279. This view dominates in the literature. Regardless of the demonstrated ambiguity, Post (1988: 38) for instance, equates “scenes” with “mental representations”: “Fillmore’s scenes are chunks of knowledge of varying size, stereotypical, simplified, representing a common sense understanding of real world situations. Scenes are intended to be mental representations of language users’ real world experiences, i.e. scenes are also experientially based.”


4.3 Relations

Finally, to determine whether Fillmore’s “frames” (or, alternatively, “cognitive scenes”) structure semantic units, it must be ensured that these bundle conventionalized back-ground assumptions and not only those that are contextually determined. Anticipating this objection, I argue in Section 6.2.1 that it is frames with default assumptions that form a semantic unit. Nevertheless, Fillmore himself speaks only very generally, as in the previous quote, of a “background” and “associations” that are bound to a “situation”. However, the status of such “backgrounds” and “associations” remains unspecified. Seana Coulson in her book on frame semantics was the first to distinguish between the standard elements (“default values”) and the emergent, contextually dependent elements of a frame (“nonce sense”) (Coulson 2001: 47).280

Semantic unit

Phonologicalunit

(a) (a)́

(b)

Symbolic unitConceptualmeaning via [semantic]

conceptualization

Phonologicalconceptualization

Usage event

Figure 7. Relations: (a) between a semantic and a phonological unit; (a)′ between a conceptual meaning and a phonological conceptualization; (b) between (elements) in a semantic unit and (elements) in a usage event

Let us assume that this distinction is given and meaningful and direct our attention once again to Langacker’s basic semiotic model. We see that two important relations have not yet been addressed:

a. the relation between a semantic and a phonological unitb. the relation between a semantic unit and its (potential) conceptual meaning.

280. “Consequently, meaning construction is best construed as occurring on a continuum that has the retrieval of standard frames and default values at one hand, and the creation of nonce sense at the other.” (Coulson 2001: 47)


To illustrate this, the two relations (a) and (b) have been added again to Figure 7 and highlighted. The illustration also shows that apart from the relation (a) there is a vari-ant (a)′ that refers to a conceptual meaning.

The concept “relation” (or “relationship” or “correlation”) is in this context only a metaphor for an inferential process that a phonological unit sets into motion in a comprehending language. This applies to the highly routinized correlation between phonological and semantic units no less than to the significantly less routinized transi-tion from a semantic unit into a conceptual meaning.

Earlier I explained the extent to which the relationship between a phonetic and a semantic unit is of a symbolic nature. Conversely, according to Langacker, coding relationships exist between (elements in) symbolic units and (elements in) usage events (Langacker 1991a: 294–298). However, it remains unclear how the transition from a semantic unit into a potential conceptual meaning occurs. Which elements in the se-mantic unit are transferred to the conceptual meaning? In what manner do elements in the semantic unit “code” the conceptual meaning? And to what extent does the latter differ structurally from a semantic unit? Similar questions arise with respect to the relationship between phonetic and semantic units, and I begin by addressing these.

4.3.1 Evoked and invoked frames (C. Fillmore)

Let us begin with the relation (a) and (a)′. One reason for making frames produc-tive for semantic and semiotic issues was Fillmore’s observation that many linguistic expressions fulfil, as it were, an indexical function. Not only the meanings of words like buy, give, earn, etc. rely on schematized experiences of human actions, but also more complex expressions like idioms and phraseologisms require knowledge of the appropriate context in which they could occur in order to be understood (Ziem and Staffeldt 2011). Fillmore’s (1976a: 25) examples of this are:

(5) Speak of the devil! (6) What a small world.

These two routinized expressions as a whole do not have to be compositionally analysed to indicate a typical social context in which they could respectively be uttered. This circumstance lends them their status as “constructions” in the technical sense explained above). The respective meanings of (5) and (6) are more than the sum of the meaning of their parts. “My point is that these expressions too are, in a sense, indexical, in that their appearance is predictive of a number of details of the situations of their performance” (Fillmore 1976a: 25). Linguistic tokens achieve a kind of stimulus function inasmuch linguistic expressions of differing complexity can prompt language users to recall the contextual relationships in which these expressions commonly occur: “[W]ords and morphemes ‘activate’ the associated schemata in the interpreter’s mind”, as Fillmore (1976a: 13) formulates it. As is the case with (5) and (6), it is not always possible to distinguish between semantic and pragmatic knowledge. Thus, although the schema


activated here provides meaning knowledge of what a devil is in (5) and a world is in (6), what is nonetheless idiosyncratic to these two sentences is that we understand them as an integrated part of a situative embedding structure.

It is phonological units that motivate such activations of schematic knowledge. As long as the link between a phonological unit and its associated meaning has the status of a construction, phonological units can certainly be of a complex nature. So, for example, a semantic unit is just as much associated with the simple phonological unit [devil] as with the very complex unit [speak of the devil]. Let us now assume that semantic units constitute a complex structural unit organized by frames. As Fillmore stresses, the activation of frames can then be considered from two perspectives:

Interpretative frames can be introduced into the process of understanding a text through being invoked by the interpreter or through being evoked by the text. A frame is invoked when the interpreter, in trying to make sense of a text segment, is able to assign it an interpretation by situating its content in a pattern that is known independently of the text. A frame is evoked by the text if some linguistic form or pattern is conventionally associated with the frame in question. (Fillmore 1985: 232)

Thus linguistic expressions “evoke” schematized meaning knowledge.281 From the point of view of a comprehending individual this is not so much a mechanical opera-tion as a process of active construal or inferencing. In this context, Fillmore speaks of “invocation”, thus referring to the interpretive accessing of texts that gives them their textual coherence.282 Thus, one could say that a phonological unit evokes a frame, whereas a language user invokes a frame.

Now in my view, Fillmore’s concept of “evocation” exactly characterizes the rela-tion marked (a) in Figure 7, whereas the concept of “invocation” corresponds to the relation (a)′.283 There are good reasons for this.

If we take Fillmore’s previous quote at face value, then a linguistic expression can only be “evoked” under one condition: the evoked frame elements must be conven-tionalized knowledge elements. At another point he states:

281. Cf. also Fillmore (1982a: 116, 122, 124; 1985: 242).

282. Fillmore uses the verb “invoke” in numerous places: Fillmore (1982a: 124; 1985: 232, 234). However, in German, it would be completely inappropriate to use the apparently equivalent term Invokation for this phenomenon due to the religious frame it evokes. For this reason, when discussing these concepts in German, I use the term Aufrufen for “evocation” and Abrufen for “invocation”.

283. As a rule, this distinction is overlooked in the literature. Thus, for instance, Beaugrande and Dressler (1981: 90) also speak of “evocation” with reference to coherence phenomena. In contrast, both Langacker (e.g., Langacker 1987: 158) and Busse (e.g., Busse 1991a: 102) appear to acknowledge the difference.


[W]e have cases in which the lexical and grammatical material observable in the text ‘evokes’ the relevant frames in the mind of the interpreter by virtue of the fact that these lexical forms or these grammatical structures or categories exist as indices of these frames … (Fillmore 1982a: 124)

If a frame is “evoked” by a word form, a grammatical structure, or category – in short, by a phonological unit – this unit performs an indexical function.284 It evokes a frame without the user being able to influence the form of the evoked frame, indeed, without them even being able to prevent the frame from being evoked.285 The reason for this lies, on the one hand, in the strong symbolic bond that links form and content to each other and, on the other, in the disposition of the evoked frame itself. This, in fact, merely consists of default values. Contextual aspects of meaning that arise from each specific contextual embedding structure of a word are by definition excluded in the case of evocation (Petruck 1996: 3). For this reason it comes as no surprise that individual sign users do not play a role from the perspective of evocation. What is decisive is the meaning that individuals link to a phonological unit simply because they are members of one and the same language community, which means they arbitrarily associate a particular phonological unit with a particular semantic unit by dint of belonging to this language community.

However, understanding a word, a sentence, or a text goes far beyond this evoca-tion process. It requires the interpretation of these linguistic expressions. Fillmore argues that, with respect to every symbolic unit,286 a sign interpreter faces the task of providing a satisfactory answer to the question of why a sign producer chose precisely this phonological unit in precisely this context.287 This gives rise to three sub-questions (Fillmore 1985: 234):

i. Which frames have already been activated within the text world at the relevant point of time?

ii. Which values (fillers, default values) have already been assigned to the slots in this frame?

284. In the sense of Searle (1979: 44). In addition, Putnam points out that, by necessity, every natural language exhibits deictic components. Only if this were not the case could semantics and pragmatics be distinctly separated.

285. Lakoff (2004: 3) provides a vivid example of this: “When I teach the study of framing at Berkeley, in Cognitive Science 101, the first thing I do is I give my students an exercise. The exercise is: Don’t think of an elephant! Whatever you do, do not think of an elephant. I’ve never found a student who is able to do this. Every word, like elephant, evokes a frame, which can be an image or other kinds of know ledge: Elephants are large, have floppy ears and a trunk, are associated with circuses, and so on. The word is defined relative to that frame. When we negate a frame, we evoke the frame”.

286. Fillmore (1985: 234) speaks of “conventional linguistic form[s]”.

287. Cf. Fillmore (1985: 234): “Why did the speaker select this form in this context?”


iii. What function could the communication settings addressed in (i) and (ii) fill in the frame that the currently given phonological form has evoked?

It is not until these questions have been answered that the transition is made from a semantic unit to a conceptual meaning. Regardless of how vague the concept of “text world” is in (i) or the relationship between the text world and the activated frame may be in (ii), and regardless of how equally unclear it is at this point how the processes required for the conceptual integration concretely operate in (iii) – there is nonetheless no doubt that in (i)–(iii) Fillmore means an interpretive process of conceptualization in the most comprehensive sense (cf. Busse 2007), since the “maximally rich inter-pretation” (Fillmore 1985: 234) aimed for by an interpreter can only be achieved by recurring to the entire range of potentially U-relevant knowledge.288 Coherence is not achieved simply by phonological units “evoking” frames, but rather, in comparison with given textual information, established knowledge is “invoked” via frames to arrive at a plausible and consistent text interpretation. And that means that language users seek answers to questions (i)–(iii).

The interpreter must … invoke the kinds of knowledge which will make it possible to integrate the answers to these questions [(i)–(iii)], for the individual lexical items and grammatical constructions found in the text, in the job of constructing an interpretation of the text as a whole. (Fillmore 1985: 234)

This results in an essential distinction between evoked and invoked frames. The latter are to a greater extent the result of inferential construal operations,289 which is why Fillmore also speaks of “interpretive” frames respectively “frames of understanding”. With their help, language users succeed in maintaining the principle of “perceptual constancy” (Hörmann 1994: 179–212), whether through the invoked frames that stem from background knowledge, or through the invoked frames that previous text ele-ments (in the sense of “knowledge of the already constituted world” in Busse’s typol-ogy) have evoked and now prove to be U-relevant.290 It therefore follows that invoked frames generate coherence. Thus Busse also states:

288. Cf. the case study in Fillmore (1981: especially 250–252), which demonstrates what con-crete form the answers to questions (i) and (ii) can take.

289. This may initially be reminiscent of the problematic distinction between “explicit” and “implicit textual information” within psycholinguistic research on inferences. Language users inferentially determine “implicit textual information”; it corresponds to the dimension of “invo-cation”. “Explicit textual information” is contained in the text itself; it corresponds to the dimen-sion of “evocation”. Nonetheless, my criticism, which also holds here, is that this also involves inferential processing by text interpreters.

290. Cf. Fillmore (1982a: 124): “Invoked frames can come from general knowledge, knowledge that exists independently of the text at hand, or from the ongoing text itself ”.


Accordingly “coherence” is not so much a property of a text as being a chain of sign-utterance sides [i.e. phonological units], but rather a property of the text worlds activated by the use of linguistic signs; a “relationship” between individual reference elements (be they “concepts” or “frames”, “scripts”, “planes”) is only given when this relationship appears to be possible in terms of the participants’ world knowledge. (Translated from Busse 1991a: 102)

“Concepts”, “frames”, “scripts”, and “plans” are particular types of invoked frames. None of them are directly motivated by the materially given and cognitively represented utterance form of a language sign (i.e. by phonological units); they do not “hang” on the words and expressions themselves, as is the case with evoked frames. Instead, they bundle inferences such as those that I discussed in connection with maximalist and situative inference theories.

This difference can be illustrated using the following example.

(7) After they had both said “I do”, they were allowed to kiss each other.

Although (7) does not expressly mention a wedding, we can easily infer this larger context on the basis of the given text elements. We then invoke the wedding frame taking recourse to our background knowledge. The motivated inference construc-tions are based on the correlation of two evoked frames, of which one structures the semantic unit [say_i_do] and the other the unit [kiss_each_other]. However, it is the evoked wedding frame that establishes the coherence between the two semantic units by interpreting them as instances of the slots ‘how’ and ‘who’.291

As illustrated in Figure 8, the context establishes a process of invocation. Thus, the script-like relationship between the units [say_i_do] and [kiss_each_other], represented by the dashed line, can only be realized with the help of the evoked wed-ding frame. For this reason, the dashed line in Figure 8 is found outside the inner box. In contrast, the content side of symbolic units is evoked. Units like [say_i_do], [kiss], [both], and [they] each evoke a frame. With Fillmore’s case grammar in mind, one could argue in addition that the two verbs also activate a case frame whose semantic role ‘agentive’ (cf. Fillmore 1968: 24–25) is occupied by the same values. However, it is important to recognize that the comprehensive semantic interpretation does not occur until frames are invoked. To do this, phonological units fulfil the function of cogni-tive stimuli as it were, even though the inferential (re-)construction of the meaning intended by the text producer is the text recipient’s own active cognitive operation.292

291. Certainly speaking of the slots ‘how’ and ‘who’ is a simplification here. In Section 6.3 and in the following example analysis, we see that the slots in a frame do have the character of ques-tions, but of a much more complex structure than simple interrogative pronouns.

292. Fillmore (1982a: 124): “An extremely important difference between frames that are evoked by material in the text and frames that are invoked by the interpreter is that in the latter case an ‘outsider’ has no reason to suspect, beyond a general sense of irrelevance or pointlessness in the text, that anything is missing”.


We can also note that Langacker’s differentiation between symbolic units and usage events is reflected in Fillmore’s distinction between the processes of evocation and invocation. A phonetic unit evokes a semantic unit. Thus, in Figure 7 the relation (a) symbolizes a process of evocation. By contrast, the conceptual meaning results from the invocation of frames arising from background knowledge. The realization of the relation (a)′ in Figure 7 places evoked frames in relation to each other; these establish coherence taking contextual aspects of meaning into consideration.

4.3.2 Meaning potentials (J. Allwood)

Let us now turn to the second relation, that between a semantic unit and its (potential) conceptual meaning. Relation (b) in Figure 7 is not addressed by Fillmore, although it could easily be explained using frame semantics. For this reason, I begin by following Langacker (1987: 65–66), who characterizes the relation (b) as a coding relationship. According to this view, semantic units serve as markers and departure points for con-struing a semantically richer meaning, that is to say, conceptual meanings that arise only in a particular communication context.

Because semantic units are only composed of conventional aspects of mean-ing, they are usually not identical with conceptual meanings (cf. Clark and Gerrig 1983: 605). Rather, semantic units remain underspecified and stand at the beginning of the conceptualization process necessary to achieve a contextually adequate meaning for an expression. Two simple examples clearly illustrate this:

[KISS_EACH_OTHER]

[BOTH] or [THEY]

[SAY_I_DO]

[WEDDING]

Who?

How? How?

evoke

invoke

Figure 8. Evoked and invoked frames, demonstrated using sentence (7)


(8) They both said “I do”. (9) They kissed each other for a long time.

How the verbs in (8) and (9) are semantically interpreted depends significantly on the given context.293 [say_i_do] can be interpreted differently if it is not a case of entering into marriage but rather of the formal consent between two contractual partners. And another contextual aspect of meaning is activated in (9) if, instead of a kiss between a bride and groom, a fraternal kiss is meant, such as that between two statesmen from the former Eastern bloc. Thus, contextual knowledge can also lead to a frame other than the wedding frame establishing the coherence context in Example (7).294 For this reason, invoked frames are prone to disappointment and subject to review to a much higher degree than evoked frames.

However, what characterizes a coding relationship? Departing from very similar premises as Langacker (1987), Jens Allwood (2003) suggests distinguishing between the meaning potential of a linguistic expression on the one hand and its actualized meaning on the other.295

It is suggested that actual meaning on the occurrence level is produced by context-sensitive operations of meaning activation and meaning determination which combine meaning potentials with each other and with contextually given informa-tion rather than by some simple compositionality operations … (Allwood 2003: 29)

Accordingly, the process characterized by Langacker as coding would be equivalent to (i) the activation of a particular semantic potential, (ii) the contextually governed realization of segments of potential, and (iii) the integration of contextual aspects of meaning. My suggestion now is that Allwood’s concept of meaning potential refers to the relation (b), that is, to the transition from a semantic unit to a conceptual mean-ing. Or more precisely, the meaning potential of an expression corresponds to the set of default values that are associated with that expression and may possibly influence a conceptual meaning.

293. Which does not mean that this is a case of semantic polysemy.

294. Imagine that (7) were embedded in the following scenario being recounted by a third person: “When Yilmaz and Olea met, they were initially uncertain about how to express their affection for each other. They were both really in love for the first time, but in their culture it is usual that men and women may only kiss each other after expressly giving their mutual consent. Finally, Yilmaz felt confident enough and asked Olea whether she would permit him to kiss her. Olea did not hesitate for a second and asked him the same question. After they had both said “I do”, they were allowed to kiss each other.”

295. Busse also states, taking recourse to Hörmann (1977) and Beaugrande and Dressler (1981): “The ‘meaning’ of an expression (in a text), a text segment, or an entire text would then be the potential for text recipients to perform a targeted activation of knowledge.” (Translated from Busse 1991a: 101–102)


To grasp the relation (b) even more precisely, let us begin by examining the way Allwood explains the concept of meaning potential on another occasion.

The meaning potential is all the information that the word has been used to convey either by a single individual or, on the social level, by the language community. The meaning potential, then, does not result from trying to find a generally valid type meaning for a word; rather, it is the union of individually or collectively remembered uses. (Allwood 2003: 43)

The meaning potential results exclusively from the (common) usage of an expression. Allwood represents a contextualist position with which he attempts from the outset to prevent tendencies towards semantic reification.296 Meaning potentials emerge from language usage and only become relevant in concrete language use. At the same time, the meaning potential is not to be misunderstood as a set of semantic features. In fact, meaning potentials only exist as long as there are language users who activate them, and meaning potentials vary insofar as expressions vary in use.

It follows that the activation of meaning potentials needs to be related to individual communicators, since different individuals do not share the same elaboration of the meaning potentials and therefore cannot always activate the same information. (Allwood 2003: 45)

Even if one cannot abstract from individual usage as the locus of the emergence and modification of a meaning potential, one and the same linguistic expression must nonetheless be recurrently used in comparable contexts in a sufficiently large language community in order to evolve a meaning potential.

Meaning potentials are thus a result of conventionalizations of semantic opera-tions meeting contextual requirements. (Allwood 2003: 50)

Since meaning potentials develop from conventionalization processes, this suggests that they should be identified with the set of all conventionalized aspects of meaning of a linguistic expression that are commonly used within a particular language com-munity at a particular time. According to Allwood, a meaning potential is associated with every word, regardless of which word class it belongs to. If this were not the case, one would have to assume there were a set of linguistic signs that were not symbolic units because their meaning side was missing. This decidedly contradicts the proposed “symbolic principle”.

296. Allwood’s motivation for speaking of “meaning potentials” is based in the movement to dis-card reifying semantic models like the theory of typical or “basic meanings” (‘Grundbedeutungen’) or abstract generalizations (‘Gesamtbedeutungen’), as he calls them. Among the latter is the se-mantic level model I have criticized, on which Allwood (2003: 30) writes: “This approach, in fact, is often equivalent to essentialism, which characterizes the essence or common denominator of a word’s uses in terms of the necessary and sufficient conditions for whatever the word denotes”.


Over and above Allwood’s observations, it is plausible to state that meaning po-tentials “hang” on linguistic expressions of every level of complexity (i.e. also complex words, phrases, idioms, etc.). Although Allwood restricts himself largely to the lexical dimension, he nonetheless stresses that the syntagmatic combination of two words forms a new, more complex meaning potential (Allwood 2003: 52–54). Put more pre-cisely, every construction is just as much associated with a meaning potential as is every symbolic unit that is composed of several constructions, except that these have emerged out of the combination of different meaning potentials and thus are more complex in form. [fraternal kiss] is just such a complex unit. The combination of several meaning potentials leads to change in some of the default values, whereas oth-ers remain the same.

Two things result from this. The meaning potential is, on the one hand, more than a frame with unfilled slots, because it contains semantic information about the common usage events of linguistic expressions. On the other hand, it is at the same time less than a conceptual meaning (i.e. than a frame whose slots have been filled with particular values and default values), because it simply remains the potential of realizable meaning aspects. The part of the potential that is actually exploited in an individual case when processing an expression is another issue, which can only be answered relative to a given context.

In my view, speaking of a “potential” seems to be crucial for another reason. If semantic units are symbolically associated with phonetic units (in Langacker’s broad sense), this means that conventional aspects of meaning must be specifically inter-preted. They are not simply there, but are themselves the results of cognitive opera-tions or “activations” (Allwood 2003: 47). What facets of knowledge a linguistic sign evokes may vary from language community to language community in this process, just as it is possible that actualizations may be unsuccessful. Consequently, there is no determination relationship but an activation relationship.297

Since slots only describe the predication potentials of a linguistic expression, for which reason I later speak of the “predication potential” of an expression, at least some of them must already be filled with values for a conceptual meaning to be of relevance. Beyond the predication potential indicated by slots, a linguistic expression also evokes concrete semantic information, namely default values, which then unfold their efficacy in the process of understanding. The relation (b) ultimately comes into play when the activation of this potential is already governed by the context. For in-stance in (9), the context gives more concrete form to what the motive of the kiss is. The actualized default value varies according to whether it is a fraternal kiss or the kiss

297. Thus the essential difference to the semantic level models is crystallized in the term “poten-tial”. Determination replaces potentiality in level semantics, and where for this model a process of semantic selection is the premise for meaning constitution, in the holistic countermodel, activation relationships prevail. The concept of potential is used similarly by Halliday (1976), Langacker (1987: 66), and Zlatev (2003).


of a bride and groom.298 Such specification is not contained in the meaning potential because it arises from the particular context. However, what is certainly contained is, for instance, information about the participants, their emotions, the body parts involved, etc. In addition, though, just like every default value, contextually induced default values have an average expected assumption about the world (in this case, for instance, that a bride and groom generally kiss because they wish to publicly express a particular emotional state).

By “fillers” I mean here the predications performed in the text, whereas by “default values” I mean inferred assumptions in the narrower sense, i.e. inferred by the text recipients on the basis of their own foreknowledge of performed predications.299 To indicate that there are a multitude of slots, default values, and fillers, the individual elements are indexicalized in Figure 9.

The relationship between frame, meaning potential, and conceptual meaning is illustrated in the following way (cf. also Figure 9):300

– The slots of an evoked frame give the predication potential of the linguistic expres-sion that evoked this frame.

– However, a linguistic expression always occurs within a particular context, and this circumstance results in the same expression disposing over a similar meaning potential in comparable contexts: Certain slots in the evoked frame are occupied by default values, whereas other slots remain irrelevant, i.e. they are neither acti-vated nor occupied by default values.

– Finally, when an expression is applied communicatively, due to given contextual data only part of its meaning potential is actually activated and thus becomes involved in the conceptual meaning of the linguistic expression in question.301 The meaning po-tential contains more default values than are represented in the conceptual meaning of an expression. In addition, the actually performed predications contained in the sentence and text (fillers) enter into the conceptual meaning. They are not reflected in the meaning potential, since fillers give the potential more concrete form.

298. The empirical implementation of the frame concept in this book is oriented towards Konerding (1993). Accordingly, the slot mentioned here would be: “What motive is there for the action?” (cf. Konerding 1993: 341). This example is only arbitrarily selected; it is just as easy to provide other default values to demonstrate different framings of the verb kiss, as the follow-ing remarks show.

299. For a more precise distinction, cf. Sections 4.4 and 4.5.

300. In this context I understand frames as cognitive units of knowledge representation, not as tools of textual analysis. For this reason, as illustrated in Figure 9 the frame of an expression encompasses both the meaning potential and the conceptual meaning. This inclusive relation-ship is explained by the fact that cognitive frames never only consist of slots, but are always also linked to fillers (implemented predications) and default values (inferred predications). A frame consisting only of slots is an analytical construct. See also Section 6.3.

301. Similarly, Allwood states (2003: 53): “Contextually determined meanings thus result from partial activation of the meaning potentials of the expressions guided by cognitive operations”.


Conceptual meaning

Frame

Default values Fillers

Meaning potential

Slotss1 s2 sn d1 d2 dn f1 f2 fn

Figure 9. The relationship between frame, meaning potential, and conceptual meaning. “s” represents slot, “d” default value, and “f ” filler

Consequently, one must imagine the relationship between frame, meaning potential, and conceptual meaning as having parts that are entirely inclusive and those that are only partially inclusive, as illustrated in Figure 9. An evoked or invoked frame com-pletely encompasses both the meaning potential and the respective conceptual mean-ing of an expression, whereas the conceptual meaning extends beyond the meaning potential, that is, the latter only partially includes the former. Although the conceptual meaning is “inherent” in the meaning potential, it nonetheless encompasses more specific information, which is why fillers do not occur in the meaning potential. The meaning potential, in turn, exploits part of the predication potential indicated by the slots but does not include any unoccupied slots, only default values.

Finally, I would like to illustrate these facts using Example (9). Once again:

(9) The kissed each other for a long time.

Three questions arise: (i) What slots are opened by the frame that structures the se-mantic unit [kiss each other]? What is the meaning potential of the expression? And what does a potential conceptual meaning consist of?

With respect to (i): Slots have the form of questions and can be determined by means of a so-called “hyperonym type reduction”.302 In our case this means that the (first nominalized) verb kiss each other can be retraced to the irreducible hyperonym action. The frames of both expressions, kiss (each other) and action, share the same slots, some of which are summarized in Table 2.303

302. A more precise explanation of the process can be found in Section 6.3.2.

303. Cf. Konerding (1993: 341–348). This lists many more slots, which I have omitted here for practical reasons.


Table 2.304 A selection of the slots in the frame evoked by the expression kiss (each other)

# Slots

(1) What motives are there for the action (of kissing each other)? (2) What preconditions are there for the action?(3) In what way is the action realized?(4) What is the essential aspect of the action for the actants?(5) What does the action announce or bear witness to? (6) In what larger functional context does the action figure?(7) How long does the action take? (8) Which actants participate in the performance of the action? (9) Do the actants act according to a role? (10) Under what conditions is the action performed?(11) Und what conditions does an actant continue the action? (12) What phases or sub-events does the action consist of?(13) Under what conditions does the manner of action characteristically change?(14) Is there a typical plan of action?(15) In what way can an actant correct a lack of success?

With respect to (ii): Now, the meaning potential encompasses the set of all default values, that is, all expected assumptions that language users potentially actualize to answer the listed questions. Since these deal with conventional aspects of meaning that are based in recurrent experience, it is to be expected that many of these questions, but not all of them, can be implicitly answered using default assumptions. Although something about the action of kissing-each-other can be predicted with respect to the slots (13)–(15) (and potentially also (12)), the corresponding default assumptions may not be reflected in the “meaning potential”. The opposite is true of the other slots. I take up three to illustrate this.

– Default value for (1): The motivation for two people kissing each other lies in a mutually felt emotional state that can range from acknowledgement to love.

– Default value for (4): What is essential for the action of kissing-each-other is that skin contact between two people occurs, usually in the mouth and cheek area.305

– Default value for (8): Generally two people are involved in the action of kissing.306

304. Lists of slots such as this are treated in the following as tables and not as figures.

305. This default value at least applies to the reflexive verb kiss each other. The transitive verb evokes another frame (more precisely: certain slots are filled with other default values).

306. It is clearly recognizable here that default values are culturally dependent, contingent units. The specification that it must be a man and a woman who share the wedding kiss is strictly speaking no longer applicable.


With respect to (iii): Let us assume that we are dealing with the kiss of a bridal couple after they have said “I do”. In this context, what is the conceptual meaning of the re-flexively used verb kiss each other? The contextual data initially provides some fillers that occupy particular evoked slots in the frame.

– Filler for (6): The action occurs in the context of a wedding.– Filler for (8): Only the bride and the groom participate in the action of kissing.– Filler for (7): The action lasts a relatively short time (more precisely: seldom longer

than five seconds).

Whereas the first two instantiated fillers relate to contextual data, the third comes from the example sentence They kissed each other for a long time, which itself predicates something about the action of kissing, namely that it took a long time. The slot (7) is also assigned a value. In addition, the given context motivates some default values, of which some shift into the foreground while others only function as background assumptions. All the default values that arise from the invoked wedding frame are profiled.307 I take up the three most important ones:

– Default value for (2): The kiss occurs in the context of an institutional, script-like procedure (thus, the kiss may not occur before saying “I do”).

– Default value for (3): The kiss is realized in a moderate manner.– Default value for (5): The kiss has a symbolic value and attests to mutual love.

At the same time, default values from the “meaning potential” are retained. This is the case with slot (4).

It is certainly possible to empirically support the hypothesis that the activated frame of an expression encompasses both the meaning potential and the conceptual meaning of the same expression and that the conceptual meaning gives concrete se-mantic form to the meaning potential (that is, the latter does not encompass the for-mer). In addition, there is evidence for Allwood’s claim that the meaning potential of an expression represents a “basic semantic unit”,308 although a systematic and com-prehensive application is most unlikely without taking recourse to frames (and their structurally constituent slots, fillers and default values).

If we abstract from this individual case analysis and return to the initial question, we can conclude the following: Although Allwood’s discussion of meaning potential

307. This is precisely what Allwood (2003: 52) meant when he wrote: “When used, a linguistic expression activates its meaning potential through cognitive operations whose function is to achieve compatibility between the meaning potential of a particular expression, the meaning potentials of other expressions, and the extralinguistic context”.

308. “In brief, I would like to suggest that the basic unit of word meaning is the ‘meaning poten-tial’ of the word.” (Allwood 2003: 43) Meaning potential is also a “basic semantic unit” because it can be applied on all levels of organization of linguistic signs and also because every meaning actualization represents an actualization of potential.


remains largely of a theoretical nature, the example analyses show that his suggestions can certainly be implemented in the context of frame semantics. In particular, the concept of meaning potential helps illustrate and make plausible the transition of a semantic unit into a conceptual meaning – that is, relation (b).

chapter 5

Frames as schemata

In Chapter 4, following an understanding of linguistic signs as constructions or sym-bolic units, I used a frame-semantic perspective to address three relations that arose from the proposed semiotic model: (i) a symbolic unit is characterized by a phonologi-cal unit evoking a semantic unit; (ii) conceptual meaning results from the interaction between default values that are invoked from memory and fillers that occur in the form of explicit predications; and (iii) the transition of a semantic unit into a conceptual meaning occurs by conceptually integrating evoked and invoked facets of knowledge. The final case example clearly illustrated that conceptual meanings are thus integrated knowledge units consisting of textually given information (fillers) on the one hand and a multitude of background assumptions (default values) on the other. Conceptual meanings prove to be construction products that are the result of the active cognitive operations of language users.

The active cognitive operations consist largely in establishing categorization re-lations, which in construction grammar terms means realizing categorization links between constructions. Conceptualization processes are categorization processes. Each of the three illustrated relations is based on a series of categorization operations.

Thus, if two semantic units (or constructions) are integrated into each other via a categorization link to form a more complex semantic unit, one of the two semantic units functions as the schema whereas the other functions as an instance of this par-ticular schema. In Chapter 4 we saw that the formation of new (symbolic, semantic, phonological) units is only possible on the basis of the schematic structure of linguistic units. Thus, for instance, the complex schema [[dog]/[-s]] emerges by the semantic unit [-s] forming an instance in the schema [dog].

Following these considerations, I now turn to a more precise specification of the cognitive relevance of schemata. Since schemata are of such key significance for the understanding of linguistic signs, as construction grammar and Cognitive Grammar suggest, then – following holistic premises – this must be reflected in the general cogni-tive functions of schemata, which extend far beyond the semiotic issues addressed here. Indeed, schemata have played a prominent role in a broad variety of cognitive-science approaches since the mid-1970s (for instance in describing memory structures, read-ing comprehension processes, visual perception, etc.).309 In this broader context the following questions are particularly relevant for frame semantics:

309. Apart from schemata, many other representational formats were under discussion above all in the 1980s. Amongst these were propositional (network) models, cyclic and macropropositional


– What role do categorizations play in human cognition? What characterizes them? What are they governed by?

– Should there be a distinction made between non-linguistic and linguistic schemata (i.e. symbolic, semantic, and phonological units)? If so, to what extent?

– To what degree are schemata involved in the construction of a semantic unit?– What features do schemata have? Do frames have the same features?– Assuming that the emergence of both semantic units and conceptual meanings

is governed by schemata, what status do schemata then have? And how can their constituents be described in terms of linguistic structure?

The following is not able to deal with all of the mentioned aspects with the same degree of detail. The final point requires separate consideration and leads us to Chapter 6, in which I engage in detail with those specific linguistic manifestations that are character-istic of the structural constituents of frames. To this extent, this chapter is to be under-stood as a sort of interim stage. It serves to arrange frame-semantic specifications in the broader cross-linguistic context following the specifications of the holistic paradigm.

5.1 Categorization

In view of the plethora of different perceptual stimuli that constantly surround an organism, how is it possible for it to orient itself in its environment? An organism can only survive if it has efficient forms of complexity reduction at its disposal. What ap-plies to simple organisms applies no less to humans. Our sensory channels must, for instance, constantly separate irrelevant from relevant data, relate these to each other, and conceptualize and evaluate them with respect to particular situational aspects (Barsalou 2003; 2005). Explaining the functional modes of such complex cognitive processes is perhaps the most fundamental and yet difficult challenge that a cognitive theory has to face.

The question of the relationship between perception and cognition is very similar in a linguistic perspective. Realizations of phonemes and graphemes represent a par-ticular type of sensory stimulus. So that language users can communicate and interact with each other, sensory stimuli must be produced and received, mentally represented, and cognitively processed. Besides linguistic input, we are simultaneously confronted with sensory data of a different quality. This includes visual and auditory impressions of the current surroundings (that is, objects we see and sounds we hear) as well as gustatory and olfactory sensations, even if these appear to play a more subordinate role to begin with. In addition, using our motor skills we render accessible haptically co-present objects (for instance, the keyboard my fingers are on typing now – in a

processing models (cf. the overviews in Christmann 1989 and Scherner 1984). In the following, I limit myself to schema theories inasfar as they are relevant for U-semantics (“understanding semantic”, Fillmore 1985; cf. Section 3.3).

Chapter 5. Frames as schemata 213

more or less motorically coordinated manner). What is more, thanks to our memories and our imagination, further information is available to us without requiring current perceptual input. Hence, it is easy for me to recall to mind the face of a friend who I met yesterday or some details of my last summer vacation.

The fact that only a small proportion of the potentially available (and thus actualiz-able) sensory data penetrate our awareness, or cross the “threshold of consciousness”, is part of the cognitive economy of our minds. Our mental processing and storage capac-ity would be hopelessly overtaxed without efficient processing strategies. In particular, important memory operations such as short- and long-term retention and relation building between the currently perceived and the remembered would barely be pos-sible (cf. Anderson 1996: 133ff. and 167ff.; Schwarz 1992b: 75–83).310

What accounts for the efficiency of our conceptual system? Our memory depends on the cognitive ability to categorize both perceived and remembered information. Categorization refers to the ability to judge on the basis of past experience whether a particular entity is an element of a particular category or not. When we employ past experiences to interpret new experiences, this just as much includes acts of catego-rization as does recognizing a current object of perception as a particular object of perception. In the former case, that of memory, the categorization operation consists in identifying relations of similarity between new and past experiences. Frederic Bartlett already described in his pioneering study “Remembering” that remembering proceeds in particular via the activation of dynamic experiential patterns:

Thus what we remember, belonging more particularly to some special active pat-tern, is always normally checked by the reconstructed or the striking material of other active settings. (Bartlett 1932: 213)

Recognizing an object of perception proceeds in a similar manner. If a line on a piece of paper is interpreted as a particular letter of the alphabet, that line thus counts as an instance of the category of precisely this letter. Without memory, such classification cannot function, because every category itself represents a product of abstraction of past experiences, and thus must be specially retrieved from memory.

The ability to categorize is thus a prerequisite of every cognitive process, regard-less of whether the categorized unit is a consciously or subconsciously perceived entity (cf. Jüttner 1980). A motorist is not consciously aware of every tree she drives past; nonetheless the tree is not withheld from her field of vision but rather the object of a visual categorization act. This fact can be generalized to the effect that the object of

310. This result applies regardless of which memory model is favoured. Schwarz (1992b) refers solely to the so-called multi-store model (cf. Atkinson and Shiffrin 1971; Tulving 1983; Wettler 1980), which posits several functionally relatively autonomous memory stores (ultra-short-term, short-term, episodic, and long-term memory). Competing with this are Bartlett’s (1932) early trace theory of memory and the related levels-of-processing model (cf. Craik and Lockhart 1972; Cermak and Craik 1979; an overview in Schnotz 1994: 78–92) as well as more recent approaches that link memory processing more strongly to situative and social cognition (e.g. Barsalou 2005).


perception always exhibits a holistic character (“holistic” is to be understood here in the sense of gestalt psychology, cf. Wertheimer 1923; Metzger 1923, 1986; with relation to language, also Lakoff 1977). For instance, to identify a group of people as such it is not necessary to be consciously aware of each individual as an instance of the category “human”. Nonetheless, every individual counts as part of the group and thus as an instance of the category “human” (and not any other category).311

A multitude of categorizations are already involved in very simple communication processes.312 Nevertheless, categorization is by no means a specifically human ability. Categorization operations can already be observed in primitive organisms. Jackendoff observes,

[W]e should note at the outset that categorization judgements need not involve the use of language: they are fundamental to any sort of discrimination task per-formed by dogs or rats …

And that means:

Thus an account of the organism’s ability to categorize transcends linguistic theory. It is central to all of cognitive psychology. (Jackendoff 1983: 77)

As soon as distinctions are made, regardless of what cognitive domain they are in, categorizations play a role. Without them, an organism could not survive, for already the most fundamental distinctions such as edible/non-edible and hostile/non-hostile include at least one act of categorization. Without categorization, humans could nei-ther understand language nor engage in interaction with their environment using other means of communication (Anderson 1996: 56–70; Barsalou 1992b: 15–51; Taylor 2003).313 As Barsalou puts it (Barsalou 1992b: 15), the categorization capacity serves as a “gateway between perception and cognition”. Regardless of their modality (visual, auditory, tactile, olfactory, etc.) sensory data only become cognitively relevant if they are identified and accordingly classified as information units of a particular class.

311. Such categorization phenomena can be explained in gestalt-psychology terms by the so-called “law of proximity” and “law of similarity” (Wertheimer 1923). Gestalts are, in general, a prime example of unconscious categorization acts.

312. This holds for all levels of a language: phonemes, morphemes, word classes, and more complex syntactic and semantic properties (including socio-cognitive, discourse-functional, and pragmatic) only fulfil their U-relevant function inasfar as they form categorical units that are served with instances on the level of the “parole” (of the type “x is a phoneme/morpheme, etc. y”, where every x has the status of a token and every y the status of a type).

313. Some authors therefore see the ability to categorize as being the central research area of cognitive science. Thus, Cohen and Lefebvre (2005b: 2) introduce their handbook of categoriza-tion with the following words: “Categorization is the mental operation by which the brain clas-sifies objects and events. This operation is the basis for the construction of our know ledge of the world. It is the most basic phenomenon of cognition, and consequently the most fundamental problem of cognitive science”.


This occurs, as already mentioned, by establishing relations of similarity to other categories.

Categorization occurs by means of comparison; under particular general condi-tions and objectives, phenomena are grouped into types on the basis of their (mutual) similarity. The formation of categories makes past perceptions the basis of interpretation and organization of new events and experiences. (Konerding 1997: 57)

Schwarz (1992b: 84) argues that, categorization also follows the principle of identity in addition to the principle of similarity.314 The principles of similarity and identity allow us to easily recognize an identical object in another place at another time. However, in my view, it remains questionable whether the two principles are in fact different in nature. Since identity in principle merely represents a special case of similarity, namely in that category and instance do not just share some but all of their properties with each other, both principles can be subsumed as one. More persuasive is Wertheimer’s (1923: 308–309) distinction between a “factor of similarity” and a “factor of proximity”.

The suggestion of positing contiguity as a second cognitive association principle alongside similarity already stems from the context of association theory (which pre-cedes gestalt psychology). Bartlett (1932: 304ff.) signifies the relevance of contiguity to categorization by taking up Charles S. Myers’ association theory and schematically rein-terpreting it. He assigns some subcategories to the principles of similarity and contiguity without, however, outlining these more closely. They are summarized in Figure 1.315

Contiguity prevails if at least two objects are associated with each other on the basis of their spatial or temporal proximity. It is, in fact, not even necessary that they share any properties with each other, that is, that they be similar to each other, which is why contiguity is classified as an autonomous principle.316 The connection between objects motivated by contiguity is solely based on previous knowledge that the objects in question occur together as a rule. This is because:

It is … our knowledge of the world that determines contiguities … Contiguity has to be considered as constituting a conceptual, extralinguistic and not an intralin-guistic relationship. (Koch 1999: 145)

314. Schwarz speaks of “equivalence” (‘Äquivalenz’) but means “similarity” in the sense used here.

315. Bartlett himself, following Charles Samuel Myers, a contemporary psychologist and col-league in Cambridge, does not speak of “categorization principles” but of “association principles”. Whether, as Bartlett suggests (following Myers), “causal contiguity” should really be considered a subtype of contiguity in time and “letters” be ascribed to “sound” remains questionable from a semiotic viewpoint.

316. Or more precisely, from a psychological perspective, contiguity and similarity are com-monly seen as so-called “primary laws of association”. By contrast, the secondary laws include the principles of “recency” and the “frequency of presentation”; cf. for instance Drever and Fröhlich (1975: 183).


Similarity

Contiguity

in meaning

in sound

in time

in space

co-ordinationsuperordinationsubordinationcontrast

in letters or syllablesin rhyme

causalverbal

Figure 1. Principles of association, according to Bartlett (1932: 305)

The perception of contiguity phenomena is based on common practice. The existence of an entity cannot logically be deduced from the existence of another object, but it can be inferred.317

Let us return once again to the principle of similarity. This is assigned an important role in more recent cognitive-psychology research, above all concerning the establish-ment of conceptual coherence (cf. Rips 1989; Hirschfeld and Gelman 1994). The shared tenor is that similarity alone is insufficient as a categorization principle for associating one entity with another. This addresses the criticism directed at traditional association theory that two entities could share a potentially infinite number of properties with each other and consequently be considered similar. Thus, apart from similarity, there has to be other factors that motivate certain inferences and categorizations (Murphy and Medin 1985). This is the context in which to evaluate the endeavour of some in-ference theorists to consider categorizations not as formal in-correlation-placements but rather as content-driven processes (Graesser, Singer, and Trabasso 1997; Graesser, Millis, and Zwaan 1997; cf. also Section 3.3.3). Thus, Anderson (1991) and Barsalou (1985) also see categorization processes as being determined by concrete, environ-mentally specific objectives. In addition, Wisniewski and Medin (1994) argue that categories constitute and organize themselves through relatively concrete experiential knowledge and do not differ significantly from it in degree of abstraction. Thus, simi-larity relationships – no less than contiguity relationships – rest significantly on the individual experiential practice of organisms.

317. Although contiguity and similarity are not the only principles that motivate categorization processes (cf. for instance the gestalt-psychology considerations in Wertheimer 1923; Metzger 1986), they are nonetheless considered to be the two most fundamental and important types of correlation between categories. In the following, I am not so much interested in a typology as in the linguistic relevance of contiguity relations, which also play a role in Bartlett’s memory theory.


Although Bartlett’s subdivision illustrated in Figure 1 is oriented above all towards visual perception,318 similarity and contiguity hold as principles of memory association that are not specific to modality. Accordingly they should also be verifiable in language-meaning relations. There is evidence for this in studies relating to word field theory (cf. Lutzeier 1992) as well as in more recent cognitive approaches (Blank 1999; Konerding 1999). When Bartlett’s specifications of the similarity and contiguity principles are applied to word semantics, numerous phenomena can be found that demonstrate the effectivity of the association principles listed in Figure 1. Let us first consider the four variants of semantic similarity relationships:

– Contrast: Semantic contrast occurs in the case of, for example, antonymic meaning relationships (black/white).

– Superordination: This correlates with the semantic relation of hyperonymy (dog/Golden Retriever).

– Subordination: This conversely concerns the relation of hyponymy (Golden Retriever/dog).

– Coordination: One can speak of semantic coordination when linguistic categories stemming from the same domain and with the same degree of abstraction correlate with each other, for instance white/red/blue; dog/cat/horse.319

Linguistic expressions can also maintain similarity relations with each other with re-spect to the form side (sound):

– Rhyme: Similarity prevails between words if on the utterance side at least their last syllables are identical.

– Letters and syllables: One recognizes letter-tokens (or tokens of syllables, words, etc.) despite varying phonetic or graphic realization and despite variable embed-ding structure (the same letter in different words, the same syllable in different words, etc.) because they belong to one and the same type.

Now let us turn to semantic contiguity relations. There are a number of these. Generally, those linguistic expressions stand in a contiguity relationship that meronymically be-long to a shared, superordinate whole. This applies just as much to spatial as to tem-poral contiguity.

– Contiguity in space: Examples of this are wall/ceiling/floor and door/window/win-dow frame, which figure as parts of a room or a building.

318. In his attempt to differentiate, Bartlett uses the “image” metaphor: “Images are, then, liter-ally details picked out of ‘schemes’ and used to facilitate some necessary response to immediate environmental conditions” (Bartlett 1932: 303). “Images” contain more details than schemata; they are similar to Fillmore’s cognitive “scenes”.

319. Co-hyponymy would thus be a triadic relation that connects subordination and coordination.


– Contiguity in time: This prevails, for instance, between linguistic expressions that evoke the same script-structured frame. Thus, order food and give a tip both evoke the frame restaurant_visit. The linear organization of language signs in the Saussurean sense is also attributable to contiguity relationships in time.320

– Causal contiguity in time: Contiguity in time is often causally motivated. An ex-ample of this can be found in the so-called perlocutionary verbs (like persuade, insult, etc.), since these describe the effect of an action on the hearer or reader. Weaker forms of causality lie in many inference constructions, for instance when it is inferred from the verbal phrase arrive too late that the process it describes may entail sanctions for the actant (cf. Section 2.2.3).

Similarity relations and contiguity relations thus not only function as categorization principles in visual perception (or in other modes of perception), but also prove to be relevant with regard to language with respect to both content- and form-side relations between symbolic signs.

5.2 Schemata

When contiguity and similarity relations are established between meanings via lan-guage categorizations, this occurs using schemata.321 Establishing semantic relations means, from a cognitive viewpoint, realizing schema instantiations. We can note the following with respect to the examples applied to word semantics in Section 5.1. (a) Two expression meanings are similar when they either fill a specific facet of knowl-edge within a schema with different values or they stand in a hyperonymic relation to each other.322 Thus, the colour of a car can be red, white, pale, etc. in an unmarked case (i.e. without integrating metaphoric or, in Goossen’s 1995 sense, metaphtonymic usage variants) but not rusty, sad, etc., which is why red, white, and pale and the su-perordinate schema colour are similar in meaning, but not red and rusty.323 (b) Two

320. It is presumably this linearity principle described by Saussure that Bartlett means when he describes a dimension of contiguity in time as “verbal”.

321. As Bartlett states (cf. 1932: 201, 207–208).

322. Konerding also observes that similarity relations are defined by the fact “that they affect the alternatives for the lexical specification of schemata, depending on the pragmatic context and the degree of heterogenous semantic granularity” (translated from 1999: 28). However, this definition remains incomplete; according to this, hyponymy and hyperonymy would not be similarity relations.

323. Or more precisely, if rusty is used metaphorically or metaphtonymically (according to Goossens 1995) to describe the colour of a car, then this expression certainly does have a rela-tion of similarity to red, as it is then interpreted as an instance of the schema colour (and not as an instance of the schema state of metal). In addition, a causal metonymic relation of contiguity is at play, in that rusty is a material state that manifests itself in a particular colour. Rust has a colour, to rust means that as the material properties, etc. change, the colour of the


expression meanings are contiguous (in the sense of contiguity in time or space) when they specify different facets of knowledge within a schema. Wall, ceiling and floor not only form instances of the schema room, but also refer to quite different slots in this schema. A similar principle applies to phenomena of contiguity in time. Complex ac-tions – take, for instance, those that refer to complex verbal phrases like go shopping, go for a meal, write a letter, etc. – imply a series of sub-actions that concern the different aspects of the superordinate action.

The use of schemata allows semantic relations to be established between expres-sions. Schemata are “abstract descriptions of classes of facts and circumstances” (trans-lated from Pohl 2002: 32) or, more precisely, they are cognitive data structures that are aggregated from individual experiences within different content areas and generalized as typical experiences on different levels of abstraction and complexity, which can then serve as an interpretation or data base for the understanding of current experiences (cf. Brewer 1999; Minsky 1975: 228ff.; Rumelhart 1980a: 61; Schnotz 1994: 61; Thorndyke and Yekovich 1980: 23–24). Schemata are employed in the interpretation of symbolic, iconic, and indexical signs. Therefore, their use is not limited to language phenomena (although these stand henceforth in the foreground). Every form of categorization encompasses a schematization process, as categories themselves are of a schematic nature: Categories are formed on the basis of past perceptions that are grouped into types with respect to similarity and contiguity. Perception, regardless of its modal-ity, is unimaginable without categorization processes (cf. the overview in Cohen and Lefebvre 2005a: Chapter 7).

In this sense, Fillmore’s case frames and cognitive scenes also function as sche-mata. They both provide standard U-relevant information in the form of evoked or invoked frames, information that stands partially in similarity relations and partially in contiguity relations to the respective case frame or cognitive scene. Elements of a schema (instances) potentially open connections to other frames, since they themselves evoke a frame, which in turn contains standard information. In the purchasing frame, for instance, contiguity relations in space and time specify the relations between the instances money, goods, seller, buyer, whereas the instance money, to name just one element, evokes a money frame that bundles information on materiality, purpose, possessor, etc.

In the following, I understand schemata as being general, structural formats of non-specific modality, whereas frames are semantic organizational units.324 The term

material also changes. Predication tests show that the lexemes rusty, sad, etc. do not stand in any similarity relation in non-metaphoric and non-metonymic usage. In a standard case, they predicate nothing about the colour: – * The car is rusty. – * The car is sad.

324. This usage corresponds to the prevailing terminology. Compare, for instance, Rumelhart (1980a: 34): “A schema, then, is a data structure for representing the generic concepts stored in memory. There are schemata representing our knowledge about all concepts: those underlying


“schema” therefore functions as a superordinate term for all complex conceptual struc-tures. Frames are also schemata, but specific ones, since they represent and structure U-relevant knowledge that is enlisted to interpret linguistic expressions.325

Since frames share all the essential properties with other, non-specific semantic schema types, it is worth including schemata in the broader reflections. I attempt to do this in Section 5.2.1 by outlining the more recent research on the role and func-tions of schemata during reception of texts, remembering, and reproducing them. First, I briefly discuss Bartlett’s psychological theory of remembering to lead into this. Sections 5.2.2 and 5.3 then show that frames exhibit hardly any structural or functional difference to schemata. I see this as further evidence of frames being firmly anchored in the holistic paradigm.

5.2.1 Schemata as representational formats of non-specific modality

Bartlett’s study of remembering laid the foundation stone for later cognitive psycho-logical and cognitive semantic studies (e.g. those of Fillmore 1968, 1975; Langacker 1987; Minsky 1975; Rumelhart 1980a; Schank and Abelson 1977).326 His concept of schema identifies key features, which have been confirmed in more recent studies (Keenan 1989). These features are the focus of the following, since it is to be expected that they are inherited by frames.

Bartlett (1932) introduced the term “schema” in the context of his theory of re-membering to explain the constructivity of human processes of understanding. He identified them in his experiments when he invited test subjects to recall a text that they had previously read. The freely reproduced texts exhibited striking differences to the original. Overall, three empirical findings led Bartlett to the assumption that observable omissions, alterations, abbreviations, and additions could be traceable to activated schemata:327

objects, situations, events, sequences of events, actions and sequences of actions. A schema contains, as part of its specification, the network of interrelations that is believed to normally hold among the constituents of the concept in question”.

325. Similarly, Fillmore (1975: 124), (1977a: 101); Barsalou (1992a); Konerding (1997). In con-trast, Minsky (1975) and some other AI researchers like Charniak (1976) and Hayes (1980) un-derstand “frames” as general representational formats of non-specific modality, in other words, as schemata in the sense just outlined.

326. Perhaps the best-known schema theorist prior to Bartlett was the philosopher Immanuel Kant; Jean Piaget also developed his own schema theory within a developmental psychology perspective. Cf. also the brief overview in Brewer (1999); Christmann (1989: 75–85); Kaiser (1982); Thorndyke and Yekovich (1980).

327. Brewer (2000: 71–74) argues in a similar fashion, citing additional schema-relevant empiri-cal findings made by Bartlett, which in my view are negligible, for instance the trivial observation that the recalled texts are briefer and less detailed than the originals.


i. Standardization. If unfamiliar phenomena occurred in the original text, when the test subjects recalled the text they tended to adjust these phenomena to facts and circumstances they knew. Such standardizations were usually accompanied by omissions of information from the original text on the one hand and the addition of new information on the other.328 Both occurred on the basis of knowledge of “ordinary regularities” (Normalfallregularitäten – Pohl 2002: 21).

ii. “Off-line” inference constructions. It is not just every standardization that depends on the addition of information. Bartlett also observed that details not occurring in the original text were added (Bartlett 1932: 84ff.). Today one would speak of inference constructions, in the sense of minimalist and maximalist approaches (cf. Section 3.3.2).

iii. Lack of awareness. Some changes exhibited in the recalled text in comparison to the original were not based on intended interpretation. In fact, the test subjects were not aware of the extent to which they had modified the original text when they recalled it (Bartlett 1932: 86ff.). Bartlett interpreted this as evidence of them constructively processing the text on the basis of schemata stored in the long-term memory and activated both during (“on-line”) and following (“off-line”) text reception. These activated schemata were, however, not introspectively accessible to the test subjects. In fact, they are analytical categories gained ex post, without which the reconstruction process of the original text could not be adequately explained.

Remembering, according to Bartlett’s general conclusion, is based not so much on the reproduction of what is known as on active, schema-driven construction processes.

“Schema” refers to an active organisation of past reactions, or of past experiences, which must always be supposed to be operating in any well-adapted organic re-sponse. That is, whenever there is any order or regularity of behaviour, a particular response is possible only because it is related to other similar responses which have been serially organised, yet which operate, not simply as individual members coming one after another, but as a unitary mass. Determination by schemata is the most fundamental of all the ways in which we can be influenced by reactions and experiences which occurred some time in the past. All incoming impulses of a certain kind, or mode, go together to build up an active, organised setting: visual, auditory, various types of cutaneous impulses and the like, at a relatively low level; all the experience connected by a common interest: in sport, in literature, history, art, science, philosophy and so on, on a higher level. (Bartlett 1932: 201)

How do schemata become cognitively relevant? According to Bartlett, cognitive process-ing operations encompass four constructive features:329 (a) Selection and typification

328. Cf. the examples in Bartlett (1932: 72, 178).

329. This list does not intend to suggest that (a)–(d) occur in a linear fashion, nor to deny that there may be interference.


of actual experiential data or stimuli on the basis of past experiences (schema recogni-tion), (b) actualization of at least one relevant schema (schema activation), (c) correla-tion of actual experience with this activated schema (schema instantiation), and finally (d) correlation of this schema with schemata evoked from other modalities (intermodal conceptualization).330

Interestingly, more recent studies in the psychology of remembering come to comparable results. Alba and Hasher (1983) argue that schema construction includes four processes: selection, abstraction, interpretation, and integration.331 (a) Selection occurs because only some, but not all, stimuli are considered to be relevant. (b) Only in the course of further abstraction does the language token become categorized (with respect to phonology, morphology, semantics, etc.). (c) That is to say, interpretation already plays a role in that foreknowledge in the form of activated schemata (catego-ries) is selectively engaged by a language user. (d) It is not until integration processes are activated that the perceived conversational sequence finally becomes coherent; for instance, evoked schemata are correlated with invoked schemata, which are in turn reconciled with data from other sensory channels. All four processes are contained in Bartlett’s concept of constructivity.

Strictly speaking, following Bartlett’s considerations, schemata constantly change with every experiential act. In fact, if schemata did not individually adapt to current experiences – that is, if they were not “active” in Bartlett’s sense – they could hardly exert any influence on current experience. This aspect is so important for Bartlett that he is initially sceptical about the schema concept, since speaking of “schemata” suggests a certain persistence instead of conveying the continuous processes of change that do in fact occur with every experience:

[The term “schema”] does not indicate what is very essential to the whole notion, that the organised mass results of past changes of position and posture are actively doing something all the time; are, so to speak, carried along with us, complete, though developing, from moment to moment. (Bartlett 1932: 201)

Conversely, it would be just as erroneous to assume that knowledge amounted to nothing more than cognitive activations and actualizations. If this were the case, then every act of remembering would already fail because there would be no actualizable

330. One form of intermodal conceptualization occurs, for instance, if the frame evoked by a deictic expression with a determinative pronoun (such as this house, that man there, etc.) is set in relation with the co-present visual perception of the respective reference object.

331. “A schema theory which asserts that all four processes occur would state that from any environmental event, only the information that is relevant and important to the currently acti-vated schema will be encoded. Of the information selected, the semantic content of the message will be abstracted and the surface form will be lost. Further, the semantic content will then be interpreted in such a way as to be consistent with the schema. The information that remains will then be integrated with previously acquired, related information that was activated during the current encoding episode.” (Alba and Hasher 1983: 204)


knowledge: Remembered knowledge is schematized knowledge (cf. Alba and Hasher 1983). Otherwise remembering would reproduce actual sensory impressions and thus encompass so much detailed information that our cognitive capacity would be vastly overtaxed.332 Schematizations avoid this. They are both an indispensible form of com-plexity reduction and the guarantee of cognitive economy. Thus, what Busse generally observes on the mutual relationship between static and dynamic knowledge also ap-plies to schemata. Knowledge is

not something static, not a fixed, given, or temporarily unchanging structure that can be held still in a quasi frozen perspective and descriptively recorded like a predetermined object. (Translated from Busse 2005a: 52)

And by the same token, it is equally true that knowledge is not merely “pure dynamics” or a “fleeting movement” (Busse 2005a: 52).333

However, positing an indissoluble oscillation between stasis and dynamics, some-thing Bartlett certainly seems to do (cf. Bartlett 1932: 85, 201), gives rise to a con-flict with Bartlett’s understanding of schemata as units of generic knowledge in the long-term memory. This point becomes particularly clear when Bartlett correlates his schema theory with an anthropological and sociological concept of conventionaliza-tion (Bartlett 1932: 280). While schemata, like conventions, also form culturally stable patterns that prove without doubt to be cognitively relevant in this function, schemata are, by contrast, identified with a “pattern of a relatively stable kind” (Bartlett 1932: 280) or, more precisely, with structural units consisting of a set of highly conventionalized elements, thus lacking any potential for the change that Bartlett considers to be so important. This is because this dynamic of change occurs solely on the basis of con-textually specific fillers.

In this context, Brewer (2000: 83–84) goes so far as to claim that no schema the-ory is capable of encompassing all forms of human knowledge, since non-generic

332. Even biographical memories of significant events (like riding a bicycle for the first time and falling off) exhibit a schematic character. In the course of “deep cognitive processing” (cf. Anderson 1996: 187–193) of the event, some facets of knowledge are memorized (such as the colour of the bicycle; the street you were riding on; the pain after the fall, etc.) and others are not (perhaps the colour of the pullover you were wearing; the time of day that the accident oc-curred, etc.). Biographical memories are in this respect no less schematic. They are subject to reinterpretation, i.e. gaps in knowledge can be filled with default values.

333. Busse (2005a) and Ziem (2006a) point out that the stability of semantic knowledge increases to the same extent that the degree of abstraction of the pertinent knowledge unit increases. As Busse puts it: “The rapidly changeable, dynamic, constantly shifting reformations of newly arranged knowledge tend to occur prototypically in the areas of knowledge and discourse for-mations of the cultural superstructure (if I may be permitted this minor, retrospective termino-logical detour). However, this can and must not blind us to the fact that there are other societal areas and thus areas of knowledge in which action and life structures, as well as the knowledge relating to them, demonstrate impressive latency and stability” (translated from Busse 2005a: 52).


knowledge plays a significant role in processes of understanding – for instance, when processing prose texts or organizing human actions – yet this form of knowledge can-not be taken account of in the analysis. In my view, Brewer’s evaluation is incorrect. He overlooks the fact that this criticism only relates to a specific problem in Bartlett’s model. The dynamics result from the categorizations (schema instantiations), and non-generic knowledge concerns those instances that are being categorized. However, since Bartlett does not distinguish schemata with respect to their two structural constituents of slots and fillers, but only pays attention to default values, the schematic values lack dynamics. This lack was not removed until the conceptions of Minsky (1975) and Rumelhart (1980a: 36f, 42),334 to which I return later.

Strictly speaking, the term “schema” does not describe a “conceptual knowledge structure in the long-term memory” (as translated from Schwarz 2000: 34), since only default values form units in the long-term memory. Slots, on the other hand, represent an analytical value, and it is highly questionable whether they have any cognitive reality at all. And concrete fillers are characterized by the fact that they are, as a rule, not stored in the long-term memory.335 Against this background, some apparent problems prove to be baseless, as is this objection to schemata not being dynamic enough to convey constant change in conceptual knowledge.

Bartlett’s core interest (like that other cognitive psychologists) concerns psycho-logical issues surrounding knowledge representation. His interest is not in semiotic considerations. What is then the relationship between schemata relevant to memory psychology and frames? With Bartlett’s study in mind, Christmann observes:

Even non-categorically linked terms can be reproduced in large number if the test subject succeeds in establishing a relationship of any kind at all between individual words. (Translated from Christmann 1989: 46)

Although similarity and contiguity are the psychologically effective association prin-ciples that establish the semantic relationships between terms that Christmann speaks of, recognizing this is not sufficient to describe U-relevant knowledge. Simply the fact that linguistic signs are subject to their own organizational principles – for instance, in forming constructions (or symbolic units) of differing complexity – makes language processing a cognitive process sui generis. The cognitive processing of language can-not be reduced to other forms of sign-conveying processes of understanding. This was accounted for in Chapter 4.

334. Rumelhart (1980a: 40) takes recourse to these structural constituents of slot and filler and despite Brewer’s criticism is able to respond: “Schemata represent knowledge at all levels of abstraction”.

335. A filler is generally not stored in the long-term memory because it forms an instance in a frame, in other words, it specifies a frame in a particular respect. At best, elements (default values) of the evoked frame could be anchored in the long-term memory.


Nonetheless, the fact that schemata are seen as uniform representational formats makes them cognitively powerful structural units in our conceptual system (Mandler 1984; Rumelhart and Ortony 1977; Rumelhart 1980a). In addition to Bartlett’s inves-tigations, there are numerous more recent psychological studies on text processing that suggest the far-reaching functional relevance of schemata for human cognition. I address this briefly here so I can than turn to frames as schematic structures in the next section.

The research results can be summarized in five points. Firstly, the findings of recent studies on textual recall, in which test subjects were required to reconstruct a previously heard or read text, indicate that the understanding and retention of texts is significantly dependent on those schemata that the recipients actualize in the pro-cess of understanding and then activate when reproducing the text (cf. for instance, Anderson and Pearson 1984; Kintsch and Greene 1978; also: Fillmore 1981; summa-rized in Mandl, Friedrich, and Hron 1987).

Secondly, assuming that schemata govern the process of understanding explains why texts are selectively registered and stored (Anderson, Pichert, and Shirey 1983; Neisser 1976): detailed information always gets lost when it either appears to be irrelevant for an activated schema or forms instances of a schema on a very low level of abstraction.

Thirdly, in this context schema theories additionally offer an explanation for how different forms of memory storage, especially short-term and long-term memory, in-teract with each other. Schemata of a higher degree of abstraction generally prove to be more persistent and stable, which is why as the degree of abstraction of a schema increases so, too, does the probability that the schema is stored in the long-term memory (Brewer and Nakamura 1984; Taylor 2002: 274–277; a critical view: Brewer 2000: 83–84).336 Thus, while data from the short-term memory (such as details of a friend’s living-room seen for the first time) function as instances in the generic concept (here: living-room or room) and are of an extremely fleeting nature, the respective superordinate schema holds stable standard information available (take, for instance, the default values ‘door’, ‘window’, ‘floor’, ‘ceiling’ in the room schema, cf. also Minsky 1975: 221–223; 1988: 249–250).

Fourthly, schemata not only play a role in the recollection or memory of personal experiences, but evidently also govern actual perception processes, for example, the reading of a text. Thus, Smith and Swinney (1992) conclude that the time taken to read a text is significantly shorter if the reader has previously activated a relevant schema. Accordingly, the influence of schemata is not merely limited to retrospective textual reconstruction. In fact, schemata are also effective to understanding during processing

336. This is explained by the principle of recurrence (cf. also Section 6.5.1): schemata that are more frequently evoked or invoked are more stable – and thus more cognitively salient. Since the actualization of a schema on a lower level of abstraction implies the activation of more super-ordinate (or, from a semantic viewpoint, more hyperonym) schemata, superordinate schemata are activated more frequently. Thus, the expression kitchen activates the room schema, as do living-room and office.


operations (in this case, during reading), and indeed on quite different levels of abstrac-tion, for instance when identifying individual letters and words or when processing whole phrases (Adams and Collins 1979).

Finally, schemata seem to obtain a cognitive reality on an even higher level of ab-straction. Numerous cognitive psychologists in addition to Rumelhart have attempted to provide evidence for the role of complex narrative schemata in the reception and production of prose texts (cf. Rumelhart 1975: 1980a). The understanding and reten-tion of a text depends significantly on the hierarchy of the text structure on the one hand and the order in which the text elements occur on the other (summarized in Christmann 1989: 80–81). Reading comprehension is improved if the hierarchically higher elements in a text are thematically central and the order of their occurrence matches the typical default information of a narrative schema.

These findings are still valid. Even among critics of schema theories there is a consensus that knowledge in the long-term memory can only become U-relevant in schematic form, that is, it can only be stored there in the form of “assumptions about typical instances of object, action, or situation categories” (translated from Schwarz 2000: 34). However, it is no coincidence that until now no schema theory has succeeded in adequately explaining the cognitive process that Alba and Hasher (1983) call “selec-tion”: There is always a potentially unlimited number of typical examples that can be assigned to the slots in a schema, and there are just as many reasons why only a small selection of typical examples are actualized in each individual case. What mechanisms ensure that every activated schema merely contains particular default assumptions whereas others play no role? Evidently numerous factors (situational demands, co- and contextual data, actual attention focus, personal motivation) inhibit potential inference constructions. Determining the interaction between such factors is an enormously difficult task, which as yet has not been resolved to its full extent.337

Additional unresolved issues result directly from the research in memory psychol-ogy. Although they do not occur in connection with frames, they are briefly listed here. One problem concerns methodological inadequacies. Although schema theories are of a highly descriptive value in their application, employing them is of little use when it is a matter of predicting what knowledge test subjects will deploy in particular situations. This is due not only to the lack of specification of the status of default values, but also, as Thorndyke and Yekovich (1980: 41) already critically noted, to the fact that many potentially relevant schemata exist to explain a single operation of understanding. Why does a test subject actualize precisely this schema and no other under particular circumstances? To date, it has not yet been possible to exhaustively specify and ex-perimentally implement the factors of scalar salience. This problem complex is closely linked to the process of schema selection outlined here. Frame-semantic approaches

337. Dan Sperber and Deirdre Wilson (1986) propose a theoretical solution with far-reaching consequences by declaring relevance to be the key cognitive principle (following Grice). Barsalou (2005) provides an overview of the current state of empirical research. Bühler’s (1934: 28) “prin-ciple of apperceptive enlargement” and “abstractive relevance” should be included here; abstract-ing according relevance roughly corresponds to the process that Alba and Hasher call “selection”.


face a similar difficulty when they address invoked frames (in the sense used in 4.3.1). I deal with this in Section 6.5.

Further, schema theories suffer from a problematic simplification. It is unclear how the process of schema application is to be precisely understood in psychological terms. For instance, the processes of selection, abstraction, interpretation, and inte-gration are themselves based on at times very complex sub-processes.338 The same applies to innumerable other processes, such as the comparison, specification, and evaluation of schemata, as well as to the linking of simpler schemata to form a single complex schema.

Finally, an additional problem complex concerns the activation of schemata. How are schemata activated? Not every sensorily perceivable stimulus that reaches an organ-ism activates a schema. However, when the schema activation precedes an estimation of the relevance of the stimuli, do schemata still then form the fundamental units (or “building blocks”, as Rumelhart 1980a puts it) of our cognition? Naturally, the estimation of relevance could also be schema-based, but what then activates these schemata? This problem seems unable to be solved without previously having precisely determined the interaction between the memory functions of processing, storing, and recalling relevant information.339

5.2.2 Shared features of frames and schemata

In the discussion of Langacker’s theory of semantic units, I pointed out the relevance of schematized knowledge for a cognitive theory of meaning. At that point, however, it was not yet possible to enter into the analytical usefulness of schemata. Now, on the basis of the previous considerations, I take up this point once again and begin by defin-ing the concept of schema to the extent that it is useful for the understanding of frames. This occurs against the background assumption that frames and schemata are of a very similar nature concerning their cognitive function and their representational properties.

What features do schemata have? In their early criticism of memory-psychology schema theories Thorndyke and Yekovich note that there is no uniform use of the schema term in the literature. In fact, they observe that it is usual to use the schema concept in a way that suits the empirical interests and demands of the respective study.

338. The process of textual interpretation is an example. Alba and Hasher (1983) discuss a human process of understanding here without being aware of the full extent of its complexity, cf. Busse (1991a).

339. Kintsch (1977) names eight processing operations that schemata seem to be significantly involved in: (a) grapheme analysis, (b) letter recognition, (c) word recognition, (d) word-meaning analysis, (e) syntactic analysis, (f) propositional analysis, (g) construction of mental coherence, and (h) contextual perception. Following semiotic considerations, according to which frames or-ganize semantic knowledge into constructions (or symbolic units), the processes (c) to (h) involve frames; graphemes and letters also occur as entities that do not have the status of symbolic units.


Because the concept of schemata as organizers of human experiences is so gen-eral, it is perhaps inevitable that each particular formulation of schemata should differ from all others in some details. As in the early research on schemata, each researcher has proposed a model that differs from all others in precisely what a schema is, how it is structured, or how it is used. (Thorn dyke/Yekovich 1980: 26)

Despite this observation, Thorndyke and Yekovich (1980: 27–28) come to the conclu-sion that all schema theories are based on five basic assumptions. Firstly, schemata represent abstract knowledge (“concept abstraction”) that, secondly, is hierarchically organized. Thirdly, slots in a schema are filled with values by “instantiation”. Fourthly, default values make schemata into prediction structures, which in turn – point five – emerge inductively. I take up these five features in the following, expanding them, however, with further specifications.340 In 5.3 we will then see that these features are also valid for frames.

a. Structural constituents: Slots, fillers, default values. Every schema is composed of three types of elements. (i) Slots show which facets of knowledge can potentially be ascribed to a schema (Rumelhart 1980a). Thus, the visual schema of a room encompasses information about the ceiling, the floor, and the walls, as well as their properties (Brewer 1999). (ii) Fillers are the data of the actual perception that occupy the slots. These data stem from an external source of stimulus, a perceived entity in the external world. (iii) Default values can additionally be inferred where there are no particular fillers to specify a schema.341 For instance, a default value in the area of visual perception would be if someone enters a room, does not perceive the ceiling, and yet assumes that there is a ceiling with particular properties. In the course of cognitive processing, the fillers and default values meld into a structural unit.

Figure 2. Schematic representation of a face

340. For explanations and criticism of Thorndyke and Yekovich (1980) cf. also Biere (1989: 79–81). Further specifications of schemata can be found in Rumelhart (1980a: 40f.); Christmann (1989: 76); Brewer (1999). These, however, remain partially incomplete. The same applies to Müller (1982), who is the only one to place frames and schemata in relation to each other.

341. The distinction between fillers and default values corresponds to the distinction com-monly made in AI between “data-driven and expectation-driven” cognitive processes, cf. Strube et al. (1996: 110f. and 161f.). The terms “default value” (Rumelhart 1980a: 36) and in German Standardwert (for instance, Fraas 1996a: 13ff.). But there are also terms that are used synony-mously, like “default assignment” (Minsky 1975: 228), and in German Ersatzannahmen (Schnotz 1994: 62) among others.


Figure 2 serves to illustrate this.342 We interpret the ensemble of depicted signs as a smiling face, which consists of three filler elements. The colon represents the eyes, the vertical line the nose, and the bracket below the smiling mouth. Now, to interpret this minimalistically reduced schematic representation as a face, at least the whole head has to be “read out” of the white background. The external deline-ation of the head must also be imagined. This added element has the status of an inferred default value without which neither the distinction between the figure of the face and the white background nor the categorical mapping ‘ensemble of signs → face’ could occur.343

b. Structural patterns. The impression of a schematic whole emerges by correlating the structural constituents of a schema with each other. As illustrated in Figure 3, three structural patterns can be distinguished. If it is a dynamic correlation, as in (i) and (ii), a relationship of sequential order exists between the default values (“d”) and the fillers (“f ”) in a schema. In the illustration, the distribution of fillers and default values is freely selected; it varies according to the given input and contex-tual data. The arrows in (i) and (ii) show the direction of the sequential relationship in time. The shading of fx in (ii) symbolizes that this is an end element.

ii. Dynamic structural pattern with goal orientation (“plan”)

i. Dynamic structural pattern without goal orientation (“script”)

iii. Static structural pattern

d1 f2

f3

d2 fx

d1 f2 d2 fx

f1

dx

f2

Figure 3. Structural patterns: Examples of possible relationships between default values d and fillers f

342. This is the typical smiley emoticon, turned on an angle of 90 degrees for clarity.

343. Of course, apart from this default value, numerous other default values are added. For in-stance, in contrast to the ensemble of signs in Figure 2, the construed form of a face has a forehead, cheeks, ears, etc. – in other words, elements that would also contrast with the white background.


In the case of (i) and (ii), the schema is organized into several structural constitu-ents that are mutually dependent by dint of their temporal contiguity. In this sense, one could understand both the visual and auditory perception of a passing car as continua of fillers f1 to fx, namely on the one hand as a sequence of images – compa-rable to a flicker book – and on the other as a sequence of sounds (cf. also Minsky 1975). The perception of movement generally follows a dynamic pattern of type (i). A more specific dynamic pattern is evident when, as in (ii) the final element in time determines the sequence of the structural constituents. This is the case in goal-oriented processes, perhaps most clearly identifiable in the application of cog-nitive problem-solving strategies or in recognizing (action) intentions (cf. Barsalou 1983, 1991; Schank and Abelson 1977: 101–130). For instance, to successfully solve the task of pealing an apple, a schema (“plan”) must be activated that, among other things, provides information about which sub-actions performed in which sequence lead to the desired goal.344 Finally, case (iii) illustrates a static structural pattern. The correlations between the structural constituents of a schema are static in that their occurrence is governed by the principle of similarity or contiguity in space. Here, “static” merely means “non-temporal” and by no means that fillers and default values are fixed once and for all. In fact, both remain variable, exchange-able, and changeable. Static relationships determine, for instance, the relationship between the structural constituents of the face perceived in Figure 2. They are motivated by contiguity in space.

c. Gestalt-like figuration. Schemata are fundamentally gestalt-like, i.e. they occur as pattern-like figurations. That means that the structural constituents of an activated schema do not exist mentally as individual, isolated elements but rather as integral components of a holistic structural framework (Metzger 1923; Wertheimer 1923). This applies to all perceptual modalities (Lakoff 1977: 246). The impression of wholeness is so strong that it is difficult to distinguish between fillers and default values on the basis of memory. They both make a schema into a severally specified perceptual unit. Unoccupied slots are not contained in the gestalt-like figure of the perceptual unit. As a result, schemata do not exist as pure structural entities, i.e. as structures with unoccupied slots.345 As soon as a schema is activated, we cannot withdraw from the pattern-like perception of, for instance, the face in Figure 2: We do not see two dots and a line and a bracket, and equally we do not see eyes and a nose and a mouth. In fact, we perceive a face.346

344. Starkly simplified, for instance, the default values: d1 = ‘enter the kitchen’, d2 = ‘take a knife out of the drawer’, d3 = ‘lead the knife to the apple’, d4 = ‘peel the apple’. Even peeling an apple itself is a complex action that would be subdivided into a sequence of sub-actions.

345. Or more precisely, a schema as a pure structure with unoccupied slots is an analytical construct (and can certainly be useful as such). But it has no cognitive reality.

346. Which in this case, of course, is also contingent on the iconic arrangement of the individual elements.


d. Dynamicity. Dynamicity is a result of recurrent schema-instance correlations. With every new experience, new fillers are assigned to the slots in an activated schema. If similar fillers appear with a high frequency within a particular time interval, they can change the existing default values. Rumelhart and Norman (1978) designate this assimilative process as “tuning”. Default values that are already stored in slots gradually adjust to the average of new experiences.

Since, as already mentioned, slots are analytical elements that emerge from the abstraction of default values, they change as well. However, the half-life of slots lies significantly higher than the half-life of default values, which in turn are, of course, significantly more persistent than fillers.

e. Prediction structures. An activated schema generates expectations or predictions with respect to the information that matches it, i.e. with respect to potential knowl-edge elements within the evoked slots (Oliver and Johnston 2000; Tannen 1979; Schnotz 1994: 61–62). Expectations arise through default values that are associated with a schema. Taking Figure 2, once the face schema has been activated, the de-fault values (in this case, ‘chin’, ‘cheek’, ‘forehead’) appear of their own accord, as it were. In this process, every default value initially has the character of a hypothesis that must first be proven (Rumelhart 1980a: 38). New data can replace default val-ues at any time, even possibly contributing to an already activated schema being rejected because another schema fits better, i.e. because it leads to a more coherent overall understanding of available data.

f. Schemata as segments of networks: Recursive embedding via instantiation. While the structural constituents of schemata do not form isolated units, schemata them-selves do. This is related to the status of the structural constituents. Since both fillers and default values are of a conceptual nature, they behave towards other ele-ments in the same way as schemata do to instances. In other words, every instance (filler, default value) forms for its part a schema for other instances (Rumelhart and Ortony 1977: 106–109). Thus, the nose (instance) of a perceived face (schema) behaves towards its components of ‘nostrils’ and ‘nose ridge’ as a schema does to its instances. This circumstance can be thus generalized that schemata are recursively embedded in each other (Barsalou 1992a: 30–35, 43). Thus, the face schema in turn forms an instance in the person schema. If one perceives a group of people, individual persons figure in turn as instances in the superordinate group sche-ma.347 So seen, schemata are segments of a comprehensive conceptual network within which they bundle multiplicities into unities.

g. Hierarchical organization: Levels of abstraction. A conceptual network features a hierarchical organizational structure. Divergences in degree of abstraction result from recurrent instantiation processes, because a schema is by definition more

347. Incidentally, the principle of recursive embedding also applies to gestalts, cf. Liebert (1992: 20).


abstract than an instance of itself. A multitude of hierarchically organized sche-mata engage with each other merely to process a sentence, which ranges from recognizing graphemes to perceiving context.348 The example of the face schema already showed us that there are many levels of abstraction, even in visual percep-tion. Similar considerations can be made in the auditory area. If two tones sound simultaneously, they form just as much a gestalt-like unit as when they are linked into a sequence of tones.

h. Production and development of schemata via induction and abduction. A schema – or more precisely, its default values and the relations between them – results abduc-tively and inductively from the intersecting set of similar individual experiences (cf. Langacker 1999b: 93). The fact that schemata evolve relative to experiences and develop on the basis of new experiences (Case 1985; also: Mandler 2005) corresponds to the “sensitivity of the cognitive system to regularities within the experienced environment” (translated from Schnotz 1994: 89). Conversely, the invocation of information from a schema occurs as a deductive process, because only default values can be invoked, and these are, in contrast to perceptual input data, of a generic nature.

In contrast to what Thorndyke and Yekovich (and others, like Christmann) suggest, the basic assumption of schema theory is, in my view, that schemata as cognitive representational formats are structurally uniform in specification. Four elements in total are sufficient to describe the representational properties of schemata: (i) slots, (ii) fillers, (iii) default values, and (iv) relations that correlate these three structural constituents of schemata. All five assumptions that Thorndyke and Yekovich evaluate as being constitutive for schemata can be retraced to these elements. Thus, processes of both concept abstraction and induction concern the transition of fillers to default values, whereas instantiation corresponds to the categorization of fillers or default values in the slots of a schema. Following the assumption that every value is simulta-neously a schema for other values on a lower degree of abstraction, the hierarchical organization of schemata is for its part a product of recurrent instantiation. Finally, schemata only create expectations or serve as predictions because activated default values give cause for this.

Therefore, in frame semantics, particular attention must be given to the structural constituents of slot, filler, and default value. Before entering into this in detail, I will begin by showing to what extent the features (a)–(h) can also be verified for frames. The analysis of a lexico-semantic example will serve to this end.

348. Cf. Kintsch (1977). Kintsch also postulates six additional levels between grapheme recogni-tion and contextual perception.


5.3 Frames as schemata: example analysis

In the following, when I conduct a frame-semantic analysis of the meaning of the word divorce within different contexts, I am essentially interested in the extent to which the frame evoked by the expression divorce exhibits a schematic structure. Following my assumption that the properties of schemata (in their function as the basic units of human cognition) are inherited by frames, the properties of frames and schemata must be very similar, if not identical.

In the same way as proved meaningful in Section 4.3.2, I begin with a hyperonym type reduction to determine the slots of the frame. Divorce can be retraced to the high-est hyperonym event. Accordingly the slots in the divorce frame correspond to the superordinate matrix frame event.349 Of the over 40 slots contained in the relevant matrix frame, I only select a few here. They are summarized in Table 1.

Table 1. A selection of slots in the frame evoked by divorce

# Slots in the matrix frame

(1) What requirements does the divorce event entail?(2) What features mark the procedure of the divorce event?(3) What can the divorce event be traced back to?(4) What is it a component of?(5) What essential phases or sub-events does the divorce event exhibit?(6) In what larger functional context does the event figure?(7) What essential co-actors/interaction partners participate in the event?(8) How are they characterized?(9) What relevant traits characterize the respective co-actors and their roles?(10) What event, etc. follows the divorce event?(11) What makes the event possible? (12) What consequences does the event have?

In addition, two alternative scenarios are given to mark the context in which the word divorce is used.

Scenario 1:Hannah tells her girlfriend about her unhappy marriage. It turns out that her husband has been having an affair for two years. This has made her decide to take matters into her own hands and finally file for a divorce. She already has an appointment with her lawyer next week.

349. Cf. Konerding (1993: 335–340). Once again, the method of hyperonym type reduction is elaborated in more detail in Section 6.3.2.


Scenario 2:Hannah and Nick are having a court battle over the custody of their children. The judge would like to get an idea of who has looked after the children in what form to date. In this context, he is also interested in how the separation came about. Nick explains that he has made mistakes, but regrets them, and he did not wish to separate from his wife. Hannah replies that she was the one who filed for divorce.

The departure point for the frame-semantic analysis here is the same: Whereas the slots in the frame show which predication types contextualize the expression divorce the most meaningfully, the given data specify the frame. This includes the language context (given in scenario 1 and 2), as well as extralinguistic perceptual data, of which none are given here. This is enhanced by inferred default information that is already associated with the evoked frame and is not replaced by the contextual data.

As already mentioned, frames and schemata share the most important structural trait of representing knowledge in the same format (consisting of the structural con-stituents of slots, fillers, and default values). To illustrate the outlined characteristics of schemata (b)–(h) I now select some fillers and default values from Table 1 that result from the two scenarios and assign them to the corresponding slots. For a better comparison between the frames of the two scenarios, I have chosen the same slots for each frame.

Scenario 1:– Filler for (3): The decision to file for divorce was taken by Hannah, and the deter-

mining factors were that her marriage was unhappy and her husband was having an affair.

– Filler for (6): Hannah’s conversation with her friend forms the superordinate con-text in which Hannah’s divorce is addressed.

– Filler for (7): At least four people are involved in the divorce event: Hannah’s lawyer, Hannah, her husband, and the woman or man that Hannah’s husband is having the affair with.350

– Default value for (1): The fact that Hannah and her husband are getting a divorce presupposes that they are married to each other, whether at a registry office, in a religious ceremony, or both. They must have been separated for at least one year at the time of the divorce (according to German law).

350. Scenario 1 gives no indication of whether there are any other people involved, such as a judge, a lawyer representing Hannah’s husband, any children the couple may have, or possibly even other women and/or men with whom Hannah’s husband has been having affairs. Therefore, these would not be fillers, but at best default values.


– Default value for (4): A divorce is part of an institutional, socially regulated prac-tice. The couple must legally file for it, usually with the mediation of a lawyer. The dissolution of the marriage takes place in court. The court may also then take a decision on other matters (maintenance payments, etc.).351

– Default value for (12): Divorce or filing for divorce fundamentally changes a person’s life situation. Apart from the formal and possibly legal conflict (for instance concern-ing maintenance payments, division of assets, possibly determining the custody of the children), the divorce also entails a psychological and emotional burden.

Scenario 2:– Filler for (3): The decision to file for divorce was taken by Hannah, and a determin-

ing factor was that Nick made mistakes in their relationship. (What these mistakes consist of would be a default value for (3).)

– Filler for (6): A legal conflict for the custody of the children forms the superordi-nate context in which the divorce is addressed.

– Filler for (7): The judge, the married couple Hannah and Nick, and their children are involved in the divorce event. (No lawyers are mentioned in scenario 2.)

– Default value for (1): The fact that Hannah and Nick are getting a divorce presup-poses that they are married to each other, whether at a registry office, in a religious ceremony, or both. They must have been separated for at least one year at the time of the divorce.

– Default value for (4): A divorce is part of an institutional, socially regulated prac-tice. It must be legally filed for by the couple, usually with the mediation of a lawyer. The dissolution of the marriage then takes place in court.

– Default value for (12): Divorce or filing for divorce fundamentally changes a per-son’s life situation. The divorce entails that Hannah and Nick must settle, among other things, who will have custody of the children and how high the maintenance payments will be.

Whereas in the two scenarios the same slots in the divorce frame are occupied by different fillers, the default values barely differ. The fact that they differ at all is due to the fact that fillers can influence the degree of concreteness of a default value. Due to the superordinate context, that is, due to filler (6), the default value for (12) in the second scenario is far more concrete than in the first.

Now, in what form do the characteristics (b)–(h) occur in the two frames? Some example detail analyses are necessary to illustrate this.

351. In the case of a church wedding, one should add that divorce is generally not possible. In exceptional cases the marriage may be at best annulled, i.e. declared void.


With respect to (b): Structural patterns. The degree to which the pattern of a frame governs the process of understanding is illustrated in the first scenario in default value (4). Here, the divorce marks the result of a complex, institutionally regulated pro-cess. It is Hannah’s goal to get a divorce from her husband and, to achieve this, a num-ber of sub-events must be performed. Since the goal determines the sequence of the necessary sub-events, we are dealing with a “dynamic pattern with goal orientation”.352 It would be possible to designate this pattern more precisely as a “plan”, since the goal is intentionally set by an acting individual (cf. Schank and Abelson 1977). Once this pattern has been activated in the first scenario, it is possible to coherently interpret the information that Hannah has an appointment with her lawyer, namely as a sub-action necessary to achieve the goal. Technically speaking, meeting a lawyer corresponds to a filler in the slot of the activated structural pattern.

The second scenario contains a “structural pattern without goal orientation” – which Schank and Abelson (1977) refer to as a “script”. Thus, the superordinate context of the custody hearing particularly profiles default value (12), the consequences of the divorce, and this in such a way that it stipulates who will be able to see the children when and for how long. Such a stipulation has the character of a script that is legally codified and thus binding for the divorced couple. From a semantic viewpoint this pattern creates coherence between the superordinate context, i.e. filler (6), and the evoked divorce frame.

A “static” pattern bundles U-relevant knowledge in both scenarios that does not concern aspects of action but knowledge requirements of another type within the divorce frame.353 Every concrete filler and default value contains at least one piece of information of this type. The first scenario contains information that refers to such varied aspects as reasons for the divorce (filler (3)), participant persons (filler (7)), the formal requirements of a divorce (default value (1)), and knowledge of the institutional and social practices of its implementation (default values (4) and (12)).

With respect to (c): Gestalt-like figuration. Recognizing the extent to which the divorce frames in each scenario diverge is the result of the outcomes of linguistic analysis. Without this, we do not have access to individual fillers or default values, but merely to a specific meaning of the expression divorce in a specific context. Although we have implicit knowledge of what the word divorce means, we would have difficulty determining the difference in meaning of the expression in each scenario and sup-plying the meaning aspects relevant to each,354 because in phenomenological terms linguistic meanings initially occur as gestalt-like figures, i.e. as unanalysed wholes (cf.

352. Cf. Figure 3, type (ii).

353. Following Fillmore, one could speak of “static structural patterns” as “scenes” to emphasize the dynamic-cognitive status of the structural constituents.

354. As already discussed, this distinction is based in particular on the filler (6) and the default values (4) and (12).


also Lakoff 1977).355 This applies to the same degree for the structural patterns ana-lysed here, which organize the information within a frame. In the first scenario, it is not the individual elements of Hannah’s “plan” to get a divorce that we recognize, but a structural whole that crystallizes as a result of the goal-setting.

With respect to (d): Dynamicity. When comparing the divorce frames of each scenario with each other it becomes apparent that it is not only the fillers that diverge. The default values instantiated in the slots also vary depending on the superordinate topical and situative circumstantial context. In the second scenario, for instance, it is constitutive that apart from Hannah and Nick the children, too, are affected by the divorce (filler (7)), and for this reason, it needs to be clarified which of the parents should have custody (default value (12)). Divergent word meanings result from the instantiation of different values in the slots of the divorce frame. Thus, it is the varying occupation of slots that constitutes the dynamics of a frame.

With respect to (e): Prediction structures. To what expect do the default values of a frame correspond to expectation or prediction structures? To determine this let us replace the last two sentences in the first scenario with two alternatives. We now have three variants:

Scenario 1:…This has made her decide to take matters into her own hands and finally file for a divorce. She already has an appointment with her lawyer next week.

Scenario 1′:…This has made her decide to take matters into her own hands and finally file for a divorce. But she realized that they hadn’t lived apart for a full year.

Scenario 1′′:…This has made her decide to take matters into her own hands and finally file for a divorce. He’d have to pay her plenty of money anyway.

Why do we have no difficulty interpreting all three text components coherently al-though at first sight it seems unclear what the evoked divorce frame has to do with a lawyer (scenario 1), living apart (scenario 1′), or paying money (scenario 1′′)? The an-swer is this: The informational content in each of the final sentences corresponds with default values in the divorce frame. Because default values (1) and (12) are already inherent in the divorce frame, the required conceptual integration into the divorce frame occurs automatically, as it were.

355. Bartlett’s finding that test subjects tend to adapt textual data to known data when recall-ing a story can also be explained by the gestalt-like nature of representations. In the process of “standardization”, the borderline between fillers and default values becomes fuzzy.


With respect to (f): Recursive embedding via instantiation. To understand the final sentence of scenarios 1, 1′, and 1′′ it is not enough to interpret it solely as an instance of a slot within the divorce frame. Let us take scenario 1: it is not the fact that a lawyer appears as an additional participant (i.e. that this is an instance in slot (7)) that makes it contextually clear why Hannah has an appointment with a lawyer, but rather, the fact that we engage knowledge about the area of activity of lawyers, their professional powers, and their institutional function. Thus, the expression lawyer itself evokes a frame that contains the corresponding information. Although not all the information in this frame is U-relevant, some is, however, significantly involved in the process of establishing coherence.

Figure 4 illustrates this. In the lawyer frame, for instance, the default value dy is not U-relevant.356 dy could, for instance, represent facets of knowledge that con-cern the outward appearance of a person, such as hair colour, shoe size, and physical build.357 By contrast, default value dx and filler fx are most certainly relevant. In this case, the default value dx in the lawyer frame is identical with the default value d6 in the divorce frame because both frames are evoked in the same superordinate context, namely within the conversation that Hannah has with her friend about her marriage. By contrast, the filler fx is not contained in the divorce frame. Filler fx could stand for the information (implied in scenario 1) that to consult a lawyer one generally makes an appointment.358 Similarly, in the divorce frame there are default values that are U-relevant but do not occur in the lawyer frame. This includes d1, namely the prerequisite that Hannah and Nick are married to each other and have been living apart for at least a year.359

356. Once more a hyperonym type reduction needs to be carried out to determine the slots in the lawyer frame. This would lead to the highest hyperonym person. Konerding (1993: 185 and 322–326) differentiates more precisely between two-person matrix frames, namely person with temporary or permanent property/disposition or in a particular state and person in a professional role. However, the slot in the second matrix frame would have to be aug-mented by the slot in the first one to cover all potential knowledge specifications.

357. In the matrix frame person with temporary or permanent property/disposition, this sort of information occupies the slot: ‘In what manner does this property occur in the per-son?’ (Translated from Konerding 1993: 322)

358. Incidentally, one piece of information that occupies the following slot in the person matrix frame: ‘What aids (for instance, organizational) does the person need to carry out their occupa-tion?’ Cf. Konerding (1993: 325).

359. In addition, in their revised theory Fauconnier and Turner (1998a) posit a generic mental space that shares elements of individual input – in this case, of the divorce and lawyer frames. In our case, the generic mental space can be ignored.


divorce frame lawyer frame

Knowledge of the role and function of a lawyer indivorce proceedings

d1 d6fx

d6’ dx’

d1’t’

dx dy

dx

Figure 4. Potential correlation between the divorce and lawyer frames; d = default value, f = filler360

A general conclusion can be drawn from this microanalysis: Both fillers and default values can themselves evoke frames, which in turn contain default values (and possibly fillers). Consequently, frames are recursively embedded in frames.

With respect to (g): Hierarchical organization. Frames that are embedded in each other differ with respect to their degree of abstraction. Frames gain in abstraction to the same degree that they lose in knowledge specification. For clarity, let us look at three levels of abstraction. The upper end of the scale is marked by matrix frames (in the sense used by Konerding 1993; see here Section 6.3.1). Thus, the matrix frame event cannot be meaningfully retraced any further to another hyperonym. At the same time matrix frames are so strongly underspecified that for a linguistic expression like event hardly any values can be found that occupy the slots of the frame associated with it without over-determining lexical meanings.

The situation is different for a frame of medium abstraction. Default values can be determined with relative ease, and only a few are sufficient to clearly identify the frame. For instance, compare the default values (3) and (12) in the divorce and accident

360. This form of illustrating processes of conceptual integration corresponds to common practice (cf. for example Fauconnier and Turner 2002). As yet there is no alternative mode of illustration for frames. However, in my view the “mental spaces” of which Fauconnier and Turner speak just as much represent schematic-semantic knowledge units and accordingly can be reconstructed in frame-semantic terms (cf. Sweetser 1999).


frames in Table 2. In contrast to the shared hyperonym event, the words divorce and accident are so-called “basic level categories” (Rosch 1978), which are, cognitively seen, particularly salient categories as they are located on a medium level of abstraction and occur fairly frequently in language use (cf. Section 6.5.1).361

Table 2. Comparison of two default values in the divorce and accident frame

# Slots in the matrix frame event

Default values for divorce Default values for accident

(3) What can the event be traced back to?

To the decision of at least one of the participants as a result of unpleasant experience.

To human or technical failure.

(12) What consequences does the event have?

Formal and legal conflicts (for instance concerning maintenance payments, division of assets, possibly determining the custody of the children), a psychological and emotional burden.

Damage of various kinds (property damage, injury, death of one or more of the participants).

If fillers specify an evoked frame more closely, beyond the activated default values, this leads to a decrease in its degree of abstraction. This is shown already in the compounds sport accident and traffic accident: Sport and traffic represent fillers that specify the ac-cident frame in such a way that they particularize the superordinate context in which the accident occurs. They occupy slot (6) in the event matrix frame. These compounds do not represent “basic level categories”. They are located on a lower level of abstraction.

With respect to (h): Production and development of frames via induction and abduc-tion. This brings us to the final feature: Do language users learn and change frames via abductive and inductive learning processes, i.e. based on concrete language data? Or are frames constituted independently of language or do they even pre-exist it? Put differently: How are schemata formed on the basis of language input? In this context, Langacker points out our fundamental, non-language-specific abstractive ability:

… [A]bstraction is the emergence of a structure through reinforcement of the commonality inherent in multiple experiences. By its very nature, this abstrac-tive process “filters out” those facets of the individual experiences which do not recur. … A schema is the commonality that emerges from distinct structures when one abstracts away from their points of difference by portraying them with lesser precision and specificity. (Langacker 1999b: 93)

Only by implementing longitudinal corpus-based studies can one expect to gain insight into which facets of knowledge are gradually “filtered out” in the course of recurrent experience. In addition, in connection with very abstract, particularly change-resistant image schemas it has transpired that it is only possible to gain a comprehensive view of

361. Thus, in the sense of prototype theory they exhibit a higher cue validity.


this problem complex through interdisciplinary access.362 At this point I can only say this much: Evidence for the enormous productivity of abductive and inductive learning on the basis of schemata has been gained from language acquisition research (Behrens 2006; Lieven et al. 2003), morphology (Bybee and Slobin 1982; Bybee 1985), semantics (Langacker 1988c; Tuggy 1993), and phonology (Bybee 2001; Taylor 2002: 143–160). It also seems that image schemas (cf. Dodge and Lakoff 2005: 60–72), such as part-whole or source-path-goal schemas, motivate relations between frame elements, thus contributing to the emergence of relatively stable structural patterns (Watters 1996; Ziem 2008b). The conceptual blending of the divorce and lawyer frames il-lustrated in Figure 4 is essentially based on a part-whole relation. The image schema path delivers, as it were, the blueprint for dynamic structural patterns.

Against the background of these example analyses, it is evident that the constitu-tive features of schemata (a)–(h) are inherited by frames. Frames are thus specifically semantic schemata that organize and structure the knowledge relevant to understand-ing a linguistic expression. We can conclude with Seana Coulson:

Frames are representations with slot/filler structure, default values, weak con-straints on the type of fillers for a given slot, and a hierarchical organization that al-lows recursive embedding of frames within frames … Frames, as representational structures, can be used to represent knowledge about a wide variety of objects, actions, and events. … [T]emporal frames, or scripts, represent sequences that extend in time … (Coulson 2001: 35)

In Chapter 6, I further develop the notion that the three structural constituents of frames play a highly prominent role.

362. This applies in particular to disciplines like philosophy, anthropology, neuroscience, psy-chology, and AI research; cf. the representative overview in Hampe (2005).

chapter 6

The structural constituents of frames

Although the terms “slot”, “filler”, and “default value” are commonly used as designa-tions for the structural constituents of schematic representations of knowledge, in many cognitive studies it remains unclear what status the entities they describe have, how they are constituted, and what method allows empirical access to them. What ap-plies to memory-psychology studies applies no less to linguistic frame theory, namely

that “conceptual structures” are spoken of in the context of the cognitive paradigm with reference to knowledge and the cognitive state of individuals, but that the nature and properties of these structures and the concepts accompanying them remains practically unresolved. (Translated from Konerding 1993: 105)

Or more precisely, in what (linguistic, cognitive, epistemic) form do slots, fillers, and default values occur in frames? And how can they be tied back to linguistic material both functionally and structurally? In this book, too, the structural constituents of frames have not yet been sufficiently reflected on; instead they have mostly been em-ployed heuristically for illustrative purposes.

The following answers to the questions posed here serve as the theoretical, linguis-tic foundation of frames. Since this cannot take place without recourse to linguistic, structural analysis, the next sections also present a proposal as to how frames can be employed as an instrument of semantic analysis.

6.1 Issues

Few frame-semantic studies fail to refer to the dynamic nature of frames, which stems from recursive schema instantiation. This makes it all the more surprising that in par-ticular in the Anglo-American literature there is generally no account given of which linguistic elements form schemata or instances or for which reasons. For comparison, take Fillmore’s early remarks:

A frame is a kind of outline figure with not necessarily all of the details filled in. … Comprehension can be thought of as an active process during which the compre-hender – to the degree that it interests him – seeks to fill in the details of the frames that have been introduced, either by looking for the needed information in the rest of the text, by filling it in from his awareness of the current situation, or from his own system of beliefs, or by asking his interlocutor to say more. (Fillmore 1976b: 29)


In other words, once a frame is activated (i.e. evoked or invoked), the language users perform the semantic construction work themselves by filling the slots of a frame with “details”. According to Fillmore, they either stem from the linguistically given data (“in-formation in the rest of the text”), from para- or non-verbal information sources (“the current situation”), from general world knowledge (“system of beliefs”), or – in the case of oral communication – from conversation partners who provide further “details”.

However, what are “details”?363 This question should be of some interest, since inferred world knowledge and knowledge supported by current perception do after all relate to quite different types of information sources, which in turn are distinct in many respects from the knowledge in the already constituted text world.364 Following construction-grammar reflections on the nature of linguistic signs, even if every sign constitution occurs inferentially, the already constituted text world remains principally transferrable into a particular set of propositions. At the same time, the number of pos-tulated assumptions can be potentially endless and evade the sign producer’s control.

Fillmore’s somewhat undifferentiated categorization of U-relevant knowledge types hints at the distinction between fillers and default values. Fillmore states in a later article that to understand a text it is necessary

to know what frames are active in the text world at this point and what values have been assigned to their slots, and to know what functions the just-introduced frame can accomplish in this setting. (Fillmore 1985: 234)

Evidently what he means here by “values” is non-processed background knowledge. This is implicitly addressed at another point in the text. Thus, Fillmore remarks that frames motivate the inferential interpretation of U-relevant knowledge: “the envision-ments get filled out and coloured in by inferences that we draw from our knowledge” (Fillmore 1984: 140). However, the distinction between textually “assigned” and in-ferred values remains vague, and at no point is it reflected on in its own right.

If slots are filled by inferred knowledge, the question remains as to how language users – and linguists like Fillmore – “know” what slots a frame has at all. In a footnote, Fillmore explains in passing using an example:

363. Fillmore posed the same question in an essay from 1977: “The first part of the text activates an image or scene of some situation in the mind of the interpreter; later parts of the text fill in more and more information about that situation. … As one continues with the text, the details of this world [which has been created by the interpreter] get filled in, expectations get set up which are later fulfilled or thwarted or left hanging…” (Fillmore 1977a: 61; my italics). As Fillmore acknowledges a few pages later (1977a: 72) this cannot be due to words alone. Or how should the words blond, glasses, handball, school specify a person frame?

364. In the description of knowledge types I orient myself towards Busse (1991a: 149f.). Cf. the detailed overview in Section 3.3.1.

Chapter 6. The structural constituents of frames 245

In my view, such words as skip, hop, leap, etc., reflect separate frames, each represen ting its own schema of pedal locomotion. There is no context-free frame within which these terms occupy different “slots”, though such a frame could easily exist if there arose, for sports purposes, say, a need for stipulating precise distinc-tions among them. (Fillmore 1985: 229)

Assuming that, the skip, hop, and leap indeed evoke different frames, and assuming that in a particular context there prove to be U-relevant meaning differences arising from the actualized frame – why should the meaning variance occur because different slots are occupied by values, as Fillmore suggests? Why is it not the converse, that different values occupy the same slots? And what slots would they occupy? Irregularities of this nature occur because it is not clear what exactly lies behind the metaphors “slot”, “default value”, and “filler”.

Lönneker (2003a: 64ff.) presents an initial approach to a solution, which indicates that frames can be structurally transferred into propositional structures. A proposition is to be understood in the Searlean sense (1979: 38–43, 48–54) as being that dimension of sentence content that pertains to the propositional content of a sentence regardless of its mode of expression. This is distinct from a pragmatic dimension, which also belongs to the sentence content, namely to the action content or the illocutionary force of a sentence. Thus, the illocution of an utterance – as in the case of a question, a statement, or an exclamation – can certainly vary, whereas the propositional content does not. Nonetheless, propositional content and illocutionary force remain abstract categories that, although able to be determined, function inseparably in the sentence utterance act, because there is no proposition without illocution;365 the same principle applies in reverse, if we set aside exceptions such as single-word sentences like Cheers! and Hello.366. To what extent, then, do frames represent the propositional content of frames? What is the connection between frames and propositions? To provide a satisfactory answer to this, we need to first clarify what propositions consist of. In a Searlean sense (1979), propositions constitute a sub-act of speech acts or, more pre-cisely, the propositional content of linguistic utterances. Something is stated about a particular entity (the subject of a sentence) in almost every sentence (cf. von Polenz 1985: 91–100). Two subcomponents play a role in this. The subject of a sentence indi-cates an entity, a reference object, which recipients must clearly identify if they are to understand the sentence. Searle (1979: 39–40) calls this the “reference”, however, not so much meaning the process of referring or the relationship between reference object and the expression referring to it as that which is being referred to. This reference object (or denotatum) is now more closely characterized or described by the predicate of a

365. “Propositional acts cannot occur alone; that is, one cannot just refer and predicate without making an assertion or asking a question or performing some other illocutionary act. The lin-guistic correlate of this point is that sentences, not words, are used to say things” (Searle 1969: 25).

366. Interjections like hmm are an additional exception. They have illocutionary force but no propositional content.


sentence: It is predicated as something. “Predication” in this sense means allocating predicates to a reference object. However, in contrast to reference, predication does not represent an autonomous speech act. Nor is it a special type of reference, which can be seen by the fact that predication always occurs in a particular illocutionary form, whereas reference is not restricted to this.

Croft (1991) makes a similar distinction, albeit without focusing on the pragmatic dimension to the same degree. He understands reference and predication as basic lin-guistic functions that are necessary to successfully move from linguistic expressions to conceptual units.367 Reference serves to “get the hearer to identify an entity as what is being talked about”; predication, by contrast, deals with “what the speaker intends to say about what he is talking about (the referent)” (Croft 1991: 52). Croft assumes that, apart from reference and predication, there is a third basic function to consider, which is missing in Searle’s approach. He calls this “modification”. Modification assumes an “accessory function to reference and predication” and “helps to fix the identity of what one is talking about (reference) by narrowing the description” (Croft 1991: 52). Attributes or relative clauses that more closely specify the reference object function as modifiers, for instance. However, in my view, these are a special form of Searlean predi-cation rather than a third basic linguistic function,368 since a modification is nothing more than a predicative (re-) specification or closer clarification of the reference object.

In the following, I assume along with Searle (1979) and Croft (1991) that the prop-ositional act occurs as an irreducible interaction between referring and predicating.369 If frames represent encyclopaedic knowledge, i.e. sets of propositions, then the two subcomponents of propositions must be able to be analytically specified from frames. My proposal for this is: An evoked frame corresponds to the referential content of a proposition, and that which is predicated on a reference object corresponds structur-ally with the fillers and default values in the frame model favoured here. Accordingly, slots show which predications can potentially occur. Until now, only Lönneker has indicated this correlation:370

367. Further, Davis (2003: 7–11) argues that in addition to referring and predicating, communi-cating also represents a basic semantic or, as he puts it, psychological act. Whether all linguistic utterances can ultimately be retraced to these three acts or sub-acts and whether, above all, each of these three acts or sub-acts can generally occur without the other two is a moot point.

368. Or more precisely, relative clauses can not only exhibit the quality of a predication, but also adopt a referential function themselves; more on this in Section 6.4.2.

369. Cf. David Schwarz (1979: 21): “For the purposes of referring, it seems to be enough that the speaker means something to the effect that a certain individual has some predicate, where this predicate is left unspecified. That is, if I start an utterance, evidencing clearly the intention of predi-cating something of an individual, but do not get the predication part out, I may still be referring”.

370. In the following, I dispense with Lönneker’s chosen term “frame name” since it is unclear what status is attributed to a “frame name”. In any case, contrary to slots, fillers, and default values, frame names do not have the status of a structural constituent inasfar as they only have


Within a frame in the model used here, the reference consists in the frame name, and the predication in two additional elements through which certain properties, abilities, or similar are ascribed or denied to a frame concept. (Translated from Lönneker 2003a: 64)

Figure 1 summarizes the terminological correlations once more.The extent to which frames may function as an instrument of semantic analysis

already suggests itself at this point: Frames can be specified on the basis of a text corpus by first identifying the occurrence of the phonological unit (that evokes the frame in question) in the corpus and on this basis determining what is predicated on this entity. The result is a large set of propositions that overall make up the semantic dimension of a frame as it is found in the base corpus.

Slot

(activated) frame

Filler/Default value

Reference Predication

PropositionSpeech-act-

theorydistinction

Distinctionsuggested here

Figure 1. Structural constituents: Comparison of terminological distinctions

The next section begins by focusing on a frame-theoretical specification of linguistic reference acts. The following sections then address the frame elements concerned, (i) slot, (ii) filler, and (iii) default value.

6.2 Reference

In the beginning was the word, or more precisely, the activation of a frame. Here “ac-tivation” means the identifying reference to an entity without which a linguistically performed communicative act could not occur.371 Any unit that is in any way quan-

to be enlisted for the analytical specification of frames. Then they would have to be external to a frame, which would exacerbate the issue of their status.

371. Cf. Vater (2005: 145) and also Vater (2005: 11; in translation): “Reference is central because no speaking occurs without referring”. Vater understands the following by “reference”: “In an utterance, the situation underlying the predication as well as the places, time periods, and objects described by the arguments are explicitly or implicitly referred to. This is ‘reference’.”


tifiable or qualifiable may be considered the potential reference object of a linguistic sign. With linguistic signs we can just as easily refer to objects (like a table or a car) as abstractions (like God or love), actions (like shopping), or even fictional entities (like unicorns or objects in a fairy tale). In this process, utterances, i.e. phonological units in Langacker’s sense, serve, so to speak, as linguistic stimuli that cause a language user to infer a reference on the basis of relevant background knowledge. They motivate linguistic reference acts.

If – as is my assumption – the activation of a frame is equivalent to the referential function of a linguistic expression, this raises two questions:

– What do linguistic expressions refer to? That means what referential relationship do frames establish?

– Which linguistic expressions are referring expressions? That means which linguis-tic expressions evoke a frame?

The next two sections engage with these questions.

6.2.1 Frames as a projection area of referentiality

Let us begin with the first question. For a long time, the view prevailed that reference objects were to be identified with real-world, i.e. extra-mental, correlates of our per-ceptual world (cf. Wimmer 1979). This position corresponds to the naive-realist notion that our mind is a mirror of nature. However:

The world to which our linguistic utterances refer cannot… be comprehended as an externally mediated world able to be accessed by our consciousness, in the sense of naive realism, but must be considered as a world constructed by the human cognitive system and thus as internally generated. (Translated from Schwarz 1992b: 169)

Without entering into the epistemological difficulties that naive realism entails (Rorty 1987), even simple psycholinguistic experiments show to what degree reference objects are the result of cognitive construction processes that very much follow their own rules. Language users do not denote things with words, but in fact conceptualize reference objects rather flexibly, depending on the contextually determined “set of alternatives” (translated from Olson 1974: 192).372

Thus Olson (1974) reports on an experiment that consisted in showing several test subjects a wooden block under which lay a golden star. The block’s surroundings change: On one occasion there is also a black, round wooden block next to it, on another a white, rectangular wooden block, and on another there are three blocks next to it (a white rectangular one, a black round one, and a black rectangular one). When the text subjects name the wooden block with the star underneath it, this always occurs

372. Olson (1974: 192ff.) uses the term Bezugsobjekt (reference object).


relative to its surroundings. Overall there are three different designations for one and the same block (the round one, the white one, the round white one).

This already indicates that language users conceptualize entities not just relative to their own experiential basis. Perhaps more importantly, they posit that this experiential basis is given or at least assume that the linguistic designation adequately describes the entity relative to its surroundings. This presumes normal conditions that must be fulfilled so that in this case, for instance, the colour descriptions and deixis can be adequately conceptualized.373 To name a few examples: To understand the preposi-tion under, it is presupposed that the test person will not do a headstand during the description, or, less absurdly, that certain assumptions about the physical properties of the star and the wooden block apply. Such assumptions exert considerable influ-ence on the process of meaning construction (cf. Brugman 1988). They are decisive for the difference in conceptualization between the complex expression the star under the wooden block and, for instance, (a) the tunnel under the wooden block. Whether the phrases (b) the ant under the wooden block or (c) the ball under the wooden block are determined to have conceptual or rather semantic similarities naturally depends on what information is given about the star or which default values are inferentially interpreted. Figure 2 illustrates the three conceptual variations.

Referentialization operations are thus based on foreknowledge, and mental pro-cessing give rise to the formation of perceptual units. Consequently, the reference of linguistic expressions does not concern the real world, but rather what Jackendoff calls the “projected world”.

We must take issue with the naive position that the information conveyed by language is about the real world. We have conscious access only to the projected world – the world as unconsciously organized by the mind; and we can talk about things only insofar as they have achieved mental representation through these processes of organisation. Hence the information conveyed by language must be about the projected world. (Jackendoff 1983: 29)

373. Cf. also the epistemological explanation given by Sellars (1999).

(a) (c)(b)

Figure 2. (a) the tunnel, (b) the ant and (c) the ball under the wooden block


Jackendoff makes reference, among others, to gestalt-psychology findings when he distinguishes the “projected world” from the “real world”. Figure 3 shows two examples from gestalt psychology.374 In the first example (a), we see four dots that we inevitably connect up as a line by dint of their arrangement. This grouping occurs automatically and unconsciously and is not motivated by the input.

(a) (b)

Figure 3. The construction of reference objects: (a) four dots that form a line, (b) a vase or the silhouettes of two faces

The well-known example (b) is similar. The object of perception changes depending on whether we interpret the black area as the figure or the background. Although these examples may seem trivial, what they demonstrate is by no means inconsequential. In (a) and (b), what we see depends crucially on what we interpret to be the surroundings of the perceived object. It is not possible to circumvent this constructive act of object construal in perception to “simply just” perceive what the input provides.

Very similar findings come from a neuropsychological perspective. When the brain processes external stimuli and we perceive a particular entity as a result, the perceived entity is not a reflection of the stimulus source due to the preceding mental processing of the stimuli.

The mind generates percepts that are perceived as real, external entities. The human mind constructs an externally projectable world, which for us is organ-ized in such a way that we experience it as an objective, independently existing structure. (Translated from Schwarz 1992a: 43)

This is how the hiatus between the “real” and the “projected” world occurs. Put more pointedly, “reality” is a construction product of our neural brain activity on the one hand and our mental power of imagination on the other.

The answer to the question concerning the reference area of linguistic expressions must therefore be formulated as follows. Linguistic expressions refer to segments of the projected world. Reference in this context means making linguistic reference to a cognitively construed representational unit that traces back to a perceived external

374. Numerous other examples can be found, for instance, in Metzger (1923) and Wertheimer (1923).


object or arises solely out of the imagination. Linguistic expressions can thus refer to perceived entities or to purely representational units that have no perceivable correlate. Again, Schwarz states:

On the one hand, we refer to the projected, yet objectively experienced world Wp, whose units we access as percepts, while on the other the representational units of our world model Wm, of whose mental character we are consciously aware, constitute the potential referents of linguistic expressions. The units of this model are representations of objects, images, and notions. These representational units are like objects in the human mind, whereas objects are like images in the external world. (Translated from Schwarz 1992a: 45)

If there are no extralinguistic correlates for a linguistic expression to refer to, and if there is consequently no intermodal connection between linguistically given informa-tion and non-linguistic data (such as visual or auditory data), then a conceptual unit alone constitutes the reference area.375 In both cases – which is the decisive point – linguistic expressions refer to cognitive units.

This result conforms to the conclusion in Section 4.2.3 that “scenes” (in the Fillmorean sense) and “situations” (in the sense of Zlatev) can only be U-relevant in the form of (propositionally structured) background assumptions.376 A linguistic expression refers to a cognitive unit by evoking a frame, which then opens a potential reference area (similar to Taylor 2002: 71–75, 347). Evocation of a frame corresponds to the cognitive act of referentialization. Frames – as units in the “projected world” – serve as projection areas for referentiality.

Against this background it becomes clear why an activated frame can only be distinguished analytically from the structural constituents of this frame. In fact, it is not possible to identify a reference object without tagging it at the same time as one among many, standing in contrast to other potential reference objects by dint of this very specificity (cf. Searle 1979: 133–138). For instance, departing from the phono-logical unit [unicorn], we could establish reference to the imagined object, i.e. to the semantic unit [unicorn], because we have implicitly qualified it so that it is sufficiently

375. As already mentioned in Section 1.3.1 in connection with the distinction between percep-tion and representation, this conceptual unit is characterized by its cognitive manipulability. If, for instance, I tell a friend about my Golden Retriever called Glamis, he may imagine a light-brown dog that is 50cm at the withers and has a bushy tail, and he can manipulate each of these characteristics (almost) at will. This type of manipulation is not possible if my friend comes to visit and sees Glamis. Cf. Schwarz (1992a: 45); Konerding (1993: 108–109).

376. Konerding observes on the role of the linguistic aspect: “A ‘situation’ – especially if it is linguistically institutionalized – can clearly only be decomposed into identifiable components by means of linguistic specifications. A specification of a situation depends on the culturally controlled textual prototypes (concepts/terms) if it does not want to be limited to idiosyncratic and arbitrary ‘interpretations’ Tokens of the latter can generally only be identified with assurance via the linguistic utterance.” (Translated from Konerding 1993: 129)


distinguishable from alternative reference objects (like [rhinoceros] or [horse]). In Busse’s words:

… the (nominally performed) act of referring already constitutes an elementary (implied, imputed) organizational structure, which epistemically seen goes be-yond the mere naming of the pre-existent. The predication adds an additional organizational structure to the reference act, thus already generating a complex epistemic structural framework within a simple sentence. (Translated from Busse 2005a: 47–48)

This means that reference and predication stand in a much closer interrelationship than Searle’s comparison appears to suggest. Thus, the predicative part of a proposition can also adopt a referential function (cf. Vater 2005: 102ff.). In any case, referentialization is an act that creates order by co-activating the default values of the evoked frame in its very performance. Using the word unicorn to refer to a particular imagined object presupposes, for example, actualized assumptions about the external properties of a unicorn.377

6.2.2 Every word evokes a frame

Turning to the second question: Which expressions are referring expressions? Searle’s answer to this question derives from his definition of referring expressions.

It is characteristic of each of these expressions that their utterance serves to pick out or identify one ‘object’ or ‘entity’ or ‘particular’ apart from other objects, about which the speaker then goes on to say something, or ask some question, etc.[…] Referring expressions point to particular things; they answer the questions “Who?” “What?” “Which?” It is by their function, not always by their surface grammatical form or manner of performing their function, that referring expres-sions are to be known. (Searle 1969: 26–27)

Searle includes proper names, nominal expressions, and pronouns amongst the set of referring expressions. However, what about other word classes, such as those that are seen as function words, as it were (like prepositions, conjunctions, modal particles, etc.)? Are these unable to adopt a referential function? If this were the case, this would deny their conceptual status. According to the basic premises, function words would therefore not evoke a frame, and that would cause the symbolic principle explained in Section 4.1.1 to topple, because there would in fact be “meaningless” signs, i.e. pho-nological units without an associated content dimension.378

377. Searle (1979: 146–149) specified this by positing conditions of success for the performance of a reference act. These illustrate the cognitive complexity of referential action.

378. Similarly, Busse states (translated from 1997b: 224): “If we deny those language units whose semantic description cannot be pressed into the concept of referential semantics their ‘meaning’


However, taking the criteria listed in Searle’s quote as the yardstick for which expressions are referring and which aren’t, doubts arise as to whether non-referring expressions in the strictest sense actually exist. Searle’s criterion is that an expression is referring if we identify it with an entity in such a way as to be able to answer the ques-tions “Who?”, “What?”, “Which?” Yet clearly all word classes can fill this criterion, at least when they occur in a nominal or nominalised form. Apart from individual words, phrasal and clausal syntagmas and even whole texts can be nominalized. Compare the following Examples (1) to (5):

(1) That car bothers me. (2) The party bothers me. (3) That redness bothers me. (4) Peter’s always-coming-too-late bothers me. (5) That “or” bothers me.

When we read or hear sentences like this, we use the subject of the sentence to identify such different objects as events, properties, complex actions, and particular linguistic expressions themselves. Within a particular contextual framework it should not be too difficult to specify each reference object according to the contextual circumstances. Thus, there may be a context in which we explain the subjects of sentences (1) and (5) in the following way:

(1′) That car bothers me. That green one that has been parked all week in our parking spot out front.

(5′) That “or” bothers me. Why can’t you just decide?

The reference areas of the two expressions car and or do indeed differ fundamentally from each other. In contrast to (1), in (5) we are dealing with a metalinguistic or metaphorical usage, which demonstrates, however, that conjunctions certainly can be ascribed specific attributes. The use of or makes it clear that we are dealing with an alternative option, and this could express – as in (5′) – the lack of decisiveness of the person who uses it remarkably often.379 Against this background I suggest that nomi-nals of the type (1), (2), and (3) (also often described as Autosemantika in the German literature) and nominals of the type (5), so-called “function words”, as the end points of

(as defined here) attributes (and instead merely speak of – as is often the case – mere ‘functions’, mere ‘co-meanings’ [Mit-Bedeutungen] in contrast to their own referentially defined meaning), then we destroy the semiotic character of these units and thus, at the same time, their language-ness per se…”

379. This does not yet say anything about the specific textual function of the conjunction or; nor is this the issue here.


a scale.380 Phrasal compounds such as those in (4) are a special case due to the fact that the verb figures as the nucleus of the nominalized syntagma. The valency-induced ref-erence points of the verb are simply co-nominalized.381 (4) is therefore a variant of (4′).

(4′) His coming bothers me.

Both (4) and (4′) are therefore located at the upper end of the scale, i.e. near (1)–(3), since conversions are frequent nominalizations and do not have a metalinguistic char-acter. In Busse’s sense, the scale shows a “more-or-less” of referentiality:

If the concept of meaning is prototypically organized, then it functions simultane-ously as a scalar concept: Then there is potentially also a “more-or-less” of referen-tiality – depending on word type. Then referentiality would perhaps be, above all, a cognitive variable that lives with the prototypical examples for which this concept was evolved, but that also becomes more problematic the more removed the refer-ence objects of the individual word types become from the referring character that defines the prototypical core category. (Translated from Busse 1997b: 223)

Accordingly, the most problematic group of referential expressions is function words, which includes particles of different types. Although their reference potential is re-duced to a minimum, it would nonetheless be wrong to deny them all referentiality, at least when they are the object of metalinguistic reflection. As soon as linguistic expressions are discussed metalinguistically, what Searle sees as characteristic of refer-ring expressions applies: The speaker identifies an entity by (implicitly) answering the questions “Who?”, “What?”, “Which?”

I do not wish to claim that all linguistic expressions in a sentence are referring expressions – a position that Vater (2005: 71ff.) seems to represent. I merely hold the view that all expressions can be made into referring expressions (for instance via meta-linguistic reference and usually accompanied by nominalization). The specific opera-tions and functions of individual word classes are therefore not being questioned. It should also be noted that metalinguistically addressed expressions lose their func-tional-pragmatic function. For instance, if someone utters the sentence You always say “mm, mm,” but you don’t understand anything at all, although the interjection mm is distinguished from other interjections, etc. to identify it, the functional-pragmatic value of the expression is lost.

The example of the preposition under has already shown (cf. Figure 2) to what extent the interpretation of the expression depends on contextual knowledge and

380. I speak with caution here of “function words” and not of “synsemantica”. The dichotomy between “autosemantica” and “synsemantica” usually entails a distinction between meaning-carrying and non-meaning-carrying words – a distinction that I specifically wish to avoid.

381. Formulated in frame-theoretical terms: Some slots in the frame coming are already oc-cupied by fillers.


presupposition.382 The same applies to modal particles. Compare the following sen-tences in German:

(6) Peter ist ein Faschist. ‘Peter is a fascist’ (7) Peter ist wohl ein Faschist. ‘Peter is probably a fascist’ (8) Peter ist vielleicht ein Faschist. ‘Peter is perhaps a fascist’

The modal particles wohl (probably) and vielleicht (perhaps) in (7) and (8) indicate attitudes that the language users take to the illocutionary and propositional content of an utterance. What is more interesting is that the same particles establish other forms of reference in other communicative contexts.

(7′) A: “Peter ist kein Faschist!” ‘Peter is not a fascist!’ B: “Peter ist WOHL ein Faschist!” ‘Peter certainly is a fascist’ = ‘He is SO!’

(8′) PETER ist vielleicht ein Faschist! Ich kenne keinen, dem man die Gesinnung schneller anmerkt.

PETER is perhaps a fascist! … ‘Peter is a real fascist! You can spot it a mile away’

Wohl and vielleicht have a discourse functional role here. In oral communication, prosodic features also assist in determining the meaning of a particle.383 If we de-nied discourse-functional properties of particles, differences such as those between (7) and (7′) or (8) and (8′) would not exist. Taylor (2002: 69) sees this as an indica-tion that particles represent conceptual units whose content certainly can vary. Vater (2005: 72) even argues that particles and modal verbs have their very own distinct area of reference, which he calls “reference to modality”.384 However, since these reference-theoretical specifications do not satisfy the criteria that Searle sets to identify referring

382. This instance may serve as a typical example. For a discussion of numerous other preposi-tions cf., for example, the volumes edited by Cuyckens and Radden (2002) and Pütz and Dirven (1996), as well as the overview in Cuyckens (1988).

383. Capital letters in (7′) and (8′) indicate intonational emphasis.

384. In total, Vater distinguishes four main reference areas, namely “reference to situa-tions” (‘Situationsreferenz’), “reference to things” (‘Dingreferenz’), “reference to locations” (‘Ortsreferenz’), and “reference to time” (‘Zeitreferenz’), as well as three marginal refer-ence areas, namely “reference to attributes” (‘Eigenschaftsreferenz’), “reference to quantity” (‘Quantitätsreferenz’), and precisely this “reference to modality” (‘Modal(itäts)referenz’) (trans-lated from Vater 2005: 71–72).


expressions and since, according to Vater, predicates (in a predicate-logical sense) also fulfil a referential function, it becomes more and more difficult within Vater’s theory of reference to adequately differentiate between reference and predication.

The fact that particles can be made into referring expressions is seen most clearly when they are metalinguistically addressed so that something can be said about them in the form of predicative attributions. In addition comparing (7) with (7′) and (8) with (8′) clarifies that the meaning of modal particles can change with respect to discourse function. Similar arguments can be provided for the other particle types.

This is where I end this discussion. What we can note is that reference objects have different manifestations in the ranking of cognitive units. In the prototypical case of “autosemantica”, linguistic expressions refer to represented things or events. Non-prototypical, but by no means less important, are pragmatic and discourse-functional references. They endow cognitive coherence.

In keeping with the semiotic premise that every linguistic sign is a symbolic unit, i.e. is a form-meaning pairing and that the content dimension can encompass semantic, pragmatic, and discourse-functional aspects, Lakoff ’s (unsubstantiated, yet succinct) dictum applies: “Every word evokes a frame” (Lakoff 2005).385

6.3 Predication potential: slots

Following the previous reflections, the question of what language units refer to is closely linked to the question of which frame evokes a linguistic expression or which potential reference points – implicit and explicit – are evoked by a token within a par-ticular context. In the following, I address this question from two perspectives. First, I explain what is to be understood by reference points (Bezugsstellen) within frame semantics. Another step is necessary for the empirical, text-analytical application: It is necessary to show a method that allows the potential reference points of a linguistic expression to be determined.

385. Cf. also Minsky (1975: 236): “Any concept can be invoked by all sorts of linguistic represen-tations. It is not a matter of nouns or verbs”. Or Petruck, from the perspective of lexicography (1995: 279): “In a frame-based organization of the lexicon, it is the frame which provides the conceptual underpinnings for related senses of a single word and semantically related words. This would necessarily include all categories of words, including nouns, verbs, adjectives, ad-verbs, prepositions, and conjunctions [sic], as well as phrases and expressions”. Reasons for the cited assumptions are given by neither Minsky nor Petruck.


6.3.1 What are slots?

The expression “Bezugsstelle” (reference point) was introduced by Peter von Polenz (1985: 116–143) in his work Deutsche Satzsemantik (German sentence semantics) to describe sentence-semantic structures. Von Polenz prefers the term “reference point” to the predicate-logical term “argument” so as to encompass not only the valency of a verb, but also “background reference objects” (‘hintergründige Bezugsobjekte’, von Polenz 1985: 130). Background reference objects are objects that are not realized or cannot be realized in the syntactic surface structure of a sentence, but are nonetheless U-relevant.386 In contrast to most valency theorists including the FrameNet project, von Polenz does not proceed from the assumption of syntactically expressible argu-ments and adjuncts, but rather “of cognitive and cultural premises that predate every communicative action” (translated from von Polenz 1985: 131). This extension appears to be only logical in view of the fact that referentialization is a fundamental linguistic action (Davis 2003), the success of which presupposes complex cognitive processes (categorizations) as well as knowledge of the language behaviour of a language com-munity. Reference points therefore concern situative, contextual, and social knowl-edge – i.e. encyclopaedic knowledge in general.

Understood this way, reference points resemble the slots of a frame in several respects. On the one hand, reference points, like slots, form a framework, i.e. they correlate with each other and are evoked by a linguistic expression as a gestalt-like whole. In this context, von Polenz (1985: 132) speaks expressly of the “frame of ref-erence” (‘Bezugsrahmen’) of a linguistic expression,387 which contains information about what a linguistic expression may potentially refer to. This characteristic seems to be largely equivalent to the function evoked frames have of defining the potential cognitive reference area of linguistic expressions (Coulson 2001: 75). In addition, they share the basic premise that word and sentence tokens are semantically radically under-specified. Without having to be expressed, the greater part of U-relevant knowledge is merely “filled in” but not expressly stated (‘mitgemeint’, ‘mitzuverstehen’, cf. von Polenz 1985: 135). The missing information is inferred by the language recipient on the basis of all the data available. Like frames, frames of reference also start at a “deep semantic” level (Busse 2000a: 42–45; 2000b: 49–51). There is, after all, a consensus that referring and predicating are the two most important object-related sub-actions of a sentence-semantic analysis (von Polenz 1985: 91). Accordingly, von Polenz characterizes frames

386. Cf. the numerous examples in von Polenz (1985: 131ff.) and Busse (1991a: 70ff.).

387. Siegfried J. Schmidt already spoke of the “reference frameworks” (‘Referenzrahmen’) of an utterance with the accent on action theory. According to Schmidt, the reference framework en-compasses such varied areas as “stereotypical actions”, “stereotypical communication situations”, “the recurrent action modes of communication partners”, and “‘conceptual structures’ such as ideologies, mythologies, scientific theories, religions, etc. that are internalized via a history of socialization” (translated from Schmidt 1980: 261–262). However, this list remains incomplete and would be of little use for meaning analytical purposes.


of reference as “predication or assertion frames” (‘Prädikations-/Aussagerahmen’, von Polenz 1985: 158). The same applies to frames as such (Busse 2007; Fraas 1996a, 1996b; Konerding 1993: 139ff.).388

Although both frames of reference and frames represent “forms of linguistically bound activation of knowledge that are tied to situative, textual, and epistemic con-texts” (translated from Busse 1991a: 77), they nonetheless differ in one crucial respect: semantically seen, von Polenz’s frames of reference remain reductionist, despite hav-ing been distanced somewhat from the syntax-oriented view of valency grammar.389 U-relevant knowledge as the key research area of semantic theory in the holistic para-digm encompasses much more than what can be rendered accessible in von Polenz’s approach to the analysis of sentence content. Measured against his own yardstick, he fails to achieve his goal of recording all the predicates and reference points that lan-guage users either explicitly or implicitly express in a sentence.

In my view, the reason for this is that von Polenz is still too oriented towards the syntactic surface structure from which he wishes to distance his in-depth analysis of sentence meaning(s). We already observed a similar phenomenon in the discussion of Fillmore’s deep-case theory (cf. Section 4.2.3, also Ziem 2014). In comparison to cogni-tive scenes, i.e. the later frames, deep case only describes a fraction of what U-relevant knowledge actually is. Deep cases (semantic roles) function in Fillmore’s conception as semantic classification grids for the reference points of verbs. Von Polenz follows on methodologically from here to ensure that access to implicit reference points is regu-lated. In so doing, he expands the list of potential deep-case types by numerous semantic roles that cannot be found in either Fillmore’s or other approaches;390 however, this does not spare him the same criticism that case theory already faced (Fillmore 2003: 466; 2006: 616). Fillmore summarizes the most significant point of criticism as follows:

Difficulties in constructing a working case grammar include the problem of know-ing when to stop in preparing the list of cases, how to construct and test a case hierarchy, and a host of worries about splitting versus lumping. … This failure to provide a well-motivated list of case notions became a central objection to the theory, even among those who felt that the ideas were worth pursuing. Many writ-ers have pointed out that one can always find both reasons for recognizing ever more refined distinctions and reasons for recognizing high-level generalization… (Fillmore 2003: 466)

388. Fraas (1996a: 5, 24–28) notes in this context that the slots of a frame specify the contex-tualization potential, whereas Busse (2007) speaks of frames as networks of predications and reference points.

389. In the sense of a semantic reinterpretation of the syntactic function of verbs (Heringer 1984: 47ff.). It is no coincidence that Heringer follows on from Fillmore’s frame concept.

390. Cf. von Polenz (1985: 170–172). The following semantic roles are new: causative (Causativ/Ursache), possessive (Possessiv/Besitz), additive (Additiv/Hinzugefügtes), privative (Privative/Entferntes), origative (Origativ/Ursprung), and temporative (Temporativ/Zeit) (in capitals in the original).


Consequently, Croft (1991: 155ff.) and Langacker (1987: 284) express fundamental doubts about whether it is possible to provide an exhaustive list of semantic roles and whether there is even a theoretical basis for doing so. As we know, as a consequence of these a priori conclusions, Fillmore abandoned case theory and the conception of semantically non-reductionist frames (cf. Fillmore 2003: 466).

Von Polenz must also face the question: Are all the reference points of an expres-sion really able to be computed using his proposal of sentence-semantic analysis?391 That the answer must be no is illustrated by this example. Using sentence fragment (9), von Polenz (1985: 133–135) demonstrates which reference points are to be considered in his analysis of the sentence structure.

(9) Der Politologe, der in der Zeit des Nationalsozialismus mit seiner jüdischen Familie nach Frankreich floh

‘The political scientist who fled with his Jewish family to France in the period of National Socialism’

What frame does the action type ‘political flight’ evoke? Von Polenz discovers seven reference points in total, which are

a. the person who fleesb. the person or persons from whom one fleesc. the reason why one fleesd. the place from which one fleese. the place to which one fleesf. the path one takes to fleeg. the time at which one flees

Von Polenz further explains that “in some contexts this reference knowledge can be completely expressed using seven reference points, from a to g”.392 Clearly, realizing all the reference points in the syntactic surface structure still plays a certain role – in prin-ciple, at least potentially – although this occurs in the rarest of cases and non-realized reference points are consequently only implied by the language producer and must be inferentially interpreted by the language recipient. However, what must be understood in addition to the seven reference points (and accordingly could be enquired after) are, for instance, facets of knowledge concerning

h. the superordinate (political, social, biographical, etc.) context in which the flight occurred

i. the role that the flight plays within this superordinate context

391. In this context it is not so relevant whether particular reference points combine to form a more general reference point or, conversely, differentiate out into several subordinate reference points.

392. Von Polenz (1985: 133); italics not in original. His example translates thus: “Because of persecutionc1 as a result of his pacifist writingc2, the authora fled from Munichd to the USAe via Switzerlandf in 1938g to escape the Gestapob” (in italics in the original).


j. the (physical, psychical, social, material, historical, biographical, etc.) prerequisites that made the flight possible

k. the (chronological and logistical) phases into which the flight can be subdividedl. those people who also fled or who were involved in the flight in some form or otherm. the roles and functions of those people who also fled or who were involved in the

flightn. the (political, social, material, psychophysical, etc.) conditions that helped or hin-

dered the flighto. the duration of the flightp. the conditions (in this context, above all political) under which the flight could

have been repeatedq. the possible (psychical, physical, social, material, emotional, etc.) consequences

entailed by the flightr. the (personal, political, etc.) meaning that the flight adopts for those who fleds. the (historical, politological, sociological, etc.) theories that address politically

motivated flight.

This is by no means an exhaustive list of reference points, alone for the reason that ad-ditional points of reference connect onto each of the reference points listed.393 Against the background of Fillmore’s (self-)criticism, quite similar questions raise themselves: Is it possible to give an exhaustive list of potential reference points? If so, when is a list exhaustive? What justifies the assumption that a list is exhaustive? And with specific reference to the case example: Why should the reference points (a) to (g) be so fun-damentally different to (h) to (s) as to justify the latter being ignored in the sentence semantics? Their lack of U-relevance cannot be an argument, because contexts can be easily constructed in which at least one of the reference points (h) to (s) proves to be constitutive to understanding. For instance, let us presume that von Polenz did not find (9) in a newspaper article in which a reporter provides background information about an interviewee, but rather the same words were the authenticated quote by a Gestapo officer who had traced the political scientist who fled. Knowledge of the consequences that such a flight might entail, i.e. (q), and possibly of which other people (apart from family members) also fled, i.e. (l), would then be highly relevant. And the fate that awaits those who fled is, of course, not able to be measured without knowledge of the superordinate context; slot (h) would also be involved.

To account for such U-relevant facets of knowledge a more radical disengagement from (form-related) sentence structures (‘Satzausdrucksstrukturen’) is required than that undertaken by von Polenz. To avoid syntactically conditioned economy and its accompanying semantic reductionism, Fillmore even draws the consequence of leaving sentence semantics entirely out of the issue and treating questions on the meaning of

393. In the sense of the thesis subscribed to here that every word evokes a frame. – for additional potential reference points for the reference frame flight cf. Konerding (1993: 435–439).


sentences only with reference to the superordinate textual embedding structure.394 If we consider frames as reference frames, then they are distinct from both the frames of reference in von Polenz’s conception of sentence semantics and Fillmore’s case frames: The slots of frames do not correspond to the slots that can solely be given on the surface level; rather, texts or discourse form the access format from which potential reference points can be derived (Fraas 1996a; Konerding 2005; Lönneker 2003a; Ziem 2008a). Thus, frames do not describe the meaning of words or the meaning of sentences, but rather at best the meaning of words in texts and the meaning of sentences in texts or discourses.395 Nonetheless, neither Fillmore’s case theory nor von Polenz’s approach loses its methodological value. Frames extend the epistemological perspective, without at the same time denying the analytical and heuristic usefulness of case frames and sentence-semantic frames of reference.396

Let us return once more to the opening question of what slots are. Putting aside the fact that, from the perspective of U-relevance theory, the classification of sentence-semantic reference points according to semantic roles falls short of the mark, there are good reasons why the potential reference points of linguistic expressions on the one hand and the slots of a frame on the other are of a very similar nature. Both indicate what type of entities can be predicatively referred to. The example of the reference points (a)–(s) clearly shows that a multitude of U-relevant knowledge references can be established on the basis of a single expression (here: flee or flight). At this point it is important to maintain the analytical separation between reference and predication. Thus, the slots (a)–(s) remain only potential reference points, and as such they only concern the referential function of language signs. In fact, as soon as language users establish an explicit or implicit knowledge reference, that is, say something about a reference object (in this case; ‘flight’), this occurs by means of a predicative attribution of the nature x is y. A predication occupies a slot, as it were, as for instance in (9) the predication the one who flees is a political scientist occupies the slot (a).397 Consequently the slots of a frame correspond to the predication potential of the expression that the frame evokes (cf. Figure 9 in Section 4.3.2). How the predication potential is exploited in the individual case, in other words, what knowledge (in the form of predicative

394. “[T]he question I am asking is whether in addition to the two branches of semantics known as lexical semantics and text semantics the empirical study of meaning within linguistics also needs sentence semantics. When I suggest that it does not, I will not mean that we will end up un-able to say anything about sentences; a sentence, after all, can be a text.” (Fillmore 1984: 125–126)

395. In other words, frame semantics is neither lexical semantics nor sentence semantics in the traditional sense. When it focuses on the investigation of lexical or sentence meanings, it does so from a text- or discourse-semantic perspective (in the sense of Busse and Teubert 1994).

396. This finding is also supported by the fact that semantic roles can be easily transferred into the slots of a frame (cf. Charniak 1981; Fillmore 1977a, c).

397. Thus, seen cognitively, every predication is a categorization (in the sense explained above).


attributions) the slots render accessible is highly variable: “we can profitably regard a lexical item as providing ‘access’ to knowledge systems of open-ended, encyclopedic proportions” (Langacker 1988b: 58).

It is now clear, as I have already indicated several times, that slots are best under-stood as questions that can be meaningfully posed with respect to a reference object. Semantic roles can also be easily reconstructed as potential question viewpoints.398 In his early frame-semantic work, Fillmore (1977a: 64) demonstrates the function that questions have within frames using the example of the verb write. If, according to his assumption, frames consist of systematically correlating concepts, it should be possible to target every concept in a question. The frame that the verb write evokes bundles conceptual knowledge about entities such as the writing person, the writing instru-ment, the surface on which is being written, and the written product.

Since I know at least that much about writing, I know that if you tell me that you have been writing, I can, talking within the frame that you have introduced into our conversation, ask you such questions asWhat did you write?What did you write on?What did you write with? (Fillmore 1977a: 64)

Answering any of these questions evokes in the respondent a potential reference point in the verb frame (referentialization) and provides a concrete value by saying some-thing about the reference object of the writing (predication); the result is a proposition. Any number of answers is conceivable, but they are by no means arbitrary. Since every predicate evokes a frame in its turn, these slots introduce additional potential connec-tions. For instance, if the answer to the first question is x is writing a letter,399 there are in turn a multitude of potential facets of knowledge one could enquire after. Fillmore (1977a: 65) names four of them:

Who are you writing to? When are you going to send it? When do you think she will get it? Do you think she will answer it? (Fillmore 1977a: 65)

As a result, we are dealing with two contiguous (in the sense of contiguity in space) and interconnected frames that can be specified or non-specified in different respects. Figure 4 summarizes Fillmore’s example once more.

398. Just to name a few examples from von Polenz’s list of semantic roles: agent ≈ Who per-formed the action?; counteragent ≈ Who does the action target?; affected object ≈ Who or what is affected by the action or process?; cause ≈ What does the cause of x consist of?, etc.

399. Here we are dealing with the semantic role of “effected object” (product, result); cf. von Polenz (1985: 171).


What did you write on?

the surface

WRITE

Who wrote? the writer

When are you going to send it?

???

What did you write with?

the implement

What did you write?

What language are you writing in?

???

Who are you writing to?

???

What does what you wrote mean?

???

When do you think s/he will get it?

???

Do you think s/he will answer it?

???

LETTER

Figure 4. Slots and fillers/default values in the verb frame write and the subframe letter, according to Fillmore 1977a: 64–65

Once more, new questions derive from the answers to the questions in the letter frame. In addition, new accents set on particular aspects within the frame can lead to questions (as well as their corresponding answers) becoming more concrete. Consider an accentuation in the temporal dimension, so that the activity of writing the letter in the future and in the past could shift into the foreground:

How many earlier letters did you write?Who did you send those letters to? (Fillmore 1977a: 65)

Apart from the recurrent embedding potential of frames in frames, I am particularly interested in the function of questions. As the degree of abstraction of frames increases, the questions that can be meaningfully posed about the respective reference object become equally more abstract. The following also applies: The number of potential questions decreases in proportion to the increase in their degree of abstraction. In principle, therefore, it seems possible to structure the potential questions and organize them into groups in such a way that will provide a manageable set of questions or slots. I take up this thought again in Chapter 7.

In addition, because it is possible to understand all facets of knowledge (that is, explicit and implicit predications or, alternatively, fillers and default values) that are bundled in a frame as answers to implicit questions, the following applies, as formu-lated by Minsky:400

400. Cf. also Charniak (1976: 360).


Examinations of linguistic discourse leads… to a view of the frame concept in which the ‘terminals’ [‘slots’] serve to represent the questions most likely to arise in a situation. To make this important viewpoint more explicit, I will spell out this reinterpretation: A Frame is a collection of questions to be asked about a hypothetical situation; it specifies issues to be raised and methods to be used in dealing with them. (Minsky 1975: 246)

Admittedly, the term “situation” is rather vague here; after the previous considera-tions, the use of the term “reference object” (or alternatively “activated frame”) seems more appropriate. Nonetheless, Minsky opens up a perspective here that channels the further argumentation. The slots of frames provide the predication potential of a linguistic expression in the form of potential reference points that can be reformulated as questions. Should gaps in knowledge or coherence occur, questions serve as instru-ments of targeted knowledge acquisition (cf. Hintikka and Hintikka 1985; Konerding 1993: 139–217). Questions help to interpret knowledge; they control or regulate the construction of U-relevant knowledge.401 The question function is thus already inher-ent in the format of propositions (in the Searlean sense). This is because every time language takes reference to an entity, this entails a predication (an assertion-about), which however by no means sufficiently specifies the reference object: Every predi-cation provokes further predications. The reference objects of linguistic expressions are notoriously underspecified, yet language users can in principle enquire about the non-known, i.e. only postulated knowledge.

In this way, [the language user] bears the problem in mind and searches for ap-propriate solutions. Therefore, when an actant solves problems and manages tasks, this features the actant performing certain actions of enquiry that govern the search for a way to solve the problem. (Translated from Konerding 1993: 144)

But how do we determine the questions bundled in a frame? Klaus-Peter Konerding (1993) has developed a procedure for this, which I explain in the next section.

6.3.2 Hyperonym type reduction: determining slots

Since both Fillmore’s case theory and its extension by von Polenz have shown them-selves to proceed in a semantically reductionist manner, we now stand at the crossroads. One could argue that attempts to classify the potential reference points of linguistic

401. “Information that is not yet recognized is transposed into reality by means of the question whose answer it constitutes. In this sense, the process of activating tacit knowledge is controlled by questions that serve to transport this information into reality.” (Hintikka and Hintikka 1985: 235, translated from the quotation in Konerding 1993: 146)


expressions are fundamentally destined to fail.402 I do not share this fundamental scep-ticism, and briefly outline my reasons here.

Can arbitrary predicates be attributed to the reference object of a linguistic ex-pression? A fundamental sceptic would have to answer in the affirmative. However, this would entail significant consequences, among others, that one could no longer sufficiently differentiate between the metaphoric and the non-metaphoric usage of words. That is to say that it is only possible to identify a metaphor under the assumption that there is a qualitative restriction on routinized predicative attributions. Thus, for instance, the size of a material reference object can be specified by an infinite number of predicates;403 but nevertheless not all properties can be equally attributed to the ref-erence object. Let us examine Example (10) (which we encountered in Section 3.1.4), as well as two other propositions.

(10) The rock is pregnant. (11) The rock is man-size. (12) The rock is climbable.

The qualitative difference between the predication performed in (10) and the two oth-ers consists in the fact that in (10) a property is attributed to the reference object of the noun rock that it cannot have due to its material-inanimate nature. Accordingly, the fact that a rock cannot be pregnant depends closely on the assumption that a rock is a material-inanimate entity or – more generally speaking – on the language user’s preconditioned theory of everyday life.404 There may be a language community, for instance, a primitive Australasian people, that does not share this assumption and would therefore not interpret (10) as a metaphorical expression because they see rocks as, say, the incarnation of female fertility. In this language community, a theory of eve-ryday life would prevail that is reflected in terminology of this type in such a way that predications of type (10) would be the norm. Other predications, perhaps those in (11) and (12), would potentially then have metaphoric status. Without contextualizing it

402. I have already mentioned Croft (1991: 155ff.) and Langacker (1987: 284) as representatives of this position. However, their criticism is first and foremost directed at access to linguistic meaning being restricted to a valency grammar approach, and consequently one could repre-sent the view that they initially “only” wished to free themselves from these – still syntactically motivated – fetters.

403. For example: x is large/ small/ a metre wide/ 21.34m3 in volume, etc.

404. Admittedly, this assumption itself has the form of a predication. Accordingly, a theory of everyday life essentially consists of a set of propositions that are considered true in a language and cultural community.


within a preconditioned ontology of everyday life, Searle described this phenomenon in the following way:405

The object [that a linguistic expression refers to] must be of a type or category such that the predicate expression or its negation could be true or false of it. Correlative with the notion of any given predicate is the notion of a category or type of objects of which that predicate could be truly or falsely predicated. For example, correla-tive with the predicate “is red” is the notion of colored (or colorable) objects. “Is red” can only predicated of objects that are colored or colourable. We can truly or falsely predicate “red” of windows, but not of prime numbers. We might put this point by saying “is red” presupposes “is colored”… (Searle 1969: 126)

“Truly” in this context means as much as “meaningful” or “normal” in Konerding’s sense, i.e. “in conformity with usual language usage and the shared knowledge estab-lished in interaction forms in a language community” (translated from Konerding 1996: 80). Therefore, what can lay claim to being a “true” or “meaningful” predication within a language community is the yardstick for where the borderline runs between metaphoric and non-metaphoric usage. If there were only “true” predications, there would be no metaphors. In the same way as one would make statements about the size and the material properties of rocks, one could also then speak of vivacious or blonde rocks. Yet this not only contradicts language usage; clearly our intuitive knowledge of which predicates can typically be attributed to a reference object or not plays a signifi-cant role in language understanding overall (cf. Konerding 1993: 164).

However, how do we determine which predicates within a language community and at a particular time have the status of “true” predicates? Konerding (1993: 139–217) has developed a linguistically based procedure, motivated by lexicography and lexicol-ogy, to systematically generate the slots of a frame in the form of questions.406 This procedure is essentially based on so-called hyperonym type reduction, which I briefly outline in the following. Including the semiotic and schema-theoretical findings from the previous chapter will help to illustrate that hyperonym type reduction stands in complete accord with the holistic-cognitive semantic conception outlined so far.

The departure point is the schema-theoretical observation that hyponyms and hy-peronyms behave as instances do to schemata. Following Langacker, I spoke of a vertical

405. When Fillmore states at one point that within the frame evoked by the word letter it is possible to address some entities but not others, he is being led by a similar thought: “Notice that if, instead, I were to ask a question like What time is it? or make a comment like I have got a bad toothache, I would not be talking within the frame you introduced; I would be changing the subject.” (Fillmore 1977a: 64)

406. Konerding’s procedure can be considered as the only systematic and linguistically grounded method to date. Other studies have either dispensed altogether with a procedure to determine slots (as is generally the case in Anglo-American literature) or the category construction is inadequately grounded (as in the studies by Wegner 1984, 1985 and Müske cf. 1992: 131ff.).


semantic dimension, or in other words, of instantiation relations in Section 4.2.1. These now become interesting for the following reason:

The conceptual import of this relationship, I suggest, is that an instantiation is fully compatible with the specifications of its schema, but is characterized in finer detail … The schema [tree], for example, defines a category that is instantiated by a variety of more specific concepts, all of them compatible with its specifications ([oak], [maple], [elm], and so on). These instantiations elaborate the schema in different ways along various parameters, to yield more precisely articulated no-tions. (Langacker 1987: 68)

The fact that an instance (hyponym) is semantically fully compatible with the specifica-tions of the schema (hyperonyms) whose instance they constitute yet that this instance contains more detail means that hyponyms and hyperonyms essentially demonstrate the same reference points, but that their predicative attributions differ with respect to the density of information. Formulated in frame-theoretical terms: although hyponyms and hyperonyms evoke the same frame, they fill its slots with fillers and default values of different degrees of abstraction and thus of specification.407 Recall the following hi-erarchy of schema instantiations from Section 4.2.1: [glamis] → [golden_retriever] → [dog] → [quadruped] → [animal] → [living_creature]. If, for instance, we are able to semantically specify the expression animal with respect to its external form, liv-ing conditions, biological properties, and material characteristics in such a way that the reference object is sufficiently distinct from the meanings of other expressions (such as human), then as far as the same facet of knowledge is concerned, it is equally pos-sible to make a semantic distinction between the reference objects of the words cat or dog. However, in the latter case, the semantic specifications would be more particular (consider, for instance, the specific external form, the specific living conditions, etc. of a dog in contrast to a cat). Consequently, if we succeeded in finding a restricted set of highest hyperonyms, it would be possible to determine the potential reference points or slots of any linguistic expressions by retracing each of them to just such a highest-level hyperonym.

With this in mind, the idea of a hyperonym type reduction can now be outlined in detail. The highest hyperonyms constitute types in that their predication potential is bequeathed to hyponyms.408 Yet, how can we determine the highest hyperonyms? Konerding describes his procedure as follows:

407. Cf. also Fillmore (1982a: 132). And Charniak notes: “[S]lots are inherited via the isa hier-archy. That is, if dogs are mammals, and mammals have a slot for representing the head of the mammal, then so do dogs, by virtue of the fact that they are mammals.” (Charniak 1981: 288)

408. Langacker (1991a: 61–64) already postulated that there were so-called “type hierarchies”. These can be understood as hierarchies of semantic inheritance in the sense just described. However, Langacker does not appear to have recognized the potential that the concept of type hierarchies holds for a theory of meaning.


The types we are seeking are found in the trends in hyperonym type reduction that occur in dictionaries…. The hyperonyms typically occur as the main ele-ment (head) of nominal phrases; if a hyperonym is identified with reference to its reading – in other words, accounting for polysemy and homonymy – the next hyperonym is identified with reference to its reading via the lemma of the hy-peronym, etc. This continues until any further reduction becomes problematic. Those hyperonyms that tend to most frequently occur as the end elements of the reduction chain are recorded as candidates for the sought for types. (Translated from Konerding 1993: 173–174)

This is essentially carried out on the basis of the second edition of the Duden Deutsches Universalwörterbuch (Duden Universal German Dictionary) (1989).409 The hyperonym typology being generated therefore relates to current standard German usage.410

How do we recognize when we have reached the final element in a hyperonym reduction? According to Konerding, a reduction becomes problematic when hypero-nyms within a reduction chain refer to each other in a circular manner (Konerding 1993: 174). This is, for example, the case in the Duden Deutsches Universalwörterbuch (1989) when the word area (‘Bereich’) is explained by the hyperonym space (‘Raum’), which leads to expansion (‘Ausdehnung’), and from there back to area. Area is thus established as the final element in the reduction chain.

Konerding conducts a large number of such hyperonym reductions, working exclusively with nouns for two reasons. First, hyperonyms usually occur in dictionary meaning explanations as the main elements of nominal phrases (cf. previous quote). Secondly, elsewhere Konerding indicates that our nominal vocabulary serves to name extralinguistic entities (Konerding 1993: 162), and that “everything that is addressed and thus can be made the object of explicit categorical and predicative analysis is nominalizable (translated from Konerding 1997: 65). This circumstance stands in close correlation with the previously explained analytical distinction between “ref-erence” and “predication”: (metalinguistically) addressed nominals in syntactically unmarked sentences serve typically to identify the reference object and thus to ac-tivate a frame.

409. Case examples illustrating the outlined procedure can be found in Konerding (1993: 173 ff.); Fraas (1996a, b); Lönneker (2003a: 84 ff.); Klein and Meißner (1999).

410. Although Konerding does not expressly state it, it can be assumed that the results of the hyperonym reduction change historically, that is, that the predication potential of an expression is also diachronically variable. Meaning change based on language usage is reflected in a change in vocabulary structures, so that the end element in a hyperonym reduction chain can differ after a sufficient interval of time. And that means that at a particular point of time z1 something that cannot be conventionally predicated on the reference object of a particular expression may under certain circumstances at a point of time z2 belong to the set of standard predications. Cf. the lexicalization examples or “de-metaphorizations” in Keller and Kirschbaum (2003: e.g. 55–56).


Table 1. End elements in a hyperonym reduction,411 according to and translated from Konerding 1993: 175–176

– Duration/time interval/period of time Location/place/situation/region/area/ field/sector (within a space, terrain)– Something Mass/substance/material Object/thing Apparatus/construction/device– Animal/plant/organism/living creature– Person/human/someone/creature– Process/event– Action/act/procedure Communication– Institution Form of human coexistence/alliance/ network/association/group/movement– Teaching/system of theories/thoughts/belief system/cognizance/knowledge– Part of/piece of/segment of/(localizable) area– Totality/stock/set/whole/unit/collation/combination/grouping– State/texture/behaviour/ability/method/propensity/inclination/manner

The result of Konerding’s hyperonym type reduction is impressive. Only a relatively small set of German nouns occur as the end elements in the reduction chain. Grouping these according to semantic similarity results in the end elements that are collated (and translated into English) in Table 1 (Konerding 1993: 175–176).412

On the basis of these empirical findings, Konerding collates various nouns into so-called “noun-types”. The result is a list of semantically similar groups, which he subdivides into primary and secondary types. Secondary types are distinct from the primary by dint of the fact that they establish meronymic relations. The noun types correspond to the highest hyperonyms, in other words, those nominals to which it is claimed that all other nominalizable words can be hyperonymically retraced. Because the resultant noun typology simultaneously serves as a typology for the (matrix) frames to be generated, it is reproduced here (in translation) (cf. Table 2).

411. Konerding’s list does not indicate, for instance, how hyperonyms like location, place, etc. i.e. expressions relating to spatial values, can be subsumed under the first end element ‘duration/time interval/period of time’, i.e. a temporal dimension. Such subsumptions are at least indicated by indentation of the second line.

412. Here I have omitted smaller interim steps, e.g. a summary of the hyperonyms of noun types that were identified; c.f. Konerding (1993: 176ff.). This provides numerous detailed explanations on the quantitative procedure, which are not relevant for my further line of argumentation.


Table 2. Typology of nouns (basis for later matrix frames), according to Konerding 1993: 178 (in translation)

Primary types: – Object (concrete object), sub-classified as: Natural mode (continuous form/discrete form) Artefact (continuous form/discontinuous form)– Organism – Person/actant– Event– Action/interaction/communication– Institution/social group– (Part of the) surroundings (of humans)

Secondary types– Part/piece (of)– Totality/stock/set/whole (of)– State/property (of)

What relevance does this typology have for our frame-semantic context? Assuming that, firstly, the predication potential of a linguistic expression (i.e. the slots in the frame that this expression evokes) is inherited by the hyperonym expressions and that, secondly, the nouns in Table 2 represent the highest irreducible hyperonyms, this results in an important consequence for frame theory: The slots in the frame that any word evokes correspond to the slots in the frame of that noun type (listed in Table 2) to which the expression can be retraced via hyperonym reduction.

The crucial second step thus consists in determining the potential reference points of the highest hyperonyms or noun types listed in Table 2. As explained above, by “ref-erence points” I understand facets of knowledge that can be meaningfully attributed to a reference object (“truly” in the Searlean sense). Subsequently, conventionally expect-able routinized predications arise as predicators, so the central task lies in finding out which predicators the ascertained noun types can be linked to. To ascertain the relevant predicator schemata, Konerding recurs to category constructions in the verb-semantic study by Ballmer and Brennstuhl (1986).413 In this study, verb schemata function as predicator schemata. The relevant predicator schema for each of the noun types in question is chosen according to polar questions of the type “Can the verb (or verb schema) Y in category Z be meaningfully predicated by a typical reference object of the noun X…?” (translated from Konerding 1993: 165). Verbal predicators in standard German are thus examined to determine which reference object of the nouns in Table 2 they may be able to predicate. Konerding collates the resultant predicator schemata in synonym groupings and weights them according to relevance. The final step consists in transforming the valency slots in the verb schemata, i.e. arguments and adjuncts,

413. Cf. the detailed outline of this in Konerding (1993: 149–173).


into questions.414 Once this has occurred, the questions are grouped thematically and arranged in (a non-obligatory) order (Konerding 1993: 190ff.).

Each final element in the hyperonym reduction (Table 2) can in this way be as-signed lists of questions through which so-called “matrix frames” emerge, which are “functions of a higher type… that can only be transferred into concrete frames by the substitution of nouns of the admissible type” (translated from Konerding 1993: 200). That means that if one reduces any arbitrary noun to its highest hyperonym by means of hyperonym reduction, one automatically extracts the slots of the frame associated with the noun, since these slots correspond to the slots or reference points of the high-est hyperonym. Potential slots in the noun frame correspond to meaningful questions that can be posed about a reference object. The predication potential of a linguistic expression that I spoke of before thus takes the form of strategically meaningful ques-tions that govern the activation of U-relevant knowledge. As Fraas (1996a: 26) puts it, frames serve as the grid for transforming knowledge into text (‘Vertextung’), as the “contextualization potential” of an expression.

Before illustrating this procedure, I would like to say a word about the matrix frames identified. As described, their number is a result of the application of the hy-peronym type reduction. According to Table 2, there should be a total of eleven ma-trix frames. However, empirical application has demonstrated that it is meaningful to make minor adjustments.415 In addition, Konerding (1993: 185) proposes refining the hyperonym type ‘person’. Apart from ‘person with temporary or permanent prop-erty/disposition’ he also defines the hyperonym type ‘person in a professional role’. In contrast, it made sense to integrate the hyperonym type ‘surroundings’ into the type ‘totality/set (of)’. Finally, the noun type ‘object’ was subdivided into three (instead of two as in Table 2) subtypes, so that ultimately twelve matrix frames remain.416 In

414. A somewhat more detailed summary of the procedure can be found in Fraas (1996a: 16ff.).

415. Cf. Lönneker 2003a: 86ff.; Konerding 1993: 181ff.. Konerding addresses issues that I am unable to elaborate on in this context. In particular, he justifies employing the matrix frame action. The fact that person constitutes its own matrix frame (and does not merge with the matrix frame organism, as one might assume) is apparently due to the importance of the person concept; on the basis of lexicon entries it cannot be hyperonymically further reduced. To the best of my knowledge, Konerding and Lönneker do not address this point.

416. Nine of the matrix frames are “primary types”, including: (1) object I: natural objects; (2) object II: artefact; (3) object III: substance; (4) organism; (5) person I: with tem-porary or permanent properties or disposition; (6) person II: in a professional role; (7) event; (8) action/interaction/communication; (9) institution/social group. The remaining three matrix frames are to be assigned to the “secondary types”: (19) part/piece of; (11) totality/stock/set (of); (12) state/property (of). What remains open in this typol-ogy is the extent to which in (9) “group” should not be understood as a plurality of persons that stand in a specific relation to one another. It is important to recall that this typology reflects a type of everyday ontology as codified in lexica. Thus, it is not evident here that, for instance, in physics “objects” are apparently understood as “states”.


the meantime, they have proven themselves empirically in Lönneker’s comprehensive corpus-based analysis.417

Nonetheless, Lönneker rightly points out that not all matrix frames are as equiva-lent as Konerding’s analysis suggests. It has been shown

… that there are some subslots [i.e. slots/sub-questions] that occur in all frames and many that are in several frames. This makes it possible to place the frames in a hierarchy, so that the greatest possible similarity in structure is guaranteed or, alternatively, so that redundancy (in the sense of duplication of frame elements) is minimized. In the hierarchy, a subframe ‘inherits’ all the information of a su-perframe or (recursively) all of its superframes… (Translated from Lönneker 2003a: 89)

This kind of frame hierarchy takes account of the fact that matrix frames do not rep-resent autonomous, self-contained units, but rather are linked with each other, or follow from each other, in a variety of ways because they share common slots. What is particularly important here are the vertical, i.e. the hierarchical, relations, since the hi-erarchically higher frames pass on their slots to hierarchically lower frames. According to Lönneker’s analysis, the frame hierarchy is structured in five levels (cf. Figure 5). The unidirectional arrows in Figure 5 indicate the heredity relations.

EventObjectof a

natural kind

Continuativum/

material

Institution/social role

Action

Whole StateArte-fact

Organism

Person in a role

Person in a

profession

Unchangingobject

Primary object

Role/perspectiveon an entity

Part

Entity

Mat

rix

fram

es

Figure 5. Translation of the frame hierarchy according to Lönneker 2003a: 93 and 2005: 134

Lönneker’s work (2003a), as well as the individual case studies by Fraas (1996a, b) and Klein and Meißner (1999), illustrate that Konerding’s procedure for determin-ing slots can easily by applied for corpus-based analysis. Nevertheless, only so-called

417. Unfortunately, the fact that the matrix frames are erroneously numbered in the typologies of both Konerding (1993: 178) and Lönneker (2003a: 86) has led to some confusion.


“autosemantica” have been the focus of empirical observation to date. This is because Konerding restricts hyperonym reduction to verbs, adjectives, and nouns, thereby excluding function words.418 Nonetheless, he notes that words of every word class can be nominalised, even if it often entails a loss of meaning.

For the following… I make the additional assumption that those non-nominal predicators that theoretically cannot be transformed into nominals due to mor-phological or syntactic rules can in any case be approximated by being para-phrased in nominal form. (Translated from Konerding 1993: 203)

The fact that function words have not been the object of corpus-based frame studies to date is not because it is impossible to nominalize them to conduct a hyperonym reduc-tion. As outlined previously, function words can be easily nominalized by addressing them metalinguistically. But precisely this seems to be the problem: In the dictionary meaning descriptions, there are very few nominal hyperonyms that explain function words, but rather almost exclusively metalinguistic references.419 It is thus highly dif-ficult or even impossible to perform a hyperonym reduction. This does not mean that function words have no meaning, but rather that additional empirical methods, such as co-occurrence analysis (cf. Fraas 2001; Heringer 1999), need to be employed to analyse their meanings and functions in discourse.

However, function words can be studied using frame analysis (in the sense pre-sented here).420 However, frame analysis cannot make use of hyperonym reduction and as a result a comprehensive list of slots (in the form of matrix frames) is not available to it. Nevertheless, it can, for instance, lead to results that provide insight into the status and semantic unit (in Langacker’s sense) of a function word. Section 6.5.3 demonstrates the use of just such a reduced frame analysis using the particle probably.

418. Konerding (1996: 80). Here, Konerding understands nominalization in Langacker’s sense (1991b: 97ff.). However, Langacker himself only addresses the nominalization of verbs, and does not state any reason why words of other word classes should not also be nominalizable. In this sense, Konerding adds at another point that, apart from verbs and adjectives, preposi-tional groups, syntagmas functioning as phrases or clauses, and entire texts can be nominalized, so that anything that is addressable in terms of syntax and function can also be nominalised (Konerding 1997: 64).

419. Cf., for instance, the metalinguistic meaning description of perhaps (‘vielleicht’) as trans-lated here from the Duden Deutsches Universalwörterbuch (1989): “…qualifies the certainty of a statement … qualifies the accuracy of the following measurement”. Most of the function words appear to be explained metalinguistically, including the conjunction or (‘oder’), which I men-tioned previously: “… expresses that of two or more statements only one of them is possible.”

420. The most difficult group in the category collectively known as “function words” is that of interjections. They can neither be subjected to hyperonym reduction nor is it possible to gain insight into their conceptual structure via co-occurrence analysis. However, it is doubtful whether this is reason enough to deny interjections any meaning substance. This is evidently more of a methodological and analytical problem.


6.3.3 Example analysis

In the following, a hyperonym type reduction will be performed on the word locust to illustrate the procedure for ascertaining the slots in a frame, as well as the relevant slots in the corresponding matrix frame. In the final chapter, I will return to the results once again within a corpus-based analysis.

For reasons of consistency, the Duden Deutsches Universalwörterbuch (Duden Universal German Dictionary) (1989) is the basis for the reduction in the following. Following the proposed procedure, I search for hyperonyms until a circular reference structure emerges. The results are summarized in Table 3.

Table 3. Translation of the hyperonym type reduction based on the Duden Deutsches Universalwörterbuch (1989, using the example locust)421

First reduction:locust: … plant-eating insect with membranous wings… → Hyperonym: insectSecond reduction:Insect: … animal belonging to the arthropods with a rigid exoskeleton surrounding its body…

→ Hyperonyms: arthropods,421 animal

Third reduction:Animal: …a living creature equipped with sensory and respiratory organs that subsists on other animal or plant organisms and is generally capable of moving freely…

→ Hyperonyms: organism, living creature

From here: circular reduction:Living creature:… organic living creature, in particular animal or human; organism: single-celled animal, plant life …

→ Hyperonyms: creature, organism

Creature: human (as creation, living creature)… → quasi-synonyms: creation, living creature

Organism: … entire system of organs: human, animal, plant organism…, …living creature, animal or plant life form…

→ Hyperonyms: living creature

The end element of the hyperonym reduction is thus the noun organism (or living creature). What matrix frame is associated with this irreducible hyperonym? Although unable to enter into a full description of Konerding’s procedure for ascertaining rel-evant slots, I would like to mention an essential interim step.

Initially the transformation of the valency slots (of those verb schemata that Konerding adopts from Ballmer and Brennstuhl 1986) results in a very comprehen-sive list of questions, which have been translated and listed in full in following Table 4.

421. The meaning description of arthropod (‘Gliederfüßer’) refers to animal (‘Tier’) (“innumer-able species of invertebrate animal…”) so that I only follow animal as a hyperonym.


Table 4. Slots in the locust frame, ascertained via the matrix frame organism, translated according to Konerding 1993: 316–321 (with minor adjustments)

– Properties How is the (external) form, configuration of the organism to be described? What does the organism look like? – under what conditions – in what phase of its existence – how long – for what reason What perceptible characteristics does the organism have? – under what conditions – in what phase of its existence – how long – for what reason What measurements does the organism have? – in what phase of its existence – how long What particular habits does the organism have? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason What determines the behaviour of the organism? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason What physical and biological characteristics does the organism have? – under what conditions – in what phase of its existence – how long – for what reason What particular properties and characteristics does the organism have? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason


– Constitutive relations Is the organism an essential constituent of a superordinate whole? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason In what functional contexts, natural processes (events, actions) does the organism play

an important role? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason What does the organism function as (in these contexts)? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason What determines the composition of the organism (body parts, organs, limbs, etc.)? – in what phase of its existence – how long – in what manner – with what consequences – for what reason– Phases of existence and distribution What (functional context) is the source of the organism? In what way does the organism emerge/appear? – under what conditions – in what manner – for what reason Under what conditions does the organism cease to exist? – how: manner (of disappearance) – for what reason Under what particular conditions does the organism cease to exist (at all)? – for what reason In what way can the organism be destroyed? – under what conditions – in what phase of its existence – for what reason What typical phases of existence does the organism undergo?

Table 4. (continued)


What conditions underlie these phases of existence? How long do the phases of existence last? At what typical places can the organism be found? – in what phase of its existence – how long – in what manner – with what consequences – for what reason What (geographical, social) distribution does the organism have? – under what conditions – in what phase of its existence – with what consequences – for what reason In what way is the organism distributed or propagated? Under what conditions does the organism normally exist? – in what phase of its existence – how long – in what manner – with what consequences – for what reason What names does the organism have? How widely known is the organism? In what (contexts of) actions contexts does the organism play a particular role? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason What does the organism function as (in these contexts)? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason What relevance does the organism have for humans? What specific use does the organism have for humans? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason



In what human work and production processes does the organism play a role? – under what conditions – in what phase of its existence – how long – in what manner – with what consequences – for what reason What organisms of another type does the organism resemble and how is it distinct

from them? In what important categories does the organism occur? How is the organism classified (contiguity concepts, etc.)? – in what phase of its existence – how long – in what manner – with what consequences – for what reason Do any special theories or more detailed (encyclopaedic) descriptions about

the organism exist? In what important theories does the organism play a role? What does the organism announce or give witness to? – under what conditions – in what phase of its existence – in what manner – for what reason

To begin with, this list of potential questions and sub-questions seems rather unsys-tematic. What is more, it exhibits a number of redundancies. And due to its length, it is of little use for the purposes of empirical analysis (cf. Klein 1999: 160). In addition, the sub-questions collated under the dashes are strictly speaking not associated with the matrix frame itself, since they propose potential ways of specifying the answers (i.e. fillers or default values) to the respective superordinate questions.

However, Konerding further systematized the sub-questions presented in Table 4 by combining thematically similar questions to form classes of superordinate predi-cators. From the over 100 questions, only 24 predicator classes remained. They are summarized in Table 5 and sorted thematically into groups.

In Chapter 7, I show that this compact number of predicator classes is empirically easy manageable, yet at the same time large enough to arrive at differentiated analytical results. However, the corpus-based analysis conducted in Chapter 7 does not use the predicator classes listed in Table 5, but rather those that Lönneker (2003a: 262–277) ex-tracted from the results of her comprehensive empirical study.422 Although Lönneker’s

422. Cf. Section 7.3.1.



list corresponds to Konerding’s in large sections, in some places it has been shown that a posited predicator class is superfluous or that a differentiation seems necessary. In principle, it is possible to obtain a more refined “granulation” by applying the sub-questions from Table 5 alongside the predicator classes.

Table 5. Predicator classes in the matrix frame organism (cf. Konerding 1993: 411–417)

– Properties

Predicators to characterize form and colourPredicators to characterize additional perceivable properties of the organismPredicators to characterize essential parts of the organismPredicators to characterize the measurements of the organismPredicators to characterize the usefulness of the organism’s properties for humansPredicators to characterize the abilities of the organismPredicators to characterize the habits or particular behaviour of the organismPredicators to characterize additional special properties of the organism

– Constitutive relations

Predicators to characterize the superordinate whole in which the organism functions as a componentPredicators to characterize natural events in which the organism functionsPredicators to characterize the roles in which the organism is identified within these events

– Phases of existence and spread

Predicators to characterize the circumstances of emergence of the organismPredicators to characterize the natural surroundings of the organismPredicators to characterize the spread of the organismPredicators to characterize the conditions of availability/existence of the organismPredicators to characterize the particular phases of existence of the organism

– Relevance of the organism for humans

Additional names for the organismPredicators to characterize the actions of humans in which the organism plays a rolePredicators to characterize the roles in which the organism is identified within these actionsPredicators to characterize the meaning of the organism for humansPredicators to characterize the production processes of humans in which the organism or parts or products of the organism play a rolePredicators to characterize similar organisms and its distinction to these, as well as to characterize general categories in which the organism is categorizedPredicators to characterize the theories in which the organism plays a rolePredicators to characterize information conveyed about the organism

The result of the example analysis is thus: The word locust can be retraced via hypero-nym reduction to the noun type ‘organism’, whose matrix frame is characterized by all those predicators summarized in Table 5. These predicators function as the slots in the frame that the word locust evokes. In other words, the sum of the listed predicators constitutes the predication potential of the expression locust.


6.4 Explicit predications: fillers

Matrix frames can be understood as schematic units par excellence. Certainly, in Konerding’s conception, matrix frames primarily function as an analytical instrument, thus not having the status of psychically real units. Nevertheless, apart from that, they share all those properties of cognitive schemata that I outlined in Section 5.2.2 follow-ing the schema theories of cognitive psychology. Rumelhart compares matrix frames with cognitive schemata in his conception of cognition:

A schema is basically a theory about knowledge. It is a theory about how knowl-edge is represented and about how that representation facilitates the use of the knowledge in particular ways. According to schema theories, all knowledge is packaged into units. These units are the schemata. … A schema contains, as part of its specification, the network of interrelations that is believed to normally hold among the constituents of the concept in question. A schema theory embodies a prototype theory of meaning. That is, inasmuch as a schema underlying a concept stored in memo ry corresponds to the meaning of that concept, meanings are en-coded in terms of the typical or normal situations or events that instantiate that concept. (Rumelhart 1980a: 34)

Matrix frames function as an epistemological instrument for empirically determining the U-relevant knowledge that is available in conceptual units. When turning to con-crete linguistic meanings (which can be determined using matrix frames) a prototype-theoretical approach – such as Rumelhart proposes here – does indeed prove to be imperative. In fact, language users only infer schema- or frame-adequate knowledge because this knowledge can be typically expected, i.e. it is conventionalized in such a way that it is generally predicatively attributed to the cognitive reference object in question. This occurs via linguistic categorizations, as dealt with at the beginning of this chapter.

In several points in this book I have already shown that it is necessary to differ-entiate more precisely between two types of knowledge in this process. Most recently, following von Polenz, I spoke of explicit and implicit predications, and previously spoke in a similar manner of fillers on the one hand and default values on the other. Before I go into fillers (or explicit predications) in more detail, I would like say a word on the distinction between fillers and default values from the perspective of inference theory.423

In Chapter 4, I criticized the fact that inference theories often posit an area of non-inferential knowledge. From a semiotic perspective this postulate proves to be extremely problematic in the following considerations. Since semiotic constitutions al-ready occur inferentially themselves, there is no “textually given information”.424 Thus,

423. Cf. also the observations made by Müske (1992: 120–121).

424. Similarly, Busse states (translated from 1991a: 82) that “one [should] not depart from a ‘given’ set of propositions, but rather the assumption that in each particular communicative situation particular respective knowledge segments are currently activated”.


on the content side, referentializations represent the language user’s own constructive mental operations: a reference object is not simply there, but rather the reference to a cognitive unit (a frame) must first be established. Nonetheless there is a qualitative difference between two types of inferences. Once a word has evoked a frame, given text elements can specify reference points. The inferential operations that participate in this have been unjustly neglected by psycholinguistic inference theory. In addition, other reference points play just as much of a U-relevant role without there being text elements to which they can be predicatively ascribed. Only those knowledge elements known here as default values form the focus of psycholinguistic inference research.

In the following, I focus on fillers to begin with. In so doing, the question of their cognitive and epistemic status remains in the foreground. We see that different linguis-tic variations can be distinguished that fulfil all the cognitive functions of perspectiv-izing U-relevant knowledge.

6.4.1 When are predications explicit?

What criterion allows explicit predications to be distinguished from implicit ones? Let us call to mind Example (9), which was discussed in the context of Von Polenz’s sentence semantics. Once again:

(9) Der Politologe, der in der Zeit des Nationalsozialismus mit seiner jüdischen Familie nach Frankreich floh

‘The political scientist who fled with his Jewish family to France in the period of National Socialism’

Clearly, in (9) the event of the flight is more precisely specified with respect to the person who flees, among other things. By providing an answer to the question ‘Who fled at time X with Y?’ this slot is occupied by the text element the political scientist.425 The status of this filler can be said to be a predicative attribution, and since predications always occur in the format of propositions, i.e. as an interaction between reference and predication, the filler is more precisely: ‘It is the political scientist who fled.’426 The pro-cess is similar for the text elements (i) in the period of National Socialism, (ii) with his Jewish family, and (iii) to France. What are qualified here are (i) the temporal dimension

425. Or more precisely, the text element is an answer to the question of who fled, and in fact when and with whom. Compare this with the matrix frame event in Konerding (1993: 335–340).

426. In this context fled (in its nominalized form: flight) evokes a frame whose agent slot is oc-cupied by the filler the ‘political scientist’, so that the following applies: ‘It is the political scientist who fled.’ Observe that, like von Polenz, I treat Example (9) as an isolated nominal phrase. Of course, the determinative pronoun the in the political scientist indicates the presupposition that the person is known in the text universe (for which reason the political scientist is in thematic position). However, if we depart solely from the information available in (9), only the predication named can be given as the filler for the agent slot.


and thus the political circumstances of the flight, (ii) the participant actors, and fi-nally (iii) the destination. When (i)–(iii) particularize the flight frame in a variety of ways, they simultaneously indicate co- and contextualization relations, since every filler then evokes another frame, which in turn contains a range of default values. In this context, Busse speaks of “overt” contextualizations, which can be motivated in a number of ways:

… consistent explicit references by means of headings, subject information, key concepts, explicit assertions, (intended) implications, expressly overt contextual-izations, etc. (Translated from Busse 2000a: 44)

Consequently, overt contextualizations are based on particular fillers (in the sense explained above). We could also say that overt contextualizations always occur when particular fillers evoke a frame. By contrast, non-overt contextualizations only occur under the condition that language users invoke a frame from memory. Such forms of contextualization seem largely identical with another contextualization type that Busse mentions, namely “non-intended, unconscious, and only analytically ascertain-able contextualizations” (translated from Busse 2000a: 44).427 I return to these in the context of default values.

The previous sections already showed that fillers represent predicates (in the pred-icate-logical sense) (von Polenz 1985: 91) and cannot be simply equated with nominal units or phrases, as many frame-semantic studies suggest.428 Attributions are epistemic ascriptions, and these are made possible by verbs. The question of the status of fillers is therefore also a question of the status of verbs. What role do verbs play within the frame-semantic conception proposed here? Are verbs fillers?

Lönneker argues for positing an additional frame element apart from slots and instances (fillers, default values) that is, as it were, an offshoot of a slot without, how-ever, being identical with it. She introduces the concept of “subslot” for this (Lönneker 2003a: 65–74), and attempts to use it to account for the structural idiosyncrasies of

427. Cf. also Barsalou (1982). Busse (2000a: 44) also names a third type: non-intended, but conscious (i.e. as consciously postulated) contextualizations. Among these he includes topicaliza-tions as well as discourse- and language-reflexively articulated contextualizations. This third type corresponds to the overt contextualization type in that knowledge addressed through this gener-ally becomes, or can become, consciously known to the language user. For example, the filler (i) in the period of National Socialism is not understood solely as a specification of the time, but also of the political circumstances. Due to language users (more or less) consciously employing such contextualizations, I see this third type as being governed by explicit predications or fillers.

428. Thus, almost all Anglo-American frame studies suggest that frame elements – whether in the form of fillers of default values – are identical with words (cf. for example Fillmore 1975; Petruck 1996). The same applies for all the studies I am aware of on mental spaces (cf. for ex-ample Fauconnier 1985; Fauconnier and Turner 2002; Sweeter 1999). Yet, how can words more closely specify a reference object? Words must appear in the form of predicates; these predicates are ascribed to an object.


predications. Predications describe a tripartite relational structure in which an at-tributive relation z is established between a reference object x and a conceptual entity y. While the reference object and the attributed entity themselves have no conceptual status, according to Lönneker, a subslot establishes the link between the two. In natural languages this function is adopted by verbs.429

Yet what justifies the assumption of a second type of slot? The term “slot” or “sub-slot” implies that we are dealing here with (sub-)types according to which verbs can be categorized in the same way as predications belonging to particular slots in a matrix frame. At one point Lönneker makes it clear that subslots and slots certainly do share characteristics in this respect. In fact, she describes subslots as:

…labeled relations from the concept frame to another (class) concept or to an individual (“instance”). The main component of a subslot is a verb; however, the relations expressed by subslots are conceptual rather than lexical. Examples of sub-slots are + server à (“+ be used for”), + causer (“+ cause”). (Lönneker 2003b: 2)

Subslots thus provide options for combining different concepts (i.e. evoked frames) with each other. If a verb particularizes a relation, then the “linking element” must also be seen as a conceptual unit, because verbs then represent fillers on the level of instances, and as we have seen, every knowledge element – i.e. every filler and every default value – itself evokes a frame in its turn.430

One objection remains: Subslots are strictly speaking not types or classes. Slots are potential sub-questions, whereas subslots are always of a concrete nature, namely individual verbs and not verb classes or types.431 Consequently, subslots and slots cannot be as categorically separated from each other as slots and fillers. Rather, the former constitute a unit, which is that of predication. This gives subslots a status that largely corresponds to that of fillers. For this reason, I henceforth dispense with the terminological distinction between subslot and filler. This distinction also adds no value to the corpus-based analysis to be carried out.

429. Cf. Langacker (1991b: 83); (1999b: 206–212). Langacker observes that verbs designate (or semantically profile) relationships or processes), whereas nouns profile things (as conceptual figurations) in the broadest sense. In Langacker (2006: 138) he adds: “A present participle profiles some internal portion of a bounded process…”, “[a] past participle views a process holistically and with some kind of terminal prominence…”, “[a]n adjective profiles a non-processual relationship with a thing as trajector and no focused nominal landmark…”, and “[a] preposition profiles a non-processual relationship with a thing as landmark”.

430. Thus verbs integrate frames in each other; from a non-cognitive perspective cf. also Heringer (1999: 185).

431. Cf. Lönneker (2003a). To create a concept or frame hierarchy, Lönneker departs from Konerding’s matrix frames and, on the basis of her empirical data from the matrix-frame slots (which are formulated there as questions), identifies verbs with which the questions can be an-swered. Thus, predications that refer to the slot ‘What relevance does the entity have for humans?’ contain verbs like use for, produce.


However, this does not mean that the category of subslot is completely useless. In preparation for the empirical analysis it highlights phenomena that the more general category of predication may have obscured. Thus, Lönneker (2003a: 66–74) points out that, for one thing, the relation the verb establishes between two concepts cannot be negated. Propositions expressed in such sentence are then denied to the reference object. For another, the same verbs can sometimes be combined with different preposi-tions. In both cases information is conveyed that should be considered in the analysis since this affects which aspects of meaning (predications) are attributed to a reference object. Lönneker sees this as a particular characteristic of subslots.432

Important to note is that fillers have the status of predications consisting of verb and – depending on the verb valency – noun phrase(s). Further, every filler contributes to a specific perspectivation of the evoked reference object.

6.4.2 Linguistic manifestations

How do fillers occur in texts? What different linguistic variations can be distinguished from one another? We have already encountered one potential variation of fillers. It corresponds to the predication performed in the (highly simplified) basic propositional structure x is y or, more generally, x+ finite verb (+ object), whereby the object details are, of course, not obligatory but depend on the valency of the verb.433 Before speaking of different variations, I explain to what extent fillers can contribute to the cognitive perspectivation of the reference object.

Every predication is accompanied by a specific epistemic perspectivation of the reference object. Consider the following examples:

(13) The light is red. (14) The window is red.

To begin with, the assumptions concerning the material, functional, and other proper-ties of the respective reference objects must already be actualized so that the property of being red can be attributed. If, for example, I did not know that (traffic) lights were lighting devices for regulating the traffic, there would be no reason not to conceptualize the being-red as in a reading of (14) as a permanent material colour property.434 This does not occur only because I dispose over appropriate world knowledge.

432. Following up on this finding, I account for both aspects in the corpus-based analysis in Chapter 7.

433. A filler also occurs when a verb does not require an object (or an object is not syntactically realized). Peter is sleeping, for instance, corresponds to the predication ‘Peter is a sleeping entity’.

434. There is a reading of (14) in which being-red is not conceptualized as a permanent material colour property. This would be, for instance, in the case of a red light being on in a room, which made the window look red (metonymically speaking) from the outside.


However, in the course of the predications performed in (13) and (14) such as-sumptions about the nature of lights and windows shift into the background. Thus, in (13) only a specific function of lights is focused on, that of being able to show red light; assumptions such as those about the size of the light, its material properties, or where it is typically located remain in the background. Similarly, in (14) the property of being-red is attributed to a part of the reference object, either the window frame or the window pane.

Thus, generally speaking, fillers

…refer to a form of representation by which the parts of an object or the elements of a complex state of affairs and their interrelations are construed and presented as if seen from a given point of view. (Graumann 2002: 25)

With the help of explicit predications, language users profile certain facets of knowl-edge about a reference object whereas others shift into the background. Every predica-tion perspectivizes the reference object in a particular manner. It is impossible to use linguistic signs without taking a refracted perspective of reference objects.435 This is because, on the one hand, it is not possible to understand signs without predication, and on the other, every predicative act specifies only one knowledge facet of the refer-ence object, even though the set of possible predications is potentially inexhaustible.

In Cognitive Grammar, Langacker addresses this phenomenon in connection with so-called “active zones”.436

Entities are often multifaceted, only certain facets being able to interact with a particular domain or play a direct role in a particular relationship. Those facets of an entity capable of interacting directly with a given domain or relation are referred to as the active zone of the entity with respect to the domain or relation in question. (Langacker 1987: 272–273)

However, what facets of knowledge belong to an “active zone” in the individual case is not merely a question of the ability for interaction between two domains (in this case between [light] or [window] and [red]). The property “red” could by all means be ascribed to the material of the light or the material of the window pane, and yet we interpret (13) otherwise. The conceptualization we arrive at does not depend first and foremost on the “interaction ability” of two domains, but rather on the degree of sali-ence of particular assumptions (Giora 1997). Nevertheless, an “active zone” remains an overt contextualization relationship in Busse’s sense.

435. Hartung (1997: 20) comes to a similar conclusion, as translated here: “Perspectivity, as an undeniable circumstance, causes that which is represented in texts to have a particular, indi-vidual, personal customization. The same applies to the mental constructs that emerge in recep-tion, or that which is being communicated (the Kommunikate). Cf. also Graumann (2002: 33).

436. Cf. also Langacker (1984). On “perspective” as a technical term in cognitive grammar, cf. Langacker (1991b: 315–318).


Turning to potential manifestations of explicit predications, apart from the basic propositional structure x is y, predications often appear in texts that are of an overt nature, but are structurally less explicit. Henceforth, I speak of quasi-explicit predica-tions if a reference object is more closely specified in a particular way, either by (i) an attribute (in the broad sense), (ii) a prepositional phrase, or (iii) a relative clause.437 All three variations of quasi-explicit predications can be easily transformed into explicit predications, and we will see that from a frame-semantic viewpoint they are structur-ally very similar.

Consider, first of all, a variant of (i), namely a complex word form whose head is more closely specified by at least one morpheme. Of the three compound types that are generally identified in terms of morphology (determinative, root/copulative, and juxtaposed compounds), only determinative compounds have a quasi-predicative structure.438 In their case, if the head of the determinative compound (determinatum) evokes a frame, then the dependent morpheme (determinans) specifies a slot in this frame. The determinans thus functions as a filler.439 It aspectualizes the contextual-ization potential of the head on the one hand and indicates a specific knowledge or contextualization context on the other (cf. Fraas 2001).

This can be illustrated using the example Finanzinvestor (financial investor), which I return to in Chapter 7: The head -investor specifies the reference object, whereas the determinans Finanz- determines what type of investor is being referred to. This occurs in the form of a predicative attribution, leading to the frame at hand becoming more specific, i.e. its predication’s scope becoming more restricted. In general terms, the more instances a frame contains, the more reduced its predication potential is.

What applies to the determinans in determinative compounds can be transferred to derivative word formations (cf. Bybee 1995). For instance, affixations (like drinkable, stupidity) have the character that their word base is semantically or grammatically and functionally more precisely specified. The base, in this case [drink-] and [stupid],

437. Cf. also Lönneker (2003a: 74–75). I consider appositions, juxtapositions, and predicative nominatives (or predicative accusatives), none of which Lönneker lists, to be special cases of (i). Thus, these can be transformed into explicit predications in the same way. Thus, for instance, That idiot Peter was late again can be transformed into ‘Peter is an idiot’ and That idiotic man Peter was late into ‘Peter is idiotic’.

438. Juxtaposed compounds (like headstand) and root (or copulative) compounds (like the red-white-and-blue) do not have a head. In the case of root compounds, the morphemes are “equiva-lent”, so that predicative attributions are missing (red does not specify white, etc.). It follows that every element has the status of a frame-evoking expression. The elements of a root compound constitute a conceptual unit solely because they form instances of the same superordinate frame (for instance, in our context with [red], [white], and [blue] as instances of [flag]).

439. Contaminations are a special case. Although these contain a predication, it is so covert (i.e. it so strongly references foreknowledge) that it is difficult to speak of quasi-explicit predica-tions any longer. For instance hidden behind the term dictocracy is the covert predication (this) democracy is (like) a dictatorship.


adopt a referential function in the sense explained above. They evoke a frame, which is then specified by an affix, thus creating complex semantic units – in this case, complex words – of the nature [[drink]/[able]] and [[stupid]/[ity]].

For the sake of completeness, it must be said that, similar to compounds, flex-ional affixes are to be equally understood as instances of a frame that the respective word stem has evoked.440 Thus, in the previously mentioned case of the semantic unit [[dog]/[-s], its morphological ambiguity emerges because the flexional form [-s] can form an instance in different schemata. Flexional and derivational morphemes can-not activate a frame. They occur exclusively as instances of a frame, since they fulfil a specifying function (with respect to the semantic, pragmatic, or discourse-functional information content of a semantic unit), but cannot themselves be specified by in-stances. From a frame-semantic perspective, this is the criterion that distinguishes affixes from lexical morphemes.

In the following, I do not elaborate any further on the many possibilities of using frame-semantic categories for morphological analysis. Suffice it to say, everything that I outline in the following sections with respect to recurrent schema instantiations and cognitive entrenchment concerns schema instantiations in general, and thus also morphologically relevant relationships. Frames constitute a uniform representational format that can be employed to explain morphological categories.

But what happens when adjectives appear as fillers? This case is subject to exactly the same schema-theoretical conditions: An adjective specifies a potential slot in the frame evoked by the expression that the adjective refers to. (13′) and (14′) are therefore only variants of (13) and (14); they can easily be transformed into explicit predications of the form x is y.

(13′) the red light (14′) the red window

Exactly as with determinative compounds, the fillers – in this case in the form of adjectival phrases – more closely specify a knowledge facet of an evoked frame. In other words, the fillers – in one case, morphemes; in the other case, lexical units in a particular syntactic function – form instances in a schema. Figure 6 illustrates this using some corpus-based evidence from Chapter 7.

Apart from financial investors (‘Finanzinvestoren’), the text corpus contains, among others, compounds such as Großinvestoren (‘large investors’) and Auslandsinvestoren (‘foreign investors’), as well as nominal phrases of the kind internationale Investoren (‘international investors’) and sprunghafte Investoren (‘erratic investors’). From a frame-semantic perspective, there is no need to distinguish between these cases, since they rely on the same cognitive act of schema instantiation illustrated in Figure 6.

440. Bybee (1985); Taylor (2002: 298–319). Bybee and Taylor do not speak of “frames” but “schemata”. Since, as outlined above, frames and schemata share all the essential characteristics, the results of Bybee and Taylor’s studies can also be useful for frame-semantic issues.


Instances

Fillers

Element to be predicated

Schema

SlotsFrame /

reference object

Speci�ed frame

Finanz - ( �nancial)

Groß - (large)

Auslands - (foreign)

internationale (international)

Sprungha�e (erratic)

…

…

-investor/en (investor/s)

Lexi

cal

inst

ance

sM

orph

emat

icin

stan

ces

Figure 6. Schema instantiation in the case of lexical and morphematic (in this case, in terms of compounding) specification of lexical units

A similar principle applies for fillers in the form of prepositional phrases (like inves-tors with criminal intent), genitive attributes (such as investors of Anglo-Saxon origin), and relative clauses (for example investors who have purchased the media company).441 Each time, the reference object specified by investor is attributively qualified such that a filler forms an instance in a slot in the evoked Investor frame. And each time, the quasi-explicit predication can be transformed into an explicit predication:

(15) foreign investors (15′) Investors come from foreign countries442

(16) investors with criminal intent (16′) Investors have criminal intent (17) investors of Anglo-Saxon origin

441. Relative clauses constitute an exception in that they always contain a finite verb that has “financial investor/s” as its subject. Thus, in this case syntactically complete sentences are transformed.

442. Determinative compounds, as is generally known, allow different readings, above all when they are verbal compounds. Which reading applies is determined from case to case.


(17′) Investors are of Anglo-Saxon origin (18) investors who have purchased the media company (18′) Investors have purchased the media company

Each of these variations contributes to a specific epistemic perspectivation of the in-vestor frame. To name a few examples: in (15) and (15′), the slot that concerns the origin of the investors is occupied by a filler. It thus contextualizes the semantic unit [investor] from the perspective of a collective “we” or “homeland”. It follows that [investor] stands under the dichotomizing auspices of “us” vs. “them”, “homeland” vs. “foreign land”. (16) and (16′) behave similarly, albeit somewhat more neutrally. The fact that in both cases no other slot in the investor frame is occupied by a filler is by no means a sign that the values in other slots are not U-relevant. Rather, they belong to that set of background assumptions that von Polenz (1985: 157) refers to as being “also intended” (‘mitgemeint’) or “also understood” (‘mitzuverstehen’), in other words, as being “filled in”. These default values are dealt with in the following.

6.5 Implicit predications: default values

By far the largest proportion of U-relevant knowledge in texts occurs in the form of neither explicit nor quasi-explicit predications. Rather, it is as Fillmore formulates:

…that much of this linking and filling-in activity depends, not on information that gets explicitly coded in the linguistic signal, but on what the interpreter knows about the larger scenes that this material activates or creates. Such knowledge depends on experiences and memories that the interpreter associates with the scenes that the text has introduced into his consciousness. (Fillmore 1977a: 75)

We can now be more concrete about what it means to activate a cognitive frame. When language users activate a frame, they establish a cognitive relationship to a frame whose slots are already partially filled with default values in the act of referentialization. This is because every identification of a reference object is preceded by attributions of properties that in their sum make the object distinguishable from other potential objects. Henceforth, I designate such attributions of properties that enable reference as “implicit predications”.

What are implicit predications? Let us approach an answer by analysing some examples. In jokes like the following it becomes clear who closely implicit predications are related to cognitive referentializations (cf. Coulson 2001: 55).

(19) Everyone was having such fun jumping off the tree into the swimming pool that we decided to put a little water into it.

In (19) the word swimming pool initially evokes a frame that is specified by an ex-plicit predication. The fact that people are jumping into the swimming pool profiles (or perspectivizes) the reference object with respect to a possible function that it can


have for humans.443 This epistemic perspectivation immediately motivates all sorts of implicit assumptions, for instance about the expected size of a swimming pool, its material properties, its location, its components, and their functions. Language users invoke corresponding default values from their memory. These specify specific slots (but not all of them) in the evoked frame. In Allwood’s sense, a meaning potential of the expression swimming pool is thus activated. The punchline is achieved by cancelling an – apparently unshakeable – implicit predication and consequently introducing a semantic reinterpretation of the main clause.

To test which additional default values the evoked frame contains, one could appropri-ately alter the subordinate clause in (19). If the explicit predication in the subordinate clause does not cancel any implicit assumptions about the swimming pool, we are not dealing with a default value. Compare the following possible subordinate clauses:

(19a) Everyone was having such fun jumping off the tree into the swimming pool that we decided to increase the size of the pool to the size of a paddling pool.

(19b) Everyone was having such fun jumping off the tree into the swimming pool that we decided to increase the water temperature from five degrees to twenty.

(19c ) Everyone was having such fun jumping off the tree into the swimming pool that we decided to increase the chlorine content.

In contrast to (19a) and (19b), the final Example (19c) obviously does not cancel an implicit predication, but rather specifies it. Consequently, information about the chlo-rine content of the water is not a default value.

The ubiquity of implicit predications can be illustrated using numerous other language phenomena that are discussed in linguistics. Among these are implicatures, presuppositions, semantic polysemy, etc. At this point I would briefly like to elabo-rate on a particularly illustrative phenomenal area, namely indirect anaphora. In so doing, I reconnect with a thought that I already formulated in another context in Section 2.2.4: Anaphoric processes operate on a frame basis, and this occurs crucially via the activation of default values. Frames constitute a descriptive format that can be used to uniformly explain anaphorization, so that, in contrast to what Schwarz sup-poses (2000: 99–117), it is by no means necessary to posit different activation types to explain the necessary processes of coherence construction.

Compare the following (translated) examples from Schwarz 2000:444

(20) I was about to unlock the door when Morelli leapt out of the bushes. I dropped my key in fright.

443. Or more precisely, swimming pool can be retraced via hyperonym reduction to the matrix frame artefact, which has, among others, the following slot: ‘In what functional contexts, events and actions does a swimming pool play an important role?’ (cf. Konerding 1993: 312).

444. Schwarz (2000: 99, 106, 111, 116).


(21) Don’t use the yellow cup. The handle is broken. (22) In those last August days … the girl called Rita Seidel awoke in a small hospital

room…. The nurse steps up to her bed. (23) Rumour is a collective bubble gum: Everyone chews on it, it grows, and in the

end the bubble bursts.

According to Schwarz, each example demonstrates a different activation type. In (29) the semantic relationship between key and unlock is motivated by the semantic role required by the verb unlock. By contrast, in (21) the relationship between handle and cup is of a meronymy-based activation type, whereas in (22) nurse and hospital stand in a schema-based relationship. Finally, in (23) the construction of coherence between bubble and bubble gum is inferentially based.

The key point is as follows: In a frame-semantic perspective, all four coherence constructions are both inferential and schema-driven in that the referential text ele-ment evokes a frame with default values, and it is precisely one of these which closes the gap in coherence. In (21) the expression handle evokes a frame that contains all the slots of the matrix frame part, since it can be retraced to this via hyperonym re-duction.445 One of its slots is:

(24) Of what entity does the handle form a part?

The highly salient default value ‘cup’ corresponds precisely to the text element indi-cated by the indirect anaphoric reference of the expression handle. Similarly, in (23) the expression bubble evokes a frame whose slots are identical with those in the matrix frame artefact. To establish reference back to the text element bubble gum (or rather its reference object), the relevant default value is that which provides an answer to the question of what the bubble consists of (Konerding 1993: 312).446

Consequently, anaphorizations, like all other forms of textual reference, succeed solely under the premise that text recipients share a common store of knowledge in the form of actualizable default values. It is important to realize that we do not dispose over “naked” frames, i.e. frames consisting of lists of potential questions or unoccupied slots.447

Frames are probably never stored in long-term memory with unassigned terminal values. Instead, what really happens is that frames are stored with weakly-bound default assignments at every terminal. (Minsky 1977: 364)

445. Cf. Konerding (1993: 450–456).

446. I already discussed (20) and (22) in Section 2.2.4.

447. Cf. also Minsky (1988: 247): “I suspect that the larger part of what we know – or think we know – is represented by default assumptions, because there is so little that we know with perfect certainty”.


The fact that slots are already filled with default values makes it possible to perform semantic categorizations in a cognitively highly efficient manner. We do not normally notice gaps in coherence such as those that occur in the case of indirect anaphora because inference constructions occur automatically, as it were.

Let us note: Default values are of major cognitive importance because, on the one hand, without the activation of default values, linguistic reference actions could not occur, neither those of a discourse-deictic nature nor those that Bühler designates as “deixis am phantasma” (i.e. referring to absent referents), and, on the other hand, no area of sign-based understanding exists without referentialization. At the same time, the fact that words evoke frames does not mean that language users do not participate in this evocation process. After all, the evocation of a frame corresponds to a reference act, and this in turn goes hand in hand with the invocation of default values, that is, with implicit predications. Since default values are of a non-overt nature, they remain the result of analytical accessing processes and thus theory-dependent constructs.

The following two sections deal with the cognitive theoretical basis of such analyti-cal processes and the possibility of their empirical operationalization.

6.5.1 Recurrent schema instantiations: token and type frequency

Which default values occupy the slots of an activated frame and for what reason? Matrix frames provide so many possibilities of knowledge specification that it seems reasonable to doubt whether a corresponding number of experiential values can be stored in the long-term memory. Alone the matrix frame artefact, which is relevant for the handle frame, encompasses over 100 questions, and even when these are bundled into predicator classes, over twenty classes remain. This is what Klein sees as a key difficulty:

The vast plethora of potential knowledge dimensions opened up by this process of extracting matrix frames raises the so far unanswered question as to how and by what means this set can be reduced to the far smaller set of default knowledge that constitutes the frame as a lexicon unit… (Translated from Klein 1999: 160)

What makes this more difficult is that often a slot is not occupied by just one default value, but that there is a range of equivalent possibilities (referring to the previous example, think of all the possible colours, forms, and sizes of handles). Which default values occupy the slots of a frame? Why do precisely these default values occupy the slots? How are default values formed, i.e. how does the transition from explicit to implicit predications proceed?

In view of these unexplained questions, one could come to the conclusion that using matrix frames does not allow us to make any cognitively realistic statements about U-relevant meaning knowledge. I do not share this view. After all, not all slots or all default values prove to be equally relevant when a word evokes a frame. In fact, in line with the graded salience hypothesis (Giora 1997), there are gradations in the


cognitive accessibility of knowledge. Only in the case of high salience do implicit predications occur. I now explain this point in terms of schema theory.

In the example with the expression handle, the problem outlined manifests itself in the following way. Let us assume that the expression handle evokes a frame whose slots are already partially specified in such a way that we could easily answer the questions in the corresponding part matrix fame. We then dispose over a semantic prototype in the form of typically expected assumptions that in total correspond to the convention-alized meaning aspects of the semantic unit [handle]. As demonstrated in Figure 7, the conventional instances of meaning include, among others, assumptions about size, part-whole relations, and material properties. A selection of possible slots is given in the rectangular boxes. According to the basic premises, default values would have to be attributed to the slots, which are represented here as ellipses. The default values (just like the expression handle) themselves evoke a frame; the dashed lines indicate this frame as a unit, but are not intended to imply it is self-contained. From the set of potential subframes, only the subframe [cup] is illustrated in Figure 7.

How do we know which values typically form instances of the slots? It is only for reasons of space that Figure 7 shows only a few default values; in reality the number of other values that could be given is almost infinite. What puts us in a position to inter-pret the expression handle so that its semantic unit [handle] stands in a part-whole relationship with the semantic unit [cup]? Or how do we know that [porcelain] is a typical value when it is a question of the material properties of handles?

Figure 7. Selection of slots and potential default values in the handle frame


The answer is clear: Because we are confronted with these values every day in (lin-guistic and extralinguistic) usage contexts and, due to this recurrence, we can distin-guish typical from non-typical assumptions. This is how prototypical frame structures, in Barsalou’s sense, emerge:

Regardless of how prototypes arise, frames naturally produce typicality effects: If an exemplar’s values occur frequently across the exemplars integrated into a frame, then it is typical … In contrast, exemplars whose values occur infrequently are atypical … Additionally, if exemplars vary in the attributes for which they have values, then an exemplar’s typicality also depends on how frequently its attributes have values across exemplars. (Barsalou 1992a: 47–48)

Barsalou’s hypothesis that not only the activation frequency of certain slots correlates with the degree of their entrenchment, but also the degree of typicality of instanti-ated values increases with their increasing frequency of occurrence448 can be applied directly to our problem.449 The hypothesis states, then, that explicit predications, by dint of their relatively high frequency of occurrence, are good candidates for future implicit predications. The large set of potentially relevant default values is thus sud-denly reduced. This applies in a similar fashion to slots: Only those slots in which values are frequently instantiated prove to be primarily U-relevant. They constitute the meaning potential of an expression in the sense explained in Section 4.3.2. The mean-ing potential encompasses only a fraction of all the possible knowledge specifications of a frame, since many slots are “faded out” or “inactive” as it were, that is, they prove not to be primarily U-relevant in the constellation of certain information contexts.

The cognitive entrenchment of certain slots in a frame, as well as certain values within these particular slots – on a lower level of abstraction – originates in recur-rent schema instantiations. The more frequently a language user establishes the same or similar schema instantiations, the more routinely and efficiently this cognitive act transpires. In this process, a slot in a frame can become entrenched. Referring to the handle example, this means that if language users are confronted with different explicit or quasi-explicit predications that relate to a particular slot significantly more often, then precisely this slot (in relation to other slots) would gain salience. In Figure 8 this corresponds to case (i). Many different values specify the size of a handle in this case, whereas comparatively few specify other slots. As a consequence, a high frequency of different tokens consolidates the slot whose instances they constitute and thus the frame to which the slot belongs. In other words, if something is frequently stated about a refer-ence object (predication), this increases the cognitive presence of the reference object.

It is equally possible, however, that in comparison to other slots a particular slot is not more often supplied with values, although one and the same value (instance)

448. Cf. also Barsalou (1985); Langacker (1988c); Bybee and Hopper (2001); Taylor (2002: 274ff.).

449. Barsalou does not speak of the “slots” in a frame, but of its “attributes”. What I call “fillers”, Barsalou calls “values”.


occurs significantly more than other values. This case is illustrated in Figure 8 under (ii). Here, the frequently instantiated value consolidates itself without the slot being entrenched to the same degree, because the set of all instances in this slot is no more frequent than the set of all instances in the other slots. In such a scenario, default values emerge. The transition from explicit to implicit predications thus occurs when fillers appear so often that language users can take this knowledge for granted and it no longer needs to be explicit (also for reasons of language economy).450

In Cognitive Grammar the consolidation of structural constituents (in this case, a filler and a slot) within a schema is discussed as “entrenchment”.451 If we consider default values as cognitive entrenchment, then aspects of representation tend to shift

450. Consider product advertising from this perspective. So that consumers quasi automatically ascribe particular predicates to a product, they must be confronted as frequently as possible with appropriate explicit predications in the vein of “KFC – It’s finger-lickin’ good” or “Beanz meanz Heinz”. When the reference object is evoked (Kentucky Fried Chicken, Heinz beans) the consumer performs the corresponding predications; cf. also Ziem (2006b).

451. Langacker (1991a: 45): “Entrenchment pertains to how frequently a structure has been invoked and thus to the thoroughness of its mastery and the ease of its subsequent activation. Along the parameter of specificity, units range from particular expressions at one extreme, through schematizations at different levels of abstraction, to maximally schematic structures representing entire grammatical categories or constructions at the opposite extreme”. See also Langacker (1987: 59f.), (1999b: 93f.).

Functions as?

…

[…]

[…][…]

[…]

Howlarge?

Functionsas? …

[…][…]

[…][…]

[…]

[…]

i. Entrenchment of a slot (high type frequency)

ii. Entrenchment of a value (high token frequency)

[…]

[…]

Howlarge?

[…]

[HANDLE]

[HANDLE]

Figure 8. Entrenchment of slots and values (represented by bold outlines) as a result of recurrent schema instantiations


into focus. However, it is important to also consider the process of entrenchment. In the next section, I engage with aspects of process.

Although this distinction between the entrenchment of a slot on the one hand and the entrenchment of values on the other is not discussed either in Cognitive Grammar or construction grammar, connections can nonetheless be made to two common ex-planations of entrenchment phenomena. Entrenchments are explained either by a high “token frequency” or a high “type frequency”. With regard to type frequency and token frequency:

Each of these gives rise to the entrenchment of different kinds of linguistic units. While token frequency gives rise to the entrenchment of instances, type frequency gives rise to the entrenchment of more abstract schemas. (Evans/Green 2006: 118)

Bybee and Slobin (1982) show in their morphological study that the frequency of oc-currence of flexion morphemes correlates directly with their degree of entrenchment. Frequently used irregular verbs like lend, for example, have retained the past participle form (lent), whereas rarely used expressions like blend allow the regular form blended alongside the irregular form blent. The distinction between the past participle forma-tions of the verbs lend and blend is explained by the entrenchment of the instance [-t] in the former case (as illustrated in (ii) in Figure 8) and by weakening of the same instance in the latter case. This is because in the latter case the two instances [-t] and [-ed] compete with each other, and the low frequency of occurrence of the two has led to neither being able to assert itself (cf. also Taylor 2002: 276–277).452

The frequency of occurrence of a linguistic category, whether in the form of an instance or a schema, yields consequences for the cognitive representation of this cat-egory as well. Increasing cognitive entrenchment increases the degree of salience.453 Once a structural constituent of a schema has been entrenched as a result of schema instantiations, it proves to be – initially – resistant to factors of contextual influence.

Well-entrenched structures, ceteris paribus, are more salient than less-entrenched structures, i.e., they occur more energetically. Entrenchment can be viewed as a kind of enduring salience, i.e., salience apart from relatively transitory effects such as directed attention or heightened activation due to contextual factors. (Tuggy 1993: 279)

452. On the relevance of type frequency and token frequency for phonology cf. Bybee (2001: 10–13), for an overview Bybee (2013).

453. In addition, the depth of processing (Anderson 1996: 187–193) influences the degree of sali-ence. The more deeply a cognitive event is processed, the more permanently it is stored, i.e. the more easily it becomes retrievable in the future (cf. also Craik and Lockhardt 1972). Accordingly, how someone processes something affects the individual memory operation. In the following, I do not engage any further with memory-theoretical aspects of processing depth.


Nevertheless, established structures can change at any time, and in fact change occurs constantly. This is because linguistic units (meant here in the broad sense of form-meaning units) establish themselves exclusively subject to language usage (Barlow and Kemmer 2000; Langacker 1988c) and are thus emergent phenomena (MacWhinney 2001). A high token frequency entrenches the instance, at the same time weakening the schema. However, a high type frequency can at any time provide a weakening of the instance, thus entrenching the schema. The transition between cases (i) and (ii) depicted in Figure 8 is therefore fluid. Between the two a continuum prevails. If, for instance, many different explicit and quasi-explicit predications, i.e. many tokens, serve one and the same slot in a frame, they contribute to the entrenchment of the slot and indirectly to that of the entire frame. If, however, a particular predication occurs frequently (relative to the others), this instance itself is strengthened, but not the frame as a whole. There are many other possibilities between these extremes. Cognitive entrenchment is a scalar phenomenon whose dynamic of change is solely determined by language use.

Let us consider the emergence process of default values once again from the per-spective of semantic prototype theory. Following its basic ideas, so-called “basic level categories” should be good candidates for default values (cf. Rosch et al. 1976; Schmid 1996; Lakoff 1987: 31–38; Taylor 2003: 48–55). From a cognitive-linguistic perspec-tive, the terms dog, cup, chair, etc. represent basic level categories, but not their hypo-nyms (like Golden Retriever, feeding cup, wooden chair) or hyperonyms (like animal, container, furniture). Basic level categories are located on a mid-level of abstraction, and their members (instances) share many attributes with each other. One could also say that many shared assumptions could be connected to basic level categories, for instance, the assumptions concerning dog are that it is a quadruped animal that can bark and wag its tail. The shared attributes or assumptions demonstrate a very high “cue validity”. “Cue validity” refers to how predictable predicative attributes are with reference to the category individual members of a category belong to. In the case of an animal that barks, for example, one can certainly assume that we are dealing with a dog with four legs and a wagging tail. The attribute bark therefore possesses a very high “cue validity” for the category ‘dog’.454

Basic level categories are fundamental cognitive units for the categorization of both linguistic and extralinguistic (visual, auditory, etc.) data. Basic level categories have a decisive cognitive function for three reasons.455 Firstly, basic level categories are distinctive because they have a highly refined internal structure, i.e. their cate-gory members (instances) have many shared attributes at their disposal. Rosch et al. (1976: 387ff.) have shown that categories on a mid-level of abstraction can be assigned eight attributes, whereas for super- and subordinate categories it is a maximum of

454. On the measurability of cue validity, cf. Lakoff (1987: 52–54).

455. Lakoff (1987: 46) and Mangasser-Wahl (2000: 38–40) name additional other features of basic level categories. Here I restrict myself to those that are relevant for default values.


three. On the other hand, basic level categories are sufficiently distinct from contigu-ous categories (for instance, cat is sufficiently distinct from dog, horse, etc.); they con-stitute relatively distinct units. This characteristic does not apply to more abstract or more concrete categories either.456 Thus, the following applies to basic level categories: “[T]hey maximize perceived similarity among category members and minimize per-ceived similarities across contrasting categories.” (Lakoff 1987: 52)

The second reason is because linguistic basic level categories have correlates in the extralinguistic world of perception. Within visual perception, stools, tables, cats, dogs, cups, glasses, etc. belong to the basic level because as a rule we do not initially perceive either products of abstraction (furniture, animals, crockery, etc.) or objects below the basic level (a wooden stool, an Angora cat, a coffee cup, etc.).457 Schmid (1996: 286–287) clarifies this point in the following way. If I face the task of drawing a picture of the object that a linguistic expression is referring to, this can be achieved easily in the case of basic level categories. By contrast, I can only illustrate super- or subordinate categories indirectly via basic level categories.

Finally, the third reason highlights the particular relationship between basic level categories and human motor functions. Rosch et al. (1976: 393–398) argue that par-ticular sequences of motor movement correlate with every object on a medium level of abstraction: By means of specifically coordinated movements, we can sit on stools and lie on beds, guided by motor routines we can lift cups and lead them to our mouths, and we can take dogs for a walk, which also involves certain coordinated motor sequences. Each of these activities can be described more precisely as a correlation of motor sub-activities. However, motor movement sequences cannot be ascribed to superordinate categories (we do not prototypically raise crockery and lead it to our mouths, nor do we prototypically sit or lie on furniture, etc.), and subordinate categories cannot be sufficiently distinguished from each other using different motor movement sequences (wicker chairs, wooden stools, and rocking chairs equally serve as seating, we raise coffee cups to our mouths just as we do moka cups and feeding cups, etc.).458

456. Fillmore (1982: 132) uses a good example to illustrate why this is the reason that basic level categories are used more often: Suppose two neighbours engaged in conversation are interrupted by a splashing noise. The one neighbour would comment that his dog has fallen into the swim-ming pool and not an animal or a Golden Retriever.

457. This is to be understood as ordinary regularity. Of course, it may be that, for instance, in a professional context certain entities are dealt with in such a way that they are perceived below the basic level, such as a cat breeder perceiving a cat as an Angora. In this case, a visual schema angora cat has emerged and become entrenched on the basis of a high token frequency.

458. Certainly, motor sequences that correlate to a basic level category can serve neither as a cri-terion for justification for nor as a means of distinction from basic level categories. Nonetheless, they clarify – in the sense of the holistic paradigm – how much extralinguistic factors participate in cognitive category construction.


If basic level categories are cognitively more salient than others for these three reasons,459 what consequences does this have for default values? The most significant consequence is that default values generally represent language categories of a medium level of abstraction. The sphere of potential default values is therefore even further reduced. Seen against the background of token frequency, it becomes clear why default values are usually located on the basic level: Basic-level language categories occur so often in our “perceptual reality” that they quickly become entrenched. The fact that basic-level categories occur frequently not only in the area of language additionally accelerates the cognitive entrenchment process. Almost all basic-level language cat-egories have a correlate in visual perception – a fact that does not apply to any (or very many) super- or subordinate categories. And we connect typical motor movement sequences with almost all of the basic-level language categories. This does not apply to this extent for more abstract or concrete categories. Consequently, basic categories are much more easily entrenched as default values than other categories.

Let us return to the initial questions. What status do default values have? And how do they emerge? The preliminary answers are as follows: Structurally, default values correspond to implicit predications, which are distinct from (quasi-)explicit predications because language users must invoke information from their memories. Default values correspond to this invoked information. They are emergent products of our language usage, because they occur under the condition that recurrent schema instantiations lead to the cognitive entrenchment of the instantiated token as a result of high token frequency. At the same time, cognitive entrenchment is to be expected in particular with those values that belong to a medium level of abstraction. However, this entrenchment of values is not necessarily permanent. The extent to which those schema instantiations that once led to the emergence of a default value no longer occur with the same frequency in language usage can lead to default values being modified or even being replaced by others. This cognitive process is decisive for lan-guage change.

It is not only values that can be entrenched. If a slot in a frame is occupied by a large number of different instances, that is, if a knowledge facet of a reference object is specified by a particularly high number of predications, then there is a high type frequency, which entrenches both the slot (the knowledge facet) and the frame (the reference object) to which the slot belongs. A high type frequency and a high token frequency together ensure that each frame does not in fact evoke an insurmountable plethora of knowledge, but rather only the small set of “default knowledge” that Klein speaks of. In other words, when language users evoke or invoke a frame – depending on their language experience – only some slots and some default values of that frame are particularly salient.

Both types of cognitive entrenchment can be empirically described using frame-semantic tools. The example in Section 6.5.3 illustrates how entrenchments occur via

459. Further evidence that basic level linguistic categories fulfil an essential cognitive function is given by Schmid (1996) and Ungerer and Schmid (1996: 60–109).


high type frequency. However, quantitatively supported corpus-based analysis is neces-sary to provide reliable statements about which default values consolidate themselves in a frame I address this in Chapter 7.

6.5.2 “Cognitive trails” as phenomena of the third kind

That the structural constituents of frames become cognitively entrenched as a result of recurrent schema instantiations is initially just a hypothesis that can be plausibly supported by schema theory. However, what lies behind the metaphor of “cognitive entrenchment”? Is it psychologically realistic to speak of “cognitive entrenchment”? Is there any (linguistic, neurological, etc.) evidence of such entrenchment processes? If so, who far does the metaphor extend? And where are its limits? Although unable to provide a comprehensive answer in the following, using two cognitive phenomena as examples I attempt to demonstrate some reasons why the theory of cognitive entrench-ment can be considered “realistic”460 in many respects. One cognitive phenomenon concerns a linguistic theory of language change, the other neurological findings on human memory processing. I take the fact that these cognitive phenomena are of an extremely different nature yet nevertheless converge in a crucial point as evidence of the feasibility of the theory of cognitive entrenchment.

The shared point of departure concerns an aspect of process, which I would first like to explain. The previous section did not deal adequately with the processual com-ponent. Strictly speaking, in contrast to what was implied, it is not the structural con-stituents of schemata (slots, values), but rather the procedural routes by which language users interpret schema-relevant knowledge that are entrenched. In this context, Taylor (2002: 43) designates schemata “principles of categorization”. What he means by this can be illustrated using an example of default values. Let us assume a (quasi-)explicit predication has occurred so often that it has become entrenched as a default value in the evoked frame. So that the language users can retrieve the corresponding informa-tion, they must establish a categorization relationship by means of which they can inferentially interpret the value. They instantiate a default value in a slot of the frame. This is a fundamental cognitive act; as already mentioned, linguistic referentializations succeed only when slots are implicitly filled with values (recall the examples [swim-ming pool] and [handle]). Symbolically associating conceptual knowledge with a phonological unit therefore means disposing over concept-specific cognitive routines, because – as Langacker puts it – “established concepts are simply entrenched routines”. Or more precisely:

460. I see a theory or hypothesis as being cognitively/psychologically realistic, in the commonly understood cognitive linguistic sense, when it does not contradict sufficiently proven insights gleaned from neuroscience and other related disciplines (like psychology, anthropology, etc.) (cf. for instance, Sweetser 1999). In this sense, adherents of cognitive linguistics have repeatedly pointed out that key hypotheses of generative grammar (such as that of a universal grammar) are psychologically unrealistic.


A complex predicate like [cat] or [banana] is more accurately viewed as a set of routines, which are interrelated in various ways … The encyclopedic characteriza-tion of a typical predicate involves many such routines (possibly quite complex), each representing a separate specification. (Langacker 1987: 162)

Particularly well-established cognitive routines prevail between hyperonym expres-sions. Let us recall the [yak] example. If one knows that [yak] is an instance of the schema [bovine], it is easy to assign a number of default values to the evoked schema. The most fundamental of these derive from the schemata of the hyperonyms animal and living creature, since default values of more abstract schemata are inherited by those schemata that the hyponym expressions evoke. Without precisely knowing what a yak is, I can claim that it is an organism that can move, take nourishment, possesses a vegetative nervous system, has a certain size, etc. The fact that such default values are less prone to disappoint is because a high token frequency entrenches the schema evoked by the default value. For instance, if cattle are said to canter, horses to gallop, and humans to run, the semantic units [canter], [gallop], and [sprint] constitute instances in the superordinate schema [move]. On the other hand, because there are only a few instances of the schemata [canter], [gallop], and [sprint], and these are not used so frequently, such schemata are less entrenched.

Cognitive routines are established not only between linguistic units.461 Following holistic premises, I introduced the schema concept at the beginning as the overarching concept for representational formats of a variety of perceptual modalities and reserved the term “frame” for specific semantic schemata. However, since all schemata consist of the same structural constituents and, what is more, the fillers and default values are connected to the slots of a schema by categorization links, cognitive routines be-come entrenched in quite varied cognitive domains – as a result of recurrent schema instantiations. For instance, they can relate to motor skills (like playing the piano, writing, driving a car, etc.)462 or to the processing of visual, auditory, and other data (recognition of objects, melodies, voices, etc.). Although in the following I restrict

461. That is, in the sense of the symbolic principle, not only (i) between semantic units, (ii) be-tween phonological units, and (iii) between symbolic units as a whole; cf. Section 4.1.2.

462. Using the example of driving a car, Busse (1991a: 169) illustrates that actions are overall schema-driven processes (here, in translation): “Using the example of driving a car, it is pos-sible to see how a learnt ‘action’ that still occurred consciously at the beginning of the learning process becomes ‘automatic’ with increasing practice: Anyone who has been driving for a long time changes gear ‘automatically’; in the individual case the person no longer thinks ‘when I want to change gear I must first put my foot on the clutch’, but they simply do it ‘automatically’. It is only when the gear box crunches that they notice they have ‘forgotten’ something. The (far more complex) process of expressing and understanding language is similarly ‘automatic’. The vast majority of the rules of language use are subconscious and since we, as individuals with developed powers of reasoning, have not learnt them through explanation (in contrast to driv-ing a car) but rather solely through practice – together with the acquisition of our powers of reasoning – these rules may possibly never have become conscious to us”.


myself to linguistic phenomena, this does not mean that similar considerations could not be equally extended to other areas (cf. for instance Ziemke, Zlatev, and Frank 2007; Zlatev 2005).

To begin with, let us illustrate cognitive routinization using the example of lan-guage change. What does language change have to do with cognitive routines? It is not until closer inspection that the connection becomes apparent: If linguistic meanings change, then the cognitive routines of the language users that lead to the actualization of these meanings must also change. According to the hypothesis of cognitive en-trenchment, there must be a close relationship between language change and a change in the usage frequency of language tokens in particular contexts (and the accompany-ing varying predications). The more seldom – in the long term – particular schema instantiations recurrently occur within a language community, the more seldom cer-tain cognitive routines are employed, and therefore the faster the respective linguistic unit changes. Apart from a decreasing token frequency, in the sense explained above, language change should also be accompanied by the entrenchment of other recurrent schema instantiations, and new cognitive routines should gradually emerge. How re-alistic is such a schema-theoretical explanation?

Language change is considered a phenomenon of the third kind. It cannot be explained as a natural phenomenon, nor does it correspond to an artefact (Keller 2003: 81ff.). Language change is not a natural phenomenon because language only changes since members of a language community use language. Nonetheless, individ-ual language users cannot plan to change a language in the same way that they are able to consciously manufacture an artefact. Language change is therefore a phenomenon of the third kind. Keller explains:

Phenomena of the third kind… are as a rule collective phenomena. They emerge through the acts of many; this is a result of the fact that the acts that generate the phenomenon have a certain uniformity, which in themselves may be irrelevant, but in their multiplicity yield certain consequences. (Translated from Keller 2003: 91)

Language change results from the use of language or, more precisely, from the fact that language is used similarly or in the same way by many language users. What is inter-esting is that Keller implicitly represents the hypothesis that language change occurs on the basis of a high token frequency. His version of the entrenchment hypothesis is:

A phenomenon of the third kind is the causal consequence of a multitude of indi-vidual intentional acts that serve at least partially similar intentions. (Translated from Keller 2003: 93; original in bold)

If we put aside the fact that Keller is explaining language change exclusively within a theory of intention,463 in one decisive point the similarity to the theory of cognitive

463. A perspective that, in my view, can become problematic if it ignores representational and cognitive aspects of language change. A linguistic action can only be successful and, conversely, can only be understood if relevant background knowledge (for instance, about its conditions of


entrenchment is striking. To be sure, cognitive entrenchment is on the one hand similar to natural phenomena in that it lies in the sphere of influence of individual language users and on the other hand similar to artefacts in that it is nonetheless a result of human action. However, because it shares properties with both artefacts and with natural phenomena, yet at the same time is distinct from them, it is a phenomenon of the third kind. This is shown in three respects:

– Similar to language change, cognitive routines are emergent phenomena that are inherent in the usage of language. As long as language is used, it changes, as do the cognitive routines that we use to understand expressions.

– Language change and cognitive routines are epiphenomena that are only difficult to predict because it is difficult to form a prognosis about the future use of linguis-tic expressions (Keller 2003: 100).

– Language change occurs within the same three-phase process that leads to the entrenchment of cognitive structures (Keller 1995: 121). (i) In the beginning there is individual language use. Keller’s assumption that on the micro level humans only perform intentional acts under certain conditions corresponds to the premise of the usage-based model that linguistic structures are abstraction products of lan-guage use.464 (ii) If the acts performed, or the actualized usage events, are uniform, they can motivate in a second phase an “invisible-hand process” or certain cogni-tive routines. (iii) Finally, on the macro level, the causal consequence is a change in language and thus a change in cognitive routines.

It is no coincidence that cognitive entrenchment phenomena are phenomena of the third kind. Every diachronic change in linguistic structures must be reflected in corre-sponding changes in cognitive structures, i.e. in the mental representation of linguistic structures. And this implies that the usage-based conditions under which the changes occur are the same.

Put simply, default values are epiphenomena, that is to say, they are unintentional results of a multitude of uniform (quasi-)explicit predications.

success, in the Searlean sense) is actualized to perform the action (or to understand the action). Although this knowledge may itself be the result of previous actions, nonetheless at the moment of the occurrence of the action (or – of the understanding) it is postulated knowledge and as such exhibits its own cognitive-representational “reality”; cf. my discussion of Zlatev in Section 4.2.2.

464. “A speaker’s linguistic knowledge is not conceived as an algorithmic constructive device giving (all and only) well-formed expressions as ‘output’. Instead, it is more modestly character-ized as an array of units (i.e. thoroughly mastered structures – cognitive routines) available to the speaker for the categorization of usage events (actual utterances in the full richness of their phonetic detail and contextual understanding). Such units arise by a process of schematization based on the reinforcement of recurrent features; thus each embodies a commonality observable across a series of usage events” (Langacker 1991a: 2; italics in bold in the original). Cf. also Langacker (1988c), (1999b: 91–145), and the overview in Barlow and Kemmer (2000).


Yet, how realistic is the theory of cognitive entrenchment from the perspective of memory theory? Or put differently: Do phenomena of the third kind have a neu-ral correlate? At one point, Keller compares the process of language change with the treading of a trail.

A simple example is provided by the theory of trails: There is a network of trails criss-crossing the lawn of our university. This network of trails is extremely clever and “set out” in an economical and sophisticated manner. Quite obviously its struc-ture is far more ingenious than the structure of the paved roads planned by the architects. What is more, on a map showing the buildings and other facilities along with their functions, but not the trails, on this map one could anticipate where these trodden trails would emerge. The system of trails could be predicted with much greater accuracy than the system of the paved routes planned by architects. (Translated from Keller 2003: 100–101)

I take up this comparison and refine it as the metaphor of “cognitive trail”, leaving aside for the moment that it is “only” a metaphor, to see how far the explanatory power of this comparison extends in (neuro-)psychological terms.

To begin with, one should note that the concept of memory itself is highly poly-semous. It refers to the neuronal preconditions for processes of encoding, storing, and retrieving information as well as to the concept of information storage that need not be of a psychological nature at all (consider such expressions from cultural science as “cultural memory”) (Tulving 2000: 36). In the following, when I speak of a “cognitive trail” I mean the entrenchment of memory structures, and memory is first and fore-most a neuronal activity.

What are, then, “cognitive trails”? Langacker sees the entrenchment hypothesis as being confirmed from a neurological perspective in that every usage event corresponds to a neurochemical reaction, which leaves a clear trace when repeatedly triggered. The metaphor “trace” is intended to indicate that a usage event may become entrenched, but it can just as easily weaken though a decreased frequency of occurrence.465

I will use the term event to designate a cognitive occurrence of any degree of complexity, be it the firing of a single neuron or a massive happening of intricate structure and large-scale architecture. We can assume that occurrence of any such event leaves some kind of neurochemical trace that facilitates recurrence. If the event fails to recur, its trace decays; recurrence has a progressive reinforcing ef-fect, however, so an event (or more properly, event type) becomes more and more deeply entrenched through continued repetition. (Langacker 1987: 100; italics in bold in the original)

465. On the metaphor of “trace” see also Langacker (1999b: 93): “The occurrence of psychological events leaves some kind of trace that facilitates their re-occurrence. Through repetition, even a highly complex event can coalesce into a well-rehearsed routine that is easily elicited and reliably executed. When a complex structure comes to be manipulable as a ‘pre-packaged’ assembly, no longer requiring conscious attention to its parts or their arrangement, I say that it has the status of a unit.”


The reinforcing effect described by Langacker is well known in cognitive psychol-ogy under the term “memory trace” (or Gedächtnisspur in the German literature) (cf. Anderson 1996: 183ff.). Ratcliff and McKoon (1981) have in fact found experimental evidence for the fact that once a memory trace is laid down, it is strengthened with every use. They were able to show in a multitude of recognition tests that both cognitive accessibility to memory content and the strength of its activation varied depending on the strength of a trace. With increasing practice, the memory trace became con-solidated, and consequently information could be more easily retrieved.466 As a result, the cognitive effort was reduced, and information retrieval became a cognitive routine. Anderson formulated the following rule from this:

The speed and probability of accessing a memory are determined by the memory’s level of activation, which in turn is determined by how frequently and how re-cently the memory has been used. (Anderson 2010: 159)

Due to the varying frequency of retrieval the concomitant varying level of activa-tion, there are innumerable potential degrees of consolidation of memory traces. The memory, understood as a sum of memory traces, is constantly developing. No input remains ineffective.467

The trail effect can be substantiated by neurological research. Meanwhile, findings show that every input-conditioned process of change in memory traces is reflected in modified signal transmission between neuronal cells (cf. Squire 2004). Every neuronal cell obtains information from several thousand other neurons via innumerable contact points, the synapses. However, not every signal that a neuron receives is so strong that it triggers this neuron to fire in turn. This only occurs above a certain signal strength. Since each neuron only disposes over a single so-called “axon” – a “transmitter cable” via which a neuron fires impulses – it only has the option of either sending a signal to another neuron or remaining silent.

For information to be accessible over a longer period of time, i.e. for it to be-come part of the long-term memory, synaptic connections between neurons have to be strengthened. The neurologist Hebb (1949) already expressed the assumption that fixed circuits, so-called neuronal networks, only emerge when the respective neurons are repeatedly stimulated so strongly that they fire. Accordingly, information is repre-sented in the mind as a pattern of neuronal activity. Recent neurobiological findings not only support Hebb’s assumptions (cf. Martinez, Barea-Rodriguez, and Derrick

466. Cf. Newell and Rosenbloom (1981). They designate this repetition effect the “power law of practice”.

467. I have already pointed out that Bartlett already represented a dynamic memory model such as this. At one point he writes that the function of memory is “not the re-excitation of innumer-able fixed, lifeless and fragmentary traces. It is an imaginative reconstruction, or construction, built out of the relation of our attitude towards a whole active mass of organised past reactions of experience, and to a little outstanding detail” (Bartlett 1932: 213).


1998: 211–217), but also provide evidence for the existence of the electrochemical processes responsible for the formation of neuronal networks.468

Today, neurological memory research posits that every perceptual experience leaves traces in the mind by changing the strength of effect in synaptic connections between neurons. This leads to a structural change in the respective neuronal network.

Every new experience that we gain is integrated into the existing network by means of new neuronal connections; it encounters old memory networks, which it ac-tivates because it represents related or, alternatively, contradictory content. The new awakens the old and through association and consolidation becomes part of the old. (Translated from Fuster 2005: 12)

The same or similar experiences consolidate already existing neuronal connections. In this process, these memory traces only become permanently stored in the case of so-called “long-term potentiation” (cf. Spitzer 1996: 45ff.). Long-term potentiation (LTP) designates the more efficient signal transmission between synapses as a result of recurrent experience. By means of LTP, synapses in the neuronal network are changed such that newly emerging patterns encode the new memory content. As a consequence, the signal transmission between synapses not only improves, but also becomes more stable and resistant to change (Martinez, Barea-Rodriguez, and Derrick 1998). In addi-tion, the respective neurons react more sensitively and faster when the same neuronal path is activated.

The converse also applies: The weaker the synaptic connections become and the less they are strengthened by new input, the more difficult it becomes to access the encoded memory content. In the extreme case, the synaptic modifications once created by learning processes decay so much that signal transmission between the neurons no longer occurs at all. In this case, we no longer have access to the memory content. This process, which we call “forgetting”, ultimately contributes to the economy of our minds. Our brain constantly adjusts to our changing experience of the environment. The input of perceptual data correlates with neuronal changes.469

Against this background, the phrase “cognitive trails” loses some of its meta-phoric character. Analogously to Keller’s comparison, our neuronal reality really does seem to be such that similar experiences are reflected in the formation and

468. Cf. Frey and Morris (1997) and the compact overview in Fields (2005). The process of firing, the signal transmission between neuronal cells, is an electrochemical process that can be more accurately described as the production of enzymes and proteins whereby the synaptic connections are either strengthened or inhibited.

469. In this context neuroscientists also speak of the “neuroplasticity” of our brain. Dependent on regular cognitive demands, certain areas of the cerebral cortex develop more, i.e. neuronal connections are particularly dense and fixed there. Elbert et al. (1995), for example, have been able to experimentally verify that the cerebral cortex area representing the left pointer finger increases in size when someone learns to play the guitar or violin to the same degree as when learning Braille. This is because in each case similar fine motor skills are required.


consolidation of neuronal network structures. And analogously to a trodden trail, neuronal networks disappear to the same degree that the frequency of their “use” declines, i.e. that new perceptual experiences do not occur to activate the respec-tive neuronal paths (cf. Bartsch 2005). However, the metaphor of the cognitive trail becomes misleading where it implies that there are individual paths instead of a rhizomatic network of paths.470

Let us note: Cognitive entrenchment has a verifiable neuronal correlate and repre-sents an emergent phenomenon – in the sense of a phenomenon of the third kind – that is the result of a multitude of uniform schema instantiations. The fact that the hypoth-esis of cognitive entrenchment and routinization is compatible with both a linguistic theory of language change and neurological insights into human memory activities is, in my view, powerful evidence for its realistic substance.

Following this excurse, I now return to semantic issues in a stricter sense. With the use of an example I demonstrate how recurrent schema instantiations entrench an evoked frame.

6.5.3 Type frequency: an example

The transition from (quasi-)explicit to implicit predications occurs gradually, and only seldom are we conscious of when the threshold to implicit knowledge has been crossed. If someone automatically reacts to the sentence Anything is possible by ver-bally uttering Toyota, we can be fairly certain that this person has learnt a predicative attribution through recurrent schema instantiations. Before turning to a complicated case in Chapter 7, which examines which recurrent schema instantiations lead to the establishment of default values through high token frequency, I would first like to take a literary example to illustrate how high type frequency can increase the cognitive presence of a reference object.471

Normally it is impossible to empirically demonstrate the gradual increase in the cognitive awareness of a reference object for one simple reason: To capture all the recurrent schema instantiations that lead to the cognitive entrenchment of a type (or schema) in a person p, the analysis of all perceptive input that concern the respective schema instantiation must be available.472 The following example is an exception. Here we do indeed have access to the entire perceptive input. That is to say, the data of the

470. In addition the metaphor does not, of course, indicate in any way what the path consists of, how long it will last once it is “established”, under what condition it changes, etc. This is where the explanatory power of the metaphor reaches its limits.

471. Cf. also Bartsch (2005). Bartsch also recurs to a literary example, namely Proust’s In Search of Lost Time, to investigate cognitive entrenchment. The cognitive-theoretical premises of her study largely correspond to those explained here.

472. A similar process applies to the entrenchment of instances through a high token frequency.


input comes from a novel, and the readers have no access to any other information source that could contribute to the entrenchment of the type or token.473

The object of analysis is the particle probably. Generally, this particle, like every function word, has the status of an instance in that it particularizes the knowledge mode of an addressed entity. However, as already argued in Section 6.2.2, some parti-cles can be qualified in more detail in a predicative form. To be sure, this case counted as a marginal phenomenon from an empirical point of view because a hyperonym reduction is not possible in the case of function words, and consequently we have no method available to determine the slots of the frame that the word probably evokes. However, for the following analysis this circumstance presents no obstacle, since it is not a matter of capturing the dimensions of knowledge that the word probably alludes to, but rather of outlining how (quasi-)explicit predications specify and entrench the semantic unit (in Langacker’s sense) of the particle.

I proceed by first quoting the relevant text passages and then analysing them with respect to their (quasi-)explicit predications. This occurs in three blocks. In order to roughly place the passages in the order of the larger narrative context of the novel, I explain the most important events – also with the necessary brevity – that have tran-spired in the meantime.

The novel in question is written by the Japanese author Haruki Murakami and is titled South of the Border, West of the Sun. The novel is set in Tokyo and tells the story of a thirty-year-old nameless first-person narrator who meets an old friend after over 18 years. As a child, the narrator felt strongly attracted to this girl, named Shimamoto. In the meantime, he is married, has two children, and earns his living running two jazz bars in the centre of Tokyo. When Shimamoto turns up at one of the bars one day, they both recall their shared childhood. After meeting each other several times, the narrator finds it more and more difficult to resist the lure of his former love for her. But suddenly, Shimamoto stops coming to the jazz bar. In this context, the narrator uses the word probably for the first time. Later, as we will see, the word gains a key interpretive function, and presumably appears for this reason in italics at the crucial points in the novel.

Passage 1 (Murakami 2003: 134; italics in the original):May came and went, then it was June. Still no Shimamoto. I was sure she had gone for ever. Probably I won’t be able to come here for a while, she’d written. It was this probably and for a while and the ambiguity inherent in them that made me suffer.

The analysis is brief. The particle probably shifts into thematic focus due to the meta-linguistic reference. Nominalization allows us to predict something about the retrieved

473. Although it is generally not impossible that cognitive entrenchment can also be retraced to the invocation of information from memory, such invocation processes alone cannot sufficiently entrench slots and values in a frame.


reference object, namely the mode of knowledge expressed.474 Three (quasi-)explicit predications occur in passage 1, and for the sake of clarification I reformulate them so that the frame-evoked expression stands in the subject position.475

(25) The (word) probably refers to Shimamoto’s temporary not-able-to-come. (26) The (word) probably adds to the suffering of the first-person narrator. (27) The (word) probably expresses vagueness.

While (26) and (27) are explicit predications, the predication in (25) is of a (quasi-)explicit nature. The next two blocks of analysis do not refer to individual passages, but to several content-related text passages.

The story takes its course. After six months’ absence, Shimamoto reappears. The following conversation occurs between her and the first-person narrator:

Passage 2 (Murakami 2003: 146; italics in the original):“But I did write you a note saying I wouldn’t be back for a while, didn’t I?”“ For a while is a phrase whose length can’t be measured. At least by the person who’s waiting,” I said…. “And probably is a word whose weight is incalculable.”

And after a time, Shimamoto herself speaks of her long absence.

Passage 3 (Murakami 2003: 149–150; italics in the original):“ Six months is a long time,” she said. “But most likely, probably, I’ll be able to come here for a while.”

“The old magic words,” I said.“Magic words?”“Probably and for a while.”…“ You’re here,” I continued. “At least you look as if you’re here…. I reach out my hand to see, but you’ve hidden yourself behind a cloud of probablys. Do you think we can go on like this for ever?”

In a holiday cottage where the two of them secretly spend the weekend together, the first-person narrator opens himself up completely to Shimamoto.

474. In addition, reference to the knowledge mode shifts into focus due to an obvious contra-diction. Although the narrator pretends to be certain that Shimamoto has disappeared for ever, at the same time he suffers because Shimamoto has written that she would probably not return.

475. Cf. Lönneker (2003a: 73). Konerding (1997: 64) notes that nominalized linguistic units can be both syntactically and functionally addressed or topicalized. I make use of this property here (and in the following); predicative attributions are thus more easily recognizable. Syntactic reformulations such as this are also legitimated by the fact that predications do not restrict themselves to the syntactically marked nominative (cf. Konerding 1993: 165).


Passage 4 (Murakami 2003: 156; italics in the original):“ Shimamoto-san,” I said, “after you left, I thought about you for a long time. Every day for six months, from morning to night…. I don’t ever want to lose you again. I don’t want to hear the words for a while any more. Or probably. …”

However, after they spend the night together, the first-person narrator wakes up to find Shimamoto has disappeared without a trace.

Passage 5 (Murakami 2003: 165; italics in the original):Once again Shimamoto had disappeared from my life. This time, though, leaving nothing to pin my hopes on. No more probablys. No more for a whiles.

In these passages, 2 to 5, there are numerous other explicit and (quasi-)explicit predica-tions. They are, in chronological order:

(28) The (word) probably is of incalculable weight to the narrator. (29) The (word) probably refers to Shimamoto’s not being able to come for a while. (30) The (word) probably is a magic word (of Shimamoto’s). (31) The probablys form a cloud. (32) The probablys serve as a hiding place for Shimamoto. (33) The (word) probably is a word that the narrator no longer wants to hear (from

Shimamoto). (34) The (word) probably is not left behind by Shimamoto. (35) The (word) probably serves as something that the narrator can hang his hopes on.

Towards the end of the novel the modal particle probably is brought up two more times. Both passages can be dealt with together.

When he arrives home, the narrator confesses to his wife. However, his thoughts constantly revolve around Shimamoto’s disappearance:

Passage 6 (Murakami 2003: 200–201; italics in the original):She just vanished, along with her secrets. No probablys or in a whiles this time – she just slipped away silently. Our bodies had become one, yet in the end she refused to open up her heart to me.

However, it becomes increasingly clear to him that he will never see Shimamoto again. Finally, he makes a conscious decision for his family – and plans to begin his new life the following day. The final passage from the novel reads:

Passage 7 (Murakami 2003: 185–186; italics in the original):I don’t know if I have the strength to care for Yukiko and the children, I thought…. it is my turn to weave dreams for others. That’s what I have to do. Such dreams may have no power, but if my own life is to have any meaning at all, that is what I have to do.Probably.


Two final predications arise from passages 6 and 7.

(36) A probably is no longer possible (concerning Shimamoto’s return). (37) The (word) probably refers to the fact that the narrator must weave dreams for

others if his life is to have meaning.

What the particle probably refers to in passage 7 is however by no means as clear-cut as the resolution in (37) suggests. At least two other alternatives can be given:476

(37a) The (word) probably refers to the fact that the life of the narrator has meaning when he weaves dreams for others.

(37b) The (word) probably refers to the fact that the dreams that the narrator weaves for others have no power.

Systematic ambiguity is inherent in the single-word sentence in passage 7. Numerous interpretations compete with each other. For the sake of simplicity, I ignore (37a) and (37b) in the following and concentrate solely on (37).

What happens to the particle probably in the course of the quoted passages? Let us recall for a moment that, according to Searle, every expression that distinguishes an entity from another entity and identifies it is a referential expression. In this sense, the nominalizations in passages 1, 3, 5, and 6 demonstrate that the modal particle probably (in contrast to other modal particles like perhaps, certainly, etc.) is metalin-guistically addressed, that is, is employed referentially. By means of the predications (25) to (37), something is said in quite different respects about the epistemic modality of the probable.

The epistemic modality – in this case, in the rank of an activated frame – is high-lighted from other modalities by predicates that are all qualitatively different to each other.477 Accordingly, the token frequency is not high, but the type frequency is. The reference object is frequently addressed and a corresponding frame is frequently evoked. The high token frequency entrenches the frame.

Each time probably is referred to in a non-metalinguistic way, this particle simul-taneously figures as a filler in a superordinate frame. This is the case in three places. Firstly, in passages 1 and 3, probably is introduced as an instance that modally specifies the case frame evoked by the verb come. Apart from (37), all the predications men-tioned belong to this superordinate case frame. Or more precisely: The predications (25) to (36) all refer to the statement made by Shimamoto that she probably won’t be able to come for a while (cf. passage 1). Table 6 illustrates this superordinate case frame.

476. In addition, if we extend the range of the modal particle, the following (unlikely, but not impossible) readings are even possible: (i) The (word) probably refers to the task of the narrator to weave dreams for others. (ii) The (word) probably refers to the fact that otherwise no one will weave dreams for others.

477. With one small exception: (25) and (29) are identical. Of course, this does not mean that there is a high token frequency.


Its slots are named according to the semantic roles in von Polenz’s classification,478 but are extended by the category “modality”, by which I mean changes made in knowledge mode due to modal verbs and/or modal particles (in the sense of Busse 1991a: 187).479

Table 6. Case frame for come, based on the sentence Shimamoto can probably not come for a while

Frame Slots Instances

come Agent ShimamotoModality can

probablynot

Temporative for a while

Only in passage 7, where probably is not metalinguistically addressed, does the particle not refer to Shimamoto’s not being able to come or being able to come. What happens here when we follow the reading in (37)? The particle forms an instance in another case frame. It does not modally relativize that Shimamoto cannot come for a while, but rather that the first-person narrator must now weave dreams for others. Table 7 summarizes the slots and instances in the case frame weave.

Table 7. Case frame for weave, based on (37) or the sentence The first-person narrator must probably find dreams to weave for others

Frame Slots Instances

weave Agent First-person narratorModality must

probablyEffected object dreams for others

The point of the use of probably in reading (37) consists in the fact that the instance probably evokes a (sub-)frame that is already occupied by instances. Now that passages 1 to 6 have particularized the meaning of the word probably for the narrator in a variety of ways, the word is, as it were, semantically charged. At least some of the predica-tive attributions (25) to (36) are inferred by the reader. Thus, in (37) a new, emergent meaning of the word probably arises, and in fact the whole of passage 7 is shifted into

478. In this context, I have chosen von Polenz’s reference points (and not those that result from the matrix frame action) because we are initially dealing with a sentence-semantic analysis of the relevant fillers.

479. Cf. Vater (2005: 72). Vater sees modality as a separate reference area. According to Vater, both modal verbs (can, must, shall, may) and particles like perhaps, certainly, of course, probably refer to the manner in which something occurs.


another light. If this should mean that the narrator must weave dreams for others – for instance, his family – and that his life will only gain meaning in this way, then the final weakening of this by probably represents far more than an epistemic-modal relativiza-tion of what has been said. We know, for instance, that the word probably contributes to the suffering of the narrator (cf. (26)) and is of incalculable weight (cf. (28)), yet it is a magic word (cf. (30)), both in a negative sense because it forms a cloud behind which one can hide (cf. (31), (32)) and a positive sense in that all hope hangs on this word (cf. (35)). Different interpretation perspectives open up depending on which of these predications the reader infers.

However, since the token frequency of the predications identified is extremely low – each predication only occurs once – none of the predications can become en-trenched. In keeping with its high type frequency, only the epistemic uncertainty that is expressed in all the usage variants of the particle probably becomes entrenched. It becomes for the narrator a metaphor for insecurity and uncertainty.

I would like to outline one final interpretation variant. If we consider passage 7 – again in the reading of (37) – in isolation to begin with, the particle probably only casts doubts on whether the narrator can really weave dreams, give his life meaning, and thus care for his family. This is the reading of probably as illustrated in frame 2 of Figure 9. Let us assume that the reader remembers that probably is a magic word (cf. (30)), but ultimately forms a cloud behind which one can hide (cf. (31), (32)). These three predications p(30), p(31), and p(32) may originate in another frame (Figure 9, frame 1), but nevertheless feed into the semantic interpretation of probably in passage 7. These two new aspects of meaning would be integrated into the emergent meaning of the particle (in Figure 9; frame 2′). In this interpretation variant, weaving dreams and giving his life a new meaning would become an unachievable goal for the narrator.480

Let us recapitulate and generalize the results. Firstly, using a small database it is rela-tively easy to establish modal particles like the word probably in the example analysed as words that evoke a frame. Three closely related linguistic means facilitate this: nomi-nalization, syntactic-functional topicalization, and metalinguistic reference. Secondly, frames that are established in this manner specify a reference object more closely using numerous predications, thus distinguishing it from potential other reference objects. Thirdly, the words that a frame evokes also figure themselves as instances of a superor-dinate frame. Should the superordinate frame change, it is possible that some instances of the subframe are also actualized. In this manner, new meanings emerge.

480. The fact that readers inferentially bridge gaps has already been dealt with in reception aesthetics (for instance, by Wolfgang Iser) from other perspectives. Thus, Iser (1975) sees readers performing important active mental operations in the normalization of the uncertainty inherent in literary texts. According to Iser, a text exhibits a set of “gaps”, which invite reader participation. In my view, this reception-aesthetic approach could be expanded and semiotically consolidated through frame semantics. Iser does not satisfactorily explain the extent to which his textual gaps (Leerstellen) and what he calls (following Roman Ingarden) “schematic perspectives” (schema-tische Ansichten) are inherent in the nature of linguistic signs.


In my analysis of the particle probably, I did not elaborate on the semantic dimension of the expression. Rather, my description focused on the gradual entrenchment of the frame that the particle evokes. This was caused by a high type frequency. However, it is not possible to make a prediction about which default values would possibly be en-trenched in the evoked frame due to the low volume of data. The token frequency was not only too low, but the tokens also exhibited maximum dispersion. Only one predica-tion occurred twice.481 Without a quantitative corpus-based analysis one can draw no conclusions about the establishment of default values. Now, in the final chapter, I con-duct a comprehensive corpus analysis to illustrate just such a cognitive entrenchment.

481. Cf. (25) and (29).

´

(instance in case frame come: Shimamotocan probably not come for a while)

probably : Frame 1(instance in case frame weave: The narrator must probably weave dreams for others)

probably : Frame 2

probably: emergent frame 2́

Figure 9. Potential emergent meaning of probably in passage 7, arising from the blending of the (quasi-)explicit predications p(30), p(31), p(32) from frame 1

chapter 7

Frames in discourseThe financial investors as locusts metaphor

In the previous section, frames were no longer being understood as representational formats; instead, they were being employed as tools for the analysis of U-relevant knowledge. This led to a change in perspective, which is also significant for this chapter.

The fact that the structural constituents of frames can easily be operationalized as categories of semantic analysis already had become apparent in the course of de-termining these constituents. If a linguistic expression – or more precisely, its form side – evokes a frame, this has, among other things, a schematic character in that it contains numerous slots that can be filled linguistically with predicates. This results in three levels of abstraction, which are of great use for linguistic analysis. On a low level of abstraction there are fillers, which correspond to the predicates attributed to the ref-erence object. Implicit predications (default values) are distinct from these in that they must be retrieved from memory. This only succeeds when the conceptual content of these predicates is already entrenched. Finally, slots constitute the most abstract units. They provide opportunities for knowledge to be specified through explicit or implicit predications. Slots have no cognitive correlate since they are purely analytical values. Precisely because of this they are of particular value for corpus analysis.

In the following, when investigating the question of how default values can be em-pirically determined, I attempt to approach an answer by means of predication analysis. The tenor of my considerations is that default values constitute themselves within a discourse (in the sense of Busse and Teubert 1994) through recurrent uniform schema instantiations. The occurrence of many different instances in a schema entrenches the schema, whereas the recurrence of similar or the same instances leads to the forma-tion of a new schema. In the former, there is a high type frequency, in the latter a high token frequency.482 Frames serve as tools, or instruments, with which type and token frequencies can be determined on the basis of large text corpora. Since frames struc-ture the content side of symbolic units and linguistic expressions (morphemes, words, phrases, sentences, argument patterns, etc.) form symbolic units of varying complexity,

482. Cf. the explanations in Chapter 6. Diverging from Peirce, by “token” I do not mean a singular sign with a particular material sign body in a particular realization context, but rather the occurrence of the same sign (i.e. the same phonological unit in Langacker’s sense) in simi-lar contexts. In a frame-semantically relevant sense this is precisely the case when the same phonological unit refers to the same reference object and specifies it with the same predicative attributions.

The FINANCIAL INVESTORS AS LOCUSTS metaphor


symbolic units of any form of complexity can be similarly analysed as required using the same frame-semantic tools.

The research focus of this case study is the metaphor locust (Heuschrecke), which has asserted itself in the German-speaking public sphere as a pejorative and stigmatiz-ing designation for a group of financial investors. The metaphor was coined by the then chairman of the German Social Democratic Party (SPD) Franz Müntefering, who in mid-April 2005 likened the behaviour of certain financial investors with swarms of locusts (Heuschreckenplagen). Müntefering’s comparison served to trigger the so-called “capitalism debate” (Kapitalismus-Debatte), a very heated controversial discussion on the necessity of politically regulating the financial conduct of private equity companies and hedge funds. On 10 May 2005, the reputable newspaper Süddeutsche Zeitung pro-nounced “No statement has led to so much controversy in recent years as this statement about the power of capital”. Indeed, in 2005 Heuschrecke even managed to gain fourth place in the annual German linguistic review “Word of the Year”.

The epistemological interest of this empirical study is, for one thing, to use a corpus-based analysis to determine the meaning of the metaphor, which gradually formed in the context of the capitalism debate. An additional goal is to apply and test the frame-semantic categories developed in the last two chapters in a corpus-based analysis. The following deals less with individual processes of understanding than with the issue of the extent to which the analysis of – in this case, written – corpora can pro-vide insight into which U-relevant background assumptions (default values) gradually become established within a discourse context. An analysis such as this has a proba-bilistic character; it posits that the frequency of occurrence of a linguistic element – in this case, predications – affects the conceptual representation of linguistic units.

7.1 Preliminaries

To clarify the methodological premises of the following corpus-based analysis, the first section explains some similarities with and differences to other studies that also em-ployed frames as instruments for corpus-based analysis. Section 7.1.2 is then devoted to the specifics of the linguistic research area “metaphor”. This explains the underpin-ning understanding of metaphors as well as their cognitive and discourse-theoretical relevance.

7.1.1 Frames as an instrument of corpus-based analysis

So far, three studies have employed matrix frames for corpus-based analyses: Fraas 1996a, Lönneker 2003a, and Klein 1999. In addition there has been a range of indi-vidual frame-based text analyses. I briefly outline these insofar as important points resemble or differ from my operationalization proposal. Sections 7.3.1 to 7.3.4 then elaborate on my method in the form of a guideline.

Chapter 7. Frames in discourse 317

Fraas (1996a) employs frames for purposes of discourse analysis.483 Using the historical change in the concepts of ‘identity’ and ‘German’ as examples, she illustrated how frames can be employed as an instrument of historical and epistemological analy-sis. Her study provides evidence that, under the influence of the discourse surrounding German unity at the end of the 1980s, the content features of the concepts changed significantly. She was able to demonstrate that the meaning variation was caused by forming part of different discourses. Fraas’ empirical procedure shows some parallels to the method I have chosen:484

– The departure point is the assumption that tacitly assumed knowledge within a language community is accessible via common predications (cf. Fraas 1996a: 5).

– In the tradition of Minsky (1975: 246–247), frames are treated as lists of meaning-ful questions that systematically govern the contextualization of the expression activating the frame. Frames are thus predication frameworks and grids (cf. Fraas 1996a: 27).

– Following Konerding, hyperonym type reduction of the frame-evoking expression can determine which slots (or questions) belong to a frame (cf. Fraas 1996a: 26).

– Predications are seen as linguistic realizations of the conceptual content of an expression, so that the cognitive presence and prominence of U-relevant facets of knowledge can be deduced on the basis of the set of (quasi-)explicit predications (cf. Fraas 1996a: 78).

– Precise statements about the semantic structure of a frame can be made via a qualitative analysis of realized predications on the one hand, while, on the other, the frequency of occurrence (in the sense of the represented token frequency) of individual predications can provide insight into which default values are in the process of emerging (cf. Fraas 1996a: 72–74).485

Alongside the frame-based analysis of conceptual knowledge, Fraas also attempts to include the attitudes of language speakers to this knowledge (Fraas 1996a: 29–30). By systematizing contextual cues (such as the use of modal particles, references to

483. What I mean by “discourse” in this context is, in the sense of Busse and Teubert (1994), a virtual set of texts that refer to each other through shared thematic reference thus constructing an epistemic-functional unit. On the connection between discourse analysis and frame seman-tics, cf. Ziem (2008a).

484. However, as previously outlined, Fraas links her conception with Bierwisch’s two-level semantic model. This causes fundamental problems; cf. Chapter 2. Nevertheless, Fraas’ orienta-tion towards a modular model has no consequences for her empirical procedure.

485. Fraas does not seem to have considered the phenomenon of high type frequency, i.e. of the cognitive entrenchment of slots. Fraas (1996a: 26) presupposes that a high token frequency determines the contextualization potential of an expression without actually justifying this as-sumption herself.


knowledge sources, etc.) she arrives at five types of speaker attitudes486 that serve her as analysis matrices. However, in my view, this socio-pragmatic extension bears the disadvantage of relativizing her methodologically sound and validated frame analysis to some extent, because typification of speaker attitudes using concrete contextual cues occurs at best heuristically and at worst arbitrarily. Since Fraas’ attitude types are gained inductively from the analysed text material alone, they partially overlap, and it can be assumed that one contextual cue can be attributed to several different attitude types. As a consequence, the attitude types identified fulfil neither the criterion of distinctivity nor that of exhaustivity. For this reason, I restrict myself exclusively to a corpus-based analysis using frames.487

Fraas (2000, 2001) returns to the frame concept in later studies to substantiate their relevance in describing knowledge of social interaction, in particular from a construc-tivist perspective. However, her focus is on theoretical issues in the broader field of socio-cognitive approaches, and where empirical case studies occur, Fraas reduces the frame analysis to co-occurrence analysis. Thus, although she arrives at contextualiza-tion and interpretation patterns (Fraas 2001: 64–65) that take on the form of frames in processes of language understanding, in comparison to other analyses (like those of Lönneker 2003a and Fraas 1996a), the co-occurrences she identifies only encom-pass a fraction of U-relevant knowledge. Without doubt, the co-occurrence analyses Fraas undertakes in following her aims do fulfil their purpose in that, with very little analytical effort, they can plausibly demonstrate with which contextualization grids word meanings are conceptualized. However, solid claims about dominant facets of knowledge (slots) or established default values are just as little to be expected as, for instance, insights into how a metaphor is conceptually formed within a discourse. If the target is to answer such questions, the analytically far more complex path via matrix frames, as proposed here, must be taken.

This path is also chosen by Lönneker (2003a), who introduces an electronic cor-pus collation program that can annotate large volumes of instances (fillers and default values) collected in the Internet. The empirical results of her study demonstrate that the predicator classes (slot types) in a frame that Konerding identified are suited for corpus-based analysis. Further, Lönneker uses the results of her analysis to make minor adjustments to the matrix frames developed by Konerding (Lönneker 2003a: 252–277; cf. Section 6.3.2). I also use these improved matrix frames for my corpus-based analy-sis. In addition to the aspects mentioned in connection to Fraas’ study (Fraas 1996a), I share the following premises with those in Lönneker’s approach:

486. Which are as follows: (a) reference to other speakers’ utterances, (b) formulation of one’s own opinion, (c) appeal to facts and their use as arguments for one’s own opinion, (d) com-mentary on the degree of validity of one’s own opinion, (e) evaluation.

487. Should one wish to include speaker attitudes, this is best achieved, in my view, by an ex-panded frame analysis (and not through a second analysis grid). This is because the contextual clues under analysis function themselves as instances within a frame. Preliminary evidence can be found in Konerding (1996) and Tsohatzidis (1993).


– The activation of a frame corresponds to a language user’s reference act. The activa-tion is triggered by a linguistic form (Lönneker 2003a: 64–66), whereas realized, i.e. (quasi-)explicit predications, are seen as fillers in frames.

– Relations between an evoked frame and invoked concepts, i.e. potential in-stances, are produced by verbs.488 Every verb refers to a particular slot (Lönneker 2003a: 77), and every verb thus forms an instance of precisely this slot.489

– Frame elements (slots, fillers, default values) are loosely connected with each other via contiguity relations (Lönneker 2003a: 60). Both contiguity relations and similarity relations (the latter not mentioned by Lönneker) group frame elements together. The correlations between frame elements form particular structural pat-terns (cf. Section 5.2.2).

– Finally, matrix frames can be arranged into a hierarchy, in contrast to what Konerding implies (Lönneker 2003a: 93; cf. Figure 5 in Section 6.3.2). This makes visible the heredity relationships of slots. The frame hierarchy is of use for em-pirical analysis in that it shows which slots are shared by different frames, and therefore, for instance as in the case of a metaphor analysis, which shared and, al-ternatively, different options of semantic specification are permitted by the source and target domains respectively (in this case, locusts and financial investors).

Nonetheless, in contrast to Fraas (1996a), Lönneker’s study remains open as to the extent that the frames identified are constituted specifically to discourse. Because her analysis is not designed for diachronic comparison, neither historical change nor vari-ance in conceptual knowledge from an individual-cognitive and/or situative perspec-tive plays a notable role.

Finally, in addition to Fraas and Lönneker, Klein (Klein 1999; Klein and Meißner 1999) also employs frames for corpus-based text analysis.490 However, for a variety of reasons the procedure introduced there differs from the method that I favour. Although Klein understands frames as a useful instrument for investigating the social distribu-tion of knowledge, his method of operationalization nonetheless differs in significant points. Klein’s research project (documented in Klein and Meißner 1999) essentially concerns empirically determining the conceptual knowledge about key economic con-cepts and conceptual relationships that young adults have at their disposal. The inves-tigation procedure occurs in two stages. Firstly, Klein and Meißner attempt to identify

488. In contrast to Lönneker I dispense with an additional frame element labelled “subslot” for the reasons stated above.

489. Thus, for instance, the verb look like is an instance of the predicator schema to ‘character-ize the form and colour’ in the matrix frame organism, or bark is an instance of the predicator schema to ‘characterize the ability’ of the organism; cf. Konerding (1993: 412).

490. Furthermore, some frame-based text analyses were conducted within the Berkeley FrameNet project (cf., for instance, Fillmore and Baker 2001; Ziem, Boas and Ruppenhofer 2014); I do not elaborate on this separately here; cf. Lönneker (2003a: 214–240), for an overview: Ziem (2014).


tacitly assumed knowledge by conducting a qualitative content analysis on the basis of approximately 300 newspaper articles along with news and current affairs broadcasts. In the second stage, to determine the extent to which young adults actually have this knowledge at their disposal, questionnaires were developed on various subject areas (such as “global competition”; “Germany as a business and investment location”) and the test subjects were asked to complete them. Finally, against the background of the tacit knowledge they identified, the qualitatively evaluated results allowed them to draw conclusions about the average expected conceptual variants.

Since the second stage of the analysis only concerns group-specific attitudes to-wards concepts and topics, this step is not necessary for my discourse-semantic meta-phor analysis. On the other hand, the first step is of key relevance. However, here I set a decidedly different accent. In contrast to Klein and Meißner, I posit along with Lönneker (2003a) and Fraas (1996a) that default values can be identified with the help of matrix frames. This analytical procedure has the advantage of not being de-pendent on purely qualitative content analysis and its concomitant interpretive heu-ristics. When investigating the globalization frame, Klein and Meißner restricted themselves to analysing a few slots like ‘subprocesses’ and ‘constitution conditions’ (cf. Klein 1999: 176); none of the other slots were taken into account. Therefore, Klein and Meißner prejudged that the selected slots exhibited a higher degree of saliency and had become cognitively entrenched in the discourse under investigation. There may be intuitive evidence for this assumption, but empirical evidence would be more convincing. Despite the disproportionately higher degree of analytical complexity, I prefer to equally include all the slots of the respectively relevant matrix frames to be able to provide statistically valid results about the cases in which a significantly higher type frequency occurs.

Apart from the corpus-based linguistic studies done by Klein and Meißner, Fraas, and Lönneker, Klein (2002b) and Holly (2001, 2002) have conducted individual text analysis with the help of matrix frames. Holly’s research focus was on a debate speech in St. Paul’s church in Chemnitz, whereas Klein investigated colonial discourse in the Wilhelminian era. Characteristic of both analyses was the use of frames as a herme-neutical tool to reconstruct the historical knowledge that is implicitly presupposed but not expressly stated in preserved texts. My own attempt to reconstruct historical knowledge in Gustav Freytag’s novel Soll und Haben using frame analysis is also com-mitted to such interpretive methods of content analysis (cf. Ziem 2005).

However, the hermeneutic employment of frames remains circular to the extent that the analysis of a text with the goal of determining typical variants of a frame already presupposes awareness of the frame – i.e. of the relevant slots and established default values.491 In other words: It is only possible to identify the knowledge that

491. Cf. also Heringer (1999: 128). In principle, this applies to any form of text analysis whose results are not based on large-scale corpus-based analysis and consequently are not largely obtained by means of qualitative content analysis. Purely qualitative content analysis to some extent always appeals to the intuitive evidence of the analytical results.


analysts themselves have at their disposal and that they themselves deploy for the interpretive and inferential processing of texts. At best, arguments for this herme-neutic circle can be validated by source analysis. However, this procedure cannot lead to quantitatively supported and thus generalizable results because, in practice, individual text analyses are conducted in such a way that a thorough reading of the pertinent literature leads to the insight that certain knowledge facets (slots) dominate more strongly than others. The dominant ones form the particular frame.492 But why do these slots in particular dominate? Are other slots entrenched in other texts? It is in this spirit that Heringer asks

…how frames and slots reveal themselves and how we recognize them. Somehow we must gain awareness of them. Of course, we have examples. The co-actors are differently coded, sometimes as an attribute, sometimes as a possessive arti-cle… But that is very little to begin with. Their selection presupposes the frame. (Translated from Heringer 1999: 128)

Heringer is incorrect about the final point. The selection of a frame need not presup-pose the frame. That is only true in the case of individual text analysis. I briefly outline the reasons for this here.

Firstly, one must differentiate between cognitively real frames and frames as instruments of analysis. Cognitive frames are activated by language users, they are usage events in Langacker’s sense and thus always specified by default assumptions (cf. Section 6.3.1). By contrast, empirical analysis is concerned with first eliciting the relevant default assumptions. All that a corpus-based analysis using frames can take recourse to are (quasi-)explicit predications. (Matrix) frames serve as an analysis grid; they are instruments to determine U-relevant knowledge, but do not represent mental entities as such. As instruments of analysis – but not as cognitive units – frames consti-tute empty formats that illustrate the potentials for specifying knowledge (predicators). Awareness of default values is thus not presupposed at all.

Matrix frames should not be employed unconditionally for individual text analysis because, on the one hand, these frames provide so many slots that individual texts could only actuate them in a highly selective manner and, on the other, because the pertinent slots could not be epistemically weighted as a result of their low type fre-quency. Only quantitatively supported analysis may provide insight into the degree of cognitive entrenchment. Text corpora are necessary for such investigations. The results arising from them can then be generalized to a certain degree (for instance, within a particular discourse context). Individual text analyses cannot live up to this require-ment. Their results have at best an indicator function.

492. For instance, Klein (2002b) comes to the conclusion that ten slots (like ‘genre, ‘agent’, ‘pa-tient’, etc.) dominate in colonial discourse. And Holly (2001) determines that a particular frame in the debate he analyses, namely that of the money-issuing bank, is preferred, in the sense of having a high type frequency. In addition, Holly identifies ten slots that contour the frame (among which are ‘key action’, ‘consequences’, ‘prerequisites’).


This finding applies particularly to frame studies of Anglo-American provenance. In the foreground is lexico-semantic analysis whose results are only seldom based on corpus analysis or psycholinguistic experiments.493 Matrix frames or similar analy-sis grids are lacking altogether. This results in glaring discrepancies in the analytical results. For instance, three studies on the meaning of the verb to run lead to starkly divergent results (cf. Langacker 1988c; Tuggy 1988; Gries 2006).

Although my empirical study follows the theoretical and methodological speci-fication that emerge from Chapter 6, and although it diverges significantly from the introspective heuristics of Langacker and others, this fact should not obscure the many shared premises that result from the holistic-encyclopaedic semantic conception. Differences concern aspects of the empirical operationalization, but not the theoreti-cal foundation on which the empirical analysis is premised.

7.1.2 Cognitive and discourse-related aspects of metaphors

In the following, when I apply the categories developed for frame-semantic analysis to metaphor analysis, this operationalization stands in close connection with the ho-listic premises of this book (cf. Chapters 3 and 4) that (i) linguistic structures are of a conceptual nature (encyclopaedic semantics) and (ii) conceptual structures are in turn anchored in our bodily experience (embodiment).494 In addition, I directly take up the schema-theoretical considerations of the previous two chapters, in particular the assumption that linguistic meanings form schematic units whose content is shaped by default values with different degrees of entrenchment.

What are the features of a metaphor? Why do metaphors in particular present an appropriate research subject for applying frames as an instrument of analysis? To begin with, I would like to sketch the theoretical premises upon which metaphors will be addressed in the following. From these, several reasons can be concluded as to why a metaphor analysis seems particularly relevant for our frame-semantic context.

In Metaphors We Live By, a study published in 1980 by George Lakoff and Mark Johnson, the authors represent the hypothesis that a large part of our conceptual system and thus (linguistic) action generally has a metaphoric character.495 According to their

493. Cf. for example Botha (1996), Watters (1996), Coulson (2001), Petruck (1996), and Sweetser (1999), among others. Baker (1999) constitutes an exception; Baker employs both psycholinguistic and lexicographic methods.

494. Cf. also Kalwa (2010); Meier (2010); Wengeler (2010), among others. Theses studies em-ploy the corpus analytical framework developed here.

495. In this sense, Sweetser represents the hypothesis that linguistic categorizations (which were treated in Section 5.1 in terms of their fundamental function for human cognition) are in part metaphorically motivated: “A word meaning is not necessarily a group of objective ‘same’ events or entities; it is a group of events or entities which our cognitive system links in appropri-ate ways. Linguistic categorization depends not just on our naming of distinctions that exist in


argumentation, metaphors play such a fundamental role for human cognition because they are manifestations not only of constitutive cognitive processes, but also of the conceptual knowledge structures with which cognitive processes operate. Example (1), for instance, shows a metaphoric mapping of conventional knowledge from the source domain foundation to the target domain theory.

(1) The theory X lacks a solid foundation.

This mapping process only succeeds if the semantic units [foundation] and [theory] form conceptual structural units with entrenched default values.

In addition, Lakoff and Johnson illustrate numerous examples of how metaphoric conceptualizations occur through the systematic pairing of source and target domains. Thus, the source domains used for the interpretation of knowledge facets in the target domain cannot be arbitrarily chosen, because a system of metaphors appears to form the basis of individual metaphoric language use. If an abstract metaphor organizes an entire system of metaphors in this sense, Lakoff and Johnson speak of a “concep-tual metaphor”.496 The metaphor foundation in (1), for example, is based on the more abstract conceptual metaphor theories are buildings,497 which motivates many other metaphors. Consequently, conceptual metaphors only exist on the “type level” (Liebert 1992: 5ff.; Pielenz 1993: 71ff.; Böke 1996: 443–444); they are abstractions of concrete metaphors (“metaphor tokens” in the sense explained above). Compare the following examples:

(2) The theory X is built on sand. (3) The theory X is in danger of collapsing. (4) Chapter X is built on the insights of chapter Y. (5) His theory is a house of cards. (If a card is missing, it collapses.)

The metaphorical language use in Examples (1) to (5) is based on the fact that the con-cepts ‘solid foundation’, ‘built on sand’, ‘collapse’, ‘is built on’, and ‘house of cards’ all figure as instances of a superordinate frame, namely that of building (cf. Liebert 1992: 6).498 Thus, entrenched knowledge structures of a higher level of abstraction form the key premise of Lakoff and Johnson’s approach. These occur on the one hand in the form of

the world, but also on our metaphorical and metonymic structuring of our perceptions of the world.” (Sweetser 1990: 9)

496. Cf. for example the explanations in Lakoff and Johnson (1980: Chapter 18).

497. In keeping with the customary notation, I use capitals to typographically indicate con-ceptual metaphors.

498. Accordingly, Liebert (1992: 7) defines conceptual metaphors as follows (in translation): “‘Conceptual metaphor’ is … to be understood as a relation between two respective concepts in which each concept is itself to be understood as a pair consisting of a set of lexemes and a structural map, as well as an allocation of lexemes to this structural map”.


conceptual metaphors. But also those knowledge facets may be considered as entrenched that pass from the source domain via metaphoric mapping into the target domain.

[M]etaphors are analyzed as stable and systematic relationships between two con-ceptual ‘domains’. … Particular elements of the source and target domains are picked out through a combination of the source language used … and the relevant conceptual metaphor, a ‘mapping‘ – presumably stored as a knowledge structure in long-term memory – which tells us how elements in the two domains line up with each other. (Grady/Oakley/Coulson 1999: 102)

Reformulated in frame-semantic terms: The source and target domains form inference bases that are structured by frames in such a way that entrenched default values in the source domain occupy the corresponding slots in the target domain. Thus, a new conceptual structure emerges – that is, a metaphoric meaning.

The counterdistinction between source and target domain already indicates that Lakoff and Johnson posit a bipolar model in their theory of conceptual metaphors.499 The main function of metaphors is to infer new knowledge about the target domain on the basis of existing knowledge held in readiness by the source domain. Let us look at an example from the corpus to illustrate this:500

(6) Die eigentlichen Heuschrecken sind die Gewerkschaften. ‘The real locusts are the trade unions’ (Die Zeit, 4 May 2005)

While the semantic unit [locust] constitutes the source domain, the metaphorical content emerges by the source domain being mapped onto the target domain [trade unions] (Kövecses 2002: 33). Instances, i.e. default values, in the locust frame are instantiated in the corresponding slot of the frame evoked by the compound trade un-ions. Bipolar metaphor models such as this have been adopted and empirically tested in numerous case studies, also in the German-speaking context (cf. for instance, Liebert 1992; Pielenz 1993; Baldauf 1997; Böke 1997; Drewer 2003).

By contrast, little attention has been given so far to the so-called “blending the-ory” or “conceptual integration (CI) theory” as methods to investigate (the usage of) metaphors in public discourse (as exception, cf. Hart 2007). CI theory was initially developed by Fauconnier and Turner (1998a, 2002), and in the meantime has found great resonance in a variety of disciplines.501 Like Lakoff and Johnson, Fauconnier and Turner also stress that metaphors have a purely conceptual status and that they emerge as a result of inferential relationships between conceptual domains. However,

499. “The essence of metaphor is understanding and experiencing one kind of thing in terms of another.” (Lakoff and Johnson 1980: 5)

500. Example (6) is a replica of Müntefering’s comparison between locusts and financial inves-tors made by FDP chairman Guido Westerwelle.

501. With a special focus on linguistic issues cf., for example, the early overview in Coulson and Oakley (2000a), and in Coulson and Oakley (2000b).


a significant difference lies in the fact that, according to blending theory, not two, but four cognitive representational units participate in the construction of a metaphorical meaning. It is not cognitive domains (which provide relatively stable, or convention-alized, knowledge) that function as representational formats, but so-called mental spaces.502 Mental spaces differ from domains (at least in the manner that Lakoff and Johnson use the term) in that they are short lived. Language users only construct mental spaces in the course of meaning actualization. Unlike domains, mental spaces are emergent representational units. If the context changes, then the activated mental spaces change. In addition, inferences need not operate unidirectionally (as Lakoff and Johnson suppose), i.e. from one representational unit (the source domain) to the other (the target domain). In fact, actualized knowledge elements in the target domain can also lead to changes in the source domain.

For example, the metaphorical use of butcher in (7) illustrates the functions of the four representational units (cf. Grady, Oakley and Coulson 1999: 105ff.).

(7) This surgeon is a butcher.

The two spaces of input, surgeon and butcher, correspond to the source and target domains in CI. The generic space encompasses abstract information units that both in-puts share, but are, however, differently specified in each (cf. Figure 1). Some informa-tion from both inputs (here, information about the agent and the activity from input I and about the ‘agent’, ‘patient’, ‘location’, and ‘goal’ from input II) feeds into the blended space (the actualized metaphorical meaning of butcher), whereas other information is not reflected (such as information about the ‘patient’, ‘location’, and ‘goal’ from input I and about the ‘activity’ from input II).

However, not all the information that is mapped onto the blended space is of equal relevance. The values in two slots play a particularly important role: on the one hand, default values that relate to the slot ‘activity’ in input I and, on the other, default values that constitute the instances in the slot ‘goal’ in input II. Thus, knowledge is mapped onto the metaphorical meaning that concerns, on the one hand from input I, the fact that butchers – in contrast to surgeons – prototypically conduct their activity in a certain way (say, with hefty blows and rough cuts) with certain tools (like cleav-ers and butcher’s knives) and, on the other from input II, that surgeons prototypically conduct their activity with the aim of healing people or easing pain, or more generally formulated, to increase the life expectancy of the patient. The conceptual blending of these aspects of meaning from inputs I and II, represented by the rectangular boxes in Figure 1, allows a new aspect of meaning to emerge: Should a surgeon’s mode of conducting their activity vary so strongly from the prototypically expected mode that the goal of healing is not met or becomes impossible, the surgeon is seen as incompe-tent, because they fail to reach the key goal of their professional activity. The fact that this implies the surgeon is incompetent is not a result of a unidirectional mapping of

502. Cf. the explanations in Section 1.2.2.


aspects of input I onto input II, but rather of a more complex process of conceptual integration that amalgamates information from both inputs with each other (cf. Evans and Green 2006: 402–403). In other words, the fact that the activity of the butcher is evaluated negatively is not inherent in input I; butchers do not prototypically conduct their activity unprofessionally. Nor does input II imply such a pejorative evaluation. The fact that the surgeon is considered professionally incompetent – in the blended space – when he conducts his activity in a particular, inappropriate manner is therefore an emergent phenomenon that cannot be “calculated” compositionally.503

Input space I: BUTCHER Input space II: SURGEON

Blended space: SURGEON AS BUTCHER

Generic space

Agent: butcher

Patient: animals

Location: abattoir

Goal: preparing meat for saleetc.

Agent

Patient

Activity

etc.Agent: surgeonPatient: patientLocation: operating

room

Goal: healingetc.

Agent: surgeon(in the role of a butcher)

Patient: patient

Activity: dressing meat

Goal: healing

Activity: severingflesh

Activity: operation

Location: operating room

Goal

Location

etc.

Figure 1. This surgeon is a butcher: The metaphor butcher according to blending theory (cf. Grady, Oakley, and Coulson 1999: 105ff.)504

503. At this point my focus is more on the aspect of emergence and less on a detailed analysis of the metaphor. Certainly we can only approach any form of approximation to the metaphoric content when further knowledge facets are taken into consideration, for instance the implica-tion that a surgeon deals with living humans as if they were dead animals, that he treats living flesh like dead flesh, that he is thus callous, unfeeling, without compassion, etc. None of these aspects are dealt with in Figure 1.

504. To clarify that elements in the generic space occur in both inputs, thus motivating cross-domain mapping, Fauconnier and Turner have additionally linked these elements with a solid line. For instance, in Figure 1, the element ‘agent: Butcher’ in input I should be linked with ‘agent:


As an interim result let us note: Metaphors are constitutive cognitive phenomena be-cause they reflect the processes of dealing with experiences linguistically, of concept building, and of interpretation and organization of knowledge (cf. Drewer 2003: 11ff.; Turner 1998: 64). If metaphors fulfil the function of understanding aspects of an en-tity (input I) using aspects of another entity (input II), then this process proceeds via cognitive operations by which information from both inputs is integrated into a new structural unit. This blending of invoked knowledge units to become a new conceptual unit can be seen as a fundamental cognitive ability that affects a variety of understanding processes.505 As we have seen in Section 6.2, linguistic reference acts already involve conceptual integration processes; referentialization is not possible with-out simultaneous conceptualization of the reference object. Conceptual integration is always an inferential process that proceeds in a highly routinized fashion, recurs to foreknowledge, has cognitive units as its departure point (inputs I and II), and has a blended space as its result.

Following Grady, Oakley, and Coulson (1999), I refer henceforth to the theories of conceptual integration and conceptual metaphor not as competing, but rather as com-plementary approaches. Whereas Lakoff and Johnson focus on recurring conceptual structures that are based on metaphorical linguistic usage, blending takes account of the creativity of ad-hoc metaphorical constructions. They are complementary because ad-hoc constructions themselves rely on stable background knowledge (such as of butchers and surgeons, as is the case here). Strictly speaking, the generic space is a conceptual cognitive domain to the extent that it contains information that is shared by both in-puts.506 ‘Agent’, ‘patient’, ‘location’, ‘activity’, and ‘goal’ – these elements would presumably belong to every space that holds generic information about professional activities.507 In addition, although the selection of elements assigned to a mental space may be governed by the current context – in (7), for instance, the location at which the butcher carries out the activity plays no role – generally only particular elements occur in the input

surgeon’ in input II. I have refrained from setting these links so that the graphical illustration remains clear.

505. Blending is constitutive for the understanding of linguistic phenomena like morphologi-cal and syntactic constructions (Mandelbit 2000), conditional clauses and complex nominal phrases (Dancygier and Sweetser 2005), and performative speech acts (Sweetser 2000), but also non-linguistic phenomena like perceived movement, rituals, and comic strips, to name a few; cf. the overview in Coulson and Oakley (2000b: 182–186) and Fauconnier and Sweetser (1996).

506. “As conceptual projection unfolds, whatever structure is recognized as belonging to both of the input spaces constitutes a generic space. At any moment in the construction, the generic space maps onto each of the inputs. It defines the current space mapping between them. A given element in the generic space maps onto paired counterparts in the two input spaces.” (Fauconnier and Turner 1998a: 143)

507. For Fauconnier and Turner (1998a), a decisive motivation for positing a generic space lay in the empirical finding that knowledge constellations in generic spaces can adopt a conventional character and then function themselves as the source for further metaphorical mappings.


regardless of the context, in the same way that only particular default predications can specify a frame. As Fauconnier and Turner (1998a: 137) themselves observe, the ele-ments of a mental space are structured by frames. This therefore implies that the generic space corresponds to those slots shared by the frames that structure both inputs during the actual construction of a metaphoric meaning. I return to this point later.

To what extent then can blending theory be applied to the example of the locust? If a person, group, institution, etc. is designated as a “locust” it does not appear to be immediately clear which elements constitute the two inputs. Thus, it remains equally uncertain which elements are represented in the generic space and which elements are mapped into the blended space. This semantic uncertainty is not specific to Example (6). When the SPD politician Franz Müntefering introduced the metaphor of the financial investors as locusts, he could not prescribe its meaning. He initially had to explain its content himself,508 quite unlike the example of This surgeon is a butcher, in which the emergent metaphorical meaning is “simply there” without having to be explained.

In relation to Example (6) this means that if trade unions are designated as locusts, the knowledge facet ‘activity’ plays such a role in the generic space that an activity of locusts a1 (from input I) conceptually blends with an activity of trade unions a2 (from input II) in the blended space (cf. Figure 2). Let us now assume that ‘activity a1’ stands for ‘eating fields bare’ and ‘activity a2’ for ‘demanding high tariff agreements from employers’, then the emergent meaning consists in trade unions destroying companies with their demands for high tariff agreements (cf. the box in Figure 2). The meaning is emergent because it cannot be deduced from either input I or input II or from combin-ing elements of input I and II. It arises solely from the conceptual integration of certain information from both inputs with the assistance of additional default assumptions. Thus, in the case at hand knowledge needs to be actualized not only about what the locusts eat bare and what consequences this entails but also about what consequences high tariff agreements can have for employers.

However, it is important to see that such a conceptualization of (6) only represents one of many interpretation variants and therefore remains speculative. For instance, it could just as easily be the locusts’ characteristic of appearing in swarms that is pro-jected into the blended space. Trade unions would then be seen as institutions whose members are weak as individuals, but who as a group can achieve something.

In both interpretations a metonymic shift is involved in the constitution of mean-ing.509 On the one hand, locusts as causes represent the effects of their activities, i.e. the destruction of a fertile field. In this respect, a causal-metonymic shift from cause to

508. Or more precisely, Müntefering used a comparison in which the “swarm of locusts” served as a tertium comparationis. He observed that financial investors pounced on companies like swarms of locusts. Explanations of the financial investors as locusts metaphor occur throughout in early textual occurrences. They only disappear once certain aspects of meaning of the metaphor have become entrenched.

509. Which is why this is strictly speaking a metaphtonymy in the sense of Goossens (1995).


effect occurs. Similarly, in the second interpretation, the occurrence of a locust swarm stands for the damage that the swarm (rather than the individual locust) may cause. Which of the two knowledge facets is projected into the blended space? Knowledge of the activities of locusts or knowledge that locusts occur in swarms? Or both? In addi-tion, it is also possible that a number of other assumptions about locusts prove to be relevant to the metaphor.

Therefore, which elements are included in the blended space in the financial investors as locusts metaphor in (6) must remain open to begin with. In contrast to the butcher metaphor in (7), the emergent meaning does not appear able to be introspectively interpreted (as is the procedure of Fauconnier and Turner).510 What is the source of the difference between the locust and the butcher metaphor? CI cannot

510. Although the butcher metaphor stands out from the financial investors as locusts metaphor in this point, they both only differ from each other in scalar (not categorical) terms. Thus, I do not claim that knowledge about butchers and surgeons has emerged outside of dis-courses, but merely that the metaphor-relevant meaning aspects of butcher and surgeon are more strongly conventionalized.

Input space I:LOCUSTS

Input space II:TRADE UNIONS

Blended space: TRADE UNIONS AS LOCUSTS

Generic space

Activity: a1

Characteristics: ???

Location: ???

etc.

Activity

LocationSizeetc.

Activity: a2

Characteristics: ???

Size: ???Size: ???

Characteristics

Activity: a2

Location: ???

Size: ???

Location: ???

Activity: a1

etc.

etc.

Figure 2. Potential application of blending theory to the metaphor trade unions as locusts


answer this question adequately, since the approach presupposes the high salience of individual elements of the input without enquiring about the reasons for the varying degrees of entrenchment of the elements. In contrast, in the context of their theory of conceptual metaphors Lakoff and Johnson address how conceptual structures are en-trenched, but they concentrate exclusively on abstract structures involved in metaphor formation. They could explain the conceptual variance of the financial investors as locusts metaphor by maintaining that in (6) there is no conceptual metaphor forming the base of use for the metaphor. However, there are two reasons why this explanation would be unsatisfactory. Firstly, a great deal speaks for the existence of a conceptual metaphor person as animal.511 More important, however, is the fact that the theory of conceptual metaphors cannot generally explain which elements of the input (or of the source and target domains) will be metaphorically mapped. This theory cannot serve with an answer to the opening question either.

I would like to contrast this inadequate explanation with an answer motivated by frame theory. The two inputs in (7), butcher and surgeon, are direct hyponyms of occupation and consequently belong to the same matrix frame (namely person in a professional role). Since the slots in the surgeon and butcher frames are there-fore identical, any slot can theoretically figure as an element in the generic space. Thus, although to begin with numerous potential metaphoric mappings present themselves, only some slots are particularly salient, namely those that concern the professional activity itself, especially the necessary manual skills and abilities. This is because, with respect to all the other slots, it would be possible that the default values in the butcher and surgeon frames could be the same; however, the slots to specify the professional activity must always be occupied with different values so that the linguistic categories butcher and surgeon remain minimally distinct. Thus, it is no coincidence that especially these elements pass into both inputs and that the values of precisely these slots are the objects of metaphoric mappings. That only some selected elements of both inputs are then relevant to the metaphor can be explained by the fact that these slots (‘agent’, ‘patient’, ‘location’, etc.) are more strongly entrenched by a high type frequency than other slots and that the instances of these slots (like ‘butcher’ vs. ‘surgeon’, ‘animals’ vs. ‘patients’, etc.) are more deeply entrenched by a high token frequency than other instances.

By contrast the frames locust and trade unions (or financial investors) only exhibit partially identical slots, namely those slots of the hierarchically superordinate frames primary object, stable object, and entity.512 Despite this overlap, the two frames do not contain the same (or similar) default values. If a value were the same, it

511. At this point I do not elaborate on this any further; cf. the explanations on this is Section 7.2.3.

512. Both frames exhibit different slots because the frame that the word locust evokes inherits the slots of the hierarchically superordinate matrix frame organism, whereas the frame that the word trade unions evokes inherits the slots in the matrix frame institution/social role; cf. the frame hierarchy illustrated in Figure 5 in Section 6.3.2.


would already be a metaphor.513 This is where the crucial difference to (7) lies. Thus, the butcher and surgeon frames can easily exhibit some shared default values without any metaphoric mappings playing a role (for instance, consider characterizations of the appearance of the person with respect to build, hair colour, or characterizations of manual skills, etc.). Consequently, many facets of knowledge from the locust frame can be metaphorically mapped onto the trade union frame, but there is a lack of prominent knowledge facets that are more likely to present themselves for mapping than others.

To sufficiently determine the metaphorical use of locust in (6) (and also in other examples in which institutions, persons, or the like are designated as locusts), the fol-lowing needs to be clarified:

– Which elements constitute the two inputs locust and trade unions (or finan-cial investors)?

– Which elements are shared by both inputs (and thus pass into the generic space)?– Which elements pass into the blended space and into an emergent metaphorical

meaning?

In the following, I represent the hypothesis that these questions can only be adequately answered if account is taken of the fact that the financial investors as locusts metaphor established itself in a particular discourse context – namely the so-called “capitalism debate”. Its meaning emerged and became entrenched within this discourse, and within this discourse, it is neither questionable which knowledge forms part of the input, nor does it remain open what the emergent meaning of the metaphor is.514 This is because only certain slots (facets of knowledge) and only certain values (pre-dicative attributions) prove to be relevant in this knowledge context.515 The degree of entrenchment of many slots and values in the locust frame emerges in the context of the capitalism debate. Outside of the capitalism debate one cannot say precisely what it means to designate financial investors, trade unions, etc. as “locusts”, since only

513. Imagine, for example, the typical distribution, activities, or places of occurrence of locusts were transferred to trade unions (or financial investors, etc.).

514. To be sure, this does not mean that the meaning of the financial investors as locusts metaphor is formed exclusively within the capitalism debate. In fact the metaphor is part of a discourse history that cannot simply be cast aside. For instance, of essential importance is the biblical “plague of locusts” (Exodus 10), to which Müntefering’s quote alludes.

515. In this context, I follow Paul Ricœur’s definition of a metaphor as an “impertinent predica-tion”. Ricœur writes: “To generate a new metaphor, it is necessary to have at least a sentence. More precisely, the metaphorical process is to be sought in the main procedure framed by the sentence, namely the predication. Now, my thesis is that the semantic extension that occurs in the word is based on a peculiar, unusual usage of the predication. The metaphor is an ‘impertinent’ predication, that is, one that breaches the usual criteria of appropriateness and pertinence in the application of the predicates.” (Translated from Ricœur 1986: vi)


nonspecific background knowledge (such as that of the biblical locust plague) can be applied to the interpretation.

In the corpus-based analysis (Sections 7.4.1 to 7.4.3), I further pursue Fauconnier and Turner’s assumption that frames structure the four representational units – generic space, inputs I and II, and blended space. I refine this assumption in three respects:

i. Elements in the generic space are to be equated with the slots that the two inputs share (Section 7.3.2).

ii. A high type frequency within the discourse ensures that only certain slots pass into the two inputs (Section 7.3.3).

iii. If an instance of precisely this slot occurs with particular frequency (in the sense of a high token frequency), then this instance is a good candidate for the metaphoric mapping onto the blended space (Section 7.3.4).

In the following, since the singular process of understanding in actu does not consti-tute the research focus, I speak henceforth of frames rather than spaces. In contrast to Fauconnier and Turner, I am concerned with the epistemological analysis of the conditions for how particular elements can become parts of a space, but not with the explanation of the emergence processes of individual metaphor meanings as such.

7.2 The “capitalism debate”

The case of the financial investors as locusts metaphor is a prime example of how we can only approach an explanation of U-relevant knowledge (in the sense above) by considering the superordinate embedding structure of a linguistic expres-sion. Section 6.3.1 already showed that frames – understood as linguistic-cognitive frames of reference – fulfil the function of transphrastic reference structures. Texts and discourse constitute the access format of frames. In this context, I noted that frames do not describe the meaning of words or the meaning of sentences, but rather at best the meaning of words in texts and the meaning of sentences in texts or discourses.

But what is a “discourse” and how can we operationalize this category of analysis for our empirical purposes? The next section addresses this question from methodo-logical viewpoints. Then, Section 7.2.2 focuses on aspects that concern the corpus of the metaphor analysis. Finally, Section 7.2.3 is devoted to the question of what role the metaphor of financial investors as locusts plays in this discourse: If discourse knowledge shapes the conceptual structure of frames, what shapes the converse rela-tionship? To what extent does the metaphor itself shape the discourse in which it occurs (in this case, the capitalism debate)? The fact that metaphors can adopt the function of discourse-constitutive elements, or, in a somewhat weaker form, can function as key “switches” in discourse, was already discussed by Busse (1997c) in connection with so-called “basic discourse-semantic figures”. My hypothesis is that the financial investors as locusts metaphor represents a basic discourse-semantic figure par excellence. I provide arguments for this based on frame theory.


7.2.1 Discourse and corpus

What is a “discourse”?516 To begin with, let us recall the results of the previous con-siderations. Disregarding the context in which the metaphor is used, the metaphori-cal use of locusts remains starkly underspecified. It is only when we involve frames in the analysis that there is hope of being able to analytically specify the metaphor more precisely. Frames structure both inputs, i.e. they specify which default values are represented in the input and which of these default values pass into the metaphorical meaning. Thus seen, a frame analysis addresses the “conditions that made it at all pos-sible (in the sense of an semantic analysis) for a particular textual component [in this case, an element in the input] to appear” (translated from Busse 1997c: 17).

Which elements appear in the input cannot be determined on either the sentence or the text level. This is because there we only find overt contextualization contexts, i.e. (quasi-)explicit predications. However, we require access to implicit predications (default values) that language users inferentially apply to processes of understanding. In other words, the meaning of the financial investors as locusts metaphor is predetermined by recurrent usages of the word that resemble each other, i.e. by recur-rent predicative attributions that are recorded in a multitude of such sentences and texts in which the metaphor was used. To gain a view of those knowledge dimensions that have emerged as a result of recurrent usage, we must therefore transcend the level of the sentence and the text. An abundance of sentences and texts then constitute the analytical research focus.

If we address the question of what a discourse is from a frame-semantic viewpoint, then the answer must be: A discourse is a virtual set of texts (or text sequences) in which slots in the frame that the metaphor locust evokes are occupied by fillers. The practical problem of designing the research corpus therefore depends crucially on the theoretical specification of the discourse concept. Following this guideline, Busse and Teubert explain the concept of discourse as follows:

With research practice in mind, we understand discourses to be virtual text cor-pora whose composition is determined by contentual (or semantic) criteria in the broadest sense…. Specific text corpora (i.e. those based on discourse-semantic investigation) are subsets of the respective discourses. (Translated from Busse and Teubert 1994: 14)

516. In the following, I do not elaborate any further on the various discourse concepts that have emerged according to different disciplinary requirements (cf. Böke 1996: 432ff.; Jung 1996: 453–454; Link 1999: 148ff.), but exclusively follow an understanding of discourse that recurs to Foucault (1974a, 1974b, 1981). In addition, I ignore methodological issues in favour of those relevant to research practice and frames. Some epistemological considerations on linking frame semantics with discourse analysis can be found in Busse (2008b) and Ziem (2008a). In the meantime, the Foucaultian discourse concept has proven effective in numerous discourse-ana-lytical studies (cf. for instance Busse 2000a; Busse and Teubert 1994; Jung 1996; Fraas 1996a: 4; Fricke 1999; Scharloth 2005a: 62ff.; Wengeler 2003, for an overview: Warnke 2007).


Whether a text or a text sequence is to be attributed to a discourse is thus determined by whether semantic relationships prevail between texts or text sequences and/or whether they share the same context of proposition, communication, function, or intention. For practical and methodological purposes, the set of texts underpinning the analysis is understood as an “open corpus” (Busse and Teubert 1994: 15), since in the individual case it may prove to be impossible for practical reasons to include relevant texts (for instance, because they are not available). In addition, constructing a corpus is, method-ologically seen, an object-constitutive act, as it first creates the research object. Strictly speaking, however, the construction of the corpus can never be concluded, since every new text in turn establishes new intertextual references. In keeping with this, Busse and Teubert therefore argue for a hermeneutical understanding of discourse, according to which discourses do not exist as objective entities, but rather represent the results of scientific processes of constitution (Busse and Teubert 1994: 17).517 The evaluation of the relevance of a text and, based on this, the selection of the texts to constitute the corpus presuppose the interpretive, comprehending intervention of the researcher.518

Conversely, discourse analyses and corpus-based frame analyses equally concur that the research aim does not consist in achieving a better understanding of individual texts or text sequences (Busse and Teubert 1994: 18). Thus, neither bears hermeneutical research objectives. When frame analysis investigates the entrenchment of conceptual structures, the object of such studies, just as in discourse analysis, is the explication of all U-relevant knowledge. This analytical perspective transcends the hermeneutical interest in the intention of the author expressed in a text. Discourse and frame analyses are directed at the prerequisite implicit knowledge that allows text production and reception in the first place.

If the following considerations are to be based on this understanding of discourse, then in my view it is necessary to determine even more precisely which criteria must be fulfilled in order to understand semantic relationships between texts. When Busse and Teubert expressly include “implicit references”, one must consider under what circumstances such semantic relationships occur and, in particular, how they can be made explicit. “Implicit references” appear to correspond to the contextualization type that Busse (2000a: 44) designates as non-intended, unconscious, and only analytically ascertainable. Busse observes that this type is problematic, since it is highly dependent on the theory applied and the descriptive language used. To identify implicit references, Busse suggests using established methods like isotopy and presupposition analysis and argumentation or topos analysis. However, these methods do not represent effective

517. At another point, Busse (2000a: 45) argues that scientific categories generally cannot be dis-engaged from the descriptive language chosen; this applies to conventional analytical categories (such as, for example, ‘morpheme’, ‘phoneme’) just as much as to ‘discourse’.

518. This point becomes particularly clear in metaphor analysis. Even when the criterion for the selection of texts is that all the texts (within a particular time frame and communication context) in which the metaphor occurs should form the corpus, nonetheless the metaphor must have been previously identified as a metaphor.


means of corpus design, since they, to some extent, take for granted that a corpus exists (in which presupposed knowledge, similar topoi, etc. can be found). The application of these methods is sometimes so complicated that it would be difficult to employ them to substantiate the discourse affiliation of a large number of texts.

I consider as the decisive criterion whether a text or text sequence may count as a component of the discourse to be solely the occurrence of the expression that evokes the frame, in this case, the metaphor locust(s). At the same time it is not necessary that the target domain financial investor/s should appear as a text element in ad-dition to the token locust. Although this is the case in the early corpus instances, once the metaphor has established itself, the target domain is increasingly presupposed. In these cases, the identity of the metaphor secures the superordinate contextualization context of the capitalism debate. In my view, there is no plausible reason for not includ-ing text instances of this kind in the corpus.

Thus, I diverge in two points from the proposals made by Busse and Teubert (1994). Firstly, the criterion of relevance (i.e. the evaluation of the importance and centrality of individual texts for the discourse) plays no role in the selection of texts. All texts in which the financial investors as locusts metaphor occurs within the set period of time and communication context constitute the text corpus and thus a segment of the discourse.519 In addition, only those texts are included in the corpus whose reference relationship is explicit by dint of the fact that the metaphor occurs in them.520 As a criterion of text selection, I exclude implicit references to minimize interpretive interference.

It should be noted that a discourse represents a unit of analysis that is to be con-sulted to explicate U-relevant knowledge. As I explained in detail in connection with von Polenz’s sentence semantics in Section 6.3.1, it is not enough to describe the predi-cation potential of a linguistic expression on the sentence or text level. The area of access is always a number of texts in which the expression under investigation occurs and selectively unfurls its predication potential. From an analytical perspective, this set of texts constitutes a discourse, and empirically seen a virtual corpus of texts. In terms of research practice, it is unavoidable that the set of texts constituting the discourse varies according to the chosen period of investigation and the area of communica-tion. The corpus remains necessarily a segment of the discourse. The discourse is an extrapolated parameter that can never be completely captured in empirical terms. Its unity (or that of the virtual text corpus) is guaranteed by content-related criteria that texts or text sequences must fulfil.

Thus, the capitalism debate includes all the texts that feature a shared thematic focus. This can be verified by means of similar macropropositions in the individual

519. For an explanation, cf. Wengeler (2003: 295–296).

520. However, this procedure can only be generalized to a limited degree. If in the period under question there were financial investors as locusts metaphors in use that were not embed-ded in the capitalism debate, this would mean there were various referential contexts requiring differentiation.


texts (Konerding 2005; Meier 2005) or by the recurrence of key textual content ele-ments (cf. for instance Böke 1996: 446ff.; Hermanns 1995; Wengeler 2003: 339ff.).521 The financial investors as locusts metaphor could represent just such a key ele-ment of textual content in the discourse of the capitalism debate.522 If the discourse of the capitalism debate derives, among other things, from the shared thematic relation-ship between corpus-constituent texts, then in practice it proves to be unnecessary to determine the thematic reference of every individual text. The same applies for many other discourses: Frequently, the discourse-constituent texts revolve, as it were, around a thematic macroproposition; in addition, there is often a key reference text that initiates the discourse to begin with, thus guaranteeing its identity, so to speak (such as the speech by Martin Walser that sparked the so-called Walser-Bubis debate on national identity and the Sloterdijk essay Regulations for the Human Zoo (‘Regeln für den Menschanpark’) in the public controversy on humanism between Sloterdijk and Habermas; cf. Steinseifer 2005). Texts that can be assigned to the capitalism debate often already refer to the topic in the article heading.

7.2.2 Investigation period, discourse development, research corpus

Let us turn from these considerations of the method of corpus design to the corpus that underpins the metaphor analysis. The period of investigation and the area of com-munication covered by the research corpus should be specified. I begin with the first point and elaborate on it by briefly outlining the progress of the discourse.

The period under investigation concerning the financial investors as locusts metaphor extends from 18 April 2005 to 6 May 2005. This rather brief time span is due to the progress of the discourse. The beginning of the public debate on whether finan-cial investors are “locusts” and if so, to what extent, falls parallel with the start of the capitalism debate, and this in turn begins with a report in the newspaper Süddeutsche Zeitung about SPD chairman Franz Müntefering’s criticism of business managers. In an interview with the newspaper Bild am Sonntag on 19 April 2005, when asked about whether he prefers socialism to capitalism Müntefering states:

No, but I don’t like capitalism either. I reject people in business and the inter-national financial markets who behave as if no barriers or rules applied to them anymore. Some financial investors don’t waste a thought on the people whose jobs they destroy. They remain anonymous, with no face, falling over companies like swarms of locusts, stripping them bare, and then moving on. It is against this form of capitalism that we are fighting. (Translated from Bild am Sonntag, 19 April 2005)

521. According to Konerding (2005) to determine the shared thematic focus of texts, three analytical steps are required. The first phase requires the main topics of the texts in the corpus to be identified. Then the shared main topics in the texts are nominalized and, finally, allocated to their generic frame type (matrix frame) via hyperonym reduction.

522. I take up this idea in the next section.


Müntefering uses the expression swarms of locusts (not locust/s) to characterize finan-cial investors. What is more, he does not use a metaphor; the metaphor of financial investors as locusts is, in fact, a later product of the (metalinguistic) reference to Müntefering’s quote. Nonetheless, this text sequence may claim the status of being a dominant text in the discourse addressed (in the sense of ‘Leittext’, cf. Teubert 1998: 148). The fact that Müntefering’s remark was frequently referred to, particularly at the beginning of the capitalism debate, shows just how controversial the likening of financial investors with locust swarms was right from the start.523

In terms of content, the quote is to be seen in close connection with a speech that Müntefering held as the then SPD party chairman on 13 April 2005 at a function introducing the new policy statement of the SPD. Since the script of the speech had been made available to the Süddeutsche Zeitung beforehand, the paper already reported on the event that same day under the title “Müntefering castigates ‘power of capital’”. Although the metaphor of financial investors as locusts does not appear in this article, it reports on Müntefering’s speech, thus establishing the thematic reference to his criticism of the increasing dominance of business investors that the metaphor later came to represent.524 In addition, the precarious state of the economy and the labour market also belong to the superordinate discourse context of the capitalism debate. Thus, in mid-April the red-green coalition had to make a downward correction in their forecasts of business activity and economic growth, while at the same time it was announced that over five million people in Germany were unemployed.

After Müntefering had made it clear that his comparison between financial in-vestors and locust swarms made in his interview with Bild am Sonntag was directed against certain private equity companies, the news journal Stern published a list of twelve private equity companies in their internet edition of 28 April 2005 under the title “The names of the locusts”. The source was given as a background paper of the SPD Parliamentary Fraction Planning Group.

Two additional events significantly influenced the further course of the discourse. At the beginning of May 2005, the metalworkers’ trade union IG-Metall published a members’ magazine that caused cross-party furore. It bore the headline “US compa-nies in Germany – sucking us dry”. It was illustrated with a caricature of a grinning

523. Cf. the textual evidence in the next section.

524. The fact that media attention is focused to such a high degree on the financial investors as locusts metaphor is surprising in that Müntefering already used the same metaphor on 22 November 2004 in a very similar context without causing any media hype (in this respect it can be seen as a prime example of a “discourse event” in the Foucaultian sense). In his speech, entitled “Freedom and Responsibility”, which Müntefering held at the Friedrich Ebert Foundation on the occasion of the SPD party policy revision, he states: “We must assist those companies that have the futurability of their companies and the interests of their employees in view in combatting the irresponsible swarms of locusts that measure their success on a quarterly basis, drain off a company’s substance, and let it fall apart once they have stripped it bare. Capitalism is not a museum item, but rather something highly topical.” (Translated from SPD 2005: 18)


mosquito wearing a pin-stripe suit and raising its top-hat (in the colours of the US flag) in greeting. The essay by the historian Michael Wolffsohn published on 3 May 2005 in the Rheinische Post newspaper led to similarly fierce discussions. Wolffsohn likened the financial investors as locusts metaphor to patterns of thought that were widespread in National Socialism.525

The sixth of May, 2005, marked a watershed in the course of the discussion. The extremely controversially conducted discussion about whether political restrictions should be applied to the capital-intensive activities of financial investors shifted in-creasingly out of the focus of public media interest. Other topics dominated the news coverage, for instance the 60th anniversary of the end of the Second World War on 9 May 2005 and the party politics of the important state elections in North Rhine-Westphalia, which were to take place on 22 May 2005. This is reflected in the low number of occurrences of the financial investors as locusts metaphor in the period following 6 May. In the following weeks, although the capitalism debate was repeatedly addressed in isolated cases – for instance on 12 May, the weekly magazine Stern headlined ironically with “The big feast – why we are all” – the continuity and intensity of the news coverage that characterized the period between 18 April and 6 May was meanwhile lacking.

The research corpus draws on the print media as its communication context. Three national daily newspapers, the Süddeutsche Zeitung, the Frankfurter Allgemeine Zeitung, and the Tagesspiegel, as well as the weekly newspaper Die Zeit are enlisted for the analysis.526 The choice of these high-circulation newspapers is intended to cover a broad political spectrum (cf. Wengeler 2003: 341; Keller 1998: 46ff.).

In keeping with the preliminary corpus-linguistic considerations, the text corpus encompasses all those texts in these papers that contain at least one occurrence of the financial investors as locusts metaphor and were published between 18 April and 6 May 2005. Of these, the analysis included those articles in which predicates were at-tributed to the reference object ‘locust’. 70 articles fulfil these conditions. Each of these articles expressly refers to the capitalism debate. In addition, there are further articles that take up the topic of the capitalism debate without employing the metaphor or without predicatively referring to the reference object ‘locust’. The newspaper articles analysed are listed in the bibliography.

525. Wolffsohn writes: “Today, 60 years ‘later’ people are once again being likened to animals that must be ‘destroyed’, ‘eradicated’ as a ‘pest’ – thus the unspoken meaning. Today this ‘pest’ is called ‘locusts’, back then ‘rats’ or ‘Jewish pigs’. Words from the dictionary of inhumanity, because humans are denied their humanity.” (Translated from Rheinische Post, 3 May 2005)

526. The articles from these newspapers form the research corpus in the narrower sense. To be sure, the source domain of the financial investors as locusts metaphor cannot be in-vestigated with this corpus because the expression locust is used in an exclusively metaphorical way within the discourse of the capitalism debate. Thus, a reference corpus must be enlisted to obtain insight into the conceptual structure of the source domain. To this end, in Section 7.4.2 I recur to the corpus data from Cosmas II, a databank made available by the Institut für Deutsche Sprache (the Institute of German Language) in Mannheim.


The restriction to print media is not commonplace, but is certainly standard lin-guistic practice in Foucaultian discourse analysis.527 In view of my research aim and the methodology applied, it suggests itself to exclusively deal with written texts because the aim of the investigation is to obtain insight into the conceptual content of the metaphor using frame-based predication analysis. The analysis of visual material would additionally necessitate including a method that allows aspects of visually represented knowledge to be coded and quantitatively evaluated, since the semantic categories that provide matrix frames for meaning analysis cannot be seamlessly transferred to visual sign carriers. Subsequently, not only would the process of investigation be made considerably more difficult, but to date it has not been clarified how linguistic and non-linguistic information units conceptually interact.528

7.2.3 Locust: a basic discourse-semantic figure

A discourse consists of a multitude of discourse elements of differing complexity. Different levels of complexity result from the levels of syntagmatic organization of linguistic signs. Catchwords and key words that are used non-metaphorically are lo-cated on a low level of complexity (cf. Jung 1997, 2000). Metaphors are more complex because their meaning recurs to the interrelation between knowledge segments from two conceptual domains. Discourse-specific knowledge in both domains can thus consolidate on the lexical level (cf. Böke 1996, 1997).529 Argument patterns solely relate to transphrastic discourse elements (cf. for instance Wengeler 2000, 2003). They concern the habitualized usage of the same patterns of reasoning in similar thematic contexts and consequently already transcend sentence boundaries on the token level. Metaphors, catchwords, and key words can in turn constitute integral components of argument patterns. Conversely, social attitudes, patterns of thought, and even the argument patterns of a language community manifest themselves in metaphors (Böke 1996: 439; Pielenz 1993). Consequently, discourse elements are intertwined (on both the syntagmatic and the paradigmatic axes).

527. However, in recent times discourse analysis takes more and more account of non-linguistic knowledge units like images and everyday objects (cf. for instance Meier 2005, 2010; Scharloth 2005a, b). This extension is only logical in view of Foucault’s open concept of discourse.

528. Initial evidence of this can be found in Scollon and Scollon (2003), Stöckl (2004), Leeuwen (2005). However, in my view it is inadequately explained how linguistic and non-linguistic means can be used to code information in a conceptually related way such that this results in a holistic cognitive information unit.

529. According to Böke, metaphor and discourse theories converge to form, on the one hand, a “view of language as a medium for reflecting on, structuring, and organizing, as well as constitut-ing socio-cultural reality in general” and on the other in the “notion(s) of language as an indicator and factor of collective consciousness (and changed in it) or more precisely, the mentality of social groups in a language community in particular” (Böke 1996: 439).


Although I restrict my analysis of the financial investors as locusts metaphor to investigating its metaphorical-conceptual structure, it could just as easily be also investigated for its argumentative function. Nevertheless, my analysis of this metaphor is not restricted to purely lexical-semantic structures. That is to say that understand-ing a metaphor as an (evoked) frame means including at least the semantic relations to those concepts that are attributed to the reference object in the linguistic form of predicates (in the sense explained above).530

Discourse elements distinguish themselves from each other within a discourse in terms of their status. Discourse elements that dominate more than others can probably be found in every discourse. This dominance is based in the regularity of their occur-rence within a discourse universe and is reflected to a high degree in the cognitive presence of precisely these elements. In this context, Busse introduced the term “basic discourse-semantic figure” (diskurssemantische Grundfigur) (Busse 1997c; 2000a: 51–53; 2003a: 28–32). Basic discourse-semantic figures are “contentual elements in a text that have become entrenched as regularities and that constitute the corpus of individual discourses” (translated from Busse 2000a: 50). Basic discourse-semantic figures do not have a uniform linguistic format; they can take effect in the form of catchwords and metaphors just as much as in the form of inferred knowledge (presuppositions, impli-catures, etc.) or argument patterns. According to Busse (2000a: 50–52), the following properties characterize basic discourse-semantic figures:

– Basic discourse-semantic figures belong to the cognitive endowment of language users.

– To take effect, they must have occurred explicitly for a time.– Once basic discourse-semantic figures have been entrenched, they generally occur

“covertly”, i.e. they are presupposed or inferred. For this reason they belong to the deep level of text semantics.

– To some extent, basic discourse-semantic figures evade the volition of speaking and comprehending persons.531

530. Conversely, Busse and Teubert observe, with lexeme analysis in mind: “However, discourse analysis is intended to investigate semantic relationships that are not only expressed through the unit of the lexeme but also transcend the lexemic unit” (translated from Busse and Teubert 1994: 19). However, understanding the meaning of a lexeme as an (evoked or invoked) frame always already includes semantic relationships that transcend the lexemic unit; predicative al-locations occur within sentences, texts, and discourses.

531. Exactly what is meant by this remains unclear in my view. My interpretation of this observa-tion is that it is not possible not to understand. In this vein, Lakoff (2004: 3) states that knowledge is often automatically activated. Subsequently, it is not possible, for example, to comply with the demand “Don’t think of locusts!” since the token locust inevitably evokes the locust frame. Therefore, both cognitive evocation processes and conventionalized elements (default values) in the evoked frame are not – or hardly – able to be voluntarily influenced.


– Basic discourse-semantic figures organize other textual content elements and can govern their occurrence at particular points in the discourse.

– Finally, basic discourse-semantic figures form a grid that can take effect across discourses, within other discourses.

The final aspect seems to me to be of crucial importance. Only under the condition that basic discourse-semantic figures exhibit a schematic structure can it be explained how they fulfil a U-relevant function without explicitly occurring (i.e. for instance in the form of a realized argumentative conclusion, a used predicate, etc.). In addition, from the perspective of memory theory, it is only the schematic character of basic discourse-semantic figures that explains how these can become part of the cognitive endowment of language users. If, for example, basic discourse-semantic figures con-tained meaning-relevant details (in the sense of a usage event), they could not become an invokable knowledge unit in the long-term memory (cf. Section 5.2.1). Finally, the fact that basic discourse-semantic figures partially evade the volition of speaking and comprehending persons is presumably a result of schematization processes, namely of the entrenchment of particular knowledge elements whose recognition as a result proceeds more routinely than other knowledge elements.

Although this suggests that the semantic form and functionality of basic discourse-semantic figures can be explained in schema-theoretical terms, Busse only mentions a connection with schemata in passing (cf. Busse 1997c: 24, 33). In addition, since he tends to specify them in functional terms, it remains unclear as to what basic discourse-semantic figures are in a semantic perspective (Scharloth 2005b: 124).532 In the fol-lowing, I would like to represent the hypothesis that only if the schematic character of basic discourse-semantic figures is taken into account is it possible to adequately describe their functionality and recognize their semantic content. Using the example of the financial investors as locusts metaphor I propose three empirical criteria to identify basic discourse-semantic figures in text corpora and attempt to show why the metaphor represents a basic discourse-semantic figure.

532. Busse (e.g., 2002a: 51) states that basic semantic figures are able to be expressed in dif-ferent forms – they occur as (i) semantic features, (ii) historical isotopy sequences, (iii) sup-port elements in a text-based derivation rule, (iv) presuppositions in the sense of linguistic pragmatics or, alternatively, parts of the implied and intended that are interpreted through inferences, (v) something that is hidden behind names, addressed persons, things, circum-stances, and thought complexes, and (vi) components of the (lexical) surface meaning of words, terms, and texts. Nevertheless, in my view such differing manifestations do not explain how basic discourse-semantic figures structurally take effect: considered in semiotic terms, do basic discourse-semantic figures occur uniformly? If this is the case, what are the features of this semiotic uniformity? How do the utterance and content sides behave in relation to each other (provided that basic discourse-semantic figures constitute symbolic units)? To what extent do basic discourse-semantic figures have a schematic character on the utterance side?


To what extent do basic discourse-semantic figures exhibit schematic properties? In an empirical case study investigating aesthetic judgements in the late 18th century, Joachim Scharloth (2005b) employs basic discourse-semantic figures as categories of analysis, but in contrast to Busse, he understands these expressly as schematic structures used to process perceptual impressions, as well as to cognitively organize knowledge bases and make them readily available (Scharloth 2005b: 124). In addition, to make them operational, Scharloth draws on structuralist observations made by Umberto Eco, according to which a cultural unit, which includes linguistic signs, is the set of those features that it shares with other cultural units and with which it distinguishes itself from them. He directly follows componential semantics by treating features as binary-coded semes that form along axes of semantic oppositions. Thus, concerning basic semantic-discourse figures he can in retrospect come to the conclusion

…that these schemata [basic discourse-semantic figures] can be conceptualized as fundamental elements in the organization of the semantic system, for instance as semantic oppositions that structure a large number of semantic fields, or as basic feature oppositions that distinguish semantic fields from each other. (Translated from Scharloth 2005b: 131)

Scharloth attempts to illustrate the added value of such a redefinition of basic dis-course-semantic figures using “natural” vs. “artificial” (‘natürlich’ vs. ‘gekünstelt’) as an example. He arrives at the result that, at the end of the 18th century, this feature opposi-tion had taken on the form of a fundamental perceptual and evaluation schema, which can be demonstrated in such varying artefacts as poems and copperplate engravings.

Apart from the empirical relevance of basic discourse-semantic figures, Scharloth is also concerned with the semiotically motivated issue of the semantic nature of basic figures. In my view, Scharloth poses the right question, but provides a problematic answer. Although it may be helpful to revive structuralist theorems for his empirical interest, he nonetheless also acquires all the difficulties and a priories that a structural-ist meaning model entails (cf. for instance Konerding 1999: 20). Not only the status of the semantic features remains unclear. It is also questionable whether constructions of binary oppositions and distinctive semantically organized field structures can do justice to the complexity of the research subject. Of greater concern, however, is the fact that the applied structuralist theorems are not compatible with either a schema concept based in cognitive psychology or cognitive processes like categorization and inference construction. Yet, it is precisely this interdependence that Scharloth’s specification of basic discourse-semantic figures implies.

The problem is that he attempts to reconcile the irreconcilable: From a cognitive perspective, semantic distinctivity can only be viewed relative to U-relevant cognitive domains or frames. Since every categorization is a knowledge-based cognitive act, neither semantically distinctive features nor clear semantic field structures exist (cf. Allwood 2003; Zlatev 2003; Langacker 2006). Knowledge units (fillers, default values) change according to language usage. Further, in contrast to semantic features, they are not inherent in linguistic units, but are rather the result of language users’ individual


processes of understanding. If we recall the results of Section 2.2.3, there are even valid reasons why including semantic features can only lead to a highly inadequate specifica-tion of U-relevant knowledge. A componential-semantically inspired reformulation of basic discourse-semantic figures as proposed by Scharloth essentially adheres to a meaning-reductionist programme. It obscures a large part of the semantic deep dimen-sion that basic discourse-semantic figures are specifically intended to render accessible.

My counterproposal is as follows: Basic discourse-semantic figures are symbolic units, i.e. form-meaning pairs of varying degrees of abstraction, whose content side has emerged specific to a particular discourse and is structured by at least one frame. The content side of a symbolic unit is schematic in that it contains knowledge seg-ments that have gradually become entrenched as conventional elements (default val-ues) within a discourse and are easily inferentially actualisable by means of cognitive routines. To achieve the status of default values, instances must have exhibited a high token frequency for a period of time in the form of (quasi-)explicit predications. If basic discourse-semantic figures exhibit numerous default values on the content side, a discourse-semantic figure governs the occurrence of other text elements by activat-ing these default values, during which every default value also opens up a variety of potential further connections. Let us assume that the metaphor locust is a basic discourse-semantic figure that features the default value, among others, that financial investors exploit and destroy small and medium-sized businesses without regard for the employees. Then, in connection with the financial investors as locusts meta-phor, the consequences of a plague of locusts may become the topic of a text without mentioning small or medium-sized businesses or their employees. In this case, it is solely the activated default values in the metaphor frame that govern the occurrence of text elements like ‘plague’ and ‘consequences of plagues’.

I give this example in advance of the results for heuristic purposes. Whether the metaphor does in fact fulfil a discourse-dominant function in the capitalism debate is yet to be demonstrated. Which criteria must be fulfilled so that we can speak of a “basic discourse-semantic figure”? How do we identify basic discourse-semantic figures? And what reasons justify denying other discourse elements the same status? When Scharloth investigates the opposition of natural/artificial, he appears to proceed by first repeatedly identifying this opposition in the course of reviewing the material, then testing its value as a provisional category of analysis in individual exemplars, and finally – based on positive findings – extrapolating from the examples the assumption of a perceptual schema that is intended to generally apply during the last third of the 18th century. Busse (1997c) evidently also proceeds in a similar manner in his empiri-cal study of the basic discourse-semantic figure ‘the native and the alien’ (das Eigene und das Fremde).

According to this procedure, what is crucial for the identification of a discourse-semantic figure is that the basic figure is already known before reviewing the material or that one can “get to know” it on the basis of the selected material. From a meth-odological viewpoint, it seems to me here that it is crucial to clarify precisely which criteria contribute to the identification of a discourse-semantic figure. For this reason, I


introduce three criteria in the following that in my view assist to reliably determine the status of discourse elements: It is possible to speak of a basic discourse-semantic figure when (i) it is a discourse element that is frequently referred to on a metalinguistic level, (ii) the same element occurs frequently at least for a period of time within a particular discourse context, and (iii) the content side of the discourse element exhibits a range of temporarily persistent default values allowing the same discourse element to take effect in other discourses. The application of these criteria presupposes a previously constituted text corpus, which conversely means that a basic discourse-semantic figure cannot be reliably identified on the basis of individual text analysis. In this point my procedure differs from that of Busse and Scharloth. Using the example of the finan-cial investors as locusts metaphor, I now attempt to substantiate the application of these three criteria.

i. Metalinguistic referentialization. Numerous studies have shown that metalinguis-tic reference (that is, reference to an expression that somebody has introduced into a discourse) is a meaningful indicator for identifying key discourse elements. Thus, Georg Stötzel, in his introduction to Kontroverse Begriffe (‘Controversial Terms’), makes it clear that every metalinguistic reference indicates the public impact and social relevance of the addressed circumstance (cf. Stötzel 1995: 3).533 Metalinguistic referentialization in public language usage is an indicator that the language participants are concerned about the semantics of linguistic expressions and wish to consciously accentuate or alternatively deny certain meaning nuances. Beyond the lexical level, argument patterns can also be considered from a meta-linguistic perspective (Wengeler 1996). In principle, linguistic units of any level of complexity, i.e. presuppositions and implicatures, as well as entire texts, can be metalinguistically addressed. Thus, it is not difficult to investigate whether not only lexical units but also more complex linguistic expressions can potentially be basic discourse-semantic figures. However, there are no grounds to assume that a concept (or concept network) has a key status until similar metalinguistic referentializations occur with significant frequency within a discourse.

The financial investors as locusts metaphor fulfils this condition within the discourse of the capitalism debate. Every one of a total of 70 articles in which the metaphor occurs makes at least one metalinguistic reference to the meta-phor.534 To illustrate this, two articles from the corpus have been chosen as ex-amples. What is characteristic at the beginning of the capitalism debate is the fact that Müntefering’s locust comparison is repeatedly quoted and commented on, as in the following example from the Süddeutsche Zeitung:

533. Cf. also Stötzel (1995: 17); Stötzel (1980: 39); Wengeler (1996: 412–413).

534. Precisely this circumstance no longer applies when the metaphor occurs in other discourses outside the capitalism debate; cf. point (iii).


Müntefering had given warning about the “increasing power of capital” and lik-ened some of the investors with “swarms of locusts” that fell upon companies, stripped them bare, and moved on. (Translated from the Süddeutsche Zeitung, 20 April 2005)535

Metalinguistic reference such as this do not only occur at the beginning of the capitalism debate, but also characterize the entire discourse. Thus, at the end of the investigation period several metalinguistic references to the metaphor can be noted in an article in Die Zeit:

On Monday, the Green faction and party leadership supported Müntefering’s gen-eral direction but rejected his ‘locust’ analogy. ‘The debate must be conducted in a far more discriminating fashion,’ said financial expert Christine Scheel to the “Welt” [newspaper] (Tuesday). [In her view] the ‘locust’ analogy frightened off companies that were not even being referred to …The chair of the Hesse SPD Andrea Ypsilanti rejected criticism of the ‘locust’ anal-ogy from the ranks of the Greens…In mid-April, Müntefering triggered the debate by using the analogy of ‘locusts’ to refer to international financial investors who take over companies to strip them of assets and then sell them off. … (Translated from Die Zeit, 3 Mai 2005)536

All four of the quoted text references deal with the conceptual content of the anal-ogy between financial investors and locusts, while the second and third quotes also concern the extent to which individual participants in the discourse agree with actualized default values. The fact that within the capitalism debate no other concept becomes the object of assertions with anywhere near the same frequency can be considered a clear indication of the key status of the analogy in discourse. The metaphor of financial investors as locusts emerges in the course of the capitalism debate against this background.

535. In the original: “Müntefering hatte vor der ‘wachsenden Macht des Kapitals’ gewarnt und einen Teil der Investoren mit ‘Heuschreckenschwärmen’ verglichen, die über Unternehmen her-fielen, sie abgrasten und weiterzögen.”

536. In the original: “Die Grünen-Fraktion und -Parteispitze hatten am Montag die Stoßrichtung Münteferings unterstützt, seinen ‘Heuschrecken’-Vergleich aber zurückgewiesen. ‘Die Debatte muss viel differenzierter geführt werden’, sagte Finanzexpertin Christine Scheel der ‘Welt’ (Dienstag). Mit dem ‘Heuschrecken’ -Vergleich würden auch Firmen verschreckt, die gar nicht gemeint seien… Auch die hessische SPD-Landesvorsitzende Andrea Ypsilanti wies die Kritik aus den Reihen der Grünen am ‘Heuschrecken’-Vergleich zurück… Müntefering hatte Mitte April die Debatte ausgelöst, als er internationale Finanzinvestoren, die Firmen übernehmen, um sie zu zerlegen und wieder abzustoßen, mit ‘Heuschrecken’ verglich…”


ii. High type frequency. The high number of metalinguistic references is a necessary, but not sufficient condition for the metaphor playing a consistently dominant role in the discourse. It would certainly be possible that, despite metalinguistic referentialization, the metaphor represented only a marginal discourse element, namely if the frame-evoking expression locust/s were (almost) exclusively meta-linguistically addressed. In absolute terms, the type frequency would then be rela-tively low and the financial investors as locusts frame would barely be able to become entrenched within the discourse, at least not to a degree that it could occur in other discourses outside the capitalism debate. In other words, the fact that there is frequent metalinguistic focus on an expression initially just means that it is more often the object of reflection relative to other expressions in the same thematic context. To fulfil this criterion in absolute terms, it is sufficient that only a few metalinguistic references occur, for instance, when in the same thematic context no other expression is metalinguistically reflected upon. In this case, the metalinguistic reference would be frequent in relative terms although the type frequency is quite low.

In fact, the empirical data indicate the opposite.537 As previously mentioned, in the articles that constitute the corpus of the capitalism debate, the metaphor or the analogy between financial investors and locusts occurs in 70 articles.538 This is a considerable sum, especially if one considers that there are a total of 173 occur-rences of the metaphor, which means that the expression locust/s occurs on aver-age more than twice in each article. Of particular note is the fact that the source domain locust/s even occurs more frequently than the target domain financial investor/s; 149 occurrences of financial investor/s are registered.

Accordingly, it should at least be clear that the financial investors as lo-custs metaphor emerging from the analogy between financial investors and lo-custs is a good candidate for a basic discourse-semantic figure, although nothing has yet been said about its conceptual content.

iii. Entrenched content structure (high token frequency). The extents to which default values become entrenched in the locust frame within the discourse is decisive for its functional effectivity both within and beyond the capitalism debate, because activated default values (i.e. those that are presupposed in texts or, alternatively, in-ferred by recipients) govern the occurrence of particular textual content elements

537. A word form counts as an occurrence if it includes the lexical morpheme locust in a complex word (e.g. in locust list, swarm of locusts, etc.) or if the word form is a lexical morpheme, which includes inflected word forms; for an explanation cf. Section 7.3.1.

538. The comparison between financial investors and locusts (or swarms of locusts) as well as (metalinguistic) references to this comparison particularly occurs in the early corpus entries. These passages are also included in the corpus since Müntefering’s comparison between financial investors and locust swarms (in his interview with the Bild am Sonntag) constitutes the departure point of the capitalism debate.


at particular points in the discourse. Without default values, this control function is not possible. In addition, the metaphor can only take effect in other discourses under the condition that a content structure has become entrenched.

I attach an entrenched content structure as the third criterion for identifying basic discourse-semantic figures because, without its inclusion, it would be pos-sible (but unlikely) that criteria (i) and (ii) could be fulfilled without the discourse element having developed a solid content structure. Such a case could occur if all (quasi-)explicit predications had such a low token frequency that no default val-ues could form. In this sense, the analysis of the particle probably in Section 6.5.3 revealed a maximum dispersion of predications. Conversely, the example of the financial investors as locusts metaphor records a concentration of some few predicators and (quasi-)explicit predications, as the next few sections concerning the course of the corpus analysis show. At this point, I would like to introduce two factors that highly favour the entrenchment of the structure of the metaphor’s meaning.

Firstly, the fact that the word locust represents a basic level category ensures the high cognitive salience of the source domain. As explained above, basic level categories are maximally distinct, and because they are “direct” objects of our experience, we have access to a range of predicates that we prototypically at-tribute to locusts. The second factor concerns the metaphorical content, or more precisely, the connection between the source domain locust/s and the target domain financial investor/s. Following Lakoff and Johnson (1980), there is strong reason to believe that the financial investors as locusts metaphor is a variant of the conceptual metaphor persons as animals (cf. Kövecses 2002: 6ff.), because there are numerous other metaphors in which animal attributes func-tion as the source domain and human characteristics as the target domain. These include, for instance:

(8) You’re a pig! (9) He’s as strong as a bear. (10) She sticks her beak into everyone’s business. (11) She has a ponytail. (12) He has a flowing mane of hair. (13) She’s as happy as a sow in muck.

These examples can only serve as an initial indicator that there is a conceptual metaphor persons as animals that motivates the systematic use of animal meta-phors. Only a large-scale corpus-bases analysis can validate whether this hypothesis is sustainable. The conceptual metaphor persons as animals would in any case explain the persistence of the financial investors as locusts metaphor to some extent. It is a fact that this is not restricted to the discourse of the capitalism debate but occurs equally in other contexts. Here are two examples to illustrate this:


Locust on the advanceThe conflict surrounding the newspaper market in the capital could be about to end: Holtzbrinck is negotiating with a British investor about the sale of the Berlin publishing house… (Translated from Der Spiegel, 10 October 2005)539

‘Locusts’ now attack casinos tooFinancial investors buy Harrah’s gaming corporation for 16.7 billion dollarsNew York – The US financial investors Texas Pacific and Apollo Management are evidently on the verge of a 16.7 billion dollar (12.8 billion euro) takeover of Harrah’s, the world’s largest casino operator, including all outstanding debts…. (Translated from Welt kompakt, 19 December 2006)540

Interestingly, the financial investors as locusts metaphor occurs exclusively in the quoted titles and is not explained at any point in the articles. This surprising finding is evidence that the structure of the metaphor’s meaning must have become entrenched. The readers must interpret the meaning of locust/s solely through the activated default values. This therefore fulfils the third criterion. Consequently, the financial investors as locusts metaphor may be considered as a basic discourse-semantic figure.

Like all linguistic signs, basic discourse-semantic figures are symbolic units whose content side is structured by frames (fillers and default values). This applies not only to metaphors, but also to basic discourse-semantic figures expressed in any form.541 It is characteristic of basic discourse-semantic figures that their default values emerge specific to a discourse but then take effect across discourses. Basic discourse-semantic figures constitute schemata of perception in that the default values woven together as content grids systematically govern the occurrence of particular discourse elements (for instance through indirect anaphorical reference to default values; cf. Ziem 2013, 2014, submitted, and Section 6.5) yet without imposing context specification, i.e.

539. In the original: “Heuschrecke im Anflug Der Streit um den Zeitungsmarkt der Hauptstadt könnte ein Ende finden: Holtzbrinck

verhandelt mit einem britischen Investor über den Verkauf des Berliner Verlags …”

540. In the original: “‘Heuschrecken’ befallen auch Kasinos Finanzinvestoren kaufen Glücksspielkonzern Harrah’s für 16,7 Milliarden Dollar New York – Die US-Finanzinvestoren Texas Pacific und Apollo Management stehen offenbar

unmittelbar davor, den weltgrößten Kasinobetreiber Harrah’s für 16,7 Milliarden Dollar (12,8 Mrd. Euro) zuzüglich ausstehender Schulden zu übernehmen…”

541. Accordingly, argument patterns or topoi in the sense of Wengeler (2003) are symbolic units of a complex nature. The content side concerns the contentual composition of individual argu-ment patterns, whereas the form side concerns the respective topos-specific reasoning processes (causal inference, analogical inference, etc.) (cf. Wengeler 2003: 277). On the schematic structure of topoi cf. Hermanns (1994: 49).


without preventing fillers from being placed in unoccupied slots. Finally, with opera-tionalization in mind, it is important that discourse elements achieve the status of basic discourse-semantic figures under three conditions. Firstly, metalinguistic reference must be frequently made to these elements within a discourse. Secondly, the discourse element must also occur so frequently that it becomes entrenched in the sense of hav-ing a high type frequency. And thirdly, only recurrently occurring (quasi-)explicit predications guarantee that default values – and thus a fixed content structure – form. Sections 7.4.1 to 7.4.3 aim to specify this structure for the metaphor’s meaning.

7.3 Methodological guidelines for the corpus-based analysis

Firstly, I would like to introduce the procedure for the corpus-based analysis forming the basis of this case study according to a methodological guideline. The categories of analysis used correspond to the structural constituents of frames as they were intro-duced in Chapter 6. The guideline is intended, on the one hand, to make the procedure as transparent as possible. In addition, it aims to allow the methodological procedure to be transferred to other research subjects.

Two structural constituents of frames form the departure point of the corpus-based analysis, namely slots and fillers. Since, among other things, the approximate degree of entrenchment of fillers is to be determined, default values are not the subject of the analysis, but the goal. Four analytical steps are necessary. (i) First, the relevant instances are annotated. (ii) These are transformed into a set of (quasi-)explicit predica-tions that specify the frame or the reference object. (iii) To be able to group this set of predications into categories in terms of the knowledge facets they address, the frame-evoking expression is reduced to the highest hyperonym using a hyperonym type re-duction; the hyperonym identified is allocated to the corresponding matrix frame. (iv) Finally, the last task consists of classifying the set of (quasi-)explicit predications according to the facets of knowledge they address (i.e. the slots of the matrix frame).

I now explain this four-stage procedure in sequence and illustrate it with instances from the corpus.

7.3.1 Annotations

A prerequisite is that a corpus is available (preferably in electronic form).542 In the following, I use the expression financial investor (‘Finanzinvestor’) to illustrate the examples. This expression evokes a frame, which I discuss in the course of the meta-phor analysis.

What word forms of the frame-evoking lexical unit financial investor should un-derpin the analysis? In principal, it makes sense to consider word forms in the singular

542. On the constitution of the text corpus cf. Section 7.2.1.


(financial investor) and the plural (financial investors), as well as all other inflected or derivative word forms.543 Although grammatical morphemes themselves do not play a role in the evaluation, since in our context we are only interested in the semantic (and not the discourse-functional or pragmatic) content of the semantic unit [financial investor], this is nonetheless consistent, because these word forms also evoke a frame and because this frame could be attributed with fillers – in the form of (quasi-)explicit predications. For the same reason, excluding compounds is equally justified, although these – at least in the case of financial investor – are not evident in the corpus.544

Once the frame-evoking expressions have been marked in the corpus, we face a further problem: What is the extent of the predication’s scope? Is it sufficient to extract those sentences in which the expression occurs? Or can frame-specific predicates also occur in neighbouring sentences? In contrast to co-occurrence analysis, it is clearly not sufficient to merely extract the lemma (in this case: financial investor/s) along with the neighbouring words, for instance, the phrase in which the lemma is embedded. The unit of the sentence is also equally unsatisfactory. The sentence boundary may not be considered as the gauge, because predicative attributions often transcend it. Pronouns (in the form of anaphors or cataphors or, alternatively, anadeictics or cata-deictics) could, for instance, refer to the frame without the frame-evoking expression Finanzinvestor/en itself appearing. It is therefore necessary to extract entire paragraphs, or at least the two previous and subsequent sentences.545

In Lönneker’s semi-automatic frame analysis, most of the cross-sentence predica-tions are lost because they cannot be resolved without the help of contextual data.546 Although this restriction does not pose a serious problem for Lönneker’s study, since the set of data is large enough to compensate for the omissions, by contrast distortions in the results can quickly arise in discourse-analytical studies that generally oper-ate with significantly smaller text corpora. Discourse-deictic references (by means of pronouns of varying kinds) should therefore essentially be resolved, i.e. be reformed as explicit predications.

Apart from discourse deictic references, there are some difficult cases that are dif-ficult to resolve. Here are two examples to illustrate this.

543. In German, an example of an inflected word form is the genitive attribute des Finanzinvestors (‘of the financial investor’) and the derivation Finanzinvestorenschaft (‘established group of fi-nancial investors’). However, these and other possible derivations are not found in the corpus.

544. In this case, the compound constructions would have to have at least three elements, like foreign financial investor/s. In contrast, compounds containing the element locusts – for instance locust list or locust rebuke – occur multiple times in the corpus.

545. Although text passages that extend beyond the two previous and the two following sentence can contain discourse-deictic references to financial investor/s, this practically never occurs.

546. Her explanation is: “The use of pronouns as fillers makes no sense in a list of proposi-tions in a frame, since it would not be possible to trace which concept it is intended to refer to.” (Translated from Lönneker 2003a: 74). This cannot be achieved with a computer-based proce-dure, and the enormous volume of Lönneker’s text corpus does not allow a manual procedure.


(14) “Für die Zielunternehmen gibt es unterschiedliche Anlässe, Finanzinvestoren an Bord zu holen. In Deutschland geht es häufig darum, die Nachfolge in Familienbetrieben zu regeln.”

‘There are difference reasons for targeted enterprises to bring financial investors on board. In Germany it is often a matter of solving succession issues in family-run businesses.’ (Süddeutsche Zeitung 3 May 2005; italics not in original)

(15) “Ausgerechnet einer der Käufer von damals, der Finanzinvestor Apax, findet sich nun auf jener Heuschrecken-Liste wieder, in der die Sozialdemokraten einige ihrer Ansicht nach besonders verwerflichen kapitalistischen Umtriebe geißeln. Apax zähle zu jenen ‘Ankäuferfirmen’, die ein Unternehmen, so der aktuelle Vorwurf der Genossen, nur erwerben, um es später zu zerschlagen oder mit üppigem Gewinn weiterzuverkaufen.”

‘Of all things, one of the previous buyers, the financial investor Apax, can now be found on the locust list, a list berating some of what the Social Democrats view as the most reprehensible capitalistic machinations. Apax counts among those ‘buy-up companies’ that, according to the current accusation made by the comrades, only acquire a company to later destroy it or to sell it off at a huge profit.’ (Süddeutsche Zeitung 3 May 2005; italics not in original)

The second sentence in (14) retrospectively explains what concrete reasons there are for bringing “financial investors on board”. A more explicit version would be:

(14′) In Germany, when financial investors are brought on board, it is often a matter of…

Once this connection is recognized, then (14′) specifies [financial investors] more closely, to the effect that financial investors can solve succession issues in family businesses.

Example (15) functions similarly. Mediated via the quasi-explicit predication ‘Apax is a financial investor’ (which is derived from the apposition the financial investor Apax), the first sentence predicates a variety of things about financial investors.547 The following sentence also contains predications that specify the frame financial inves-tors without the frame-evoking expression financial investors having to appear. In fact, it is sufficient that the company name “Apax” is mentioned again. A theme-rheme analysis may assist here in explaining the reference structure.

It should be noted at this point that the predication’s scope extends far beyond the sentence boundary and that the corpus-based analysis must take this into account by, as a rule, including in the annotation at least two sentences before and two sen-tences after the one containing the frame-evoking expression financial investor/s. In individual cases it is possible that more than two sentences have to be recorded. Thus, this should be determined from case to case depending on the pronominal reference to the frame-evoking expression.

547. That is: ‘financial investors are buyers’ and ‘financial investors can now find themselves on the locust list’.


7.3.2 Predication analysis

From the extracted textual records, we are then interested in the set of all predications by which the frame in question (here: financial investors) can be more precisely specified. These correspond to the set of fillers that form instances in this frame. As in Example (15), most of the textual records contain several predications. In the following, when I refer to examples of how quasi-explicit predications are to be transferred into explicit predications, I follow the specifications in Section 6.4.2.

To simplify the predication analysis, it is advisable to re-form all the extracted sentences so that the frame-evoking expression (in our case, financial investor/s) only occurs in the subject position (cf. Lönneker 2003a: 73). This provides structural uni-formity and better empirical manageability. If the expression financial investor/s func-tions as the object of the sentence, two cases must be distinguished from each other. Either the expression itself constitutes the nominal phrase or it only functions as an attribute within the nominal phrase.548 Compare the following Examples (16) and (17):

(16) “Müntefering hatte Finanzinvestoren kritisiert, die es oft an jeder Verantwortung fehlen ließen.”

‘Müntefering had criticized financial investors who often lacked any sense of responsibility.’ (Frankfurter Allgemeine Zeitung, 4 May 2005; my italics)

(17) “In diesem Zusammenhang wies die SPD-Abgeordnete Nina Hauer auf den Kampf angelsächsischer Finanzinvestoren um die Führung der Deutschen Börse hin.”

‘In this context SPD member Nina Hauer drew attention to the battle among Anglo-Saxon financial investors for the leadership of the German stock exchange.’ (Frankfurter Allgemeine Zeitung, 22 April 2005; my italics)

In the first case the most obvious course is to convert the main clause into a passive construction.

(18) Financial investors are criticized by Müntefering.

Example (17) is more difficult to resolve. Here, financial investors occurs as a subjective genitive within a nominal phrase. In such cases, the predication’s scope is restricted to the complements of the nominal phrase. Accordingly, the resolution of the phrase “battle among Anglo-Saxon financial investors for the leadership of the German stock exchange” reads as:

(19) Anglo-Saxon financial investors battle for the leadership of the German stock exchange.

548. Lönneker also names a third case, namely phrases with object attributes (like financial investors find the rebuke inappropriate), by which she understands phrases with “non-verbal predicates” (Lönneker 2003a: 73), in this case: inappropriate. However, phrases with object at-tributes only occur very rarely in the corpus.


In this I am following von Polenz’s observation that predicative expressions can be expressed through nouns (here: battle) in verbal phrases because the more significant sentence-semantic function of nouns (excluding proper nouns) is not referentialization but rather predication (von Polenz 1985: 110).

The predication analysis of (16) and (17) does not end at this point since the relative clause in (16) anaphorically picks up the frame-evoking expression financial investors using a relative pronoun.549 In such cases, the relative clause is amended to become a syntactically complete sentence:

(20) Financial investors often lack any sense of responsibility.

An analogous procedure is used for infinitive and gerund constructions.550 To conclude the predication analysis of (17), it is only now necessary to resolve the nominal phrase Anglo-Saxon financial investors:

(21) Financial investors are of Anglo-Saxon origin.

Apart from nominal phrases of the type Anglo-Saxon financial investors, multiple at-tributions like large, international financial investors can be found in the corpus. In such cases, of course, there are multiple predications that must therefore be resolved; thus, for instance the latter nominal phrase is resolved into two predications:

(22) Financial investors are large. (23) Financial investors are of international origin.

All of the completed predications are to be individually listed and collated in this vein.It is important that only explicit and quasi-explicit predications are taken into

account in the sense outlined above and not presuppositions, implicatures, or other-wise inferred knowledge.551 This is because the analytical criteria for implicit predi-cations, as we have already ascertained, then become “problematic because they are highly theory-dependent and contingent on description” (translated from Busse 2000a: 44–45). To achieve maximum objectivity, inferences (or more precisely: invoked

549. In the analysis, I do not take into account the perspectivation of (16) that may be expressed in German by the use of a relative clause with the subjunctive (in distinction to a that-clause or an infinitive clause). The chosen variant could indicate that the author agrees with Müntefering’s criticism to some degree. However, in my view this remains mere speculation. In any case, its frame-semantic relevance tends to be marginal.

550. Further evidence and differentiation can be found in Lönneker (2003a: 74–77).

551. Thus, on the basis of the corpus entry Financial investors buy up companies one could con-clude that financial investors are rich or become sole owners due to the purchase. However, such inferences are the result of foreknowledge not specific to the discourse, and there is evidence to suggest that a divergent concept is activated within the discourse surrounding the capitalism debate. In fact, in the corpus there is a significant number of text entries, which suggests that the money comes from pension funds, i.e. that it is borrowed money, as it were.


knowledge) must be ignored, even when they positively obtrude and they seem without doubt to be of vital importance to the process of understanding.552

The result of step (ii) is a structurally uniform set of propositions. The frame-evoking expression specifies in every proposition what the reference object is that particular properties are being assigned to by means of predication.

7.3.3 Hyperonym type reduction

The set of predications (in the predicate-logic sense) identified in the second ana-lytical step corresponds to all the instances (fillers) in the frame under investigation that are to be found in the text corpus. Of course, this data does not state any-thing about the meaning of the frame-evoking expression financial investor/s that has emerged in the discourse of the capitalism debate. More precise information about the U-relevant facets of knowledge is only evident once the predicates we have identified have been quantitatively evaluated with respect to their frequency of occurrence in slots. But which slots does the frame in question exhibit? Or more precisely: Which matrix frame does a hyperonym type reduction allocate the expres-sion financial investor/s to?

In Section 6.3.3, I already demonstrated how a hyperonym type reduction is con-ducted using the lexeme locust as an example. To begin with, let us recall the results without having to elaborate on the details: Locust can be successively retraced to the highest hyperonym organism, whose matrix frame bequeaths all the slots to the frame evoked by the expression locust.

To maintain a compact overview of the slots without the frames losing analytical power, it is useful to envisage the slots in the matrix frame organism in the form of predicator classes (cf. Section 6.3.3). However, this corpus-based analysis does not follow the predicator classes proposed by Konerding (1993: 399–462), but rather Lönneker’s (2003a: 262–277) empirically validated classifications. They are summa-rized in Table 1.553 The predicators are in turn thematically divided into classes.

552. Knowledge that is produced specific to the discourse and is presupposed within it should be as distinct as possible from the knowledge of the researcher; cf. Sandra and Rice (1995), who ironically ask: “Mirroring whose mind – the linguist’s or the language user’s?”

553. As is evident in the frame hierarchy of Lönneker (2003a: 93; see Figure 5 in Section 6.3.2) the matrix frame organism inherits all the slots of the higher frame in the hierarchy (in this case, ‘discontinuous object’, ‘primary object’, ‘stable object’, ‘entity’). Accordingly, the predicators listed in Table 1 are taken from the frames discontinuous object (Lönneker 2003a: 266), primary object (Lönneker 2003a: 265), stable object (Lönneker 2003a: 264) and entity (Lönneker 2003a: 263). More precise details on the predicators can be found in Lönneker (2003a) at the given references.


Table 1. Predicators in the matrix frame organism, based on and translated from Lönneker 2003a: 262–277

– Definition

Additional names/designations for the organismPredicators to characterize similar organismsPredicators to characterize superordinate categories that are attributed to the organismPredicators to characterize the theories in which the organism plays a role

– Relevance of the organism for humans

Predicators to characterize the relevance of the organism for humansPredicators to characterize the usefulness that the behaviour of the organism has for humans and to characterize the consequences that the behaviour of the organism can have for humansPredicators to characterize the entity of which the organism is a sign/evidencePredicators to characterize the degree to which the organism is knownPredicators to characterize the production processes in which the organism participates

– Constitutive relations I: Part

Predicators to characterize the superordinate whole of which the organism is a partPredicators to characterize the nature of the relationship of the organism to the superordinate whole

– Constitutive relations II: Totality

Predicators to characterize the composition of the organismPredicators to characterize the properties of the parts that constitute the organismPredicators to characterize the arrangement of these partsPredicators to characterize the functions of these parts

– Constitutive relations III: Process

Predicators to characterize the origin of the organismPredicators to characterize the phases of existence of the organismPredicators to characterize the conditions under which the organism dies and to characterize the reasons for this deathPredicators to characterize the nature of the emergence and manifestation of the organismPredicators to characterize the prerequisites for the emergence of the organismPredicators to characterize the conditions of existence of the organismPredicators to characterize the entities which manifest themselves before the organism appearsPredicators to characterize the entities which manifest themselves after the organism appearsPredicators to characterize the conditions under which other entities appear after the organism disappears

– Properties

Predicators to characterize the dimensions and proportions of the organismPredicators to characterize other properties of the organismPredicators to characterize invisible properties of the organismPredicators to characterize the whereabouts of the organism or where it appearsPredicators to characterize the particular division properties of the organismPredicators to characterize the form of the organismPredicators to characterize the appearance of the organism


Predicators to characterize the physical and biological properties of the organismPredicators to characterize the abilities of the organismPredicators to characterize the habits of the organismPredicators to characterize the behaviour of the organism

– Constitutive relations IV: Part of an event and/or action

Predicators to characterize the events and/or actions in which the organism plays a rolePredicators to characterize the function that the organism adopts in these events and/or actions

– Phases of existence and distribution

Predicators to characterize the conditions for the non-existence of the organismPredicators to characterize the potential of the organism to destroyPredicators to characterize the (geographical and social) distribution of the organismPredicators to characterize the nature of the propagation of the organism

In the capitalism debate, the expression locust becomes a metaphor by referring to the input [financial investor]. What about the predication potential of this semantic unit? In the following, I initially conduct a hyperonym type reduction of the expres-sion financial investor. In contrast to the case of locust in Section 6.3.3, my outline here is brief, especially since a glance at the list of matrix frames accounted for already provides an idea of the result.

The fact that the lexical unit [[financial]/[investor]] is a compound and thus a complex semantic unit first of all raises the question of which part of the compound should be the object of the hyperonym type reduction. The following must be said about this: The complex word form financial investor is a determinative compound. The semantic unit of the first constituent (determinans, or modifier) [financial] specifies the semantic unit of the second constituent (determinatum, or head) [investor]; it thus constitutes an instance in the schema [investor]. Whereas the determinatum investor specifies the reference object, the determinans financial attributes a property to it. This occurs in the form of a predicative attribution, leading to the frame in question becoming more specific, i.e. its predication’s becoming more restricted because one of its slots is occupied by a value.554 The frame-evoking expression is thus investor, and it alone is subjected to the hyperonym type reduction.

This is reflected in the hyperonym type reduction. That is to say, the first step in the reduction consists in reducing financial investor to its hyperonym investor. Nevertheless, these considerations were not superfluous, since there is no entry for the expression

554. In general terms, the more instances a frame contains, the fewer the occurrences to be found in the corpus. For this reason, to guarantee as broad a data base as possible in a corpus-based analysis, it is advisable to give priority to simple word forms (like investor) as research objects rather than complex word forms or nominalized syntagma, etc. Although with increasing specification the number of occurrences is lower, the complex word form financial investor forms the basis of this analysis because it (and not investor) figures as the source domain.



Finanzinvestor (financial investor) in the Duden Deutsches Universalwörterbuch (1989).555 All further steps in the reduction are summarized in Table 2.

Table 2. Translation of the hyperonym type reduction of investor based on the Duden Deutsches Universalwörterbuch (1989)556

Second reduction:investor: …someone who commits capital for financial gain (jmd. der anlegt); financier (Anleger)…

→ Hyperonym: someone (jemand); synonym: financier (Anleger)556

Third reduction:someone: …designates a particular person not known to or more closely specified by the speaker…

→ Hyperonym: person

We have already reached the final element in the reduction chain after the third step: Person is one of the categories accounted for by Konerding that cannot be further retraced to other hyperonyms.

However, in Section 6.3.2 I pointed out that is has proven to be empirically prac-tical to refine the hyperonym type ‘person’ such that a distinction is made between ‘person with a temporary or permanent property or disposition’ and ‘person in a pro-fessional role’ (Konerding 1993: 185). Accordingly, the frame that financial investor evokes exhibits those slots that are inherent in the matrix frame person in a profes-sional role.557 Here, too, I orient myself towards the predicator classes devised by Lönneker (2003a: 262–277), which are listed in Table 3. The predicators are in turn thematically divided into classes.558

555. For reasons of consistency, since Konerding’s hyperonym type reduction is based on the dictionary Duden Deutsches Universalwörterbuch (1989), this is also taken as the basis here.

556. Although in contrast to financier (‘Anleger’), someone (‘jemand’) is a pronoun (and not a noun), it nonetheless has a hyperonymic character. If we follow financier further, a circular ref-erential structure becomes evident at this stage already; In Duden (1989), financier is explained by the entry investor.

557. In analogy with the matrix frame organism, the matrix frame person in a professional role inherits all the slots of the hierarchically higher frame (in this case, ‘role of/perspective on an object’, ‘stable entity’, ‘entity’). Thus, the predicators listed in Table 1 are a set of predications from the frames role of/perspective on an object (Lönneker 2003a: 269), stable entity (Lönneker 2003a: 264), and entity (Lönneker 2003a: 263). More precise details on the individual predicators can be found in Lönneker (2003a) at the given references.

558. Despite following Lönneker (2003a: 267–277) in this respect, I nonetheless undertake some corrections. Thus, Lönneker does not seem to have consistently grouped the predicators in the frame role of/perspective on entity (2003a: 269) in terms of topic. Some of the predicators listed (for instance predicators to characterize the entities that appear before the appearance of the person) should correctly belong to the class ‘constitutive relations III’.


Table 3. Translation of predicators in the matrix frame person in a professional role, based on Lönneker 2003a: 262–277; “Person” is short for “person in a professional role”

– Definition

Additional names/designations for the personPredicators to characterize people with similar professionsPredicators to characterize superordinate categories that the person is assigned toPredicators to characterize the theories in which the person plays a role

– Relevance of the person (or profession) for humans

Predicators to characterize the relevance of the person for humansPredicators to characterize the usefulness of the person (in the professional role) for humansPredicators to characterize the entity of which the person is a sign/evidencePredicators to characterize the degree to which the person is knownPredicators to characterize the entities which manifest themselves with the appearance of the personPredicators to characterize the status of the profession in society


Predicators to characterize the superordinate whole of which the person is partPredicators to characterize the nature of the relationship of the person to the superordinate whole


Predicators to characterize the “composition” (body parts, etc.) of the personPredicators to characterize the properties of the parts that the person consists ofPredicators to characterize the arrangement of these partsPredicators to characterize the functions of these parts


Predicators to characterize the origin of the personPredicators to characterize the phases of existence of the personPredicators to characterize the conditions under which the person dies and to characterize the reasons for this deathPredicators to characterize the nature of the emergence and appearance of the personPredicators to characterize the prerequisites for conducting this professionPredicators to characterize the prerequisites for achieving this professionPredicators to characterize the entities which manifest themselves before the appearance of the personPredicators to characterize the entities which manifest themselves after the disappearance of the personPredicators to characterize the conditions under which other entities appear after the person disappears

– Properties

Predicators to characterize the dimensions and proportions of the personPredicators to characterize other properties of the personPredicators to characterize the knowledge the person has of their profession



Predicators to characterize the events and/or actions in which the person plays a rolePredicators to characterize the function that the person adopts in these events and/or actions

– Manifestation, form of expression of the property, of the state/part, of the whole, and the Constitutive relations

Predicators to characterize the consequences for the event in which the person plays a rolePredicators to characterize the consequences for other people who play a role in the event

– Activities

Predicators to characterize the actions in which the person participatesPredicators to characterize the co-participants in the actionPredicators to characterize the necessary and used meansPredicators to characterize the duties and obligations of the person

It is no coincidence that some of the predicators in the matrix frame person in a professional role also occur in the matrix frame organism. The reason for this is that there are two frames that are higher in the hierarchy (so-called “superframes” or “top-level frames”; cf. Lönneker 2003a: 89; Minsky 1975: 274, 221–222) that bequeath their slots to both matrix frames.559

As a result of step (iii) a grid for semantic analysis in the form of predicators is available with which the explicit predications identified in step (ii) can be grouped together and thus quantitatively evaluated from a semantic perspective.

7.3.4 Classification of explicit predications

The quantitative analysis of the explicit predications occurs in the final step of the corpus-based analysis. Every explicit predication extracted in step (ii) is to be matched to a predicator in the corresponding matrix frame. That means that all the predicative specifications of the reference object ‘locust’ are matched with the predicators in the matrix frame organism. Then, all the predicative specifications of the reference object ‘financial investor’ are classified with respect to the predicators in the matrix frame person in a professional role.

How does this process of allocating explicit predications to a predicator proceed? Can each predication be allocated to precisely one class or are there multiple options? These questions are closely related. Part of an explicit predication is the finite verb that establishes a conceptual relation. It connects the frame (that the expression in subject position evokes) with particular fillers. Thus, for every predicator verbs can be specified with which typical predications of the corresponding predicator can be performed.560 (24) to (25) are examples translated from the text corpus:

559. These superframes are stable object and entity; cf. Figure 5 in Section 6.3.2.

560. Cf. Lönneker (2003a: 66–74, 262–277). On the basis of her corpus, Lönneker provides such typical verbs for every predicator.


(24) Ein Finanzinvestor heißt Carlyle Group. ‘One financial investor is called Carlyle Group’ (25) Finanzinvestoren stammen aus England. ‘Financial investors come from England’

(26) Finanzinvestoren treten als Konsortium auf. ‘Financial investors act jointly as a consortium’ (27) Finanzinvestoren sanieren marode Unternehmen. ‘Financial investors clean up run-down companies’

The copular nominative in (24), constituted in German through the copula verb heißt and in English through the copula verb is together with the passive clause comple-ment called, serves as a typical predicator for ‘additional names/designations for the person’, whereas in (25) the complex verb form stammen aus (come from) represents a good example of a ‘predicator to characterize the origin of the person’. And in (26) the complex verb act jointly as (‘auftreten als’) is also a typical ‘predicator to character-ize a superordinate category”, just as in (27) the verb clean up (‘sanieren’) figures as a ‘predicator to characterize the duties and obligations of the person’ (cf. Table 3). With the two initial questions in mind, these examples show that by using typical verbs (i.e. verbs through which the nature of a predicator class is particularly well reflected) it is possible to unambiguously match a particular predication to a particular predicator. At the same time, such typical verbs simplify the classification of predications that consist of other verbs. They are prototypical representatives of predicators and thus help in evaluating whether a performed predication belongs to a particular predicator or not. For example, the copular nominatives expressed by the verbs to name oneself (‘sich nennen’) and to denote onself (‘sich bezeichnen’) can be relatively quickly identified as predicators that allow the attribution of alternative names or designations by being compared with the typical copula is (‘heißen’) – as in (24).

However, although in this sense (24) to (27) can be seen as prototypical examples, many other predications cannot be so unambiguously matched to a predicator. Their classification is more difficult. To illustrate this let us examine (28) and (29):

(28) Finanzinvestoren kaufen Unternehmen auf. ‘Financial investors buy up companies’ (29) Finanzinvestoren zerstören Arbeitsplätze. ‘Financial investors destroy jobs’

To begin with, it is clear that the verb buy (‘kaufen’) addresses an activity of financial investors. The verb makes a statement about an action in which financial investors participate.561 However, apart from this there is no reason that buy cannot also be clas-sified as a ‘predicator to characterize the prerequisites for conducting a profession (cf.

561. Buy thus belongs to the predicator that ‘characterizes the actions in which the person participates’; cf. Table 3.


Table 3); financial investors could hardly conduct their profession without appearing in the role of buyers. Thus, in this case multiple allocations can be made.

However, such multiple allocations are only conditionally possible. It is important to recall that implicit predications, regardless of their form, must not motivate alloca-tion to a predicator. Therefore, in (28) the predication ‘buy companies’ may not serve as a ‘predicator to characterize the consequences for other persons…’ despite the fact that this is a prima facie implication. This is because the fact that the purchase of a company by financial investors entails consequences for the employees of a company is an inference that we draw on the basis of background knowledge. The verb buy does not establish a conceptual relation to the consequences that the action of buying may entail; such a correlation rests solely on invoked knowledge. In fact, the verb merely specifies which activity financial investors are carrying out. Therefore, evoked, but not invoked knowledge, may have an influence on the classification of predications.562

This becomes particularly clear when (28) and (29) are contrasted. The verb buy in (28) and destroy (‘zerstören’) in (29) belong to those predicators that characterize an action of financial investors. In contrast to (28), however, the predication destroy jobs (‘Arbeitsplätze zerstören’) has consequences for other persons. Conversely, the destruction of jobs is not (prototypically) a prerequisite for conducting the profession of a financial investor, but the purchase of companies certainly is.

To conclude, I collate the final insights in one criterion for governing the alloca-tion of explicit predications to predicators and use examples to illustrate it. Konerding proposed employing so-called “function-oriented polar questions” in his procedure for hyperonym type reduction. The relevant passages are quoted once again:

‘Can the verb … Y in category Z be meaningfully predicated by a typical reference object of the noun X…?’ or: ‘Can one meaningfully claim of a typical table that it is part of something, figures in something, is involved in something?’… (Translated from Konerding 1993: 165)

Applied to our circumstances, that means that when deciding whether a performed predication should be allocated to a predicator (or a predicator class) it is necessary to analogously ask whether the predication in question can be meaningfully allocated to a predicator. For instance, is it possible to meaningfully claim that the destruction of jobs characterizes a prerequisite for conducting the profession of a financial investor? And is it possible to meaningfully claim that the destruction of jobs states something about the consequences for other people? Using such polar questions clarifies to which predicator the predication destroy jobs should be allocated. It also becomes clear that, in contrast to the case in Lönneker (2003a), explicit predications must be manually and interpretively classified. These must be decided on a case-by-case basis to be able to fulfil the specified criteria and produce replicable and reliable results.

The classification of all of the explicit predications is followed by the quantita-tive and qualitative analysis of the results. As formulated by Fraas (translated from

562. For more on the distinction between invoked and evoked knowledge, cf. Section 6.3.1.


1996a: 78): “On this basis [the basis of the quantitative survey] it is possible to sys-tematically compare the mode of realization of particular concepts within particular periods of time or areas of communication.” The next section shows how the meaning of the financial investors as locusts metaphor is realized in detail in the discourse of the capitalism debate.

7.4 Empirical results

The corpus-based analysis orients itself conceptually towards four representational units that, according to CI (Fauconnier and Turner’s blending theory), participate in the constitution of metaphors:

i. Generic frame. The generic frame represents facets of knowledge that both inputs (the source and target domains of the metaphor) possess. This area of overlap naturally does not concern instances (fillers, default values), but rather possibilities of knowledge specification (slots).

ii. Input frame I. The source domain locust only contains those knowledge segments (fillers, default values) that potentially feed into the meaning of the metaphor.

iii. Input frame II. It also applies that not all knowledge segments, but only those that are potentially relevant to the metaphor, feed into the target domain financial investor/s.

iv. Metaphor frame. In the metaphor meaning, selected knowledge segments from both inputs are conceptually blended together to form a unit.

Which facets of knowledge (slots) and/or knowledge segments (default values, fill-ers) feed into these four representational units? The following sections investigate this question.

7.4.1 The generic frame

According to CI, facets of knowledge shared by both inputs – in this case, the frames locust/s and financial investor/s – are relevant to the formation of the metaphor (Fauconnier and Turner 1998a: 137–138). Together these constitute the generic frame (or “generic space”). Note that we are dealing with facets of knowledge; how these are respectively specified varies in both inputs. Regarding the example This surgeon is a butcher given in 7.1.2, Fauconnier and Turner state that both the input frame surgeon and the input frame butcher share knowledge facets like ‘agent’, ‘patient’, ‘location’, and ‘activity’, but that all of these aspects are specified differently. Whereas in the surgeon frame a surgeon (agent) operates (activity) on a patient (patient) in an operating theatre (location), in the butcher frame a butcher (agent) prepares for sale (activity) the meat of dead animals (patient) in the butcher’s shop (location).


Proceeding from the frame-semantic approach developed in the previous chapters, we have effective means for concretely specifying the elements in the generic frame. Bear in mind that on the one hand locust can be retraced to organism and on the other financial investor to person (in a professional role) by means of hyperonym reduction. Since the slots of the matrix frame of the same name correspond to the slots that both input frames share with each other, the generic knowledge facets can easily be identi-fied using the frame hierarchy introduced in 6.3.2. As illustrated in Figure 3, there are two so-called “superframes” that the matrix frames organism and person in a professional role share, namely stable object and entity. The set of slots in these superframes form the generic frame. They encompass all the meaning aspects that are potentially relevant to the metaphor. Which elements of the meaning potential actually enter into the metaphor meaning remains open at this point.

Thus, since the number of meaning aspects that are potentially relevant to the metaphor meaning is restricted to the intersecting set of slots between the matrix frames organism and person in a professional role (or, alternatively, the locust and financial investor frame), the slots in the superframe object of a natural kind and primary object, as well as role/perspective on the entity are hence-forth irrelevant. A brief look at the individual slots in this superframe reveals why. For instance, the financial investor frame can be particularized with respect to the co-actors in an action in which a financial investor is involved, whereas this speci-fication option is missing in the locust frame (and in the matrix frame organism). Even when locust is used metaphorically and the target domain allows predicative

Object of anatural kind

Organism

Person in aprofessional role

Stableobject

Primaryobject

Role/perspective onthe entity

Entity

Inpu

t-sp

eci�

csu

perf

ram

e G

ener

icsu

perf

ram

e

Figure 3. Translation of generic and input-specific superframes for the matrix frames organism and person in a professional role (based on Lönneker’s frame hierarchy, 2003a: 93)


attributions concerning the participants in an action (as in the case of financial inves-tor immediately above), this facet of knowledge cannot become part of the metaphor meaning since it is not expressible through the source domain.563 This applies similarly to the other input-specific slots.

Table 4 summarizes the slots occurring in the superframes entity and stable object that are relevant to the metaphor analysis. The names of the matrix frames have already been replaced by the frame-evoking expressions locust and financial investor.

Table 4. Shared slots in the frames locusts and financial investors (resp. the matrix frames organism and person in a professional role), based on and translated from Lönneker 2003a: 262–277

– Definition

Other names/designations for locusts and financial investorsPredicators to characterize superordinate categories that the locusts and investors are allocated toPredicators to characterize similar entitiesPredicators to characterize the theories in which locusts and financial investors play a role

– The relevance of locusts and financial investors for humans

Predicators to characterize the relevance of locusts and financial investors for humansPredicators to characterize the usefulness of locusts and financial investors for humans and to characterize the consequences that locusts and financial investors can have for humansPredicators to characterize the entity of which locusts and financial investors are a sign/evidencePredicators to characterize the degree to which the locusts and financial investors are known


Predicators to characterize the superordinate whole of which the locusts and financial investors are partPredicators to characterize the nature of the relationship of locusts and financial investors to the superordinate whole


Predicators to characterize the “composition” of locusts/financial investorsPredicators to characterize the properties of the parts that locusts and financial investors consists ofPredicators to characterize the arrangement of these partsPredicators to characterize the functions of these parts

563. Bear in mind that slots are an analytical construct that can be empirically deduced from a large set of similar predications. Thus seen, a slot is a routinized predication option. Slots arise relative to a prevailing ontology of everyday life; if areas change in the ontology of everyday life, then the slots in a frame will – possibly – change. (Consider, for example, the consequences the Copernican turn entailed for the prevailing ontology of everyday life; without doubt it caused fundamental changes in numerous frame structures.)



Predicators to characterize the origin of the locusts and financial investorsPredicators to characterize the phases of existence of the locusts and financial investorsPredicators to characterize the conditions under which locusts die or financial investors cease to carry out their professionPredicators to characterize the reasons for thisPredicators to characterize the nature of emergence and the manifestation of the locusts and financial investorsPredicators to characterize the prerequisites for the “emergence” of locusts and financial investorsPredicators to characterize the conditions of existence of locusts/financial investors

– Properties

Predicators to characterize the dimensions and proportions of locusts and financial investorsPredicators to characterize other properties of locusts and financial investors


Predicators to characterize the events and/or actions in which the locusts and financial investors play a rolePredicators to characterize the function that the locusts and financial investors adopt in these events and/or actions

Not all of the predicators (or slots) listed in Table 4 are relevant for the metaphor addressed. The following sections show that many only have a marginal semantic relevance, whereas others, by contrast, appear to be all the more important for the metaphor meaning.

7.4.2 The input frames locust/s and financial investor/s

The slots with concrete values (predications) listed in Table 4 are in the input frames. Some of these values enter into the metaphor meaning, whereas others are, as it were, in quarantine, i.e. although they are represented in the input, they do not constitute an aspect of meaning relevant to the metaphor.564 Before discussing the metaphor mean-ing, or blended space, in the next section, I first analyse the two inputs.

The knowledge domain locust constitutes the first input. To specify this domain, a corpus is required in which locust is not used metaphorically – in contrast to its use in the capitalism debate. That means a reference corpus is required. To constitute a reference corpus I have recurred to the databank Cosmas II provided by the Institut für Deutsche Sprache (Institute of German Language in Mannheim). From the data bank I

564. This is the case in our example This surgeon is a butcher with respect to details about the person (age, looks, clothing, etc.). Other information units only feed into the blended space from one input, such as information about the location (in this case, the operating theatre, not a butcher shop).


selected the years 1997–1999 of the national newspapers Frankfurter Rundschau, Die Zeit, and Berliner Morgenpost. In these three years there are a total of 182 occurrences of the lexeme Heuschrecke (locust) in a non-metaphorical reading.

The procedure was similar for the second input, the knowledge domain financial investor. Overall, the newspapers in question yielded 122 occurrences of the lexeme financial investor (‘Finanzinvestor’) in these three years. I also took recourse to the reference corpus for the second input, because it can be assumed that the meaning of the lexeme financial investor undergoes a discourse-specific mintage within the capital-ism debate. Under certain circumstances, such semantic traces may be interesting for the analysis of the metaphor meaning, as we later see.

For both knowledge domains the pertinent text references were annotated and reformulated as explicit predications, which were then allocated to the predicators (slots) in the relevant matrix frame (cf. Section 7.3). The result was information about how often the individual predicators and predicator classes were “served”.

To initially gain an overview of the dominance of certain classes of slots at a higher level of abstraction, Table 5 does not list individual predicators. Rather, the percentages indicate how high the proportion of a predicator class is in the set of all predications.565 Both inputs are determined on the basis of the reference corpus, as shown in the second and third columns. The column on the extreme right refers to the financial investor frame in the capitalism debate.

What conclusions can be drawn from the data in the reference corpus? Against the background that selected information from both inputs is reflected in the metaphor meaning, it is particularly interesting to know which facets of knowledge are domi-nant in the inputs and whether there are knowledge facets that are significantly more dominant in the locust frame than in the financial investor frame (or vice versa). If there are facets of knowledge that are more dominant, they should be of particular relevance for the metaphor. A comparison between the locust and financial inves-tor frames shows, however, that from a quantitative perspective none of the predicator classes of either frame are significantly more dominant than its counterpart in the other frame. In the locust frame some properties and processual aspects (including, for instance, ‘origin’ and ‘phases of existence’ of an entity; cf. Table 5) are more frequently specified than in the financial investor/s frame, but the difference is negligible. The fact that the quantitative weighting of individual predicator classes in both frames is very similar seems to substantiate what was shown at the beginning. At least on the level of the predicator classes, there is no facet of knowledge in the financial in-vestors as locusts metaphor that is more salient in one input than the other such that it motivates metaphorical mapping.566 The concrete arrangement of the semantic

565. The percentages have been rounded up or down to one digit after the decimal point.

566. By contrast, it can be assumed that the two inputs in the butcher metaphor differ more starkly from each other; however, the correctness of this assumption can only be confirmed by a (quantitatively supported) corpus-based analysis.


dimension of the metaphor locust is thus all the more shaped by discourse knowledge surrounding the capitalism debate.

Table 5. Percentage of predicator classes in the sum of all predications in the locust and financial investor frames

Slots or predicator classes Reference corpus Corpus on the capitalism debate

locust frame financial investor frame

financial investor frame

Definition 12.1% 14.8% 9.6%Relevance for humans 19.5% 14.8% 31.8%Constitutive relations I: Part 20.2% 21.1% 16.8%Constitutive relations II: Totality 5.6% 7.7% 5.5%Constitutive relations III: Process 6.2% 3.0% 4.9%Properties 2.3% 0.5% 0.2%Part of an event and/or action 34.2% 38.1% 31.2%

A comparison between the financial investor/s frame in the reference corpus and the financial investor frame in the capitalism debate corpus reveals that there are only very few significant deviations. However, one is quite striking: In the discourse context of the capitalism debate, financial investors are addressed particularly fre-quently (relative to other facets of knowledge) with respect to their ‘relevance for hu-mans’. Compared with the reference corpus, there are on average more than twice as many predications in the capitalism debate concerning, for instance, the influence of financial investors on society (or parts of society). Examples for this are references such as the following:567

(30) Finanzinvestoren vernichten Arbeitsplätze. ‘Financial investors destroy jobs’ (Frankfurter Allgemeine Zeitung, 18 April 2005) (31) Finanzinvestoren gefährden unsere Demokratie. ‘Financial investors endanger our democracy’ (Der Tagesspiegel, 22 April 2005) (32) Finanzinvestoren helfen Rentenkasse aus der Not. ‘Financial investors help public pension fund out of difficulties’ (Süddeutsche Zeitung, 3 May 2005)

567. The corpus entries given here and in the following are not quotes in the strictest sense – that means that text sequences are not literally quoted from the newspapers, but rather that the propositions resulting from the analytical steps described in Section 7.3 are given.


The fact that this facet of knowledge plays a significantly more important role within the capitalism debate than outside of it may well be an example of how discourse structures become reflected in lexical meanings. Further, the fact that the influence of financial investors within the capitalism debate is more dominant than in the refer-ence corpus seems to support the assumption that a metonymic shift is involved in the constitution of the metaphor meaning. At least it is clear that financial investors are addressed with significantly more frequency in terms of the consequences of their activities in the capitalism debate. This promotes a causal-metonymic usage of financial investor/s, according to which the causes (activities of financial investor/s) represent the effects.

In addition, by far less significant, yet nonetheless worth mentioning, is the fact that predications concerning “definitions” (in the broadest sense) of financial inves-tors occur on average more frequently in the reference corpus. Here we find instances such as:

(33) Finanzinvestoren sind “Private Equity”-Firmen wie Bridge Investment und Deutsche Effecten- und Wechselbeteiligungs-AG.

‘Financial investors are “private equity” companies like Bridge Investment and Deutsche Effecten- und Wechselbeteiligungs-AG’

(Frankfurter Rundschau, 18 May 1999) (34) Finanzinvestoren nennen sich einige Mitarbeiter der Deutschen Bank ‘Financial investors is what some employees of the Deutsche Bank call them-

selves’ (Berliner Morgenpost, 14 November 1998)

Evidently in the context of the capitalism debate it is less important to precisely specify what financial investors actually are, whereas in the reference corpus individual fi-nancial investors are more often addressed. Thus, this may explain why events and actions in which financial investors are involved play a greater role in the reference corpus. The reference corpus deals less with financial investors in general and more with the concrete actions of particular financial investors.568 On the other hand, it may be that the events and actions in which financial investors are involved are al-ready presupposed as given in the capitalism debate and therefore need no longer be (quasi-)explicitly addressed. However, whether predicators of this nature actually are presupposed as default values would only be able to be validated by means of further corpus-based analysis.

Before turning to the metaphor financial investors as locusts, I wish to il-lustrate the results of the predicator class analysis on a lower level of abstraction. Are

568. The following propositions from the Frankfurter Rundschau can be considered as typical: “Financial investors will sell off 600,000 shares cheaply in the next few months” (‘Finanzinvestoren geben in den nächsten Monaten 600.000 Aktien ab’, 2 May 1998); “Financial investors buy Kohap Emtec” (‘Finanzinvestoren kaufen Kohap Emtec’, 3 July1999); “Financial investors include the investment management company Schroder Ventures and DB investor” (‘Finanzinvestoren sind die Kapitalanlagegesellschaften Schroder Ventures sowie DB Investor’, 24 December 1999).


there particular predicators (slots) in the predicator classes that are significantly more dominant than others in the same class? To answer these questions, I select particular predicators from the predicator classes to serve as examples (cf. Table 4).

Indeed, there are some predicators that are very dominant within their own predi-cator class. Of those listed in Table 4, this is the case for the predicators to characterize the superordinate whole in which locusts or, alternatively, financial investors occur. A typical verbal phrase for these predicators is to be part of/an item of something else. The following examples occur in the corpus:

(35) Heuschrecken sind in Afrika Teil des Essens. ‘Locusts are an item of food in Africa’ (Frankfurter Rundschau, 17 May 1997) (36) Heuschrecken gehören zu den exotischen Delikatessen. ‘Locusts are an exotic delicacy’ (Frankfurter Rundschau, 17 March 1999) (37) Heuschrecken leben in Biotopen. ‘Locusts live in biotopes’ (Berliner Morgenpost, 11 June 1999)

Conversely, there are numerous predicators that play practically no role within the capitalism debate, such as those that concern the characterization of the dimensions and proportions or the “composition” of locusts or financial investors.569

Overall it is important to note that the predications identified from the corpus have very much the same quantitative distribution across the predicator classes (Table 5) as particular predicators (slots) do (Table 6). In this respect, the analysis of predica-tors provides no insights that were not already evident from the investigation of the predicator classes. For this reason I dispense with any further elucidation of the results summarized in Table 6.

One result that it is important to note is that in the reference corpus knowledge facets (predicators/predicator classes) shared by the frames financial investor/s and locust/s are equally strong in quantitative weighting. Accordingly, on the abstraction level of slots and slot classes, there are no facets of knowledge either from the locust/s input or the financial investor/s input that are particularly salient for constituting the metaphor, since no knowledge facet dominates more significantly in one input than the other. This finding is, however, somewhat relativized by the fact that recurrently occurring predications (in the sense of a high token frequency) can most certainly prove to be metaphor-relevant.

A second finding concerns the relationship between a discourse and its discourse elements, in this context, especially lexical units. Evidently the discourse context in which a linguistic expression recurrently appears influences the semantic structure of the linguistic expression. At least this is suggested by the finding that within the

569. Although the term “composition” is ambiguous here, I have nonetheless adopted it from Lönneker (2003a: 263). The relevant aspects in this sense are, in the case of locusts, the (body) parts of the insect, and in the case of financial investors, the components that belong to their profession.


capitalism debate one predicator class in the financial investor frame is more domi-nant than outside of the debate. The discourse surrounding the capitalism debate thus appears to semantically profile the financial investor frame. The high type fre-quency (here, with respect to the predicator class ‘relevance for humans’) indicates that the facets of knowledge in question are more relevant within the discourse than others.

From a frame-semantic view, the financial investor and locust frames ana-lysed here constitute the meaning potential for the metaphor. Following Allwood, I introduced the term “meaning potential” to describe the transition from a semantic unit (in Langacker’s sense) to a conceptual meaning (cf. Section 4.3.2). What is char-acteristic of the meaning potential of a linguistic expression is that, on the one hand, it “prestructures” the evoked frame while, on the other hand, only part of it is tapped into to feed into the conceptual meaning. Referring to our example, this means first of all that not all of the slots in the financial investor and locust frames feed into the input, but only those that are shared. Further, both frames are prestructured such that the slots and slot classes do not have the equivalent status. The different quantitative weighting shows that some are more relevant than others.

Table 6. Percentage of selected predicators in the sum of all predications in the locust and financial investor frames

Predicator classes and slots Reference corpus Capitalism debate corpus

locust frame financial investor frame


Definition:Name 3.7% 4.4% 2.9%Relevance for humans:Usefulness/consequences of the entity for humans 9.2% 8.8% 14.3%Constitutive relations I: Part, superordinate whole in which the entity plays a role

20.2% 21.1% 16.8%

Constitutive relations II: Totality, composition 2.0% 6.6% 4.1%Constitutive relations III: Process, phases of existence 1.2% 3.0% 1.4% Properties:Dimensions, proportions 1.8% 0.3% 0.2%Part of an event/action:Event/action in which the entity plays a role

17.2% 19.5% 16.2%


7.4.3 The metaphor frame

Which information units from the two inputs financial investor/s and locust/s enter into the metaphor meaning? Which facets of knowledge dominate in the meta-phor frame? To what extent do the knowledge facets accentuated in the metaphor meaning diverge from the target domain, i.e. from the financial investor frame in the capitalism debate? And to what extend is the metaphor frame distinct from the locust frame in the reference corpus? In the following, I attempt to approach an answer to these questions.

Once all occurrences of the metaphor have been identified and annotated in the corpus according to the specified criteria and the explicit predications have been re-corded and allocated to the corresponding slots,570 a quite differentiated data table emerges, which is summarized in Table 7. As before, the percentages indicate how high the proportion of a predicator class is in the set of all predications. In addition to the data in the locust frame, the data in the financial investor frame was also determined to enable a comparison between the two.

Table 7. Percentage of predicator classes in the sum of all predications in the locust and financial investor frames

Predicator classes and predicators belonging to it Corpus on the capitalism debate

locust frame (metaphor)


– Definition 18.0% 9.6%

Additional names/designationsSimilar entitiesSuperordinate categoriesTheories in which the entity plays a role

8.5% 0.2% 9.3% –

2.9% 1.6% 4.7% 0.4%

– Relevance for humans 14.2% 31.9%

Relevance for humansUsefulness of entity for humans/consequences that the entity can have for humansEntities of which locusts/financial investors are signs/evidenceDegree to which the entity is known

6.8% 7.0%

0.4% –

13.3%14.3%

3.3% 1.0%

– Constitutive relations I: Part 21.8% 16.8%

Superordinate whole of which the entity is partNature of the relationship of the entity to the superordinate whole

21.8%

–

16.8%

–

570. As a reminder, the set of relevant slots corresponds to those predicators that the locust and financial investor frames (or, respectively, the matrix frames organism and person in a professional role) share with each other.


Predicator classes and predicators belonging to it Corpus on the capitalism debate

locust frame (metaphor)


– Constitutive relations II: Totality 8.7% 5.5%

“Composition” of the entityProperties of the parts which make up the entityArrangement of these partsFunctions of these parts

7.0% 1.7% – –

4.1% 0.8% 0.6% –

– Constitutive relations III: Process 2.1% 5.0%

Origin of the entityPhases of existence of the entityConditions for the end of the entity (or for the profession no longer being performed)/reasons for the deathEmergence and manifestation of the entityPrerequisites for the emergence of the entity (or for acquiring the profession)Conditions of existence of the entity

0.4% – 0.3%

0.6% 0.4% 0.4%

0.2% 1.4% 0.4%

1.6% – 1.4%

– Properties 1.3% 0.2%

Dimensions and proportions of the entityOther properties of the entity

0.4% 0.9%

0.2% –

– Constitutive relations IV: Part of an event and/or action 33.8% 31.2%

Events/actions in which the entity plays a roleFunctions that the entity adopts in these events/actions

16.9%16.9%

16.7%15.5%

Table 7 records all the predicator classes and the predicators that make up each class. However, for some predications (and even for some predicator classes) there are no predications in the corpus, and in other cases there are only very few predications. This particularly applies to the predicator class ‘properties’. In the entire corpus on the capitalism debate there are only two propositions whose predicates specify the dimen-sions and proportions of locusts (as financial investors):571

(38) Die Heuschrecke hat einen dürren Leib. ‘The locust has a scrawny body’ (Die Zeit, 4 May 2005) (39) Die Heuschrecke ist klein, aber groß. The locust is small, but large (Die Zeit, 21 April 2005)

Predicators and predicator classes with a low type frequency can be ignored in the further analysis, since the type frequency must reach a minimum to be reflected in the

571. ‘Dimensions and proportions’ stands for ‘predicators to characterize the dimensions and proportions of an entity’ (cf. Table 4). This applies analogously to all the other listed predications.



corresponding knowledge facet in the metaphor meaning. Accordingly, in addition to the predicator class ‘properties’, at least the class ‘constitutive relations III: process’ and, depending on demarcation, ‘constitutive relations II: totality’ would not be relevant;572 some other predicators are also equally irrelevant.573

Given the case that the metaphor locust is indeed – as argued at the begin-ning – a basic discourse-semantic figure, which knowledge facets then become so entrenched that they can take effect beyond the discourse surrounding the capitalism debate? To begin with, if we consider in isolation the locust frame (in the corpus of the capitalism debate), the dominance of the predicator class ‘constitutive relations IV’ is striking. More than one third of all predicative attributions in the corpus concern an event or an action in which financial investors are involved, or more precisely, in which particular knowledge facets of locusts play such a role that they can be meta-phorically mapped onto certain activities and properties of financial investors. These include, for instance:

(40) Heuschrecken fallen über die Wirtschaft her. ‘Locusts pounce on businesses’ (Der Tagesspiegel, 22 April 2005) (41) Heuschrecken fressen alles leer und ziehen dann ab. ‘Locusts eat everything bare and then leave’ (Süddeutsche Zeitung, 30 April 2005) (42) Heuschrecken schlachten Unternehmen aus. ‘Locusts strip businesses down’ (Frankfurter Allgemeine Zeitung, 2 May 2005)

Locusts (or financial investors as locusts) are addressed in (40) to (42) with respect to particular (results of) actions that they perform. Belonging to the same predicator class are predications that more strongly highlight the function financial investors (as locusts) adopt in actions or events:

(43) Heuschrecken nutzen Unternehmen. ‘Locusts use companies’ (Süddeutsche Zeitung, 25 April 2005) (44) Heuschrecken sanieren Unternehmen. ‘Locusts clean up companies’ (Frankfurter Allgemeine Zeitung, 2 May 2005) (45) Heuschrecken sind Käufer von Unternehmen. ‘Locusts are buyers of companies’ (Der Tagesspiegel 3 May 2005)

572. The question of precisely where to draw the line remains open at this point. It should no doubt be answered from case to case. Thus, the predicator ‘composition’ is certainly of little interest within the predicator class ‘constitutive relations II: totality’, since it exhibits a relatively high type frequency.

573. In particular predicators to characterize similar entities and predicators to characterize such entities of which locusts and financial investors are signs/evidence.


There are numerous predicative attributions of this nature in the text corpus.574

It can be no coincidence that, seen quantitatively, the predicator class ‘constitutive relations I: part’ is quite dominant. More than one fifth of all predications concern the superordinate whole of which financial investors (as locusts) are a part. The following corpus records illustrate this:

(46) Heuschrecken treten in Schwärmen auf. ‘Locusts appear in swarms’ (Die Zeit, 22 April 2005) (47) Heuschrecken bilden Teil der Investoren. ‘Locusts form a part of the investors’ (Süddeutsche Zeitung, 30 April 2005)

Crucial for the metaphor is not alone the fact that locusts appear in swarms, but that they (can) cause damage in this form. If the knowledge facet of ‘swarm’ is mapped onto the knowledge domain financial investor/s, then this inherits the notion that swarms potentially cause damage. Accordingly, the possibility of a metonymic shift from cause to effect is already inherent in the predicate appear in swarms.

The dominance of these two predicator classes – ‘part of an event/action’ and ‘part of a superordinate whole’ – can be explained, on the one hand, by the fact that the same predicator classes quantitatively dominate in the reference corpus. That means that the dominance of particular knowledge facets in the locust frame in the non-metaphorical reading is evidently bequeathed to the metaphor frame, even though, of course, there is no representational relationship overall.575 A significant proportion of the metaphori-cal projection therefore consists of the dominance relationships of the knowledge facets in the input locust being transferred to the metaphor frame. This does not mean, however, that the same predications (fillers, default values) enter into the metaphor frame, but simply that the same knowledge facets prove to be U-relevant there.

The fact that these particular facets of knowledge are dominant may well be a discourse effect. The unity of the discourse in the capitalism debate is motivated by the Müntefering quote. As already indicated, metalinguistic reference is repeatedly made to this quote. Thus, the knowledge facets addressed in the Müntefering quote are recurrently addressed, and these are, on the one hand, the events and actions in

574. It would be necessary to extend the analysis to include topoi if the argumentative use of the concept were to be integrated beyond the analysis of the conceptual structure of the financial investors as Locusts metaphor. This would take far greater account of the respective argu-mentative contexts in which the metaphor occurs. Pragmatic variables like speaker perspectives and attitudes would thus shift into focus. It seems that such an extension could easily be made on the basis of the frame-semantic categories developed here. Nonetheless, I limit myself here to the analysis of semantic-conceptual structures.

575. Cf. Table 5. Although approximately every fifth predication in the reference corpus refers to the superordinate whole of which the locust forms a part, and every third predication concerns an event or action in which a locust is involved, there are nonetheless deviations with respect to other predicator classes.


which locusts are involved as well as their consequences (pick clean, leave) and, on the other hand, the superordinate whole of which locusts form a part (swarms of locusts).

The emergence of default values stands in close relation to the dominant knowl-edge facets (predicators and their classes). Only where a high type frequency prevails is a high token frequency possible.576 Accordingly, default values form particularly within the predicator classes just analysed. The following predications allocated to the predicator class ‘constitutive relations IV: part of an event/action’ occur several times within the corpus; the figures in brackets indicate their frequency of occurrence.

(48) Locusts pick companies clean. (10 times) (49) Locusts devour companies. (6 times)

In addition there are some predicates that occur less frequently but are nonetheless similar to the predicates in (48) and (49).

(50) Locusts pounce on companies. (3 times) (51) Locusts ruin companies. (2 times) (52) Locusts strip companies bare. (1 time) (53) Locusts leave purchased company to bleed dry. (1 time)

I subsume the predicates ‘pick companies clean’, ‘devour companies’, ‘pounce on com-panies’, ‘ruin companies’, ‘strip companies bare’, and ‘bleed companies dry’ under the superordinate predicate ‘destroy companies’. Due to the high token frequency of this predicate, it makes a good candidate for a later default value.

If default values (just like fillers) are themselves of a conceptual nature and if they thus consist of a network of conceptual elements,577 then one element in the concept of destroying companies proves to be particularly relevant: the consequences of this action. Destroying a company implies that it is or will be no longer functional (just as ‘pick clean’ implies that there is nothing left, ‘devour’ that the devoured object is dead and gone, ‘pounce on’ that someone or something is attacked, etc.). Consequently, the (potential) default value ‘destroy companies’ profiles the knowledge facet ‘negative consequences of actions of financial investors’. This semantic profile construction may

576. This is because a high token frequency, i.e. the specification of the reference object through many different predicates, implies a high type frequency (referentialization of “locust” or, in other words, activations of the locust frame).

577. The recognition, dating back to Minsky (1975), that every filler in turn has the status of a subframe and thus opens further opportunities for semantic connection also makes it possible to investigate fillers (like pounce on, ruin, strip bare, etc.) using frame analysis. Thus, strip bare activates the knowledge that only what is valuable and useful is “exploited”, whereas the valueless and the useless is what remain. Further data-based analysis would be required to systematically arrive at empirically validated results. An investigation of subframes is basically not included in the corpus-based study presented here.


well motivate causal metonymies in which the causes (actions of financial investors) represent the effects (destruction of companies).

In the predicator class ‘constitutive relations I: part (of a superordinate whole)’, the predicate ‘appearing in swarms’ occurs the most frequently. In the Müntefering quote, this is already inherent in the form of an implicit predication in the phrase “swarms of locusts”. Whereas this predicate arrives in the metaphor frame from the input locust, by contrast other predications such as (55) and (56) originate in the input financial investor.

(54) Locusts appear in swarms. (15 times) (55) Locusts are part of capitalism. (7 times) (56) Locusts count among international financial investors. (4 times)

In the metaphor frame a conceptual blending occurs between the predicate ‘appear in swarms’ on the one hand and ‘part of capitalism’ or ‘count among international financial investors’ on the other. Financial investors are “agents” of capitalism578 that appear in swarms. It is only against the background of this conceptual blending that one can understand why financial investors as locusts are dangerous and threatening, since neither individual locusts nor individual financial investors are dangerous and prototypically a group of financial investors poses equally no threat; however, what is threatening is a group of financial investors conceptualized as a “swarm of locusts”.

In sum, it remains to be said that at least the predicates ‘destroy companies’, ‘be part of capitalism’ and ‘appear in swarms’ qualify as default values. Usage events of the financial investors as locusts metaphor outside of the capitalism debate can serve to empirically test this finding. In addition to the two examples presented in Section 7.2.3, here is a third:

The piranha strategyBuy companies and plunder them: that is what locusts do. Locusts buy and plun-der: that is new….579 (Der Tagesspiegel, 12 April 2007)

In contrast to the two previously quoted examples, in this quote the metaphor is ex-plained once again: “Locusts” plunder companies. That they appear in swarms is not said but presupposed. An indication of the relevance of this facet of knowledge may be that in all the text samples only the plural locusts is used. In my view, this “sampling test” confirms my assumption that the two identified default values are U-relevant across discourses.

578. As in Die Zeit, 28 April 2005.

579. In the original: “Die Piranha-Strategie…Firmen kaufen und plündern: Das tun Heuschrecken. Heuschrecken kaufen und plündern: Das ist neu…”


7.5 Frame semantics and discourse analysis: some conclusions

Certainly the metaphor analysis could not duly consider all the aspects that could be derived from the data collection. The analytical focus lay on the conceptual investi-gation of the four representational units that, in keeping with CI, participate in the process of metaphor constitution. What is, however, crucial, is that all four represen-tational units (generic frame, input frame I, input frame II, metaphor frame) can be analysed using the frame-semantic means developed here. Whereas Fauconnier and Turner (and other adherents of blending theory) illustrate processes of CI solely on the basis of introspectively gathered data, the implementation of frame semantics as a corpus-based analytical instrument serves to identify the elements of the representa-tional units and explain their occurrence. Blending theory is not able to achieve this be-cause it does not have the means at its disposal for determining conceptual structures.

Finally, beyond the empirical results obtained here, I would like to draw some general conclusions concerning the relationship between frame semantics and dis-course analysis. Frames, understood here as corpus-linguistic “tool” and a semantic representation format, allow a systematic description of the influence of discourses on the meaning of linguistic discourse elements to be made. Neither the two inputs of the analysed metaphor nor the metaphor meaning itself would be explainable with-out involving discourse-related knowledge. In addition, there are grounds to assume that the sphere of influence of discourses extends to all levels of the organization of linguistic signs (i.e. to symbolic units of varying degrees of abstraction), because the content side of signs is structured through frames and at the same time varies to the extent that the structural elements of frames (fillers, default values) change depending on the discourse context.

Frame-semantic categories can thus be employed as a tool for text and discourse analysis, not only for lexical and sentence semantic studies, but also for the investiga-tion of discourse elements like argument patterns as complex form-meaning pairs. This brings a historical dimension of meaning constitution into focus that can be demonstrated with the quantitative support of large text corpora. If such corpus-based investigations recur to frame-semantic categories of analysis, then the structural con-stituents slots, fillers, and default values are particularly relevant in this process. Slots can be addressed on different levels of abstraction, thus on the one hand on the level of predicator classes and on the other on the level of individual predicators. In both cases, the dominant facets of knowledge can be identified on the basis of quantitative analysis.

However, we should not lose sight of the limits of a frame-based discourse analysis. Only probabilistic assertions concerning the degree of cognitive entrenchment can be made on the basis of the frequency of occurrence of predications (fillers). In contrast, the cognitive salience or, respectively, the mental representation, of particular linguis-tic categories can only be determined by means of psycholinguistic studies (cf. Baker 1999; Klein and Meißner 1999). In addition, the higher the degree of abstraction of the analysed frame structures is, in particular their integral default value, the more difficult it may be to determine them using frame analysis, since default values can


only be indirectly identified via empirical corpus-based studies. However, if they are so fundamental to the identity resp. mentality of a language community that they can hardly be expressly addressed, measuring them empirically and quantitatively naturally becomes difficult as well.

Evidently Foucault, whose discourse concept (in a slightly modified form) forms the basis of metaphor analysis, targets such fundamental knowledge structures. As is generally known, however, Foucault had to restrict his horizon of analysis from the very broad episteme (Foucault 1974a) to comparatively brief discourses.

Now, even briefer discourse formations are studied using frame semantics. Thus, the set of all the slots in a frame that are served with instances corresponds to the discourse-specific “contextualization potential” (Fraas 1996a: 5) of a frame-activating linguistic expression. Predications that cannot be allocated to one of these predication types (slots) are, of course, not impossible, but do not contribute to the discourse, i.e. they are “not heard” and are marginalized in the public discourse. In functional terms, this certainly seems to approximate a historical a priori as defined by Foucault, namely “the conditions of emergence of statements” (Foucault 1981a: 127)580. Accordingly, the set of all values instantiated in these slots represents the actual discourse, that is, the exploited meaning potential. In discourse, regularities are ultimately prominent where knowledge facets (i.e. slots in the form of predication classes) become entrenched due to a high type frequency or default values emerge on the basis of a high token frequency. In this process, the historical a priori unfurls its normative power to the extent that the degree of entrenchment displays the cognitive salience of the knowledge facets and default values.

The aspects in question are only to be understood as a selection of potential points of contact between frame semantics and Foucaultian discourse analysis. In any case, using frames makes it possible to empirically study the extent to which the discourse displays its own “reality” in the course of understanding signs thus exhibiting its own cognitive presence, which not least is able to be demonstrated in processes of meaning constitution.

580. By “statements”, Foucault does not mean the propositional content of individual sentences, but rather the contentual and epistemic convergence points between different sentences, texts, and non-linguistic knowledge elements (like architectures and practices). For this reason, Busse (for instance in 2007b) speaks of “knowledge segments”.

chapter 8

Conclusions and future research perspectives

In this book, I have developed the basic principles of a frame theory and discussed a holistic approach to cognitive semantics in a larger context. Following Fillmore’s con-ception of a semantics of understanding (U-semantics) (Fillmore 1985), at the basis of my approach lies a double understanding of frames. On the one hand, frames are considered as conceptual knowledge structures that both motivate the use of linguistic expressions and allow linguistic meanings to be understood. On the other hand, frames serve as analytical instruments for the empirical investigation of conceptual knowledge structures. Both perspectives have been equally taken into account.

With respect to linguistics, this study is to be understood as a large-scale plea for a conceptualist approach to semantics that attempts to account for the fact that humans use language in concrete contexts and that in order to understand language they take recourse to all the knowledge resources available to them. According to this understanding, linguistic ability is closely linked to general cognitive abilities, such as being able to categorize, schematize, and make foreground/background distinctions.

This study focused on the cognitive issues surrounding the comprehension of linguistic meanings. The most important results can be summarized as follows:

1. One fundamental cognitive issue concerns the question of how it is possible that our cognitive system can instantly comprehend linguistic meanings as gestalt-like units, as it were, without a large set of (potentially) U-relevant facets of knowledge being lost. The answer offered in this book is that we dispose over a multitude of well-structured conceptual knowledge units, so-called “frames”, that are readily available without entailing any significant cognitive effort because they are stored in a typified form in the long-term memory. These frames consist of three structural constituents, “slots”, “default values”, and “fillers”.

2. Frames create coherence between apparently uncorrelated elements. Activated frames thus ensure that during language processing “sense constancy” (Hörmann 1994: 179–212) is not only maintained, but can emerge at all. An activated frame provides a wealth of knowledge elements (default values) that can be potentially U-relevant. The frame car, for instance, opens further opportunities for semantic connection, which can relate to, for example, the typical components of cars, the typical material properties and size, typical functions, or typical (material and non-material) value of cars. Once the car frame has been activated, the expressions in italics in Examples (1) to (5) can be easily understood:


(1) The fuel cap is open. (Component) (2) The accident only caused metal damage. (Material property) (3) The parking spot is too small. (Size) (4) It only takes me five minutes to get to work. (Function) (5) My dream is a Ferrari (but I can’t afford one). (Non-material and material value)

Each of the expressions in italics accentuates at least the knowledge facet (slot) in the car frame that has been added in brackets. At the same time, each of the expressions itself activates a frame. Thus, taking the first example, the frame fuel cap is in turn specified with respect to slots like ‘material properties’, ‘size’, ‘function’, ‘material value’, and many others.

3. To comprehend U-relevant knowledge of this nature, it is not necessary to pro-cess “levels” of knowledge organization, as adherents of modular theories of meaning maintain. In fact, the results of both empirical and basic theoretical studies suggest that in our “search for meaning” (Bartlett 1932) we already activate frames during the reception of linguistic expressions (“on-line”), thus having access to a large variety knowledge right from the start. This holistic position is supported by the fact that it is not possible to draw a distinct dividing line between linguistic knowledge and world knowledge. All the criteria that are listed in the literature to legitimate the special status of linguistic knowledge prove to be untenable (cf. Chapter 3.1).

4. The following observation leads to basic scepticism towards a syntax-oriented approach to semantic and semiotic issues: In a modular approach, if we posit that, on a level of purely linguistic knowledge, the so-called “syntactic combinatory potential” structurally predetermines that which then must merely be further differentiated on the level of conceptual knowledge, then a large proportion of (potentially) U-relevant world knowledge cannot be comprehended, because this cannot be fitted into the predetermined structure. The possibilities that so-called “semantic representations” accordingly provide for conceptual differentiation are not then sufficient to explain how a linguistic expression is understood. This insight can be generalized such that semantically relevant knowledge facets are left out of consideration insofar as a theory of meaning orients itself towards syntactically and component-semantically motivated specifications such as “selection restriction”, which limit the sphere of U-relevant knowledge.

5. In contrast, the frame theory presented in this book does not exclude any reference points of linguistic expressions, not even those that concern presupposed, background knowledge facets. In connection with the problematic founding of semantics on syntax, Jackendoff (2003: 654) speaks of a “syntactocentric fallacy”. Fillmore’s case grammar is also a weak form of “syntactocentrism” in that it focuses on the valency-governed deep structures (“deep cases”) of linguistic expressions; the same holds for FrameNet frames since they are defined as valence patterns (Ziem 2014). Even von Polenz (1985) conducts his study on sentence semantics to some extent syntactocentrically, although

Chapter 8. Conclusions and future research perspectives 381

his approach specifically does not orient itself towards the complements and specifica-tions that a verb requires. Nonetheless, von Polenz fails in his aim of detecting all the reference point that are explicitly and implicitly expressed in a sentence because he merely augments Fillmore’s list of potential deep cases by adding some semantic roles.

In contrast, Fillmore (1985) implemented the “semantic turn” with his programme of an “understanding-semantics” (U-semantics respectively “interpretative seman-tics”). From 1975–1985 Fillmore (1975, 1976a, b, 1977a, 1982a, 1985) developed a frame concept that does not take syntactic structures and categories as a point of departure for a non-reductionist theory. It is this frame concept that underlies the frame-semantic model introduced in the present study.

6. I have attempted to more precisely describe the holistic countermodel to the modular semantic-level theory by using the metaphor of the “space of understand-ing”, which traces back to the philosopher Christoph Demmerling (2002). The entire knowledge that a language user has at their disposal at a particular time constitutes their field of access for understanding a linguistic expression. The space of under-standing is a multidimensional network of premises within which language under-standing becomes possible. Thus, language processing presupposes our ability to perceive situations with all our senses and correlate linguistic expression with all sorts of objects of our perceptual experience. Equally important is the ability to relate affectively to linguistic expressions and involve experiences that stand directly in connection with our bodily disposition (anatomically, physiologically, neurologically, and ontogenetically).

7. There is empirical evidence from the psycholinguistic study of inference process-ing that the space of understanding represents the relevant reference parameter in the interpretation of linguistic expressions. I have outlined the significant empirical results of inference processing research that demonstrate that people do not only take recourse to background knowledge when there are local gaps in understanding (like co-referential connections) to be bridged. In fact, inferences equally concern global textual coherence (intentions of the author, thematic references, etc.) as well as situative factors (like person, time, and place deixis). Graesser, Singer, and Trabasso (1994) and Graesser, Millis, and Zwaan (1997) differentiate between a whole range of inference types on the basis of the results of psycholinguistic studies. In a text linguistic study, Busse (1991a) arrives at a comparable typology of U-relevant knowledge. The identified knowledge types are integral components of the space of understanding. However, in this space of understanding they do not constitute an amorphous mass, but instead are organized in frames in such a way that they can be efficiently “accessed”.

Nevertheless, in psycholinguistic inference-processing theories I see a deficit in the fact that until now the research results have rarely been used to solve genuine semantic issues. Consequently, there is a failure to reflect on their own procedure in linguistic and semiotic terms. For instance, the question remains open as to which basic semiotic model forms the basis of the studies and which linguistic and cognitive theoretical consequences the empirical results may entail in detail.


8. Beyond the assumptions of psycholinguistic inference-processing theories, I have attempted to illustrate that the constitution of a linguistic sign also occurs inferen-tially. This is shown against the background of the so-called “symbolic principle” of Langacker’s Cognitive Grammar (1987). According to the symbolic principle, the structure of human language can be exhaustively described as an inventory of form-meaning pairs (of varying degrees of abstraction), of so-called “symbolic units”. On the one hand, an inferential relationship pertains to the connection between a linguistic form and a linguistic meaning. Thus, for example, the phonological form [dog] is con-nected with the semantic content [dog] on the basis of knowledge of the appropriate convention. In addition, if we consider only the meaning side of the symbolic unit, syn-tagmatically more complex units like the word dogs emerge inferentially. When [dog] is combined with the plural-marking semantic unit [-s], the complex unit [[dog]/[s]] emerges. A third inferential relationship pertains to paradigmatic relations. The semantic content [dog] – just like the meaning of the complex unit [[dog]/[s]] – is inferentially constituted since it maintains relationships to numerous other semantic units, such as the units [animal] or [bark]. Frames are the units that structure the semantic side of a symbolic unit.

9. Following Langacker’s basic semiotic model, I have differentiated between two cognitive processes that are involved in the meaning constitution. On the one hand, the form side of a linguistic sign evokes a frame. Fillmore calls this process “evoca-tion”. What is evoked is a set of default values that specify a frame. Thus, the word car evokes a frame that features, for instance, the default value ‘metal’ in the slot ‘material properties’ and the default value ‘engine’ in the slot ‘component’.

The process of evocation must be distinguished from that of invocation. If two (or more) frames are linked with each other, new slots emerge that need to be filled with default values. These default values are recalled, or invoked, from memory. Thus, the connection between the frames car and dent in (6) leads to the default value ‘accident’ being invoked.

(6) The car has a dent.

Evocation and invocation may be understood as cognitive processes that are consti-tutive to understanding. A language user’s active mental processing is greater in the process of invocation.

10. Every linguistic expression “has” a (variable) meaning potential. From a frame-semantic perspective, the meaning potential of a linguistic expression corresponds to the set of default values in an evoked frame that may potentially feed into the concep-tual meaning of the expression. Thus, for instance, the default values ‘fuel cap’ (com-ponent), ‘fast mode of transport’ (function), and ‘metal’ (material property) belong to the evoked car frame. However, the default values ‘fast mode of transport’ and ‘fuel cap’ clearly do not play a significant role in understanding sentence (6). Thus, only part of the meaning potential of a linguistic expression is ever exploited in language processing.


11. Frames represent schematic units par excellence. Numerous examples were used to demonstrate that frames share all the essential properties of schemata. This in-cludes, among others, that frames are gestalt-like, they change dynamically, and they are acquired via inductive and/or abductive inferential processes. Further, frames are connected to other frames via categorization links and are thus part of a comprehen-sive conceptual network. This network exhibits a hierarchical organization structure that linguistically speaking has the character of recursive hyperonymy or, in turn, hyponymy relations.

12. The hierarchical organization of the network was able to be used to systematically identify the slots in a frame. To do this, Konerding (1993) developed a procedure called “hyperonym type reduction”. The identification of the slots in a frame is important for the implementation of frames as an analytical instrument.

Slots are understood as potential reference points with respect to which the refer-ence object of a linguistic expression can be more precisely specified in predicative terms. In this process, slots have the character of questions that can be meaningfully posed with respect to the reference object. Thus, concerning the reference object “car”, it is possible to meaningfully ask what material properties a car has, what non-material value it has (for a person), or what function it fulfils for someone. Conversely, questions about, for instance, the activities or the customary actions of a car are not meaningful.

13. When a linguistic expression evokes a frame, this act of frame evocation cor-responds to an act of cognitive reference. For instance, understanding the word car means establishing a reference to the respective conceptual unit or car frame. Acts of reference only succeed under the condition that particular slots in the evoked frame are occupied by default values. Thus, in our case, the referentialization would be unlikely to succeed if, for example, no default values for the slots ‘size’ and ‘material properties’ were actualized. Linguistically, default values occur as implicit predica-tions. An implicit predication concerning the size of a car would be, for example is four metres long; an implicit predication concerning its material properties would be made of metal. I call such predications “implicit” because they must be invoked from memory in the form of default values during the act of referentialization. Conversely, a predication is “explicit” when it is linguistically performed, like the predication has a dent in Example (6).

14. Drawing on the “entrenchment hypothesis”, which has been widely discussed in cognitive linguistics, I was able to demonstrate how default values emerge. Default values are cognitively entrenched predications. Thus, they are “phenomena of the third kind”, in Keller’s (2003) sense. Metaphorically speaking one can also say that default values are “cognitive trails”.

Two types of cognitive entrenchment can be distinguished. A high type frequency is the case when a slot (and thus the entire frame) becomes entrenched as a result of many different tokens being instantiated in the slot. Therefore, if the non-material value of a car was addressed in different ways over and over again within a language


community, this slot would become entrenched. Conversely, in the case of a high token frequency, one and the same instance would recurrently occupy the slot ‘non-material value’ and thus become entrenched as a default value.

15. To date, research has not adequately taken account of the fact that frames con-stitute formats for knowledge representation and at the same time instruments for knowledge analysis. The comprehensive corpus-based analysis of the financial in-vestors as locusts metaphor was able to demonstrate that both of these aspects are closely linked together and that the categories of frame analysis are suitable as an analytical instrument. It was able to demonstrate the entrenchment of some slots in the metaphor frame (in the sense of a high type frequency) as well as the emergence of default values through a high token frequency. Methodologically speaking, this shows, on the one hand, that implementing the frame-semantic categories developed here is useful for operationalizing and extending CI (or blending theory) for corpus-based analysis. On the other, a frame-supported corpus-based analysis can make an important contribution to discourse analysis. Thus, it is possible to describe how the meanings of linguistic expressions can systematically vary within a discourse (under-stood as the set of texts sharing the same topic) in that new default values gradually emerge in particular frames. The frame theory I have developed may be understood in this sense as form of semantic-epistemological linguistic analysis.

Thus, overall this study is just as relevant for basic theoretical considerations on the (cognitive) nature of linguistic meanings as it is for empirical individual case studies. One advantage of frame semantics, as I see it, is that many linguistic phenomena can be uniformly explained, i.e. using the same methodological instruments and under the same cognitive and linguistic premises. In this book, the phenomena I discussed included reference, predication, coherence, (indirect) anaphors, language change, basic discourse-semantic figures, and metaphors. In addition, Section 6.5.3 mentions the extent to which frames could also be used to investigate literary texts. In my view, the frame-semantic model I expound here opens a promising perspective for the develop-ment of cognitive aesthetics (cf. also Freeman 2007; Ziem 2009). Moreover, it turned out to be useful for didactic purposes, such as language teaching (Ziem 2011).

Whereas this study was essentially concerned with developing the basic prin-ciples for an empirically applicable frame semantics, one task of future studies is to systematically discuss the linguistic phenomena previously mentioned. In addition, research is required on the extent to which argument patterns can be incorporated into semantic analysis. In this context, it seems logical to understand argument pat-terns as complex form-meaning pairs whose content side is structured by frames. In addition, the question remains open as to the extent to which interjections constitute symbolic units. If one does not wish to deny them their semiotic character, then the content side of these signs must also be organized via frames. Furthermore, one inter-esting field of research concerns the acquisition of linguistic meanings. The hypoth-esis needs to be tested whether children and adults learn new meanings (or meaning potentials) by acquiring a system of potential predications that can be linked to the


form side of an expression. Accordingly, acquiring a meaning or meaning potential would mean succeeding in transforming (quasi-)explicit into implicit predications. Finally, corpus-based analysis – such as the metaphor analysis in Chapter 7 – should be conducted alongside studies on individual language processing and comprehen-sion. Psycholinguistic priming experiments could serve in this context. Thus, it would also be possible to make empirically solid statements about the extent to which results gathered from corpus-based analysis demonstrate a cognitive “reality” for individual language users or groups of users.

The linguistic expressions used in this study also evoke frames. At one point, Minsky notes that the fine art of an author consists in activating particular processes in the readers’ minds, processes that the author knows just as little about as the readers in whose minds these processes occur. Thus, if in this vein this study activates cogni-tive processes, let us hope that Minsky is proven right when, quoting Marcel Proust, he writes:

Each reader reads only what is already inside himself. A book is only a sort of optical instrument which the writer offers to let the reader discover in himself what he could not have found without the aid of the book. (Minsky 1988: 247)

References

Text corpus for the metaphor analysis in Chapter 7

[Interview by Michael Schmidt with Hans-Ulrich Wehler]. (2005). “Das hat was von einer Dreck-schleuder”. Der Tagesspiegel, 4 May 2005, 2.

[Interview by Moritz Döbler with Paul Nolte]. (2005). “Die Debatte bewegt sich auf armseligem Niveau.” Der Tagesspiegel, 28 April 2005, 4.

[no author given]. (2005). Am Tag der Arbeit Eier auf Müntefering. Die Gewerkschaften greifen die Kapitalismuskritik auf. “Den Worten Taten folgen lassen”. Frankfurter Allgemeine Zei-tung, 2 May 2005, 1.

[no author given]. (2005). Angelsachsen über Heuschrecken-Vergleich irritiert. Medien rügen Populismus und Sozialneid. “Keine gute Zeit, in Deutschland zu sein”. Frankfurter Allge-meine Zeitung, 6 May 2005, 11.

[no author given]. (2005). Clement hat genug. Wolffsohn: Aus dem Wörterbuch des Unmen-schen. SPD: Scharfmacher. Frankfurter Allgemeine Zeitung, 4 May 2005, 1.

[no author given]. (2005). Das letzte. Die Zeit, 4 May 2005, 53.[no author given]. (2005). Der Staat soll es richten. Müntefering rückt von seiner Kritik am

Kapitalismus nicht ab. Industrie-Chef reagiert empört. Der Tagesspiegel, 28 April 2005, 4.[no author given]. (2005). Die “Heuschrecken” wehren sich. Beteiligungsgesellschaften entsetzt

über SPD-Liste. “Neue Investoren werden abgeschreckt”. Frankfurter Allgemeine Zeitung, 2 May 2005, 13.

[no author given]. (2005). Die “Heuschrecken” wehren sich. Der Tagesspiegel, 3 May 2005, 16.[no author given]. (2005). Die Grünen nehmen die Wirtschaft in Schutz. Fraktionschefin

Göring-Eckardt: Unternehmen müssen Gewinne machen dürfen. Frankfurter Allgemeine Zeitung, 4 May 2005, 11.

[no author given]. (2005). Die Kapitalismuskritik bringt der SPD keine Punkte. Deutschland-trend: Die Debatte über die Rolle der Wirtschaft finden viele richtig, aber sie ärgern sich über Münteferings Wortwahl. Der Tagesspiegel, 6 May 2005, 4.

[no author given]. (2005). Entsetzen über Kapitalismuskritik der SPD. Gesamtmetall-Chef Kan-negiesser fordert Rückkehr zur wirtschaftspolitischen Vernunft. Frankfurter Allgemeine Zeitung, 18 April 2005, 11.

[no author given]. (2005). Hedge-Fonds als Firmenjäger. Wen die SPD wohl meinte, aber nicht nannte. Frankfurter Allgemeine Zeitung, 2 May 2005, 15.

[no author given]. (2005). Hundt findet Kapitalismus-Debatte “zum Kotzen”. Streit von Wirtschaft und Politik geht weiter, weil die Deutsche Bank trotz hohen Gewinns Jobs ab-baut. Der Tagesspiegel, 30 April 2005, 16.

[no author given]. (2005). Kapitalismus-Kritik. Die Plage der Neuzeit. Der Tagesspiegel, 1 May 2005, 8.


[no author given]. (2005). Keine Erleichterungen für Hedge-Fonds. Ministerium rudert zurück. Verhaltene Bilanz ein Jahr nach dem Start in Deutschland. Frankfurter Allgemeine Zeitung, 4 May 2005, 21.

[no author given]. (2005). Koalition streitet über Kapitalismuskritik. Frankfurter Allgemeine Zeitung, 4 May 2005, 1.

[no author given]. (2005). Lafontaine flirtet mit der neuen Linkspartei. Podiumsdiskussion im Landtagswahlkampf in Nordrhein-Westfalen. WASG erwartet Empfehlung. Frankfurter Allgemeine Zeitung, 23 April 2005, 1.

[no author given]. (2005). Müntefering verkaufte an “Heuschrecken”. Der Tagesspiegel, 3 May 2005, 1.

[no author given]. (2005). Münteferings Heuschrecken. Der Tagesspiegel, 3 May 2005, 6.[no author given]. (2005). Nicht unbedingt: Zwischen Marx und Murks. Frankfurter Allgemeine

Zeitung, 25 April 2005, 2.[no author given]. (2005). Plageregister. “Wie die Nazis gegen Juden”. Kapitalismusstreit außer

Rand und Band. Frankfurter Allgemeine Zeitung, 4 May 2005, 1.[no author given]. (2005). SPD erstellt Liste mit “Heuschrecken”. Frankfurter Allgemeine Zeitung,

30 April 2005.[no author given]. (2005). SPD gibt den Heuschrecken Namen. Fraktion listet Private-Equity-

Unternehmen auf. Arbeitgeberchef nennt Debatte “zum Kotzen”. Frankfurter Allgemeine Zeitung, 4 May 2005, 13.

[no author given]. (2005). SPD im Streit mit den Kapitalisten. Der doppelte Franz. Der Tages-spiegel, 24 April 2005, 8.

[no author given]. (2005). Und sie fraßen alles, was im Lande wuchs. Franz Müntefering, die Heuschrecken und die achte Plage über Ägyptenland. Frankfurter Allgemeine Zeitung, 4 May 2005, 9.

[no author given]. (2005). Unternehmenschef Stark von der schlechteren Stimmung für Aktien nicht entmutigt. Weiterverkauf an Finanzinvestor keine Alternative. Frankfurter Allgemeine Zeitung, 4 May 2005, 16.

[no author given]. (2005). Wall Street befürchtet gedämpftes Investitionsklima. Anlagebanker rechnen langfristig aber trotz Heuschrecken-Debatte mit steigendem Engagement. Frank-furter Allgemeine Zeitung, 6 May 2005, 12.

[no author given]. (2005). Wolffsohn und die Heuschrecken. Zu weit gehüpft. Der Tagesspiegel, 4 May 2005, 2.

[no author given]. (2005). Aus dem Gleichgewicht. Frankfurter Allgemeine Zeitung, 20 April 2005, 1.[no author given]. (2005). Debatte über Kapitalismus. SPD macht Druck auf Merkel. Süddeutsche

Zeitung, 3 May 2005, 1.[no author given]. (2005). Kapitalismus-Kritik. Süddeutsche Zeitung, 3 May 2005, 2.[no author given]. (2005). Kritik am Kapitalismus: Heftige Debatte im Spannungsfeld von Eigentum

und Gemeinwohl. Warnung vor “entfesseltem Markt”. Süddeutsche Zeitung, 30 April 2005, 6.[no author given]. (2005). Neue Empörung über Wolffsohn. Süddeutsche Zeitung, 4 May 2005, 6.[no author given]. (2005). Wirtschaft will mit Schröder Klartext reden. BDI-Präsident trifft sich

mit dem Kanzler. Der Tagesspiegel, 27 April 2005, 4.Bannas, Günter. (2005). Die lancierte Programmrede. Politik ist immer auch Wahlkampf –

Münteferings Kapitalismuskritik. Frankfurter Allgemeine Zeitung, 20 April 2005, 4.Beikler, Sabine. (2005). Ein bisschen Klassenkampf vor dem Roten Rathaus. Der Tagesspiegel,

2 May 2005, 10.

References 389

Bickerich, Sebastian, & Meisner, Matthias. (2005). Kapitalismus – jetzt will die SPD handeln. Müntefering macht konkrete Vorschläge. Der Tagesspiegel, 6 May 2005, 1.

Bovensiepen, Nina, & Blechschmidt, Peter. (2005). “Bei manchen Unternehmen stimmt die Ethik nicht”. Müntefering geht Manager frontal an. Süddeutsche Zeitung, 18 April 2005, 1.

Bruns, Tissy. (2005). 1. Mai und die Wirtschaft. Aussitzen reicht nicht. Der Tagesspiegel, 2 May 2005, 8.

Casdorff, Stephan-Andreas. (2005). Vorfahrt für Vernunft. Zum Parteitag: Die FDP täte gut daran, wirklich liberal zu sein. Der Tagesspiegel, 4 May 2005, 8.

Eubel, Cordula. (2005). Müntefering: Marktradikal und asozial. Der Parteichef verstärkt seinen Angriff gegen die Wirtschaft – und trifft offenbar den Nerv der Genossen. Der Tagesspiegel, 22 April 2005, 4.

Fried, Nico, & Bovensiepen, Nina. (2005). Debatte über Kapitalismuskritik. Grüne auf Distanz zu SPD-Chef. Süddeutsche Zeitung, 19 April 2005, 5.

Haselberger, Stephan. (2005). Clement: Schluss mit den Schlagworten. Wirtschaftsminister fordert mehr Sachlichkeit bei Kapitalismuskritik. Der Tagesspiegel, 4 May 2005, 1.

Hesse, Martin. (2005). Feindbild ohne Feinschliff. Nicht alle Finanzinvestoren sind so schlimm, wie sie Müntefering darstellt – oft profitieren Unternehmen vom Einstieg ausländischer Beteiligungsgesellschaften. Süddeutsche Zeitung, 3 May 2005, 2.

Hesse, Martin. (2005). Vom Nutzen der Heuschrecke. Süddeutsche Zeitung, 23 April 2005, 23.Heuser, Uwe Jean. (2005). Der Barbar im Haus. Franz Münteferings Kritik am Kapitalismus

verunsichert ausländische Investoren und schadet Deutschland. Die Zeit, 4 May 2005, 9.Hummel, Thomas. (2005). Unionspolitiker attackieren Müntefering für Kapitalismuskritik. Süd-

deutsche Zeitung, 20 April 2005, 10.Jacobs, Stefan. (2005). Blaumänner allein sind nicht genug. Die IG Metall kämpft gegen Mitglie-

derschwund – bei neuen Zielgruppen. Der Tagesspiegel, 1 May 2005, 22.Jahberg, Heike, Monath, Hans, & Rosenfeld, Dagmar. (2005). SPD: Banken lassen deutsche

Firmen im Stich. Sozialdemokraten kritisieren Kreditvergabe und planen schärfere Gesetze gegen “räuberische Fonds”. Der Tagesspiegel, 1 May 2005, 1.

Joffe, Josef. (2005). Müntismus. Der SPD-Parteichef erklärt dem Kapitalismus den Kampf und damit der eigenen Regierung. Die Zeit, 21 April 2005, 1.

Kister, Kurt. (2005). Politik als Krawallritual. Süddeutsche Zeitung, 3 May 2005, 4.Lohr, Bernhard. (2005). “In die falsche Ecke gestellt”. Gewerbe-Präsident Wickenhäuser zu

Müntefering-Kritik. Süddeutsche Zeitung, 26 April 2005, 41.Meisner, Matthias. (2005). Jetzt muss die SPD auch handeln. Gewerkschaftschef Sommer fordert

von Müntefering nach Kapitalismuskritik Taten gegen “nackte Gier”. Der Tagesspiegel, 2 May 2005, 1.

Monath, Hans, & Haselberger, Stephan. (2005). SPD fordert Solidarität von den Grünen. Der Streit um Münteferings Kapitalismuskritik trübt das Klima in der Koalition. Der Tagesspiegel, 4 May 2005, 4.

Piper, Nikolaus. (2005). Die Logik der Märkte. Süddeutsche Zeitung, 4 May 2005, 25.Piper, Nikolaus. (2005). Unter Heuschrecken. Hundert Jahre deutsche Kritik am Kapitalismus.

Süddeutsche Zeitung, 25 April 2005, 15.Riechelmann, Cord. (2005). Das arme, verängstigte Tier. Ihr Wandertrieb beruht auf Stress: Hat

die SPD denn keine Ahnung vom Leben der Heuschrecke? Frankfurter Allgemeine Zeitung, 4 May 2005, 33.

Rudzio, Kolja. (2005). Achtung, Müntewelle! Schlichte Parolen sind die eigentliche “Plage”. Die Zeit, 4 May 2005, 23.


Schäfer, Ulrich. (2005). Die Heuschrecken-Liste der SPD. Franz Müntefering und seine Ge-nossen attackieren nicht nur die Deutsche Bank. Süddeutsche Zeitung, 30 April 2005, 25.

Schäfer, Ulrich. (2005). Willkommene Plage. Wie die SPD nach der Regierungsübernahme 1998 selbst “Heuschrecken”-Investoren anlockte. Süddeutsche Zeitung, 3 May 2005, 2.

Schlegel, Matthias. (2005). Zwei Linke fühlen sich gelinkt. Der Tagesspiegel, 22 April 2005, 4.Schmid, Klaus-Peter, & Niejahr, Elisabeth. (2005). Die SPD hat eine Schwarze Liste von Private-

Equity-Firmen ins Internet gestellt. Was plant sie als Nächstes? Die Zeit, 4 May 2005, 9. Schmidt, Thomas E. (2005). Insekten der Republik. Schönes Wetter führt regelmäßig zu er-

wünschten politischen Mitnahmeeffekten: Die Sonne wirkt deeskalierend, zumindest sorgt sie für eine gerechte Verteilung. Die Zeit, 4 May 2005, 11.

Siemons, Mark. (2005). Die Verheißung. Bei der Kapitalismusdebatte geht es um den Spielraum des Staates. Frankfurter Allgemeine Zeitung, 2 May 2005, 35.

Soldt, Rüdiger. (2005). Münteferings Formeln. Der SPD-Vorsitzende erwärmte sich erst spät für die Agenda 2010. Frankfurter Allgemeine Zeitung, 20 April 2005, 4.

Steinfeld, Thomas. (2005). Das Kapital. Franz Müntefering und der Wortschatz des Klassen-kampfes. Süddeutsche Zeitung, 20 April 2005, 13.

Steltzner, Holger. (2005). Vom Nutzen der Heuschrecken. Frankfurter Allgemeine Zeitung, 2 May 2005, 1.

Storn, Arne, Heusinger, Rober, & Brost, Marc von. (2005). Kapitalismus: Angst vor Investoren. Was ist an Deutschland so verlockend? Die Zeit, 28 April 2005, 23.

Ulrich, Bernd, & Niejahr, Elisabeth. (2005). Der neue Klassenkampf? Zurück zum Klassen-kampf. Die Zeit, 23 April 2005, 23.

Sources

Duden Deutsches Universalwörterbuch. (1989). Second ed. Fully revised and greatly expanded. Published and edited by the research council and staff of the Duden editorial team under Günther Drosdowski. Mannheim: Bibliographisches Institut.

Murakami, Haruki. (2003). South of the border, west of the sun. London: Vintage Books.SPD. (2005). Programmheft. [Accessible at: http://www.partei.spd.de/servlet/PB/show/1043150/

221204_programmheft_1.pdf; last accessed: 1 July 2008]Christa Wolf. (2003). They Divided the Sky. Translation by Luise von Flotow of Der geteilte

Himmel. Ottawa: University of Ottawa Press.

Secondary literature

Adams, Marilyn, & Collins, Allan. (1979). A schema-theoretic view of reading. In Roy Freedle (Ed.), New directions in discourse processing (Vol. II) (pp. 1–22). Norwood: Ablex.

Aitchison, Jean. (1994). Words in the mind: An introduction to the mental lexicon. Oxford: Basil Blackwell.

Alba, Joseph W., & Hasher, Lynn. (1983). Is memory schematic? Psychological Bulletin, 93(2), 203–231. DOI: 10.1037/0033-2909.93.2.203

Allwood, Jens. (2003). Meaning potentials and contexts: Some consequences for the analysis of variation in meaning. In Hubert Cuyckens, René Dirven, & John R. Taylor (Eds.), Cogni-tive approaches to lexical semantics (pp. 29–45). Berlin & New York: Mouton de Gruyter.

DOI: 10.1515/9783110219074.29

http://dx.doi.org/10.1037/0033-2909.93.2.203

http://dx.doi.org/10.1515/9783110219074.29

References 391

Anderson, John R. (1983). The architecture of cognition. Cambridge: University Press.Anderson, John R. (1991). The adaptive nature of human categorization. Psychological Review,

98, 409–429. DOI: 10.1037/0033-295X.98.3.409Anderson, John R. (1996). Kognitive Psychologie: eine Einführung. Heidelberg: Spektrum.Anderson, John R. (2010). Cognitive psychology and its implications (7th ed.). New York: Worth

Publishers.Anderson, Richard C., & Pearson, P. David. (1984). A schema-theoretic view of basic processes in

reading comprehension. In P. David Pearson (Ed.), Handbook of reading research (pp. 255–291). New York & London: Longman.

Anderson, Richard C., Pichert, James W., & Shirey, Larry L. (1983). Effects on the reader’s schema at different points in time. Journal of Educational Psychology, 75, 271–279.

DOI: 10.1037/0022-0663.75.2.271Aschenberg, Heidi. (1999). Kontext in Texten. Umfeldtheorie und literarischer Situationsaufbau.

Tübingen: Niemeyer.Atkins, Sue, Fillmore, Charles J., & Johnson, Christopher R. (2003). Lexicographic relevance:

Selecting information from corpus evidence. International Journal of Lexicography, 16(3), 251–280. DOI: 10.1093/ijl/16.3.251

Atkinson, Richard C., Shiffrin, Richard M. (1971). The control of short-term memory. Scientific American, 224, 82–90. DOI: 10.1038/scientificamerican0871-82

Baker, Collin F. (1999). Seeing clearly: Frame Semantic, psycholinguistic, and cross-linguistic ap-proaches to the semantics of the English verb see. Ph.D. dissertation, University of California, Berkeley. [Accessible at: http://www.icsi.berkeley.edu/~collinb/indexedDiss.pdf, last ac-cessed: 10 April 2014]

Baker, Collin F., Fillmore, Charles J., & Lowe, John B. (1998). The Berkeley FrameNet project. In Proceedings of the COLING-ACL, Montreal, Canada. [Accessible at: http://framenet.icsi.berkeley.edu/papers/acl98.pdf]

Baldauf, Christa. (1997). Metapher und Kognition. Grundlagen einer neuen Theorie der Alltags-metapher. Frankfurt a.M. et al.: Lang.

Ballmer, Thomas T., & Brennstuhl, W. (1986). Deutsche Verben. Eine sprachanalytische Untersu-chung des deutschen Verbwortschatzes. Tübingen: Narr.

Barclay, J. Richards, Bransford, John D., Franks, Jeffrey J., McCarrell, Nancy S., & Nitsch, Kathy. (1974). Comprehension and semantic flexibility. Journal of Verbal Learning and Verbal Behavior, 13, 471–481. DOI: 10.1016/S0022-5371(74)80024-1

Barlow, Michael, & Kemmer, Suzanne (Eds.). (2000). Usage-based models of language. Stanford: CSLI.

Barsalou, Lawrence W. (1982). Context-independent and context-dependent information in concepts. Memory & Cognition, 10(1), 82–93. DOI: 10.3758/BF03197629

Barsalou, Lawrence W. (1983). Ad hoc categories. Memory & Cognition, 11, 211–227. DOI: 10.3758/BF03196968Barsalou, Lawrence W. (1985). Ideals, central tendency, and frequency of instantiation as de-

terminants of graded structure of categories. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 629–654. DOI: 10.1037/0278-7393.11.1-4.629

Barsalou, Lawrence W. (1987). The instability of graded structure: Implications for the nature of concepts. In Ulric Neisser (Ed.), Concepts and conceptual development: Ecological and intellectual factors in categorization (pp. 101–140). Cambridge: Cambridge University Press.

Barsalou, Lawrence W. (1991). Deriving categories to achieve goals. In Gordon H. Bower (Ed.), The psychology Of learning and motivation: Advances in research and theory (Vol. 27, pp. 1–64). San Diego: Academic Press.

http://dx.doi.org/10.1037/0033-295X.98.3.409

http://dx.doi.org/10.1037/0022-0663.75.2.271

http://dx.doi.org/10.1093/ijl/16.3.251

http://dx.doi.org/10.1038/scientificamerican0871-82

http://dx.doi.org/10.1016/S0022-5371(74)80024-1

http://dx.doi.org/10.3758/BF03197629

http://dx.doi.org/10.3758/BF03196968

http://dx.doi.org/10.1037/0278-7393.11.1-4.629


Barsalou, Lawrence W. (1992a). Frames, concepts, and conceptual fields. In Lehrer, Adrienne, & Kittay, Eva Feder (Eds.), Frames, fields and contrasts. new essays in semantics and lexical organisation (pp. 21–74). Hillsdale, New Jersey: Erlbaum.

Barsalou, Lawrence W. (1992b). Cognitive psychology: An overview for cognitive scientists. Hills-dale, New Jersey: Erlbaum.

Barsalou, Lawrence W. (2003). Situated simulations in the human conceptual system. Language and Cognitive Processes, 18, 513–562. DOI: 10.1080/01690960344000026

Barsalou, Lawrence W. (2005). Situated conceptualization. In Henri Cohen, & Claire Lefebvre 2005a, 619–650.

Bartlett, Frederic. (1932). Remembering: A study in experimental and social psychology. Cambridge: Cambridge University Press.

Bartsch, Renate. (1998). Dynamic conceptual semantics: A logico-philosophical investigation into concept formation and understanding. Stanford: CSLI.

Bartsch, Renate. (2002). Consciousness emerging: The dynamics of perception, imagination, ac-tion, memory, thought and language. Amsterdam: John Benjamins. DOI: 10.1075/aicr.39

Bartsch, Renate. (2005). Memory and understanding: Concept formation in Proust’s “A la recherche du temps perdu”. Amsterdam: John Benjamins. DOI: 10.1075/aicr.63

Beaugrande, Robert-Alain, & Dressler, Wolfgang Ulrich. (1981). Einführung in die Textlinguistik. Tübingen: Niemeyer. DOI: 10.1515/9783111349305

Behrens, Heike. (2005). Wortarten-Erwerb durch Induktion. In Clemens Knobloch, & Burkhard Schaeder (Eds.), Wortarten und Grammatikalisierung: Perspektiven in System und Erwerb (pp. 177–197). Berlin: de Gruyter.

Behrens, Heike. (2006). Input-output relationship in first language acquisition. Language and Cognitive Processes, 1/2/3, 2–24. DOI: 10.1080/01690960400001721

Bergen, Benjamin, & Chang, Nancy. (2005). Embodied construction grammar in simulation-based language understanding. In Jan-Ola Östman, & Mirjam Fried (Eds.), Construction grammars: Cognitive grounding and theoretical extensions (pp. 147–190). Amsterdam: John Benjamins. DOI: 10.1075/cal.3.08ber

Bergenholtz, Henning, & Kaufmann, Uwe. (1996). Enzyklopädische Informationen in Wörter-büchern. In Nico Weber (Ed.), Semantik, Lexikographie und Computeranwendungen (pp. 168–182). Tübingen: Niemeyer.

Biere, Bernd Ulrich. (1989). Verständlich-Machen. Hermeneutische Tradition – Historische Praxis – Sprachtheoretische Begründung. Tübingen: Niemeyer. DOI: 10.1515/9783111631844

Bierwisch, Manfred. (1969). Strukturelle Semantik. Deutsch als Fremdsprache, 6, 2, 66–71.Bierwisch, Manfred. (1971). On classifying semantic features. In Danny D. Steinberg, & Leon

A. Jakobovits (Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psychology (pp. 410–435). Cambridge: Cambridge University Press.

Bierwisch, Manfred. (1979). Wörtliche Bedeutung – Eine pragmatische Gretchenfrage. In Inger Rosengren (Ed.), Sprache und Pragmatik. Lunder Symposium 1978 (pp. 63–85). Lund: Gleerup.

Bierwisch, Manfred. (1982). Formal and lexical semantics. Linguistische Berichte, 80, 3–17.Bierwisch, Manfred. (1983a). Semantische und konzeptuelle Repräsentation lexikalischer

Einheiten. In Rudolf Ruzicka, & Wolfgang Motsch (Eds.), Untersuchungen zur Semantik (pp. 61–99). Berlin: Akademie-Verlag.

Bierwisch, Manfred. (1983b). Psychologische Aspekte der Semantik natürlicher Sprachen. In Wolfgang Motsch, & Dieter Viehweger (Eds.), Richtungen der modernen Semantikforschung (pp. 15–64). Berlin: Akademie-Verlag.

http://dx.doi.org/10.1080/01690960344000026

http://dx.doi.org/10.1075/aicr.39

http://dx.doi.org/10.1075/aicr.63

http://dx.doi.org/10.1515/9783111349305

http://dx.doi.org/10.1080/01690960400001721

http://dx.doi.org/10.1075/cal.3.08ber

http://dx.doi.org/10.1515/9783111631844

References 393

Bierwisch, Manfred. (1986). On the nature of semantic form in natural language. In Friedhart Klix, & H. Hagendorf (Eds.), Human memory and cognitive capabilities – mechanisms and performances (pp. 765–784). Amsterdam & New York: North Holland Publishers.

Bierwisch, Manfred. (1987). Linguistik als kognitive Wissenschaft – Erläuterungen zu einem Forschungsprogramm. Germanistische Linguistik, 8, 645–667.

Bierwisch, Manfred. (1997). Mindest- und Höchstbedingungen für lexikalische Semantik. Paper presented at the first annual conference of the Dgfs (German Linguistic Society) in Düsseldorf, February 1997.

Bierwisch, Manfred, & Kiefer, Ferenc. (1970). Remarks on definitions in natural language. In Ferenc Kiefer (Ed.), Studies in syntax and semantics (pp. 55–79). Dordrecht: Reidel.

Bierwisch, Manfred, & Lang, Ewald (Eds.). (1987). Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven. Berlin: Akademie-Verlag.

Bierwisch, Manfred, & Robert Schreuder. (1992). From concepts to lexical items. Cognition, 42, 23–65. DOI: 10.1016/0010-0277(92)90039-K

Blank, Andreas. (1999). Kognitive Linguistik und Bedeutungswandel. In Inge Pohl (Ed.), Inter-disziplinarität und Methodenpluralismus in der Semantikforschung: Beiträge der Konferenz “Interdisziplinarität und Methodenpluralismus in der Semantikforschung” an der Universität Koblenz-Landau/Abteilung Landau 1998 (pp. 125–148). Frankfurt a.M.: Lang.

Blutner, Reinhard. (1995). Systematische Polysemie: Ansätze zur Erzeugung und Beschränkung von Interpretationsvarianten. In Manfred Bierwisch, & Peter Bosch (Eds.), Semantic and conceptual knowledge. Sprachtheoretische Grundlagen für die Computerlinguistik (pp. 33–48). Working papers of Collaborative Research Centre (SFB) 340.

Boas, Hans C. (2003). A constructional approach to resultatives. Stanford: CSLI.Boas, Hans C. (2010). The syntax-lexicon continuum in construction grammar: a case study of

english communication verbs. Belgian Journal of Linguistics, 24, 57–86. DOI: 10.1075/bjl.24.03boaBoas, Hans C. (2013). Cognitive construction grammar. In Thomas Hoffmann, & Graeme

Trousdale (Eds.), The Oxford handbook of construction grammar (pp. 233–254). Oxford: Oxford University Press.

Bobrow, Daniel G., & Collins, Allan (Eds.). (1975). Representation and Understanding. New York: Academic Press.

Böke, Karin. (1996). Überlegungen zu einer Metaphernanalyse im Dienste einer “parzellierten” Sprachgeschichtsschreibung. In Karin Böke, Matthias Jung, & Martin Wengeler (Eds.), Öffentlicher Sprachgebrauch (pp. 431–452). Opladen: Westdeutscher Verlag.

Böke, Karin. (1997). Die “Invasion” aus den “Armenhäusern Europas”. Metaphern im Einwan-derungsdiskurs. In Matthias Jung, Martin Wengeler, & Karin Böke (Eds.), Die Sprache des Migrationsdiskurses. Das Reden über “Ausländer” in Medien, Politik und Alltag (pp.163–192). Opladen: Westdeutscher Verlag.

Börkel, Manfred. (1995). Conceptual interpretation of systematic polysemy. In Manfred Bierwisch, & Peter Bosch (Eds.), Semantic and conceptual knowledge. Sprachtheoretische Grundlagen für die Computerlinguistik (pp. 69–78). Working papers of Collaborative Re-search Centre (SFB) 340.

Botha, Willem J. (1996). The cognitive frame of a set of cricket terms. In Eugene H. Casad (Ed.), Cognitive linguistics in the redwoods: The expansion of a new paradigm in linguistics (pp. 237–247). Berlin & New York: Mouton de Gruyter.

Bransford, John D., Barclay, J. Richards, & Franks, Jeffrey J. (1972). Sentence memory: A con-structive versus interpretive approach. Cognitive Psychology, 3, 193–209.

DOI: 10.1016/0010-0285(72)90003-5

http://dx.doi.org/10.1016/0010-0277(92)90039-K

http://dx.doi.org/10.1075/bjl.24.03boa

http://dx.doi.org/10.1016/0010-0285(72)90003-5


Brewer, William F. (1999). Schemata. In Frank C. Keil, & Robert A. Wilson (Eds.), The MIT encyclopedia of the cognitive sciences (pp. 729–730). Cambridge: MIT Press.

Brewer, William F. (2000). Bartlett’s concept of the schema and its impact on theories of knowl-edge representation in contemporary cognitive psychology. In Akiko Saito (Ed.), Bartlett, culture and cognition (pp. 69–89). Hove: Psychology Press.

Brewer, William F., & Nakamura, Glenn V. (1984). The nature and functions of schemas. In Robert S. Wyer, & Thomas K. Srull (Eds.), Handbook of Social Cognition (Vol. 1, 119–160). Hillsdale, New Jersey: Erlbaum.

Brisard, Frank. (2002). Introduction: The epistemic basis of deixis and reference. In Frank Brisard (Ed.), Grounding. The epistemic footing of deixis and reference (pp. xi–xxxiv). Berlin & New York: Mouton de Gruyter. DOI: 10.1515/9783110899801

Brugman, Claudia. (1988). The story of “Over”: Polysemy, semantics, and the structure of the lexicon. New York: Garland.

Bühler, Karl. (1934). Sprachtheorie. Die Darstellungsfunktion der Sprache. Jena: Fischer.Busse, Dietrich. (1987). Historische Semantik. Analyse eines Programms. Stuttgart: Klett-Cotta.Busse, Dietrich. (1988). Kommunikatives Handeln als sprachtheoretisches Grundmodell der

historischen Semantik. In Ludwig Jäger (Ed.), Zur historischen Semantik des deutschen Gefühlswortschatzes. Aspekte, Probleme und Beispiele seiner lexikographischen Erfassung (pp. 247–272). Aachen: Rader.

Busse, Dietrich. (1991a). Textinterpretation. Sprachtheoretische Grundlagen einer explikativen Semantik. Opladen: Westdeutscher Verlag.

Busse, Dietrich. (1991b). Wortbedeutung und sprachliches Handeln. Überlegungen zu den Grund-lagen der Bedeutungstheorie. Unpublished manuscript. [Accessible at: http://www.phil-fak.uni-duesseldorf.de/germ1/mitarbeiter/busse/mat/busse-up-1991-01.pdf; last accessed: 10 April 2014]

Busse, Dietrich. (1994). Kommunikationsmodelle und das Problem des Sprachverstehens. Zur Verwendung technischer Metaphern in der Sprachwissenschaft. In Rudolf Hoberg (Ed.), Technik in Sprache und Literatur (pp. 207–234). Darmstadt: Verlag der TH Darmstadt.

Busse, Dietrich. (1995). Hermeneutik und Sprachverstehen. Zu den verstehenstheoretischen Grundlagen einer modernen Linguistik. In Kjell S. Johannessen, & Tore Nordenstm (Eds.), Culture and value. Philosophy and the cultural sciences. Proceedings of the 18th International Wittgenstein Symposium 13–20 August 1995, Kirchberg am Wechsel (pp. 616–622). Vienna: Hölder-Pichler-Tempsky.

Busse, Dietrich. (1997a). Semantisches Wissen und sprachliche Information. Zur Abgrenzung und Typologie von Faktoren des Sprachverstehens. In Inge Pohl (Ed.), Methodologische Aspekte der Semantikforschung (pp. 13–34). Frankfurt a.M. et al.: Lang.

Busse, Dietrich. (1997b). Wortarten und semantische Typen. Überlegungen zu den Grundlagen der lexikalisch-syntaktischen Wortarten-Klassifikation. In Christa Dürscheid, Karlheinz Ramers, & Monika Schwarz (Eds.), Sprache im Fokus (pp. 219–240). Tübingen: Niemeyer.

Busse, Dietrich. (1997c). Das Eigene und das Fremde. Zu Funktion und Wirkung einer dis-kurssemantischen Grundfigur. In Matthias Jung, Martin Wengeler, & Karin Böke (Eds.), Die Sprache des Migrationsdiskurses (pp. 17–35). Opladen: Westdeutscher Verlag.

Busse, Dietrich. (1998). Bedeutung und Bedeutungen. Ein Essai. In Inge Pohl, & Jürgen Pohl (Eds.), Texte über Texte. Interdisziplinäre Zugänge (pp. 531–541). Frankfurt a.M.: Lang.

Busse, Dietrich. (2000a). Historische Diskurssemantik. Ein linguistischer Beitrag zur Analyse gesellschaftlichen Wissens. In Anja Stukenbrock, & Joachim Scharloth (Eds.), Linguistische Diskursgeschichte (= Sprache und Literatur in Wissenschaft und Unterricht 31, Vol. 86, pp. 39–53).

http://dx.doi.org/10.1515/9783110899801

References 395

Busse, Dietrich. (2000b). Öffentliche Sprache und politischer Diskurs. Anmerkungen zu einem prekären Gegenstand linguistischer Analyse. In Hajo Diekmannshenke, & Iris Meißner (Eds.), Politische Kommunikation im historischen Wandel (pp. 31–55). Wiesbaden: West-deutscher Verlag.

Busse, Dietrich. (2003a). Begriffsgeschichte oder Diskursgeschichte? Zu theoretischen Grund-lagen und Methodenfragen einer historisch-semantischen Epistemologie. In Carsten Dutt (Ed.), Herausforderungen der Begriffsgeschichte (pp. 17–38). Heidelberg: Winter.

Busse, Dietrich. (2003b). Diskursanalyse in der Sprachgermanistik. In Ulrike Haß, & Christoph König (Eds.), Literaturwissenschaft und Linguistik (pp. 175–187). Marbach: Wallstein.

Busse, Dietrich. (2005a). Architekturen des Wissens – Zum Verhältnis von Semantik und Epi-stemologie. In Ernst Müller (Ed.), Begriffsgeschichte im Umbruch (= Archiv für Begriffsge-schichte, Special Edition 2004) (pp. 43–57). Berlin: Meiner.

Busse, Dietrich. (2005b). Sprachwissenschaft als Sozialwissenschaft? In Dietrich Busse, Thomas Niehr, & Martin Wengeler (Eds.), Brisante Semantik. Neuere Konzepte und Forschungsergeb-nisse einer kulturwissenschaftlichen Linguistik, (pp. 21–43). Tübingen: Niemeyer.

DOI: 10.1515/9783110918328Busse, Dietrich. (2006). Text – Sprache – Wissen. Perspektiven einer linguistischen Epistemo-

logie als Beitrag zur Historischen Semantik. Scientia Poetica, 10, 101–137.Busse, Dietrich. (2007). Diskurslinguistik als Kontextualisierung: Methodische Kriterien.

Sprachwissenschaftliche Überlegungen zur Analyse gesellschaftlichen Wissens. In Ingo Warnke (Ed.), Diskurslinguistik nach Foucault. Theorien und Gegenstände (pp. 81–105). Berlin & New York: de Gruyter. DOI: 10.1515/9783110920390.81

Busse, Dietrich. (2008a). Linguistische Epistemologie. Zur Konvergenz von kognitiver und kul-turwissenschaftlicher Semantik am Beispiel von Begriffsgeschichte, Diskursanalyse und Frame-Semantik. In Heidrun Kämper, & Ludwig M. Eichinger (Eds.), Sprache – Kognition – Kultur. Sprache zwischen mentaler Struktur und kultureller Prägung (pp. 73–114). Berlin & New York: de Gruyter.

Busse, Dietrich. (2008b). Begriffsgeschichte – Diskursgeschichte – Linguistische Epistemologie. Bemerkungen zu den theoretischen und methodischen Grundlagen einer Historischen Se-mantik in philosophischem Interesse anlässlich einer Philosophie der Person. In Alexander Haardt, & Nikolaj Plotnikov (Eds.), Der Diskurs der Personalität: Philosophische Begriffe im interkulturellen Umfeld (pp. 113–139). Munich: Fink.

Busse, Dietrich. (2009b). Textbedeutung und Textverstehen - aus Sicht einer linguistischen Epistemologie. In Andrea Bachmann-Stein, Stephan Merten & Christine Roth (Eds.), Per-spektiven auf Wort, Satz und Text. Semantisierungsprozesse auf unterschiedlichen Ebenen des Sprachsystems. (pp. 45–56). Trier: WVT Wissenschaftlicher Verlag Trier.

Busse, Dietrich. (2012). Frame-Semantik. Ein Kompendium. Berlin/New York: de Gruyter. DOI: 10.1515/9783110269451Busse, Dietrich, & Bickes, Hans. (1984). Kommunikatives Handeln und die Rekonstruktion von

Handlungsmustern. In Epistimoniki Epetirida Tis Philosophikis Scholis. Tmima Germani-kis Glossas Ke Philologias. Aristoteleio Panepistimio Thessalonikis. Thessaloniki 1984–1986. (= Yearbook of the Faculty of Philosophy. Department of German Language and Philology. New issue, Department 6, Vol. 1. Thessaloniki, 1984–86), 235–269.

Busse, Dietrich, & Teubert, Wolfgang. (1994). Ist Diskurs ein sprachwissenschaftliches Objekt? Zur Methodenfrage der historischen Semantik. In Dietrich Busse, Fritz Hermanns, & Wolfgang Teubert (Eds.), Begriffsgeschichte und Diskursgeschichte. Methodenfragen und Forschungsergebnisse der historischen Semantik (pp. 10–28). Opladen: Westdeutscher Verlag.

http://dx.doi.org/10.1515/9783110918328

http://dx.doi.org/10.1515/9783110920390.81

http://dx.doi.org/10.1515/9783110269451


Bybee, Joan L. (1985). Morphology: A study of the relation between meaning and form. Amster-dam: John Benjamins. DOI: 10.1075/tsl.9

Bybee, Joan L. (1995). Regular morphology and the lexicon. Language and Cognitive Processes, 10, 425–455. DOI: 10.1080/01690969508407111

Bybee, Joan L. (2001). Phonology and language use. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511612886

Bybee, Joan L. (2013). Usage-based theory and exemplar representation of constructions. In Thomas Hoffmann, & Graeme Trousdale (Eds.), The Oxford Handbook of Construction Grammar (pp. 49–69). Oxford: Oxford University Press.

Bybee, Joan L., & Hopper, Paul (Eds.). (2001). Frequency and the emergence of linguistic structure. Amsterdam, et al.: John Benjamins. DOI: 10.1075/tsl.45

Bybee, Joan L., & Slobin, Dan I. (1982). Rules and schemas in the development and use of English present tense. Language, 58, 265–289. DOI: 10.1353/lan.1982.0021

Case, Robbie. (1985). Intellectual development: Birth to adulthood. Orlando: Academic Press.Cermak, Laird S., & Craik, Fergus I.M. (Eds.). (1979). Levels of processing in human memory.

Hillsdale, New Jersey: Erlbaum.Charniak, Eugene. (1976). A framed painting: The representation of a common sense knowledge

fragment. Cognitive Science, 1, 355–394. DOI: 10.1016/S0364-0213(77)80015-3Charniak, Eugene. (1981). The case-slot-identity theory. Cognitive Science, 5, 285–292. DOI: 10.1207/s15516709cog0503_4Chomsky, Noam. (1965). Aspects of the theory of syntax. Cambridge: MIT Press.Chomsky, Noam. (1975). Reflections on language. New York: Random House.Chomsky, Noam. (1980). Rules and representations. New York: Columbia Press.Chomsky, Noam. (1986). Changing perspectives on knowledge and use of knowledge. Leuvense

Bijdragen, 75, 1–71.Christmann, Ursula. (1989). Modelle der Textverarbeitung. Textbeschreibungen als Textverstehen.

Münster: Aschendorff.Clark, Herbert H. (1977). Inferences in comprehension. In David Laberge, & S. Jay Samuels

(Eds.), Basic processes in reading: Perception and comprehension (pp. 243–263). Hillsdale, New Jersey: Erlbaum.

Clark, Herbert H. (1978). Inferring what is meant. In William J.M. Levelt, & Giovanni B. Flores d’Arcais (Eds.), Studies in the perception of language (pp. 295–337). Chichester, et al.: Wiley.

Clark, Herbert H. (1983). Making sense of nonsense. In Giovanni B. Flores D’Arcais, & Robert J. Jarvella (Eds.), The process of language understanding (pp. 297–331). Chichester, et al.: Wiley

Clark, Herbert H. (1996). Using language. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511620539Clark, Herbert H. (1997). Dogmas of understanding. Discourse Processes, 23, 567–598. DOI: 10.1080/01638539709545003Clark, Herbert H., & Gerrig, Richard J. (1983). Understanding old words with new meanings.

Journal of Verbal Learning and Verbal Behavior, 22, 591–608. DOI: 10.1016/S0022-5371(83)90364-XClausner, Timothy C., & Croft, William. (1999). Domains and image schemas. Cognitive Lin-

guistics, 10(1), 1–31. DOI: 10.1515/cogl.1999.001Cohen, Henri, & Lefebvre, Claire. (2005b). Bridging the category divide. In H. Cohen, &

C. Lefebvre (Eds.), 2005a, 1–15.Cohen, Henri, & Lefebvre, Claire (Eds.). (2005a). Handbook of categorization in cognitive science.

Amsterdam, et al.: Elsevier.

http://dx.doi.org/10.1075/tsl.9

http://dx.doi.org/10.1080/01690969508407111

http://dx.doi.org/10.1017/CBO9780511612886

http://dx.doi.org/10.1075/tsl.45

http://dx.doi.org/10.1353/lan.1982.0021

http://dx.doi.org/10.1016/S0364-0213(77)80015-3

http://dx.doi.org/10.1207/s15516709cog0503_4


http://dx.doi.org/10.1080/01638539709545003

http://dx.doi.org/10.1016/S0022-5371(83)90364-X

http://dx.doi.org/10.1515/cogl.1999.001

References 397

Coseriu, Eugenio. (1980). Textlinguistik. Eine Einführung. Tübingen: Narr.Coulson, Seana. (1995). Cognitive science. In Jef Verschueren, Jan-Ola Östman, & Jan Blommaert

(Eds.), Handbook of pragmatics (pp. 123–140). Amsterdam: John Benjamins. DOI: 10.1075/hop.m.cog5Coulson, Seana. (2001). Semantic leaps: Frame-shifting and conceptual blending in meaning

construction. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511551352Coulson, Seana, & Oakley, Todd (Eds.). (2000a). Cognitive Linguistics, 11(3/4). Special Issue:

Conceptual Blending.Coulson, Seana, & Oakley, Todd. (2000b). Blending basics. Cognitive Linguistics, 11(3/4),

175–196.Craik, Fergus I.M., & Lockhart, Robert S. (1972). Levels of processing: A framework for memory

research. Journal of Verbal Learning and Verbal Behavior, 11, 671–684. DOI: 10.1016/S0022-5371(72)80001-XCroft, William. (1991). Syntactic categories and grammatical relations. Chicago: University of

Chicago Press.Croft, William. (2001). Radical construction grammar: Syntactic theory in typological perspec-

tive. Oxford: Oxford University Press. DOI: 10.1093/acprof:oso/9780198299554.001.0001Croft, William. (2009). Toward a social cognitive linguistics. In Vyvyan Evans, & Stephanie

Pourcel (Eds.), New directions in cognitive linguistics (pp. 395-420). Amsterdam: John Benjamins.

Croft, William. (2013). Radical construction grammar. In Thomas Hoffmann, & Graeme Trousdale (Eds.), The Oxford handbook of construction grammar (pp. 211–232). Oxford: Oxford University Press.

Croft, William, & Cruse, D. Alan. (2004). Cognitive linguistics. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511803864

Crothers, Edward J. (1978). Inference and coherence. Discourse Processes, 1, 51–71. DOI: 10.1080/01638537809544428Cruse, D. Alan. (1988). Word meaning and encyclopedic knowledge. In Werner Hüllen, & Rainer

Schulze (Eds.), Understanding the lexicon: Meaning, sense and world knowledge in lexical semantics (pp. 73–83). Tübingen: Niemeyer.

Cuyckens, Hubert. (1988). Spatial prepositions in cognitive semantics. In Werner Hüllen, & Rainer Schulze (Eds.), Understanding the lexicon: Meaning, sense and world knowledge in lexical semantics (pp. 316–328). Tübingen: Niemeyer.

Cuyckens, Hubert, Dirven, René, & Taylor, John. (2003). Introduction: New directions in cog-nitive lexical semantic research. In H. Cuyckens, R. Dirven, & J. Taylor (Eds.), Cognitive approaches to lexical semantics (pp. 1–28). Berlin: Mouton de Gruyter.

DOI: 10.1515/9783110219074.1Cuyckens, Hubert, & Radden, Günter (Eds.). (2002). Perspectives on prepositions. Tübingen:

Niemeyer. DOI: 10.1515/9783110924787Dancygier, Barbara, & Sweetser, Eve. (2005). Mental spaces in grammar: Conditional construc-

tions. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511486760Danks, Joseph H., & Glucksberg, Sam. (1980). Experimental psycholinguistics. Annual Review

of Neuroscience, 7, 391–417.Davis, Wayne A. (2003). Meaning, expression, and thought. Cambridge: Cambridge University

Press.Deane, Paul D. (1996). On Jackendoff ’s conceptual semantics. Cognitive Linguistics, 7(1), 35–91.

DOI: 10.1515/cogl.1996.7.1.35

http://dx.doi.org/10.1075/hop.m.cog5


http://dx.doi.org/10.1016/S0022-5371(72)80001-X

http://dx.doi.org/10.1093/acprof:oso/9780198299554.001.0001


http://dx.doi.org/10.1080/01638537809544428

http://dx.doi.org/10.1515/9783110219074.1

http://dx.doi.org/10.1515/9783110924787


http://dx.doi.org/10.1515/cogl.1996.7.1.35


Demmerling, Christoph. (2002). Sinn, Bedeutung, Verstehen. Untersuchungen zu Sprachphiloso-phie und Hermeneutik. Paderborn: Mentis.

Dieckmann, Walther. (1989). Vielleicht ist Wortsemantik doch noch erlaubt? Zeitschrift für Germanistische Linguistik, 17, 220–225. DOI: 10.1515/zfgl.1989.17.2.162

Diessel, Holger. (2004). The acquisition of complex sentences. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511486531

Dijk, Teun A. van. (1980). Textwissenschaft: eine interdisziplinäre Einführung. Tübingen: Niemeyer. DOI: 10.1515/9783110954845

Dijk, Teun A. Van, & Kintsch, Walter. (1983). Strategies of discourse comprehension. London: Academic Press.

Dirven, René. (1994). Cognition and semantic structure: The experiential basis of the semantic structure of verbs of body contact. In Monika Schwarz (Ed.), Kognitive Semantik. Ergebnisse, Probleme, Perspektiven (pp. 131–145). Tübingen: Narr.

Dirven, René, & Radden, Günter. (1977). Semantische Syntax des Englischen. Wiesbaden: Athenaion.

Dirven, René, Roslyn, Frank, & Pütz, Martin (Eds.). (2003). Cognitive models in language and thought: Ideology, metaphors and meanings. Berlin & New York: de Gruyter.

DOI: 10.1515/9783110892901Dirven, René, & Taylor, John R. (1988). The conceptualisation of vertical space in English: The

case of tall. In Brygida Rudzka-Ostyn (Ed.), Topics in cognitive linguistics (pp. 379–402). Amsterdam: John Benjamins. DOI: 10.1075/cilt.50.15dir

Dodge, Ellen, & Lakoff, George. (2005). Image schemas: From linguistic analysis to neural grounding. In Beate Hampe (Ed.), From perception to meaning: Image schemas in cognitive linguistics (pp. 57–91). DOI: 10.1515/9783110197532.1.57

Dölling, Johannes. (1995). Lexikalische Einträge und systematische Variationen im Nominal-bereich. In Manfred Bierwisch, & Peter Bosch (Eds.), Semantic and conceptual knowledge. Papers from a joint workshop of the Arbeitsgruppe Strukturelle Grammatik and the Institut für Logik und Linguistik, Berlin, 21–23 April, 1994 (pp. 127–138). Stuttgart, Tübingen, Heidelberg: IBM Deutschland.

Dölling, Johannes. (2001a). Aspektuelle Anpassungen. In Johannes Dölling, & Tatjana Zybatow (Eds.), Ereignisstrukturen. Linguistische Arbeitsberichte 76 (pp. 321–353). Universität Leipzig.

Dölling, Johannes. (2001b). Systematische Bedeutungsvariationen: Semantische Form und kon-textuelle Interpretation. Linguistische Arbeitsberichte 78. Universität Leipzig.

Dölling, Johannes. (2005). Semantische Form und pragmatische Anreicherung: Situationsaus-drücke in der Äußerungsinterpretation. Zeitschrift für Sprachwissenschaft, 24(2), 159–225. DOI: 10.1515/zfsw.2005.24.2.159

Drever, James, & Fröhlich, Werner D. (1975). dtv-Wörterbuch der Psychologie. Munich: dtv.Drewer, Petra. (2003). Die kognitive Metapher als Werkzeug des Denkens. Zur Rolle der Analogie

bei der Gewinnung und Vermittlung wissenschaftlicher Erkenntnisse. Tübingen: Narr.Elbert, Thomas, Pantev, Christo, Wienbruch, Christian, Rockstroh, Brigitte, & Taub, Edward.

(1995). Increased cortical representation of the fingers of the left hand in string players. Science, 270, 305–307. DOI: 10.1126/science.270.5234.305

Elman, Jeffrey L., Bates, Elizabeth A., Johnson, Mark H., Karmiloff-Smith, Annette, Parisi, Domenico, & Plunkett, Kim. (1996). Rethinking innateness: A Connectionist perspective on development. Cambridge: MIT Press.

http://dx.doi.org/10.1515/zfgl.1989.17.2.162


http://dx.doi.org/10.1515/9783110954845

http://dx.doi.org/10.1515/9783110892901

http://dx.doi.org/10.1075/cilt.50.15dir

http://dx.doi.org/10.1515/9783110197532.1.57

http://dx.doi.org/10.1515/zfsw.2005.24.2.159

http://dx.doi.org/10.1126/science.270.5234.305

References 399

Epstein, Richard. (2002). Grounding, subjectivity and definite description. In Frank Brisard (Ed.), Grounding: The epistemic footing of deixis and reference (pp. 41–82). Berlin & New York: Mouton de Gruyter.

Evans, Vyvyan, & Green, Melanie. (2006). Cognitive linguistics: An introduction. Edinburgh: Edinburgh University Press.

Fauconnier, Gilles. (1985). Mental spaces: Aspects of meaning construction in natural languages. Cambridge: MIT Press.

Fauconnier, Gilles. (1990). Domains and connections. Cognitive Linguistics, 1(1), 151–174. DOI: 10.1515/cogl.1990.1.1.151Fauconnier, Gilles. (1997). Mappings in thoughts and language. Cambridge: Cambridge Univer-

sity Press. DOI: 10.1017/CBO9781139174220Fauconnier, Gilles. (1999). Methods and generalizations. In Theo Janssen, & Gisela Redeker

(Eds.), Cognitive linguistics: Foundations, scope, and methodology (pp. 94–127). Berlin & New York: Mouton de Gruyter.

Fauconnier, Gilles, & Sweetser, Eve. (1996). Spaces, worlds, and grammars. Chicago: University of Chicago Press.

Fauconnier, Gilles, & Turner, Mark. (1998a). Conceptual integration networks. Cognitive Science, 22(2), 133–187. DOI: 10.1207/s15516709cog2202_1

Fauconnier, Gilles, & Turner, Mark. (1998b). Principles of conceptual integration. In Jean-Pierre Koenig (Ed.), Discourse and cognition (pp. 285–296). Stanford: Centre for the Study of Language and Information.

Fauconnier, Gilles, & Turner, Mark. (1999). Metonymy and conceptual integration. In Klaus-Uwe Panther, & Günter Radden (Eds.), Metonymy in language and thought (pp. 77–90). Amsterdam: Mouton de Gruyter. DOI: 10.1075/hcp.4.05fau

Fauconnier, Gilles, & Turner, Mark. (2000). Compression and global insight. Cognitive Linguis-tics, 11(3/4), 283–304.

Fauconnier, Gilles, & Turner, Mark. (2002). The way we think: Conceptual blending and the mind’s hidden complexities. New York: Basic Books.

Felder, Ekkehard. (2006). Semantische Kämpfe in Wissensdomänen. Eine Einführung in Benen-nungs-, Bedeutungs- und Sachverhaltsfixierungs-Konkurrenzen. In Ekkehard Felder (Ed.), Semantische Kämpfe. Macht und Sprache in den Wissenschaften (pp. 13–46). Berlin & New York: de Gruyter. DOI: 10.1515/9783110193282

Felix, Sascha, Habel, Christopher, & Rickheit, Gert (Eds.). (1994). Kognitive Linguistik. Reprä-sentation und Prozesse. Opladen: Westdeutscher Verlag. DOI: 10.1007/978-3-663-05399-6

Fields, R. Douglas. (2005). Wie Erinnerungen haften bleiben. Spektrum der Wissenschaft, 9, 62–69.

Fillmore, Charles F., & Baker, Collin F. (2001). Frame semantics for text understanding. In Pro-ceedings of WorldNet and Other Lexical Resources Workshop, NAACL, Pittsburgh, June, 2001.

Fillmore, Charles J. (1968). The case for case. In Emmon Bach, & Robert T. Harms (Eds.), Uni-versals in linguistic theory (pp. 1–88). New York: Holt, Rinehart and Winston.

Fillmore, Charles J. (1970). The grammar of breaking and hitting. In Roderick A. Jacobs, & Peter S. Rosenbaum (Eds.), Readings in English transformational grammar (pp. 120–133). Waltham: Ginn.

Fillmore, Charles J. (1971a). Types of lexical information. In Danny D. Steinberg, & Leon A. Jakobovits (Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psy-chology (pp. 370–392). Cambridge: Cambridge University Press.




http://dx.doi.org/10.1075/hcp.4.05fau

http://dx.doi.org/10.1515/9783110193282

http://dx.doi.org/10.1007/978-3-663-05399-6


Fillmore, Charles J. (1971b). Verbs of judging: An exercise in semantic description. In Charles J. Fillmore, & D. Terence Langendoen (Eds.), Studies in linguistic semantics (pp. 272–289). New York: Holt, Rinehart and Winston.

Fillmore, Charles J. (1971c). Some problems for case grammar. In Reister J. O’Brien (Ed.), Georgetown university round table in languages and linguistics (Monograph Series on Lan-guages and Linguistics 23) (pp. 35–56). Washington: Georgetown University Press.

Fillmore, Charles J. (1973). A grammarian looks to socio-linguistics. In Roger W. Shuy (Ed.), Report of the Twenty-Third Annual Round-Table Meeting on Linguistics and Language Studies. (Georgetown University Monographs on Languages and Linguistics Number 25) (pp. 273–287). Washington: Georgetown University Press.

Fillmore, Charles J. (1975). An alternative to checklist theories of meaning. In Cathy Cogen, Henry Thompson, Graham Thurgood, Kenneth Whistler, & Wright, James (Eds.), Proceed-ings of the First Annual Meeting of the Berkeley Linguistics Society (pp. 123–131). Berkeley: Berkeley Linguistics Society.

Fillmore, Charles J. (1976a). The need for a frame semantics within linguistics. In Hans Karlgren (Ed.), Statistical Methods in Linguistics 12 (pp. 5–29).

Fillmore, Charles J. (1976b). Frame semantics and the nature of language. Annals of the New York Academy of Sciences: Conference on the Origin and Development of Language and Speech (Vol. 280, pp. 20–32).

Fillmore, Charles J. (1977a). Scenes-and-frames semantics. In Antonio Zampolli (Ed.), Linguistic structures processing (Vol. 5, pp. 55–81). Amsterdam, New York, Oxford: North Holland.

Fillmore, Charles J. (1977b). Topics in lexical semantics. In Roger W. Cole (Ed.), Current Issues in Linguistic Theory (pp. 76–138). Bloomington & London: Indiana University Press.

Fillmore, Charles J. (1977c). Schemata and prototypes. Lecture notes of a symposium held at Trier University, 1977. In René Dirven, & Günter A. Radden (Eds.), Fillmore’s case grammar. A reader (pp. 99–106). Heidelberg: Groos.

Fillmore, Charles J. (1977d). The case for case reopened. In Peter Cole, & Jerrold M. Sadock (Eds.), Syntax and semantics. Vol. 8. Grammatical relations (pp. 59–81). New York & Lon-don: Academic Press.

Fillmore, Charles J. (1981). Ideal readers and real readers. In Deborah Tannen (Ed.), Analyzing discourse: Text and talk. Georgetown University Round Table in Languages and Linguistics (pp. 248–270). Washington: Georgetown University Press.

Fillmore, Charles J. (1982a). Frame semantics. In The Linguistic Society of Korea (Ed.), Linguis-tics in the morning calm (pp. 111–137). Seoul: Hanshin Publishing Company.

Fillmore, Charles J. (1982b). Towards a descriptive framework for spatial deixis. In Robert J. Jarvella, & Wolfgang Klein (Eds.), Speech, place, and action (pp. 31–59). Chichester, et al.: Wiley.

Fillmore, Charles J. (1984). Lexical semantics and text semantics. In James E. Copeland (Ed.), New directions in linguistics and semiotics, (pp. 123–147). Houston: Rice University Studies. DOI: 10.1075/cilt.32.12fil

Fillmore, Charles J. (1985). Frames and the semantics of understanding. Quaderni di Semantica, 6(2), 222–254.

Fillmore, Charles J. (1988). The mechanisms of “Construction Grammar”. Berkeley Linguistics Society, 14, 35–55.

Fillmore, Charles J. (2003). Valency and semantic roles: The concept of deep structure case. In Vilmos Ágel, et al. (Eds.), Dependenz und Valenz. Ein internationales Handbuch/Depend-ency and Valency. An International Handbook of Contemporary Research (pp. 457–475). Berlin & New York: de Gruyter.

http://dx.doi.org/10.1075/cilt.32.12fil

References 401

Fillmore, Charles J. (2006). Frames semantics. In Keith Brown (Ed.), Encyclopedia of linguistics and language (Vol. 4, pp. 613–620). Amsterdam: Elsevier.

Fillmore, Charles J. (2013). Berkeley construction grammar. In Thomas Hoffmann, & Graeme Trousdale (Eds.), The Oxford handbook of construction grammar (pp. 111–132). Oxford: Oxford University Press.

Fillmore, Charles J., & Atkins, Sue. (1992). Toward a frame-based lexicon: The semantics of RISK and its neighbors. In Adrienne Lehrer, & Eva F. Kittay (Eds.), Frames, fields, and con-trast: New essays in semantics and lexical organisation (pp. 75–102). Hillsdale, New Jersey: Erlbaum.

Fillmore, Charles J., & Atkins, Sue. (1994). Starting where the dictionaries stop: The challenge for computational lexicography. In Sue Atkins, & Antonio Zampolli (Eds.), Computational approaches to the lexicon (pp. 349–393). Oxford: Oxford University Press.

Fillmore, Charles J., Kay, Paul, & O’Connor, Mary C. (1988). Regularity and idiomaticity in gram-matical constructions: The case of “let alone”. Language, 64, 501–538. DOI: 10.2307/414531

Fischer, Kerstin, & Stefanowitsch, Anatol. (2006). Konstruktionsgrammatik: Ein Überblick. In Anatol Stefanowitsch, & Kerstin Fischer (Eds.), Konstruktionsgrammatik: Von der Anwend-ung zur Theorie (pp. 3–17). Tübingen: Stauffenburg.

Fisher, Kim. (1997). Locating frames in the discursive universe. Sociological Research Online, 2(3). [Accessible at: http://www.socresonline.org.uk/2/3/4.html; last accessed: 10 April 2014] DOI: 10.5153/sro.78

Fodor, Jerry A. (1983). The modularity of mind: An essay on faculty psychology. Cambridge: MIT Press.

Fodor, Jerry A., Garrett, Merrill F., Walker, Edward C.T., & Parkes, Cornelia H. (1980). Against definitions. Cognition, 8, 263–367. DOI: 10.1016/0010-0277(80)90008-6

Foucault, Michel. (1971). The archaeology of knowledge and the discourse on language. Translated from the French by A.M. Sheridan Smith. New York: Pantheon Books.

Foucault, Michel. (1974a). Die Ordnung der Dinge. Frankfurt a.M.: Suhrkamp.Foucault, Michel. (1974b). Die Ordnung des Diskurses. Munich: Hanser.Foucault, Michel. (1981). Archäologie des Wissens. Frankfurt a.M.: Suhrkamp.Fraas, Claudia. (1996a). Gebrauchswandel und Bedeutungsvarianz in Textnetzen. Die Konzepte

“Identität” und “Deutsche” im Diskurs zur deutschen Einheit. Tübingen: Narr.Fraas, Claudia. (1996b). Bedeutungskonstitution in Texten – Das IDENTITÄTs-Konzept im

Diskurs zur deutschen Einheit. In Edda Weigand, & Franz Hundsnurscher (Eds.), Lexi-cal structures and language use. Proceedings of the International Conference on Lexicology and Lexical Semantics (Vol. 2, pp. 39–51). Münster, September 13–15, 1994. Tübingen: Niemeyer.

Fraas, Claudia. (2000). Begriffe – Konzepte – kulturelles Gedächtnis. Ansätze zur Beschreibung kollektiver Wissenssysteme. In Horst Dieter Schlosser (Ed.), Sprache und Kultur (pp. 31–45). Frankfurt a.M.: Lang.

Fraas, Claudia. (2001). Usuelle Wortverbindungen als sprachliche Manifestation von Bedeu-tungswissen. Theoretische Begründung, methodischer Ansatz und empirische Befunde. In Henrik Nikula, & Robert Drescher (Eds.), Lexikon und Text (pp. 41–66). Vaasa: Saxa.

Freeman, Margaret H. (2007). Cognitive linguistic approaches to literary studies: State of the art in cognitive poetics. In Dirk Geeraerts, & Hubert Cuyckens (Eds.), The handbook of cogni-tive linguistics (pp. 1175–1202). Oxford, et al.: Oxford University Press.

Frey, Uwe, & Morris, Richard. (1997). Synaptic tagging and long-term potentiation. Nature, 385, 533–536. DOI: 10.1038/385533a0

http://dx.doi.org/10.2307/414531

http://dx.doi.org/10.5153/sro.78

http://dx.doi.org/10.1016/0010-0277(80)90008-6

http://dx.doi.org/10.1038/385533a0


Fricke, Matthias. (1999). Diskussion neuerer Foucault-basierter Verfahren der Diskursanalyse anhand von empirischen Analysen von Printmedientexten. Oldenburg. [Accessible at: http://docserver.bis.uni-oldenburg.de/publikationen/dissertation/1999/friemp99/friemp99.html; last accessed: 10 April 2014]

Fritz, Gerd. (1998). Historische Semantik. Stuttgart & Weimar: Metzler.Fuster, Joaquín. (2005). Im Netzwerk der Erinnerungen. Spektrum der Wissenschaft, 2. Special

edition: Gedächtnis, 10–15.Gansel, Christina. (2002). Zur kognitiven Erfassung der Welt durch sprachliche Bedeutungsstruk-

turen. In Akten des X. Internationalen Germanistenkongresses Wien 2000. Vol. 2: Entwick-lungstendenzen der deutschen Gegenwartssprache – Lexikologie und Lexikographie (Jahrbuch für Internationale Germanistik, Reihe A) (Vol. 54, pp. 277–282). Berlin & Frankfurt a.M. et al.: Lang.

Gansel, Christina, & Jürgens, Frank. (2002). Textlinguistik und Textgrammatik. Eine Einführung. Wiesbaden: Westdeutscher Verlag.

Gardner, Howard. (1989). Dem Denken auf der Spur. Der Weg der Kongnitionswissenschaft. Stutt-gart: Klett-Cotta.

Garnham, Alan. (1980). Default values, criteria and constructivism. Cognitive Science, 4, 427–433. DOI: 10.1207/s15516709cog0404_4

Garrod, Simon C. (1985). Incremental pragmatic interpretation versus occasional inferencing during fluent reading. In Gert Richkheit, & Hans Strohner (Eds.), Inferencing in text pro-cessing (pp. 161–181). Amsterdam: North-Holland. DOI: 10.1016/S0166-4115(08)62736-3

Geeraerts, Dirk. (1993). Cognitive semantics and the history of philosophical epistemology. In Richard A. Geiger, & Brygida Rudzka-Ostyn (Eds.), Conceptualization and mental process-ing in language (pp. 54–79). Berlin & New York: Mouton de Gruyter.

Giora, Rachel. (1997). Understanding figurative and literal language: The graded salience hy-pothesis. Cognitive Linguistics, 8(3), 183–206. DOI: 10.1515/cogl.1997.8.3.183

Goffman, Erving. (1974). Frame Analysis: An Essay on the Organization of Experience. New York: Harper and Row. [Germ: Goffman, Erving. (2000). Rahmenanalyse. Ein Versuch über die Organisation der Alltagserfahrung. Frankfurt a. M.: Suhrkamp.]

Goldberg, Adele. (1995). Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press.

Goossens, Louis. (1995). Metaphtonymy: The interaction of metaphor and metonymy in figurative expressions for linguistic action. In Louis Goossens, Paul Pauwels, Brygida Rudzka-Ostyn, Anne-Marie Simon-Vanderbergen, & Johan Vanparys (Eds.), By word of mouth: Metaphor, metonymy and linguistic action in a cognitive perspective (pp. 159–174). Amsterdam: John Benjamins. DOI: 10.1075/pbns.33.06goo

Grady, Joseph E., Oakley, Todd, & Coulson, Seana. (1999). Blending and metaphor. In Gerard Steen, & Raymond Gibbs (Eds.), Metaphor in cognitive linguistics (pp. 101–124). Amster-dam: John Benjamins. DOI: 10.1075/cilt.175.07gra

Graesser, Arthur C., & Bower, Gordon H. (Eds.). (1990). Inferences and text comprehension. San Diego: Academic Press.

Graesser, Arthur C., Millis, Keith K., & Zwaan, Rolf A. (1997). Discourse comprehension. Annual Review of Neuroscience, 46, 163–189.

Graesser, Arthur C., Singer, Murray, & Trabasso, Tom. (1994). Constructing inferences during narrative text comprehension. Psychological Review, 101(3), 371–395.

DOI: 10.1037/0033-295X.101.3.371


http://dx.doi.org/10.1016/S0166-4115(08)62736-3


http://dx.doi.org/10.1075/pbns.33.06goo

http://dx.doi.org/10.1075/cilt.175.07gra

http://dx.doi.org/10.1037/0033-295X.101.3.371

References 403

Graumann, Carl F. (2002). Explicit and implicit perspectivity. In Carl F. Graumann, & Werner Kallmeyer (Eds.), Perspective and perspectivation in discourse (pp. 25–39). Amsterdam: John Benjamins. DOI: 10.1075/hcp.9

Gries, Stefan T. (2006). Corpus-based methods and cognitive semantics: The many meanings of to run. In Stefan Gries, & Anatol Stefanowitsch (Eds.), Corpora in cognitive linguistics: Corpus-based approaches to syntax and lexis (pp. 57–98). Berlin, Heidelberg, New York: Mouton de Gruyter. DOI: 10.1515/9783110197709

Gries, Stefan T., & Stefanowitsch, Anatol (Eds.). (2006). Corpora in cognitive linguistics: Corpus-based approaches to syntax and lexis. Berlin, Heidelberg, New York: Mouton de Gruyter. DOI: 10.1515/9783110197709

Habel, Christopher, Kanngießer, Siegfried, & Rickheit, Gert (Eds.). (1996). Perspektiven der Kognitiven Linguistik. Opladen: Westdeutscher Verlag. DOI: 10.1007/978-3-663-07678-0

Haiman, John. (1980). Dictionaries and encyclopedias. Lingua, 50, 329–357. DOI: 10.1016/0024-3841(80)90089-3Halliday, Michael A.K. (1976). The form of a functional grammar. In Günther Kress (Ed.), Hal-

liday: System and function in language (pp. 7–25). Oxford: Oxford University Press.Hampe, Beate (Ed.). (2005). From perception to meaning: Image schemas in cognitive linguistics.

Berlin: Mouton de Gruyter. DOI: 10.1515/9783110197532Hampton, James A., & Moss, Helen E. (2003). Concepts and meaning: Introduction to the special

issue on conceptual representation. Language and Cognitive Processes, 18(5/6), 505–512. DOI: 10.1080/01690960344000161

Harras, Gisela, Hermann, Theo, & Grabowski, Joachim. (1996). Aliquid stat pro aliquo – aber wie? In Joachim Grabowski, Gisela Harras, & Theo Hermann (Eds.), Bedeutung – Konzepte – Bedeutungskonzepte. Theorie und Anwendung in Linguistik und Psychologie (pp. 3–15). Opladen. DOI: 10.1007/978-3-322-86544-1

Hart, Christopher. (2007). Critical discourse analysis and conceptualisation: mental spaces, blended spaces and discourse spaces in the British national party. In Christopher Hart, & Dominik Lukeš (Eds.), Cognitive linguistics in critical discourse analysis (pp. 107-131). Newcastle: Cambridge Scholars Publishing.

Hartmann, Peter. (1958). Sprache und Erkenntnis. Zur Konstitution des explizierenden Erkennens. Heidelberg: Winter.

Hartung, Wolfdietrich. (1997). Text und Perspektive. Elemente einer konstruktivistischen Tex-tauffassung. In Gerd Antos, & Heike Tietz (Eds.), Die Zukunft der Textlinguistik: Tradi-tionen, Transformationen, Trends. Tübingen: Niemeyer.

Hayes, Patrick J. (1980). The logic of frames. In Dieter Metzing (Ed.), Frame conceptions and text understanding. Berlin & New York: de Gruyter.

Hebb, Donald Olding. (1949). The organization of behaviour: A neuropsychological theory. New York: Wiley.

Heinemann, Wolfgang, & Viehweger, Dieter. (1991). Textlinguistik. Eine Einführung. Tübingen: Niemeyer. DOI: 10.1515/9783111376387

Heringer, Hans Jürgen. (1984). Neues von der Verb-Szene. In Gerhard Stickel (Ed.), Pragmatik in der Grammatik. Yearbook 1983 of the IdS [Institute for German Language] (pp. 34–64). Düsseldorf: Schwann.

Heringer, Hans Jürgen. (1999). Das höchste der Gefühle. Empirische Studien zur distributiven Semantik. Tübingen: Stauffenberg.

http://dx.doi.org/10.1075/hcp.9

http://dx.doi.org/10.1515/9783110197709

http://dx.doi.org/10.1515/9783110197709

http://dx.doi.org/10.1007/978-3-663-07678-0

http://dx.doi.org/10.1016/0024-3841(80)90089-3

http://dx.doi.org/10.1515/9783110197532

http://dx.doi.org/10.1080/01690960344000161

http://dx.doi.org/10.1007/978-3-322-86544-1

http://dx.doi.org/10.1515/9783111376387


Hermanns, Fritz. (1994). Linguistische Anthropologie. Skizze eines Gegenstandsbereiches lin-guistischer Mentalitätsgeschichte. In Dietrich Busse, Fritz Hermanns, & Wolfgang Teubert (Eds.), Begriffsgeschichte oder Diskursgeschichte? Methodenfragen und Forschungsergebnisse der historischen Semantik (pp. 29–59). Opladen: Westdeutscher Verlag.

Hermanns, Fritz. (1995). Sprachgeschichte als Mentalitätsgeschichte. Überlegungen zu Sinn und Form und Gegenstand historischer Semantik. In Andreas Gardt, Klaus J. Mattheier, & Oskar Reichmann (Eds.), Sprachgeschichte des Neuhochdeutschen. Gegenstände, Methoden, Theorien (pp. 69–101). Tübingen: Niemeyer.

Herweg, Michael. (1988). Zur Semantik einiger lokaler Präpositionen des Deutschen: Überlegun-gen zur Theorie der lexikalischen Semantik am Beispiel von ‘in’, ‘an’, ‘bei’ und ‘auf ’. Stuttgart: IBM Deutschland.

Hess, David J., Foss, Donald J., & Carroll, Patrick. (1994). Effects of global and local context on lexical processing during language comprehension. Journal of Experimental Psychology: General, 124(1), 62–82. DOI: 10.1037/0096-3445.124.1.62

Hintikka, Jaakko, & Hintikka, Merril B. (1985). Sherlock Holmes in Konfrontation mit der modernen Logik. Zu einer Theorie des Informationserwerbs durch Befragung. In Umberto Eco, & Thomas A. Sebeok (Eds.), Der Zirkel oder im Zeichen der Drei. Dupin. Holmes. Peirce (pp. 231–251). Munich: Fink.

Hirschfeld, Lawrence A., & Gelman, Susan A. (Eds.). (1994). Mapping the mind: Domain specific-ity in cognition and culture. New York: Cambridge University Press.

DOI: 10.1017/CBO9780511752902Hoffmann, Thomas, & Trousdale, Graeme (Eds.). (2013). The Oxford handbook of construction

grammar. Oxford: Oxford University Press. DOI: 10.1093/oxfordhb/9780195396683.001.0001Holly, Werner. (2001). ‘Frame’ als Werkzeug historisch-semantischer Textanalyse. Eine Debat-

tenrede des Chemnitzer Paulskirchen-Abgeordneten Eisenstuck. In Hajo Diekmannshenke, & Iris Meißner (Eds.), Politische Kommunikation im historischen Wandel (pp. 125–146). Tübingen: Stauffenburg.

Holly, Werner. (2002). ‘Tschechen’ und ‘Deutsche’ in der Böhmen-Debatte der Paulskirche. Ein frame-analytischer Beitrag zur Geschichte der sprachlichen Konstruktion deutsch-tschechischen Beziehungen. In Dieter Cherubim (Ed.), Neue deutsche Sprachgeschichte: mentalitäts-, kultur- und sozialgeschichtliche Zusammenhänge (pp. 349–378). Berlin & New York: de Gruyter. DOI: 10.1515/9783110903447.349

Hoppe-Graff, Siegfried. (1984). Verstehen als kognitiver Prozeß. Psychologische Ansätze und Beiträge zum Textverstehen. LiLi, 55, 10–37.

Hörmann, Hans. (1977). Psychologie der Sprache. Berlin, Heidelberg, New York: Springer. DOI: 10.1007/978-3-662-02286-3Hörmann, Hans. (1994). Meinen und Verstehen (4th ed.). Frankfurt a. M.: Suhrkamp.Huck, Geoffrey J., & Goldsmith, John A. (1998). On comparing linguistic theories: Further

notes on the generative semantics/interpretive semantics debate in American linguistics. Historiographia Linguistica, 25(3), 345–372. DOI: 10.1075/hl.25.3.07huc

Husserl, Edmund. (1980). Logische Untersuchungen (Vol. II/1). Tübingen: Niemeyer.Iser, Wolfgang. (1975). Die Appellstruktur der Texte. In Rainer Warning (Ed.), Rezeptionsästhe-

tik. Theorie und Praxis (pp. 228–252). Munich: Fink.Jackendoff, Ray S. (1983). Semantics and cognition. Cambridge: MIT.Jackendoff, Ray S. (1990). Semantic structures. Cambridge: MIT.Jackendoff, Ray S. (1996). Conceptual semantics and cognitive linguistics. Cognitive Linguistics,

7(1), 93–129. DOI: 10.1515/cogl.1996.7.1.93

http://dx.doi.org/10.1037/0096-3445.124.1.62


http://dx.doi.org/10.1093/oxfordhb/9780195396683.001.0001

http://dx.doi.org/10.1515/9783110903447.349

http://dx.doi.org/10.1007/978-3-662-02286-3

http://dx.doi.org/10.1075/hl.25.3.07huc


References 405

Jackendoff, Ray S. (1997). The architecture of the language faculty. Cambridge: MIT Press.Jackendoff, Ray S. (2002). Foundations of language. Brain, meaning, grammar, evolution. Oxford:

Oxford Press.Jackendoff, Ray S. (2003). Précis of foundations of language: Brain, meaning, grammar, evolu-

tion. Behavioral and Brain Sciences, 26, 651–707.Jakobson, Roman. (1936). Beiträge zur allgemeinen Kasuslehre. Gesamtbedeutung der russi-

schen Kasus. In Roman Jakobson (Ed.), Selected Writings II (pp. 23–71). Den Haag: Mouton.Johnson, Christopher. (1999). Metaphor vs. conflation in the acquisition of polysemy: The case of

“see”. In Masako K. Hiraga, Christopher Sinha, & Shem Wilcox (Eds.), Cultural, psychologi-cal and typological issues in cognitive linguistics (pp. 155–169). Amsterdam: John Benjamins. DOI: 10.1075/cilt.152.12joh

Johnson, Mark. (1987). The body in the mind: The bodily basis of meaning, imagination, and reason. Chicago: University of Chicago Press.

Johnson, Mark. (1999). Embodied reason. In Gail Weiss, & Honi Fern Haber (Eds.), Perspectives on embodiment: The intersections of nature and culture (pp. 143–162). New York: Routledge.

Johnson-Laird, Philipp N. (1983). Mental models. Cambridge: Cambridge University Press.Jung, Matthias. (1996). Linguistische Diskursgeschichte. In Karin Böke, Matthias Jung, & Martin

Wengeler, (Eds.), Öffentlicher Sprachgebrauch. Praktische, theoretische und historische Per-spektiven. Georg Stötzel zum 60. Geburtstag gewidmet. [Dedicated to Georg Stötzel on his 60th birthday] (pp. 453–472). Opladen: Westdeutscher Verlag.

Jung, Matthias. (1997). Lexik und Sprachbewußtsein im Migrationsdiskurs. Methodik und Ergebnis wortbezogener Untersuchungen. In Matthias Jung, Martin Wengeler, & Karin Böke (Eds.), Die Sprache des Migrationsdiskurses. Das Reden über “Ausländer” in Medien, Politik und Alltag (pp. 194–213). Opladen: Westdeutscher Verlag.

Jung, Matthias. (2000). Diskurshistorische analyse als linguistischer Ansatz. In Sprache und Literatur in Wissenschaft und Unterricht, 86, 20–38.

Jüttner, Caspar. (1980). Nichtbewußtes Wissen, Mustererkennung und der Entwurf eines Gedächtnismodells. Psychologische Beiträge, 22, 70–87.

Kaiser, Michael. (1982). Die Schematheorie des Verstehens fiktionaler Literatur. Bemerkungen zur Forschungssituation. Deutsche Vierteljahrsschrift für Literaturwissenschaft und Geistes-geschichte, 56. Special edition: Kultur, Geschichte und Verstehen. Forschungsberichte und Beiträge zu den Themen ‘Kulturwandel’, ‘Ästhetisierung der Historie’ und ‘Schematheorie des Verstehens’, 226–248.

Kalwa, Nina. (2010). Islam-Konzepte im Kölner Moscheebaudiskurs: Eine frame-semantische Analyse zum Islambegriff. Aptum. Zeitschrift für Sprachkritik und Sprachkultur, 6(1), 55-75.

Kay, Paul. (1997). Construction grammar. In Paul Kay (Eds.), Words and the grammar of context (pp. 123–131). Stanford: CSLI.

Kay, Paul, & Fillmore, Charles J. (1999). Grammatical constructions and linguistic generaliza-tions: The What’s X doing Y? construction. Language, 75 (1), 1–33. DOI: 10.2307/417472

Keenan, Janice M. (1989). Beyond Bartlett: Issues in the study of comprehension. In Hans W. Dechert, & Manfred Raupach (Eds.), Psycholinguistic models of production (pp. 239–243). Norwood: Ablex.

Keil, Frank C. (2003). Categorisation, causation, and the limits of understanding. Language and Cognitive Processes, 18(5/6), 663–692. DOI: 10.1080/01690960344000062

Keller, Jörg, & Leuninger, Helen. (2004). Grammatische Strukturen – Kognitive Prozesse. Ein Arbeitsbuch. Tübingen: Narr.

http://dx.doi.org/10.1075/cilt.152.12joh

http://dx.doi.org/10.2307/417472

http://dx.doi.org/10.1080/01690960344000062


Keller, Reiner. (1998). Müll – Die gesellschaftliche Konstruktion des Wertvollen. Die öffentliche Diskussion über Abfall in Deutschland und Frankreich. Opladen & Wiesbaden: Westdeut-scher Verlag. DOI: 10.1007/978-3-322-99391-5

Keller, Rudi. (1995). Zeichentheorie: zu einer Theorie semiotischen Wissens. Tübingen & Basel: Francke.

Keller, Rudi. (2003). Sprachwandel. Von der unsichtbaren Hand in der Sprache. (3rd ed.). Tübingen & Basel: Francke.

Keller, Rudi, & Kirschbaum, Ilja. (2003). Bedeutungswandel. Eine Einführung. Berlin & New York: de Gruyter. DOI: 10.1515/9783110895315

Kindt, Walther. (2001). Neue Wege der Inferenzforschung. In Lorenz Sichelschmidt, & Hans Strohner (Eds.), Sprache, Sinn und Situation. Festschrift für Gert Rickheit (pp. 109–124). Wiesbaden: Deutscher Universitäts-Verlag. DOI: 10.1007/978-3-663-08962-9_7

Kintsch, Walter. (1977). Memory and cognition. New York, et al.: Wiley.Kintsch, Walter. (1980). Learning from text, levels of comprehension, or: Why anyone would

read a story anyway. Poetics, 9, 87–98. DOI: 10.1016/0304-422X(80)90013-3Kintsch, Walter. (1988). The role of knowledge in discourse comprehension, or: Why anyone would

read a story anyway. Psychological Review, 95, 163–182. DOI: 10.1037/0033-295X.95.2.163Kintsch, Walter, & Greene, Edith. (1978). The role of culture-specific schemata in the comprehen-

sion and recall of stories. Discourse Processes, 1, 1–13. DOI: 10.1080/01638537809544425Kintsch, Walter, & van Dijk, Teun A. (1978). Toward a model of text comprehension and produc-

tion. Psychological Review, 85(5), 363–194. DOI: 10.1037/0033-295X.85.5.363Kleiber, George. (1998). Prototypensemantik. Eine Einführung. Tübingen: Narr.Klein, Josef. (1999). Frame als semantischer Theoriebegriff und als wissensdiagnostisches In-

strumentarium. In Inge Pohl (Ed.), Interdisziplinarität und Methodenpluralismus in der Semantikforschung (pp. 157–183). Frankfurt a.M.: Lang.

Klein, Josef. (2002a). Metapherntheorie und Frametheorie. In Inge Pohl (Ed.), Prozesse der Be-deutungskonstruktion (pp. 179–185). Frankfurt a.M. et al.: Lang.

Klein, Josef. (2002b). Topik und Frametheorie als argumentations- und begriffsgeschichtliche In-strumente, dargestellt am Kolonialdiskurs. In Dieter Cherubim (Ed.), Neue deutsche Sprach-geschichte: mentalitäts-, kultur- und sozialgeschichtliche Zusammenhänge (pp. 167–181). Berlin & New York: de Gruyter DOI: 10.1515/9783110903447.167

Klein, Josef, & Meißner, Iris. (1999). Wirtschaft im Kopf. Begriffskompetenz und Einstellungen junger Erwachsener bei Wirtschaftsthemen im Medienkontext. Frankfurt a.M.: Lang.

Knobloch, Clemens. (1994). Sprache und Sprechtätigkeit. Sprachpsychologische Konzepte. Tübingen: Niemeyer.

Koch, Peter. (1999). Frame and contiguity: On the cognitive basis of metonymy and certain types of word formation. In Klaus-Uwe Panther, & Günter Radden (Eds.), Metonymy in language and thought (pp. 139–167). Amsterdam: John Benjamins. DOI: 10.1075/hcp.4.09koc

Konerding, Klaus-Peter. (1993). Frames und lexikalisches Bedeutungswissen. Untersuchungen zur linguistischen Grundlegung einer Frametheorie und zu ihrer Anwendung in der Lexikographie. Tübingen: Niemeyer. DOI: 10.1515/9783111674926

Konerding, Klaus-Peter. (1996). Konzept, Bedeutung und sprachliche Handlung. Grundzüge einer methodisch fundierten Framesemantik für Sprechaktverben. In Edda Weigand, & Franz Hundsnurscher (Eds.), Lexical structures and language use. Proceedings of the In-ternational Conference on Lexicology and Lexical Semantics. Münster, September 13–15. Tübingen: Niemeyer.

http://dx.doi.org/10.1007/978-3-322-99391-5

http://dx.doi.org/10.1515/9783110895315

http://dx.doi.org/10.1007/978-3-663-08962-9_7

http://dx.doi.org/10.1016/0304-422X(80)90013-3

http://dx.doi.org/10.1037/0033-295X.95.2.163

http://dx.doi.org/10.1080/01638537809544425

http://dx.doi.org/10.1037/0033-295X.85.5.363

http://dx.doi.org/10.1515/9783110903447.167

http://dx.doi.org/10.1075/hcp.4.09koc

http://dx.doi.org/10.1515/9783111674926

References 407

Konerding, Klaus-Peter. (1997). Grundlagen einer linguistischen Schematheorie und ihr Einsatz in der Semantik. In Inge Pohl (Ed.), Methodologische Aspekte der Semantikforschung. Pro-ceedings of the international conference on “Methodologische Aspekte der Semantikforschung” at the Universität Koblenz-Landau/Abteilung Landau 1996 (pp. 57–84). Frankfurt a.M. et al.: Lang.

Konerding, Klaus-Peter. (1999). Merkmale, Bedeutungspostulate, Prototypyen, Schemata – As-pekte der Kompatibilität prominenter Modelle in der Semantik. In Inge Pohl (Ed.), In-terdisziplinarität und Methodenpluralismus in der Semantikforschung: Proceedings of the international conference on “Interdisziplinarität und Methodenpluralismus in der Semantik-forschung” at the Universität Koblenz-Landau/Abteilung Landau 1998 (pp. 19–41). Frank-furt a.M.: Lang.

Konerding, Klaus-Peter. (2005). Themen, Diskurse und soziale Topik. In Claudia Fraas, & Michael Klemm (Eds.), Mediendiskurse als Bausteine gesellschaftlicher Wissenskonstruk-tion (pp. 9–38). Frankfurt a.M.: Lang.

Kövecses, Zoltán. (2002). Metaphor: A practical introduction. Oxford: Oxford University Press.Krämer, Sybille. (2001). Sprache, Sprechakt, Kommunikation. Sprachtheoretische Positionen des

20. Jahrhunderts. Frankfurt a.M.: Suhrkamp.Kumashiro, Fumiko. (2000). Phonotactic interactions: A non-reductionist approach to phonology.

Ph.D. dissertation. University of California, San Diego.Lakoff, George. (1971). On generative semantics. In Danny D. Steinberg, & Leon A. Jakobovits

(Eds.), Semantics: An interdisciplinary reader (pp. 232–296). Cambridge: Cambridge Uni-versity Press.

Lakoff, George. (1977). Linguistic gestalts. In Woodford A. Beach, Samuel E. Fox, & Shulamith Philosoph (Eds.), Papers from the Thirteenth Regional Meeting Chicago Linguistic Society (pp. 236–286), April 14–16, 1977. Chicago: Chicago Linguistic Society.

Lakoff, George. (1983). Categories: An essay in cognitive linguistics. In The Linguistic Society of Korea (Ed.), Linguistics in the morning calm (pp. 139–194). Seoul: Hanshin Publishing Company.

Lakoff, George. (1987). Women, fire, and dangerous things: What categories reveal about the mind. Chicago: University of Chicago Press. DOI: 10.7208/chicago/9780226471013.001.0001

Lakoff, George. (1988). Cognitive semantics. In Umberto Eco, Marco Santambrogio, & Patrizia Violi (Eds.), Meaning and mental representations (pp. 119–154). Indiana: Indiana University Press.

Lakoff, George. (1990). The Invariance Hypothesis: Is abstract reason based on image-schemas? Cognitive Linguistics, 1(1), 39–74. DOI: 10.1515/cogl.1990.1.1.39

Lakoff, George. (1991). Cognitive versus generative linguistics: How commitments influence re-sults. Language and Communication, 11(1/2), 53–62. DOI: 10.1016/0271-5309(91)90018-Q

Lakoff, George. (2004). Don’t think of an Elephant! Know your values and frame the debate: The essential guide for progressives. Vermont: Chelsea Green.

Lakoff, George. (2005). Simple framing: An introduction to framing and its uses in politics. [Accessible at: http://www.rockridgeinstitute.org/projects/strategic/simple_framing/view, last accessed: 10 April 2014]

Lakoff, George, & Johnson, Mark. (1980). Metaphors we live by. Chicago: University of Chicago Press.

Lakoff, George, & Johnson, Mark. (1999). Philosophy in the flesh: Embodied mind and its challenge to western thought. New York: Basic Books.

http://dx.doi.org/10.7208/chicago/9780226471013.001.0001


http://dx.doi.org/10.1016/0271-5309(91)90018-Q


Lakoff, George, & Turner, Mark. (1989). More than cool reason: A field guide to poetic metaphor. Chicago: University of Chicago Press. DOI: 10.7208/chicago/9780226470986.001.0001

Lang, Ewald. (1985). Symmetrische Prädikate: Lexikoneintrag und Interpretationsspielraum. Eine Fallstudie zur Semantik der Personenstandslexik. Linguistische Studien, Reihe A, 75–112.

Lang, Ewald. (1988). Primary perceptual space and inherent proportion schema: Two interacting categorization grids underlying the conceptualization of spatial objects. Duisburg: L.A.U.D.

Lang, Ewald. (1991). A two-level approach to projective prepositions. In Gisa Rauh (Ed.), Approaches to prepositions (pp. 127–167). Tübingen: Narr.

Lang, Ewald. (1994). Semantische vs. konzeptuelle Struktur: Unterscheidung und Überschnei-dung. In Monika Schwarz (Ed.), Kognitive Semantik: Ergebnisse, Probleme, Perspektiven (pp. 25–40). Tübingen: Narr.

Langacker, Ronald W. (1979). Grammar as image. Linguistic Notes from La Jolla, 6, 87–126.Langacker, Ronald W. (1984). Active zones. Berkeley Linguistics Society, 10, 172–188.Langacker, Ronald W. (1987). Foundations of cognitive grammar. Volume 1: Theoretical prereq-

uisites. Stanford: Stanford University Press.Langacker, Ronald W. (1988a). An overview of cognitive grammar. In Brygida Rudzka-Ostyn

(Ed.), Topics in cognitive linguistics (pp. 3–48). Amsterdam: John Benjamins. DOI: 10.1075/cilt.50.03lanLangacker, Ronald W. (1988b). A view of linguistic semantics. In Brygida Rudzka-Ostyn (Ed.),

Topics in cognitive linguistics (pp. 49–90). Amsterdam: John Benjamins. DOI: 10.1075/cilt.50.04lanLangacker, Ronald W. (1988c). A usage-based model. In Brygida Rudzka-Ostyn (Ed.), Topics in

cognitive linguistics (pp. 127–161). Amsterdam: John Benjamins. DOI: 10.1075/cilt.50.06lanLangacker, Ronald W. (1991a). Foundations of cognitive grammar. Volume 2: Descriptive applica-

tion. Stanford: Stanford University Press.Langacker, Ronald W. (1991b). Concept, image, and symbol: The cognitive basis of grammar.

Berlin & New York: Mouton de Gruyter. DOI: 10.1515/9783110857733Langacker, Ronald W. (1994). Culture, cognition, and grammar. In Martin Pütz (Ed.), Language

contact: Language conflict (pp. 25–53). Amsterdam & New York: John Benjamins.Langacker, Ronald W. (1997). The contextual basis of cognitive semantics. In Jan Nuyts, & Eric

Pederson (Ed.), Language and conceptualization (pp. 229–252). Oxford: Oxford University Press. DOI: 10.1017/CBO9781139086677.010

Langacker, Ronald W. (1999a). Assessing the cognitive linguistic enterprise. In Theo Janssen, & Gisela Redeker (Eds.), Cognitive linguistics: Foundations, scope, and methodology (pp. 13–60). Berlin & New York: Mouton de Gruyter. DOI: 10.1515/9783110803464.13

Langacker, Ronald W. (1999b). Grammar and conceptualization. Berlin: Mouton de Gruyter. DOI: 10.1515/9783110800524

Langacker, Ronald W. (2000). A dynamic usage-based model. In Michael Barlow, & Suzanne Kemmer (Eds.), Usage-based models of language (pp. 24–63). Stanford: CSLI.

Langacker, Ronald W. (2005). Construction grammars: Cognitive, radical, and less so. In Francisco J. Ruiz de Mendoza Ibáñez, & M. Sandra Peña Cervel (Eds.), Cognitive linguis-tics: Internal dynamics and interdisciplinary interaction (pp. 101–159). Berlin & New York: Mouton de Gruyter.

Langacker, Ronald W. (2006). On the continuous debate about discreteness. Cognitive Linguistics, 17(1), 107–151. DOI: 10.1515/COG.2006.003

http://dx.doi.org/10.7208/chicago/9780226470986.001.0001

http://dx.doi.org/10.1075/cilt.50.03lan



http://dx.doi.org/10.1515/9783110857733

http://dx.doi.org/10.1017/CBO9781139086677.010

http://dx.doi.org/10.1515/9783110803464.13

http://dx.doi.org/10.1515/9783110800524

http://dx.doi.org/10.1515/COG.2006.003

References 409

Langlotz, Andreas. (2006). Idiomatic creativity: A cognitive-linguistic model of idiom-represen-tation and idiom-variation in English. Amsterdam: John Benjamins. DOI: 10.1075/hcp.17

Leeuwen, van Theo. (2005). Introducing social semiotics: An introductory textbook. London: Taylor & Francis.

Liebert, Wolf-Andreas. (1992). Metaphernbereiche in der deutschen Alltagssprache. Kognitive Linguistik und die Perspektiven einer Kognitiven Lexikographie. Frankfurt a.M. et al.: Lang.

Lieven, Elena, Behrens, Heike, Spears, Jenny, & Tomasello, Michael. (2003). Early syntactic creativity: A usage based approach. Journal of Child Language, 30, 333–370.

DOI: 10.1017/S0305000903005592Link, Jürgen. (1999). Diskursive Ereignisse, Diskurse, Interdiskurse: Sieben Thesen zur Ope-

rationalisierung der Diskursanalyse, am Beispiel des Normalismus. In Hannelore Bublitz, Andrea D. Bührmann, Christine Hanke, & Andrea Seier (Eds.), Das Wuchern der Dis-kurse. Perspektiven der Diskursanalyse Foucaults (pp. 148–161). Frankfurt a.M. & New York: Campus.

Löbach, Brigitte. (2000). Semantikerwerb. Ein Beitrag zu einer empirisch-naturalistischen Bedeu-tungstheorie. Tübingen: Niemeyer. DOI: 10.1515/9783110929898

Lönneker, Birte. (2003a). Konzeptframes und Relationen. Extraktion, Annotation und Analyse französischer Corpora aus dem World Wide Web. Berlin: Aka.

Lönneker, Birte. (2003b). Semantic frame elements in annotated concept frames. In Eva Haijcová, A. Kotescová, & Jiri Mírovkský (Eds.), Proceedings of the XVII International Congress of Linguists (CIL17), Prague: Matfyzpress, MFF, UK. CD-ROM. [Accessible at: www1.uni-hamburg.de/story-generators//Birte/Semantic_Frame_Elements.pdf; last accessed: 10 April 2014]

Lönneker, Birte. (2005). Weltwissen in Textannotationen mit Konzeptframes: Modell, Meth-ode, Resultate. In Sabine Braun, & Kurt Kohn (Eds.), Sprache (n) in der Wissensgesells-chaft. Proceedings of the 34th annual meeting of the Gesellschaft für Angewandte Linguistik (pp. 127–142). Frankfurt a.M. et al.: Lang.

Lutzeier, Peter Rolf. (1985). Linguistische Semantik. Stuttgart: Metzler.Lutzeier, Peter Rolf. (1991). Major pillars of German syntax: An introduction to CRMS-theory.

Tübingen: Niemeyer. DOI: 10.1515/9783111356532Lutzeier, Peter Rolf. (1992). Wortfelder als kognitive Orientierungspunkte? In Peter Rolf Lutzeier

(Ed.), Studien zur Wortfeldtheorie (pp. 102–214). Tübingen: Niemeyer.MacWhinney, Brian. (2001). Emergentist approaches to language. In Joan L. Bybee, & Paul

Hopper (Eds.), Frequency and the emergence of linguistic structure (pp. 449–468). Amster-dam, et al.: John Benjamins. DOI: 10.1075/tsl.45.23mac

MacWhinney, Brian (Ed.) (1999). The emergence of language. Mahwah: Erlbaum.Makkai, Adam. (1971). Degrees of nonsense, or transformation, stratification, and the contextual

adaptability principle. Chicago Linguistic Society, 7, 479–492Mandelblit, Nili. (2000). The grammatical marking of conceptual integration: From syntax to

morphology. Cognitive Linguistics, 11, 197–251.Mandl, Heinz, Friedrich, Helmut F., & Hron, Aemilian. (1987). Theoretische Ansätze zum Wis-

senserwerb. Deutsches Institut für Fernstudien an der Universität Tübingen, Research Re-port no. 41.

Mandler, Jean M. (1984). Stories, scripts, and scenes: Aspects of schema theory. Hillsdale, New Jersey: Erlbaum.

Mandler, Jean M. (2004). The foundations of mind: Origins of conceptual thought. Oxford, et al.: Oxford University Press.

http://dx.doi.org/10.1075/hcp.17

http://dx.doi.org/10.1017/S0305000903005592

http://dx.doi.org/10.1515/9783110929898

http://dx.doi.org/10.1515/9783111356532

http://dx.doi.org/10.1075/tsl.45.23mac


Mandler, Jean M. (2005). How to build a baby: III. Image schemas and the transition to verbal thought. In Beate Hampe (Ed.), From perception to meaning: Image schemas in cognitive linguistics (pp. 137–163). Berlin & New York: Mouton de Gruyter.

DOI: 10.1515/9783110197532.2.137Mangasser-Wahl, Martina. (2000). Von der Prototypentheorie zur empirischen Semantik: darges-

tellt am Beispiel von Frauenkategorisierungen. Frankfurt a.M. et al.: Lang.Martin, Willy. (1997). A frame-based approach to polysemy. In Hubert Cuyckens, & Britta

Zawada (Eds.), Polysemy in cognitive linguistics. Selected Papers from the Fifth International Cognitive Linguistics Conference (pp. 56–81). Amsterdam: John Benjamins.

Martinez, Joe L., Barea-Rodriguez, Edwin J., & Derrick, Brian E. (1998). Long-term potentia-tion, long-term depression, and learning. In Joe L. Martinez, & Raymond P. Kesner (Eds.), Neurobiology of learning and memory (pp. 211–246). San Diego, et al.: Academic Press.

DOI: 10.1016/B978-012475655-7/50007-1McCawley, James D. (1968). Concerning the base component of transformational grammar.

Foundations of Language, 4, 243–269.McKoon, Gail, & Ratcliff, Roger. (1986). Inferences about predictable events. Journal of Experi-

mental Psychology: Learning, Memory, and Cognition, 12, 82–91. DOI: 10.1037/0278-7393.12.1.82McKoon, Gail, & Ratcliff, Roger. (1992). Inferences during reading. Psychological Review, 49,

25–42. DOI: 10.1146/annurev.psych.49.1.25Meier, Stefan. (2005). Zeichenlesen im Netzdiskurs – Überlegungen zu einer semiotischen Dis-

kursanalyse multimedialer Kommunikation. In Claudia Fraas, & Michael Klemm (Eds.), Mediendiskurse (pp. 123–141). Frankfurt a. M. et al.: Lang.

Meier, Stefan. (2010). Bild und Frame – Eine diskursanalytische Perspektive auf visuelle Kommunikation und deren methodische Operationalisierung. In Anna Duszak, Juliane House, & Łukasz Kumięga (Eds.), Globalization, discourse, media: In a critical perspec-tive/Globalisierung, Diskurse, Medien: eine kritische Perspektive (pp. 371–392). Warszawa: Wydawnictwa Uniwersytetu Warszawskiego.

Metzger, Wolfgang. (1923). Figural-Wahrnehmung. In Wolfgang Metzger (Ed.), Handbuch der Psychologie, Vol. 1: Allgemeine Psychologie. Der Aufbau des Erkennens (pp. 693–744). Göttingen: Verlag für Psychologie.

Metzger, Wolfgang. (1986). Die Entdeckung der Prägnanztheorie. In Gestalt-Psychologie: Aus-gewählte Werke aus den Jahren 1950–1982 (pp. 145–181). Frankfurt a.M.: Kramer.

Metzing, Dieter (Ed.). (1979). Frame conception and text understanding. Berlin, New York: de Gruyter.

Meyer, Ralf. (1994a). Konstruktion von Komposita-Repräsentationen in einem modifizierten DRT-Ansatz. In Monika Schwarz (Ed.), Kognitive Semantik: Ergebnisse, Probleme, Perspek-tiven (pp. 61–79). Tübingen: Narr.

Meyer, Ralf. (1994b). Probleme von Zwei-Ebenen-Semantiken. Kognitionswissenschaft, 4, 32–46.Minsky, Marvin. (1975). A framework for representing knowledge. In Patrick H. Winston (Ed.),

The psychology of computer vision (pp. 211–277). New York: McGraw-Hill. [Extract printed in: Dieter Metzing (Ed.) (1980), Frame Conceptions and Text Understanding. Berlin & New York: de Gruyter, 1–25. Germ. in: Dieter Metzing (Ed.) (1992). Kognitionswissenschaft. Grundlagen, Probleme, Perspektiven (pp. 92–133). Frankfurt a.M.: Suhrkamp.]

Minsky, Marvin. (1977). Frame-system theory. In Philip Johnson-Laird, & Peter C. Wason (Eds.), Thinking: Readings in cognitive science (pp. 355–376). Cambridge: University Press.

Minsky, Marvin. (1988). The society of mind. New York, et al.: Touchstone. [Germ.: Minsky, Marvin. (1990). Mentopolis. Stuttgart: Klett-Cotta.]

http://dx.doi.org/10.1515/9783110197532.2.137

http://dx.doi.org/10.1016/B978-012475655-7/50007-1

http://dx.doi.org/10.1037/0278-7393.12.1.82

http://dx.doi.org/10.1146/annurev.psych.49.1.25

References 411

Minston, Patrick (Ed.). (1975). The psychology of computer vision. New York: McGraw-Hill.Müller, Klaus. (1982). Frame-Struktur und Alltagskommunikation. Zur Wechselbeziehung zwi-

schen kognitiven Prozessen und pragmatischem Verstehen. Grazer Linguistische Studien, 16, 105–130.

Murphy, Gregory L. & Medin, Douglas L. (1985). The role of theories in conceptual coherence. Psychological Review, 92, 289–316. DOI: 10.1037/0033-295X.92.3.289

Müske, Eberhard. (1991). Frame-Struktur im narrativen Diskurs. Anmerkungen zu einem dy-namischen Modell des literarisch-künstlerischen narrativen Diskurses. Hallesche Studien zur Wirkung von Sprache und Literatur, 19, 4–16.

Müske, Eberhard. (1992). Diskurssemiotik. Zur funktionellen Integration des Frame-Konzepts in ein dynamisches Modell literarisch-künstlerischer Texte. Stuttgart: Akademie-Verlag.

Neisser, Ulric. (1967). Cognitive psychology. New York: Meredith.Neisser, Ulric. (1976). Cognition and reality: Principles and implications of cognitive psychology.

San Francisco: Freeman.Newell, Allen, & Rosenbloom, Paul S. (1981). Mechanisms of skill acquisition and the law of

practice. In John R. Anderson (Ed.), Cognitive skills and their acquisition (pp. 1–51). Hills-dale, New Jersey: Erlbaum.

Newmeyer, Frederick J. (1986). Linguistic theory in America. San Diego: Academic Press.O’Connor, Mary C. (1994). The marking of possession in Northern Pomo: Private opposition

and pragmatic reference. In Chris Johnson et al. (Eds.), Proceedings of the Twentieth Meet-ing of the Berkeley Linguistics Society (pp. 387–401). Berkeley: Berkeley Linguistics Society.

Oliver, Pamela E., & Johnston, Hank. (2000). What a good idea! Frames and ideologies in social movement research. Mobilization, 5, 50–65.

Olson, David R. (1974). Sprache und Denken. Aspekte einer kognitiven Theorie der Semantik. In Helen Leuninger et al. (Eds.), Linguistik und Psychologie. Ein Reader. Vol. 1: Psycholin-guistische Untersuchungen sprachlicher Performanz (pp. 179–207). Frankfurt a. M.: Fischer.

Palmer, Stephen E. (1978). Fundamental aspects of cognitive representation. In Eleanor Rosch, & Barbara B. Lloyd (Eds.), Cognition and categorization (pp. 259–303). New York & Hills-dale: Erlbaum.

Peeters, Bert. (2000). Setting the scene: Some recent milestones in the lexicon-encyclopedia debate. In Bert Peeters (Ed.), The lexicon-encyclopedia interface (1–52). Oxford: Elsevier Science.

Peirce, Charles S. (1985). Über die Klarheit unseres Gedankens. Meiner: Hamburg.Peirce, Charles S. (1993). Phänomen und Logik der Zeichen. Frankfurt a.M.: Suhrkamp.Petruck, Mirjam R.L. (1986). Body part terminology in Hebrew: A study in lexical semantics.

Ph.D. dissertation, University of California, Berkeley.Petruck, Mirjam R.L. (1996). Frame semantics. In Jef Verschueren, Jan-Ola Östman, Jan

Blommaert, & Chris Bulcaen (Eds.), Handbook of pragmatics (pp. 1–13). Amsterdam: John Benjamins.

Petruck, Mirjam R.L. (1995). Frame semantics and the lexicon: Nouns and verbs in the body frame. In Masayoshi Shibatani, & Sandra Tompson (Eds.), Essays in semantics and pragmat-ics: In honor of Charles Fillmore (pp. 279–297). Amsterdam: John Benjamins.

Petruck, Mirjam R.L., & Boas, Hans C. (2003). All in a day’s week. In Eva Hajicova, A. Kotesovcova, & Jiri Mirovsky (Eds.), Proceedings of CIL 17. CD-ROM. Prague: Matfyzpress. [Accessible at: http://framenet.icsi.berkeley.edu/papers/weekday.pdf; last accessed: 10 April 2014]

Piaget, Jean. (1983). Dialogue on the psychology of thought. In Robert W. Rieber (Ed.), Dialogues on the psychology of language and thought (pp. 107–125). New York: Plenum Press.

Pielenz, Michael. (1993). Metapher and argumentation. Tübingen: Narr.

http://dx.doi.org/10.1037/0033-295X.92.3.289


Pohl, Inge. (2002). Inferenzkonzept und Bedeutungskonstruktion. In Inge Pohl (Ed.), Prozesse der Bedeutungskonstruktion (pp. 17–37). Frankfurt a.M.: Lang.

Pohl, Inge (Ed.). (1997). Methodologische Aspekte der Semantikforschung: Beiträge der Konfe-renz “Methodologische Aspekte der Semantikforschung“ an der Universität Koblenz-Landau/Abteilung Landau. (1996). Frankfurt a. M. u.a.: Lang.

Polenz, Peter von. (1985). Deutsche Satzsemantik. Grundbegriffe des Zwischen-den-Zeilen-Lesens. Berlin & New York: de Gruyter.

Pollard, Carl, & Sag, Ivan A. (1994). Head-driven phrase structure grammar. Chicago: University of Chicago Press.

Pörings, Ralf, & Schmitz, Ulrich (Eds.). (2003). Sprache und Sprachwissenschaft: Eine kognitiv orientierte Einführung. Tübingen: Narr.

Post, Michal. (1988). Scenes-and-frames semantics as a neo-lexical field theory. In Werner Hüllen, & Rainer Schulze (Eds.), Understanding the lexicon: Meaning, sense and world knowledge in lexical semantics (pp. 36–47). Tübingen: Niemeyer.

Putnam, Hilary. (1975). The meaning of ‘meaning’. In Keith Gunderson (Ed.), Language, mind and knowledge. Minnesota Studies in the Philosophy of Science (Vol. 7, pp. 131–193). Minneapolis: University of Minnesota Press.

Putnam, Hilary. (1988). Vernunft, Wahrheit und Geschichte. Frankfurt a.M.: Suhrkamp.Pütz, Martin, & Dirven, René (Eds.). (1996). The construal of space in language and thought.

Berlin & New York: Mouton de Gruyter. DOI: 10.1515/9783110821611Quine, Williard van Orman. (1960). Word and object. Cambridge: MIT Press.Quine, Williard van Orman. (1979). Zwei Dogmen des Empirismus. In Williard van Orman

Quine (Ed.), Von einem logischen Standpunkt. Neun logisch-philosophische Essays (pp. 27–50). Frankfurt a.M., Berlin, Vienna: Ullstein.

Radden, Günter. (2000). How metonymic are metaphors? In Antonio Barcelona (Ed.), Metaphor and metonymy at the crossroads: A cognitive perspective (pp. 93–108). Berlin & New York: Mouton de Gruyter.

Raskin, Victor (Ed.). (1985). Round table discussion on frame/script semantics. Quaderni di Semantica, 6(2).

Ratcliff, Roger A., & McKoon, Gail. (1981). Priming in item recognition: Evidence for the propo-sitional structure of sentences. Journal of Verbal Learning and Verbal Behavior, 17, 403–417. DOI: 10.1016/S0022-5371(78)90238-4

Reischer, Jürgen. (2002). Die Sprache – Ein Phänomen und seine Forschung. Berlin & New York: de Gruyter. DOI: 10.1515/9783110896183

Reisigl, Martin. (1999). Sekundäre Interjektionen. Frankfurt a.M.: Lang.Revlin, Russell, & Hegarty, Mary. (1999). Resolving signals to cohesion: Two models of bridging

inference. Discourse Processes, 27, 77–102. DOI: 10.1080/01638539909545051Rice, Sally. (1996). Prepositional prototypes. In Martin Pütz, & René Dirven (Eds.), The construal

of space in language and thought (pp. 135–165). Berlin & New York: Mouton de Gruyter.Rickheit, Gert, Sichelschmidt, Lorenz, & Strohner, Hans. (2004). Psycholinguistik. Tübingen:

Stauffenburg.Rickheit, Gert, & Strohner, Hans. (1990). Inferenzen. Basis des Sprachverstehens. Die Neueren

Sprachen, 89, 532–545.Rickheit, Gert, & Strohner, Hans. (1993). Grundlagen der kognitiven Sprachverarbeitung. Tübingen

& Basel: Francke.

http://dx.doi.org/10.1515/9783110821611

http://dx.doi.org/10.1016/S0022-5371(78)90238-4

http://dx.doi.org/10.1515/9783110896183

http://dx.doi.org/10.1080/01638539909545051

References 413

Rickheit, Gert, & Strohner, Hans. (1999). Textverarbeitung: Von der Proposition zur Situa-tion. In Angela D. Friederici (Ed.), Enzyklopädie der Psychologie. Vol. 2: Sprachrezeption, (pp. 271–306). Göttingen, et al.: Hogrefe.

Rickheit, Gert, & Strohner, Hans. (2003). Inferenzen. In Gert Rickheit, Theo Herrmann, & Werner Deutsch (Eds.), Psycholinguistik: Ein internationales Handbuch (pp. 566–577). Ber-lin: de Gruyter.

Rickheit, Gert, & Wachsmuth, Ipke (Eds.). (2006). Situated communication. Berlin: de Gruyter. DOI: 10.1515/9783110197747

Ricœur, Paul. (1986). Die lebendige Metapher. Munich: Fink.Rips, Lance J. (1989). Similarity, typicality, and categorization. In Stella Vosniadou, & Andrew

Ortony, (Eds.), Similarity and analogical reasoning (pp. 21–59). Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511529863.004

Ritter, Nancy. (2005). On the status of linguistics as a cognitive science. The Linguistic Review, 22, 117–134. DOI: 10.1515/tlir.2005.22.2-4.117

Rohrer, Tim. (2005). Image schemata in the brain. In Beate Hampe (Ed.), From perception to meaning: Image schemas in cognitive linguistics (pp. 165–196). Berlin & New York: Mouton de Gruyter. DOI: 10.1515/9783110197532.2.165

Roos, Eckhard. (2001). Idiom und Idiomatik: Ein sprachliches Phänomen im Lichte der kognitiven Linguistik und Gestalttheorie. Aachen: Shaker.

Rorty, Richard. (1987). Der Spiegel der Natur: eine Kritik der Philosophie. Frankfurt a.M.: Suhrkamp.

Rosch, Eleanor. (1977). Human categorization. In Neil Warren (Ed.), Studies in cross-cultural psychology (pp. 1–49). London, New York, San Francisco: Academic Press.

Rosch, Eleanor. (1978). Principles of categorization. In Eleanor Rosch, & Barbara B. Lloyd (Eds.), Cognition and categorization (pp. 27–48). Hillsdale & New York: Erlbaum.

Rosch, Eleanor, Mervis, Carolyn B., Gray, Wayne D., Johnson, David M., & Boyes-Braem, Penny. (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382–439.

DOI: 10.1016/0010-0285(76)90013-X Rudzka-Ostyn, Brygida (Ed.). (1988). Topics in cognitive linguistics. Amsterdam: John Benjamins.

DOI: 10.1075/cilt.50Rumelhart, David E. (1975). Notes on a schema for stories. In Daniel G. Bobrow, & Allan Collins

(Eds.), Representation and understanding. Studies in Cognitive Science (pp. 211–236). New York: Academic Press. DOI: 10.1016/B978-0-12-108550-6.50013-6

Rumelhart, David E. (1980a). Schemata: The building blocks of cognition. In Rand J. Spiro, et al. (Eds.), Theoretical issues in reading comprehension. Perspectives from cognitive psychology, linguistics, artificial intelligence, and education (pp. 33–58). Hillsdale, New Jersey: Erlbaum.

Rumelhart, David E. (1980b). On evaluating story grammars. Cognitive Science, 4(3), 313–316. DOI: 10.1207/s15516709cog0403_5

Rumelhart, David E., & Norman, Donald A. (1978). Accretion, tuning, and restructuring: Three modes of learning. In John W. Cotton, & Roberta L. Klatzky (Eds.), Semantic factors in cognition (pp. 37–53). Hillsdale, New Jersey: Erlbaum.

Rumelhart, David E., & Ortony, Andrew. (1977). The representation of knowledge in memory. In Richard C. Anderson, Rand J. Spiro, & William E. Montague (Eds.), Schooling and the acquisition of knowledge (pp. 99–135). Hillsdale, New Jersey: Erlbaum.

Sandra, Dominiek. (1998). What linguists can and can’t tell you about the human mind: A reply to croft. Cognitive Linguistics, 9(4), 361–378. DOI: 10.1515/cogl.1998.9.4.361

http://dx.doi.org/10.1515/9783110197747

http://dx.doi.org/10.1017/CBO9780511529863.004

http://dx.doi.org/10.1515/tlir.2005.22.2-4.117

http://dx.doi.org/10.1515/9783110197532.2.165

http://dx.doi.org/10.1016/0010-0285(76)90013-X

http://dx.doi.org/10.1075/cilt.50

http://dx.doi.org/10.1016/B978-0-12-108550-6.50013-6




Sandra, Dominiek, & Rice, Sally. (1995). Network analysis of prepositional meaning: Mirroring whose mind – the linguist’s or the language user’s? Cognitive Linguistics, 6, 89–130.

DOI: 10.1515/cogl.1995.6.1.89Sanford, Anthony J., & Garrod, Simon C. (1981). Understanding written language: Explorations

in comprehension beyond the sentence. Chichester, et al.: Wiley.Saussure, Ferdinand de. (2001). Grundfragen der allgemeinen Sprachwissenschaft. Edited by Charles

Bally, & Albert Sechehaye. Berlin & New York: de Gruyter. DOI: 10.1515/9783110870183Schank, Roger C. (1999). Dynamic memory revisited. Cambridge: Cambridge University Press.

DOI: 10.1017/CBO9780511527920Schank, Roger C., & Abelson, Robert P. (1977). Scripts, plans, goals and understanding: An inquiry

into human knowledge systems. Hillsdale, New Jersey: Erlbaum.Scharloth, Joachim. (2005a). Sprachnormen und Mentalitäten. Sprachbewusstseinsgeschichte in

Deutschland im Zeitraum von 1766–1785. Niemeyer: Tübingen. DOI: 10.1515/9783110910247Scharloth, Joachim. (2005b). Die Semantik der Kulturen. Diskurssemantische Grundfiguren als

Kategorien einer linguistischen Kulturanalyse. In Dietrich Busse, Thomas Niehr, & Martin Wengeler (Eds.), Brisante Semantik. Neuere Konzepte und Forschungsergebnisse einer kul-turwissenschaftlichen Linguistik (pp. 119–135). Tübingen: Niemeyer.

Scherner, Maximilian. (1975). ‘Sprachkenntnis’ oder ‘Weltkenntnis’. Zum Problem des Textver-stehens aus linguistischer Sicht. Der Deutschunterricht, 27(6), 59–79.

Scherner, Maximilian. (1979). Nichttextualisierte Verstehensvoraussetzungen als sprachwis-senschaftliches Problem. In Edeltraud Bülow, & Peter Schnitter (Eds.), Integrale Linguistik. Festschrift für Helmut Gipper (pp. 329–357). Amsterdam: John Benjamins.

Scherner, Maximilian. (1984). Sprache als Text. Ansätze zu einer sprachwissenschaftlich begrün-deten Theorie des Textverstehens. Forschungsgeschichte – Problemstellung – Beschreibung. Tübingen: Niemeyer.

Scherner, Maximilian. (1994). Textverstehen als ‘Spurenlesen’. Zur texttheoretischen Tragweite dieser Metapher. In Peter Canisius, Clemens-Peter Herbermann, & Gerhard Tschauder (Eds.), Text und Grammatik. Festschrift für Roland Harweg zum 60. Geburtstag (pp. 316–340). Bochum: Brockmeyer.

Schlobinski, Peter. (2003). Grammatikmodelle. Positionen und Perspektiven. Wiesbaden: West-deutscher Verlag.

Schmid, Hans-Jörg. (1996). Basic level categories as basic cognitive and linguistic building blocks. In Edda Weigand, & Franz Hundsnurscher (Eds.), Lexical structures and language use. Pro-ceedings of the International Conference on Lexicology and Lexical Semantics (pp. 285–295). Münster, September 13–15. Tübingen: Niemeyer.

Schmidt, Siegfried J. (1973). Texttheorie. Probleme einer Linguistik sprachlicher Kommunikation. Munich: UTB.

Schmidt, Siegfried J. (1980). Grundriss der Empirischen Literaturwissenschaft. Vol. 1. Gesellschaft-licher Handlungsbereich Literatur. Braunschweig: Vieweg.

Schnelle, Helmut. (1970). Zur Entwicklung der theoretischen Linguistik. Studium Generale, 23, 1–28.

Schnotz, Wolfgang. (1994). Aufbau von Wissensstrukturen. Untersuchungen zur Kohärenzbildung bei Wissenserwerb mit Texten. Weinheim: Beltz.

Scholz, Oliver. (1999). Verstehen und Rationalität. Frankfurt: Klostermann.Schulze, Rainer. (1991). Getting round to (a)round: Towards the description and analysis of a “spa-

tial” predicate. In Gisa Rauh (Ed.), Approaches to Prepositions (pp. 253–274). Tübingen: Narr.


http://dx.doi.org/10.1515/9783110870183


http://dx.doi.org/10.1515/9783110910247

References 415

Schwarz, David S. (1979). Naming and referring: The semantics and pragmatics of singular terms. Berlin & New York: de Gruyter. DOI: 10.1515/9783110851328

Schwarz, Monika. (1992a). Kognitive Semantiktheorie und neuropsychologische Realität: repräsen-tationale und prozedurale Aspekte der semantischen Kompetenz. Tübingen: Niemeyer.

DOI: 10.1515/9783111353883Schwarz, Monika. (1992b). Einführung in die kognitive Linguistik. Tübingen & Basel: Francke.Schwarz, Monika. (1994). Kognitive Semantik – State of the Art und Quo vadis? In Monika

Schwarz (Ed.), Kognitive Semantik: Ergebnisse, Probleme, Perspektiven (pp. 9–21). Tübingen: Narr.

Schwarz, Monika. (2000). Indirekte Anaphern in Texten. Prinzipien der domänengebundenen Referenz im Deutschen. Tübingen: Niemeyer. DOI: 10.1515/9783110912517

Scollon, Ron, & Scollon, Suzie. (2003). Discourse in place: Language in the material world. London & New York: Routledge. DOI: 10.4324/9780203422724

Searle, John R. (1969). Speech acts: An essay in the philosophy of language. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9781139173438

Searle, John R. (1979). Sprechakte. Ein sprachphilosophischer Essay. Frankfurt a. M.: Suhrkamp.Searle, John R. (1992). The rediscovery of mind. Cambridge: MIT Press.Searle, John R. (1998). Wörtliche Bedeutung. In John R. Searle (Ed.), Ausdruck und Bedeutung

(pp. 139–159). Frankfurt a. M.: Suhrkamp.Sellars, Wilfrid. (1999). Empirismus und die Philosophie des Geistes. Paderborn: Mentis.Seuren, Pieter A.M. (1972). Autonomous versus semantic syntax. Foundations of Language, 8,

237–265.Singer, Murray, & Ferreira, Fernanda. (1983). Inferring consequences in story comprehension.

Journal of Verbal Learning and Verbal Behavior, 22, 437–448. DOI: 10.1016/S0022-5371(83)90282-7Sinha, Chris. (1999). Grounding, mapping and acts of meaning. In Theo Janssen, & Gisela

Redeker (Eds.), Cognitive linguistics: Foundations, scope, and methodology (pp. 223–255). Berlin & New York: Mouton de Gruyter. DOI: 10.1515/9783110803464.223

Sinha, Chris. (2004). Biology, culture and the emergence and elaboration of symbolization. In Anjum P. Saleemi, Albert Gjedde, & Ockey-Schewn Bohn (Eds.), In Search of a language for the mind-brain: Can the multiple perspectives be unified? (pp. 311–335). Aarhus: Aarhus University Press.

Smith, Edward E., & Swinney, David A. (1992). The role of schemas in reading text: a real-time examination. Discourse Processes, 15, 303–316. DOI: 10.1080/01638539209544814

Sperber, Dan, & Wilson, Deirdre. (1986). Relevance: Communication and cognition. Oxford: Blackwell.

Spitzer, Manfred. (1996). Geist im Netz. Modelle für Lernen, Denken und Handeln. Heidelberg, Berlin, Oxford: Spektrum Akademischer Verlag.

Squire, Larry R. (2004). Memory systems of the brain: A brief history and current perspective. Neurobiology of Learning and Memory, 82, 171–177. DOI: 10.1016/j.nlm.2004.06.005

Steinseifer, Martin. (2005). Die ‘Sloterdijk-Debatte’. Linguistische Diskursanalyse eines (Print-)Medienereignisses und seiner argumentativen Ordnung. In Martin Wengeler (Ed.), Sprach-geschichte als Zeitgeschichte. Konzepte, Methoden und Forschungsergebnisse der Düsseldorfer Sprachgeschichtsschreibung für die Zeit nach 1945 (pp. 498–526). Hildesheim & New York: Olms.

Stöckl, Hartmut. (2004). Die Sprache im Bild – Das Bild in der Sprache. Zur Verknüpfung von Sprache und Bild im massenmedialen Text. Konzepte – Theorien – Analysemethoden. Berlin & New York: de Gruyter. DOI: 10.1515/9783110201994

http://dx.doi.org/10.1515/9783110851328

http://dx.doi.org/10.1515/9783111353883

http://dx.doi.org/10.1515/9783110912517

http://dx.doi.org/10.4324/9780203422724


http://dx.doi.org/10.1016/S0022-5371(83)90282-7

http://dx.doi.org/10.1515/9783110803464.223

http://dx.doi.org/10.1080/01638539209544814

http://dx.doi.org/10.1016/j.nlm.2004.06.005

http://dx.doi.org/10.1515/9783110201994


Stötzel, Georg. (1980). Konkurrierender Sprachgebrauch in der deutschen Presse. Sprachwis-senschaftliche Textinterpretation zum Verhältnis von Sprachbewusstsein und Gegenstand-skonstitution. Wirkendes Wort, 30, 39–53.

Stötzel, Georg. (1995). Einleitung. In Georg Stötzel, & Martin Wengeler (Eds.), Kontroverse Begriffe. Geschichte des öffentlichen Sprachgebrauchs in der Bundesrepublik Deutschland, (pp. 1–17). Berlin & New York: de Gruyter. DOI: 10.1515/9783110881660.1

Strohner, Hans. (1995). Kognitive Systeme. Eine Einführung in die Kognitionswissenschaft. Opladen: Westdeutscher Verlag.

Strohner, Hans. (2006). Textverstehen aus psycholinguistischer Sicht. In Hardarik Blühdorn, Eva Breindl, & Ulrich H. Waßner (Eds.), Text – Verstehen. Grammatik und darüber hinaus (pp. 187–204). Berlin & New York: de Gruyter.

Strube, Gerhard et al. (Eds.) (1996). Wörterbuch der Kognitionswissenschaft. Stuttgart: Klett-Cotta.Sucharowski, Wolfgang. (1996). Sprache und Kognition. Opladen: Westdeutscher Verlag. DOI: 10.1007/978-3-322-95664-4Sweetser, Eve. (1990). From etymology to pragmatics: Metaphorical and cultural aspects of seman-

tic structure. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511620904Sweetser, Eve. (1999). Compositionality and blending: Semantic composition in a cognitively

realistic framework. In Theo Janssen, & Gisela Redeker (Eds.), Cognitive linguistics: Foun-dations, scope, and methodology (pp. 129–162). Berlin & New York: Mouton de Gruyter.

DOI: 10.1515/9783110803464.129Sweetser, Eve. (2000). Blended spaces and performativity. Cognitive Linguistics, 11, 305–333.Tabossi, Patricia. (1988). Effects of context on the immediate interpretation of unambiguous

words. The Quarterly Journal of Experimental Psychology, 14, 153–162.Talmy, Leonard. (1988). Force dynamics in language and cognition. Cognitive Science, 12, 49–

100. DOI: 10.1207/s15516709cog1201_2Talmy, Leonard. (2000). Toward a cognitive semantics. Vol. I: Conceptual structuring systems.

Cambridge: MIT Press.Tannen, Deborah. (1979). What’s in a frame? Surface evidence for underlying expectations. In

Roy Freedle (Ed.), Discourse production and comprehension (pp. 137–181). Norwood: Ablex.Taylor, John R. (1994). The two-level approach to meaning. Linguistische Berichte, 149, 3–26.Taylor, John R. (1995). Approaches to word meaning: The network model (Langacker) and the

two-level model (Bierwisch). In René Dirven, Johan Vanparys (Eds.), Current approaches to the lexicon (pp. 3–26). Frankfurt a.M. et al.: Lang.

Taylor, John R. (1996). On running and jogging. Cognitive Linguistics, 7(1), 21–34. DOI: 10.1515/cogl.1996.7.1.21Taylor, John R. (2002). Cognitive grammar. New York: Oxford University Press.Taylor, John R. (2003). Linguistic categorization. Prototypes in linguistic theory (3rd ed.). Oxford:

University Press.Taylor, John. (2012). The mental corpus: How language is represented in the mind. Oxford: Oxford

University Press. DOI: 10.1093/acprof:oso/9780199290802.001.0001Tesnière, Lucien. (1959). Elements de Syntaxe Structurale. Paris: Klincksieck.Teubert, Wolfgang. (1998). Korpus und Neologie. In Wolfgang Teubert (Ed.), Neologie und

Korpus (pp. 129–170). Tübingen: Narr.Thorndyke, Perry W., & Yekovich, Frank R. (1980). A critique of schema-based theories of

human story memory. Poetics, 9, 23–49. DOI: 10.1016/0304-422X(80)90011-XTomasello, Michael. (2000). First steps toward a usage-based theory of language acquisition.

Cognitive Linguistics, 11(1/2), 61–82.

http://dx.doi.org/10.1515/9783110881660.1

http://dx.doi.org/10.1007/978-3-322-95664-4


http://dx.doi.org/10.1515/9783110803464.129



http://dx.doi.org/10.1093/acprof:oso/9780199290802.001.0001

http://dx.doi.org/10.1016/0304-422X(80)90011-X

References 417

Tomasello, Michael. (2003). Constructing a language: A usage-based account of language acquisi-tion. Cambridge: Harvard University Press.

Tomasello, Michael. (2006). Konstruktionsgrammatik und Erstsprachenerwerb. In Kerstin Fischer, & Anatol Stefanowitsch (Eds.), Konstruktionsgrammatik. Von der Anwendung zur Theorie. Tübingen: Stauffenburg.

Trabasso, Tom, & Suh, Soyoung. (1993). Using talk-aloud protocols to reveal inferences during comprehension of text. Discourse Processes, 16, 3–34. DOI: 10.1080/01638539309544827

Tsohatzidis, Savas L. (1993). Scenes and frames for orders and threats. In Richard A. Geiger, & Brygida Rudzka-Ostyn (Eds.), Conceptualization and mental processing in language, 731–740. Berlin & New York: Mouton de Gruyter.

Tuggy, David. (1988). Náhuatl causatives/applicatives in cognitive grammar. In Brygida Rudzka-Ostyn (Ed.), Topics in cognitive linguistics (pp. 587–618). Amsterdam: John Benjamins.

DOI: 10.1075/cilt.50.21tugTuggy, David. (1993). Ambiguity, polysemy, and vagueness. Cognitive Linguistics, 4(3), 273–290.

DOI: 10.1515/cogl.1993.4.3.273Tulving, Endel. (1983). Elements of episodic memory. New York: Oxford University Press.Tulving, Endel. (2000). Concepts of memory. In Endel Tulving, & Fergus I.M. Craik (Ed.), The

Oxford handbook of memory (pp. 33–43). Oxford: Oxford University Press.Turner, Mark. (1998). Figure. In Albert N. Katz, Cristina Cacciari, Raymond W. Gibbs, & Mark

Turner (Eds.), Figurative language and thought (pp. 44–87). New York & Oxford: Oxford University Press.

Ungerer, Friedrich, & Schmid, Hans-Jörg. (1996). An introduction to cognitive linguistics. London: Longman.

Vater, Heinz. (2005). Referenz-Linguistik. Munich: Fink.Warnke, Ingo (Ed.). (2007). Diskurslinguistik nach Foucault. Theorien und Gegenstände. Berlin

& New York: de Gruyter. DOI: 10.1515/9783110920390Watters, James K. (1996). Frames and the semantics of applicatives in Tepehua. In Eugene H.

Casad (Ed.), Cognitive linguistics in the redwoods: The expansion of a new paradigm in lin-guistics (pp. 971–996). Berlin & New York: Mouton de Gruyter.

Wegener, Philip. (1885). Untersuchungen ueber die Grundfragen des Sprachlebens. Halle, Saale: Niemeyer.

Wegner, Immo. (1979). Frame-Theorie und lexikalische Semantik. Deutsche Sprache, 7, 298–314.Wegner, Immo. (1984). Frame-Simulation mit dem Computer. In Klaus Oehler (Ed.), Zeichen

und Realität: Akten des 3. semiotischen Kolloquiums der Deutschen Gesellschaft für Semiotik e.V (pp. 547–555). Tübingen, Stauffenburg.

Wegner, Immo. (1985). Frame-Theorie in der Lexikographie. Untersuchungen zur theoretischen Fundierung und computergestützten Anwendung kontextueller Rahmenstrukturen für die lexikographische Repräsentation von Substantiven. Tübingen: Niemeyer.

Wengeler, Martin. (1996). Sprachthematisierungen in argumentativer Funktion. Eine Typolo-gie. In Karin Böke, Matthias Jung, & Martin Wengeler (Eds.), Öffentlicher Sprachgebrauch (pp. 413–430). Opladen: Westdeutscher Verlag.

Wengeler, Martin. (2000). ‘Gastarbeiter sind auch Menschen.’ Argumentationsanalyse als dis-kursgeschichtliche Methode. Sprache und Literatur in Wissenschaft und Unterricht, 86, 54–69.

Wengeler, Martin. (2003). Topos und Diskurs. Begründung einer argumentationsanalytischen Methode und ihre Anwendung auf den Migrationsdiskurs (1960–1985). Tübingen: Niemeyer. DOI: 10.1515/9783110913187

http://dx.doi.org/10.1080/01638539309544827

http://dx.doi.org/10.1075/cilt.50.21tug


http://dx.doi.org/10.1515/9783110920390

http://dx.doi.org/10.1515/9783110913187


Wengeler, Martin. (2005). ‘Das Szenario des kollektiven Wissens einer Diskursgemeinschaft entwerfen’. Historische Diskurssemantik als ‘kritische Linguistik’. Aptum, 1(3), 262–282.

Wengeler, Martin. (2010). “Noch nie zuvor“. Zur sprachlichen Konstruktion der Wirtschaftskrise 2008/2009 im Spiegel. Aptum. Zeitschrift für Sprachkritik und Sprachkultur, 6(2), 138–156.

Wertheimer, Max. (1923). Untersuchung zur Lehre von der Gestalt. Psychologische Forschung, 4, 301–351. DOI: 10.1007/BF00410640

Wettler, Manfred. (1980). Sprache, Gedächtnis, Verstehen. Berlin & New York: de Gruyter.Wiegand, Herbert Ernst. (1988). Was eigentlich ist Fachlexikographie? Mit Hinweisen zum

Verhältnis von sprachlichem und enzyklopädischem Wissen. In Horst-Haider Munske et al (Eds.), Lexikologische Studien (pp. 729–790). Berlin & New York: de Gruyter.

Wierzbicka, Anna. (1972). Semantic primitives. Frankfurt: Athenäum.Wiese, Heike. (1999). Die Verknüpfung sprachlichen und konzeptuellen Wissens: Eine Dis-

kussion mentaler Module. In KogWiss 1999. Proceedings of the fourth symposium of the Gesellschaft für Kognitionswissenschaft Bielefeld (pp. 92–97). 28 September–1 October. St. Augustin: Infix-Verlag.

Wilson, Margaret. (2002). Six views of embodied cognition. Psychonomic Bulletin and Review, 9, 625–636. DOI: 10.3758/BF03196322

Wimmer, Rainer. (1979). Referenzsemantik. Untersuchungen zur Festlegung von Bezeichnungs-funktionen sprachlicher Ausdrücke am Beispiel des Deutschen. Tübingen: Niemeyer.

Wisniewski, Edward J., & Medin, Douglas, L. (1984). On the interaction of theory and data in concept learning. Cognitive Science, 18, 221–281. DOI: 10.1207/s15516709cog1802_2

Wittgenstein, Ludwig. (1989). Philosophische Untersuchungen. Frankfurt a.M.: Suhrkamp.Ziem, Alexander, Boas, Hans, & Ruppenhofer, Josef. (2014): Semantische Frames und gram-

matische Konstruktionen für die Textanalyse. In Jörg Hagemann, & Sven Staffeldt (Eds.), Syntaxtheorien. Vergleichende Analysen (pp. 297–333). Tübingen: Stauffenburg.

Ziem, Alexander, & Staffeldt, Sven. (2011). Compositionality and embodied meanings of phra-seologisms: The case of somatisms. In Doris Schönefeld (Ed.), Converging evidence: Meth-odological and theoretical issues for linguistic research (pp. 195–219). Amsterdam: John Benjamins. DOI: 10.1075/hcp.33.13zie

Ziem, Alexander, & Lasch, Alexander. (2013). Konstruktionsgrammatik. Konzepte und Grundla-gen gebrauchsbasierter Ansätze. Berlin: de Gruyter. DOI: 10.1515/9783110295641

Ziem, Alexander. (2005). Begriffswissen. Ein linguistischer Beitrag zur sprachlichen Bedeu-tungskonstruktion in literarischen Texten. In Martin Roussel, Markus Wirtz, & Antonia Wunderlich (Eds.), Eingrenzen und Überschreiten. Verfahren in der Moderneforschung (pp. 272–285). Würzburg: Königshausen & Neumann.

Ziem, Alexander. (2006a). Begriffe, Topoi, Wissensrahmen: Perspektiven einer semantischen Analyse gesellschaftlichen Wissens. In Martin Wengeler (Ed.), Sprachgeschichte als Zeitge-schichte. Konzepte, Methoden und Forschungsergebnisse der Düsseldorfer Sprachgeschichts-schreibung für die Zeit nach 1945 (pp. 315–348). Hildesheim & New York: Olms.

Ziem, Alexander. (2006b). Wie Werbung Sinn macht. Bedeutungskonstruktionen und -korrek-turen durch Text-Bildbeziehungen. In Martin Wengeler (Ed.), Linguistik als Kulturwissen-schaft. Neuere empirische Analysen zur öffentlichen Kommunikation in Werbung, Politik, und Medien (pp. 45–67). Hildesheim, Zürich, New York: Olms.

Ziem, Alexander. (2006c). Mental Spaces, Idealized Cognitive Models, Frames, or Domains? Toward a Unified Theory of Semantic Representations. Paper presented at the Second In-ternational Conference of the Gesellschaft für Kognitive Linguistik, Munich, 7 October 2006.

http://dx.doi.org/10.1007/BF00410640

http://dx.doi.org/10.3758/BF03196322


http://dx.doi.org/10.1075/hcp.33.13zie

http://dx.doi.org/10.1515/9783110295641

References 419

Ziem, Alexander. (2008a). Frame-Semantik und Diskursanalyse – Skizze einer kognitionswis-senschaftlich inspirierten Methode zur Analyse gesellschaftlichen Wissens. In Ingo Warnke, &Jürgen Spitzmüller (Eds.), Diskurslinguistik nach Foucault. Methoden (pp. 89–117). Berlin & New York: de Gruyter.

Ziem, Alexander. (2008b). Universale Prägung und kulturelle Varianz. Überlegungen zu einem integralen Bedeutungsmodell im kognitiven Paradigma. In Zeitschrift für Literatur und Linguistik (Vol. 118, pp. 185–198).

Ziem, Alexander. (2009). Konzeptuelle Integration als kreativer Prozess: Prolegomena zu einer kognitiven Ästhetik. In Martin Huber, & Simone Winko (Eds.), Literatur und Kognition. Bestandsaufnahmen und Perspektiven eines Arbeitsfeldes (pp. 63–84). Paderborn: Mentis.

Ziem, Alexander. (2011). Cognitive science meets language pedagogy: Towards a schema-theo-retical framework for teaching vocabulary. In Markus Callies et al. (Eds.), Bi-directionality in the cognitive sciences (pp. 249–268). Amsterdam: John Benjamins.

DOI: 10.1075/hcp.30.16zieZiem, Alexander. (2013). Beyond the sentence: Towards a cognitive-linguistic approach to tex-

tual reference. Yearbook of the Cognitive Linguistic Association, 1, 39–58.Ziem, Alexander. (2014). Von der Kasusgrammatik zum FrameNet: Frames, Konstruktionen und

die Idee eines Konstruktikons. In Alexander Ziem, & Alexander Lasch (Eds.), Grammatik als Inventar von Konstruktionen? Sprachwissen im Fokus in der Konstruktionsgrammatik (pp. 351–388). Berlin & New York: de Gruyter. DOI: 10.1515/9783110353693

Ziem, Alexander. (2014). FrameNet, Barsalou frames and the case of associative anaphora. In Thomas Gamerschlag et al. (Eds.), Frames and concept types: Applications in cognition, language, and science (pp. 351–388). Berlin: Springer.

Ziem, Alexander. (submitted). Frames and Constructions Enhance Text Coherence: the Case of DNI Resolution. In Yearbook of the German Cognitive Linguistics Association.

Ziemke, Tom, Zlatev, Jordan, & Frank, Roslyn (Eds.). (2007). Body, language, and mind I: Embodiment. Berlin & New York: Mouton de Gruyter.

Zlatev, Jordan. (1997). Situated embodiment: Studies in the emergence of spatial meaning. Stock-holm: Gotab.

Zlatev, Jordan. (2003). Polysemy or generality? Mu. In Hubert Cuyckens, René Dirven, & John R. Taylor (Eds.), Cognitive approaches to lexical semantics (pp. 447–494). Berlin & New York: Mouton de Gruyter. DOI: 10.1515/9783110219074.447

Zlatev, Jordan. (2005). What’s in a schema? Bodily mimesis and the grounding of languet thiage. In Beate Hampe (Ed.), From perception to meaning: Image schemas in cognitive linguistics (pp. 323–342). Berlin: Mouton de Gruyter.

http://dx.doi.org/10.1075/hcp.30.16zie

http://dx.doi.org/10.1515/9783110353693

http://dx.doi.org/10.1515/9783110219074.447

Name index

AAbelson, Robert P. 13–14, 142Alba, Joseph W. 222–223, 226Allwood, Jens 181, 202–210Anderson, John R. 87, 305

BBaker, Collin F. 16Ballmer, Thomas T. 270Barclay, J. Richards 139Barsalou, Lawrence W. 46–47,

214, 216, 294Bartlett, Frederic 11, 213,

215–218, 220–225Bartsch, Renate 33Biere, Bernd Ulrich 136Bierwisch, Manfred 38–39,

55–65, 73, 76, 107–114, 116, 120Blank, Andreas 47Bransford, John D. 139Brennstuhl, W. 270Brewer, William F. 223–224Bühler, Karl 1–2, 157, 292Busse, Dietrich 17, 96, 118,

129–136 143–145, 200–201, 223, 252, 254, 282, 332–335, 340–343, 381

Bybee, Joan L. 296

CChomsky, Noam 29, 58–59, 111,

114–115Christmann, Ursula 224Clark, Herbert H. 141, 148Coulson, Seana 15, 72, 196, 241,

324–327Croft, William 41, 153, 155–158,

163, 246, 259

DDemmerling, Christoph

121–128, 147, 183–185

Dijk, Teun A. Van 138, 140Dirven, René 89Dölling, Johannes 39, 64,

73, 78

EEvans, Vyvyan 22–24, 296

FFauconnier, Gilles 15–16,

26–28, 32, 40, 89–90, 324, 362, 377

Felder, Ekkehard 43–44Fillmore, Charles J. 2, 11, 15,

24–25, 28, 41, 69–70, 101–102, 112, 116–117, 120–121, 126, 157–158, 160, 177, 188–202, 243–245, 258–263, 289

Fischer, Kerstin 160–161Foucault, Michel 378Fraas, Claudia 16, 45–46, 100,

271–273, 316–320Franks, Jeffrey J. 139

GGansel, Christina 46Garrod, Simon C. 142, 149Goldberg, Adele 15, 42,

156–158, 160Grady, Joseph E. 324–327Graesser, Arthur C. 139–144,

147–148Green, Melanie 22–24, 296

HHaiman, John 106, 108, 111,

114–115Hasher, Lynn 222, 226Hebb, Donald Olding 305Heringer, Hans Jürgen 321Hermanns, Fritz 348Herweg, Michael 52, 63, 89

Holly, Werner 320Hörmann, Hans 32, 125

JJackendoff, Ray S.

44, 55–56, 86–88, 91–97, 137, 214, 249–250

Johnson, Mark 50–51, 106, 322–325, 327, 330

KKay, Paul 25, 41, 157–158, 160Keller, Rudi 117–118, 302–304Kintsch, Walter 138, 140–141Klein, Josef 16, 272, 292,

319–320Konerding, Klaus-Peter 16–17,

215, 243, 264, 266–280, 361, 383

LLakoff, George 20–23, 25–26,

28, 35, 40–42, 50–51, 54–55, 106, 117, 155, 157, 256, 322–325, 330

Lang, Ewald 59–60Langlotz, Andreas 163Langacker, Ronald W. 15, 23–

25, 41–42, 77, 91–98, 104–106, 152–159, 161, 168–182, 196–197, 202–203, 240, 259, 267, 285, 300–301, 304–305

Liebert, Wolf-Andreas 323Lönneker, Birte 16, 245–247,

272, 278, 282–284, 316, 318–320, 350, 354–359, 363–365

MMakkai, Adam 115McKoon, Gail 138Medin, Douglas, L. 216Meißner, Iris 272, 319–320


Minsky, Marvin 1, 13–14, 18–21, 28, 263–264, 291, 385

OOakley, Todd 324–327Olson, David R. 115–116, 248

PPetruck, Mirjam R. L. 15, 34,

189Polenz, Peter von 257–261,

289, 353, 380–381

QQuine, Williard Van Orman

88, 110–111

RRatcliff, Roger 138, 305Rice, Sally 181Rickheit, Gert 30, 142, 148–149Rosch, Eleanor 34, 297–298Rumelhart, David E. 226, 231,

280

SSandra, Dominiek 181Sanford, Anthony J. 142

Saussure, Ferdinand de 104, 155, 161

Schank, Roger C. 13–14, 142, 186, 236

Scharloth, Joachim 342–343Scherner, Maximilian 187Schlobinski, Peter 172Schnotz, Wolfgang 147–149,

232Schwarz, Monika 30, 32–34,

38–39, 52, 55, 73, 77–88, 95, 97–98, 138, 215, 224, 226, 248, 250–251, 290–291

Searle, John R. 123, 183, 245–246, 252–255, 266, 311

Singer, Murray 139–140, 143–144, 147–148, 381

Slobin, Dan I. 296Stötzel, Georg 344Sweetser, Eve 26, 100

TTalmy, Leonard 12, 41, 74Taylor, John R. 23, 31–32, 67,

77, 123, 152, 255, 300Thorndyke, Perry W. 226–228Tomasello, Michael 122Trabasso, Tom 139–140, 144,

147–148, 381

Turner, Mark 15–16, 26, 324, 328, 332, 362, 377

VVater, Heinz 254–256

WWegner, Immo 16Wengeler, Martin 344Wiese, Heike 60, 100Wisniewski, Edward J. 216Wittgenstein, Ludwig

130, 182, 184

YYekovich, Frank R. 226–228

ZZiem, Alexander 42, 380, 384Zlatev, Jordan 179–188

Subject index

Aabduction 232, 240–241, 383ability 36–39, 54–55, 92,

111–112, 165, 188, 213–214, 240, 379, 381

abstraction 216–217, 219, 222–228, 231–232, 239–240, 323level of 131, 161–163,

225–226, 315action frame 83activation 84, 86, 197–199,

201–206, 247–248, 289–292actualization 59–60, 177–179,

203, 225–226, 302, 325adjuncts 257, 270agent 80–84, 325–327artificial intelligence 13, 29ambiguity 311anaphor 84–87, 290–291argument structure 52–53, 61,

63, 65, 74, 80, 402association 125, 188–190, 196,

199, 215–217attribution 261, 282, 289

attribute 297

Bbackground 24, 75, 179–188,

195, 196, 228–229background assumption

89, 117, 195–196, 289, 316background frame 15, 18,

119–120background information,

see background knowledgebackground knowledge

3–4, 24–26, 28, 33, 188, 200–202, 327, 380–381

base 18, 25 see also figure/ground

basic discourse–semantic figure 339–344, 346–349

basic level category 297–299, 347

blended space 324–332, 365see also conceptual

integration, blendingblending 26–28, 241, 314,

324–325, 376–377see also conceptual

integration, blended spacebodily experience 124, 126, 322

see also embodimentbody 51–54, 111–112

Ccase frame 11–12, 191–195, 261,

311–312case grammar 11–13, 258, 380categorization 34–35, 93,

212–216, 224categorization link

42, 161–170, 180category 21–23, 41–42, 55, 152,

181, 191, 212–216, 295–299, 330, 384

causality 74–75, 83, 111, 218, 328–239

code 136, 176, 197, 202–203 coding relationship

197, 202–203cognition 29–30, 36–37, 53–54,

56–57, 87, 92–93, 212–213model of 41, 49, 86, 93,

96–98see also holism, modularism

cognitive model 30–32, 49, 89see also idealized cognitive

modelcognitive economy 37, 213, 223Cognitive Grammar 41–42,

104–105, 151–152, 295–296, 382

cognitive linguistics 38–42, 79, 88, 122

cognitive operation 26, 28, 30, 87, 89–91, 201, 211see also abstraction,

anaphor, categorization, schematization

cognitive presence 294, 307, 317, 340, 378

cognitive reality 39, 79, 224, 226, 230

cognitive representation, see representation

cognitive resources, see knowledge resource

cognitive routine 68, 145, 155, 166, 300–301, 343see also cognitive trail,

entrenchmentcognitive science 29–30cognitive semantics 10, 43–44,

379cognitive theory 20, 36–37, 65,

77, 95–96, 120, 166, 212, 227see also modularism, holism

cognitive trail 179, 300–301, 383

coherence 19, 137–140, 198, 200–203, 290–292, 379construction of 291establishment of 137, 140

communication 7, 55, 124, 129–135, 139–340communication context

334–338 communication situation,

see situation communicative purpose

39, 113–114, 170, 179competence 57, 127, 150–151component structure 38, 90,

165component, see semantic

component


compositionality 39, 56, 58, 72, 79, 80, 113, 160–161, 203, 326

compound 85, 89, 133, 174, 240, 286–288, 350, 356see also determinative

compoundconcept 29–35

see also conceptual meaning, representation, schema

conceptual blending, see blending

conceptual integration 26–28, 70–72, 324–328, 399see also blending, blended

space, conceptual blending

conceptual knowledge 21, 30, 35–36, 38–39, 57–61, 80–83, 101–106, 115–117, 121, 379–380see also conceptual meaning,

holismconceptual meaning 169–172,

175–178, 182–183, 187–188, 196–197, 202–203, 205–207, 209–212, 382see also usage event

conceptualization 84–85, 91–93, 104–105, 169–172, 176–177, 183, 323, 327–328

conceptual metaphor 51, 92, 95–96, 99, 323–324, 327, 330, 347

conceptual semantics 86–87, 91–92

conceptual unit 8–9, 24, 28, 47, 73–74, 80, 251, 286

consciousness 30threshold of 213

construal operations 200construction 150, 151–170construction grammar 15–17,

41–42, 92, 150–156, 159, 164–167

container 22–23, 50–52, 136of meaning 58, 136

content side 154, 162–163, 169–170, 177, 179, 343–344see also semantic unit

context 82–83, 89–90, 100–102, 108–109, 115–116, 136, 171–172, 177, 179, 187, 203–206

contextual dependence 9–10, 89, 171

contextual meaning 60, 171–172, 196

contextualization 89, 282, 285, 334 potential 100, 258, 271, 286,

317, 378contiguity 215–219, 230, 319convention 150, 154–155, 165,

168–173, 177, 180conventional meaning

150, 171–178, 180, 183conventionalization 118–119,

172, 180, 204, 223core focus 132core meaning 59–60, 80–83,

107core, see semantic corecorpus 247–248, 333–339, 344,

346, 349, 365–366created prime 73, 109–112cue validity 173, 297cultural knowledge 101–102

see also encyclopedic knowledge, conceptual knowledge

Ddecomposition 63, 70, 77,

79–80, 96–97, 112, 115–116, 132deep case 11–12, 18, 192–194,

258, 380–381default assumption 19, 98, 208,

226, 328default predications, see implicit

predicationdefault value 10, 176–188,

205–210, 227–228, 231, 250, 289–292see also implicit predication,

default assumptiondeixis 133, 141, 157, 199, 292, 350 determinative compound

173, 286–288, 356discourse 130–131, 180, 261,

322–324, 333–339discourse analysis 317, 333–334,

339–340, 377–378, 384discourse element 339–340,

346–349, 349, 377

discourse history 337–338domain 23–28, 80–84, 192,

323–324, 365–366basic 23abstract 25

dualism 42, 50dynamic 14, 40, 47, 55, 223–

224, 229

Eembedding structure 127, 134,

187, 217, 261, 332embodiment 40, 54emergent meaning 312–314,

328–331see also blending, conceptual

blendingemergence 27, 100, 171, 297–

300, 312, 314, 325–329, 375encyclopaedic knowledge

13, 59–61, 62, 78–79, 101–121, 131–134, 186–189, 257, 380

entrenchment 294–297, 299–300, 302–304, 307, 321, 331, 347, 377–378, 383–384

epiphenomenon 151, 303epistemology 17, 30, 88, 102–

103, 128–136, 154essence 107–109event, see usage eventeveryday knowledge 13, 57–58evocation potential 164, 171evoke 1–2, 164, 191–193,

197–202, 218–220, 233–335, 252–256, 299, 307–314see also invoke

experience 19–22, 31–33, 53–55, 94–96, 111–112, 181–183, 219, 221–222, 231

experiential data 11, 19, 30, 222experientialist cognition

54–56, 93–94explanatory adequacy 122explanatory power 49, 77explicit predication 280–289,

308–309, 352–353, 359–361, 366, 383see also filler

extralinguistic 13, 59, 76, 125, 141, 177, 186–188 see also non–linguistic

Subject index 425

Ffacet of knowledge 46–47, 53,

171–175, 259–260, 270, 285, 362–363, 366–367see also slot, sub–question

figure/ground 25, 75, 93, 191, 228–299

filler 9–10, 69, 84–85, 206–210, 242, 228–239, 280–289 see also explicit predication

financial investor 288–289, 349–351, 356–359, 365–370

fixed prime 73–74, 80, 109–113 focal adjustment 35, 68focus field 132

see also core focusforeground 25, 68–69, 75, 126,

159, 167see also figure/ground

form–meaning pair 4, 40–41, 150–158, 162, 165–167, 169, 343see also symbolic unit

frame 5–48, 152–156, 158, 161, 168–210, 218–314, 315–322

frame element 14, 51, 65, 193–194, 198, 247, 282, 319

frame hierarchy 272, 283, 319, 354, 363

frame of reference 257–258, 332, 260–261see also frame

frames of understanding, see frame, U–semantics, U–relevant knowledge

FrameNet 2, 15, 18, 319frequency 294–314, 343–349,

375see also type frequency, token

frequencyfrequency of occurrence 294,

296, 304, 316–317, 354, 375, 377

Ggenerative grammar 36–38, 56,

11, 150–152, 162generic 145, 219, 223–225, 232

generic frame 336, 362–365generic space 325–332, 362

gestalt 11, 21–22, 25, 94, 155, 214–216, 230, 236–237, 250, 257, 379, 383

gestalt psychology 25, 56, 214–216, 250

goal 236–237

Hhabitualized 252–253, 339holism 35–39, 43–55, 87–98,

99–106, 103, 120–121, 152, 164 hyperonymy 162–163, 173–174,

207, 217–218, 233, 238–240, 267–279, 301

hyperonym type reduction 207–208, 233, 238, 264–274, 349, 354–359, 361

hyponymy 173, 217–218, 266–267, 297, 301, 330, 383

IICM (idealized cognitive

model) 20–21, 25–26image schema 22–23, 50, 68,

124, 183, 240–241implicit predication

280, 289–307, 315, 353, 361, 376, 383see also default value, default

assumption, explicit predication

indirect anaphor 84–86, 290–291

induction 189, 232, 240 inference 7–8, 64, 135–150,

164, 166–167, 221, 280–281, 381–382

inference base 141 inference construction

136, 141–142, 146, 221inference theory 128–129,

138–141, 148, 164–166information processing 31

see also language processinginput 31, 212–213, 240, 250,

305–306, 325–332see also conceptual

integration, conceptual blending

input frame 365–370instance 8–9, 14–15, 162–163,

173–174see also filler, default value

instantiation 34, 84, 162–163, 175, 228, 231–232, 243–244, 307

integration, see conceptual integration

interactionsocial interactio 66, 101,

120, 124, 133, 144, 318interaction of knowledge

(generative grammar) 39, 55–56, 57–58, 60, 77–78, 113, 120

interpretative semantics 381see also U–semantics,

understanding semantics, semantics of understanding

interpretation 51–54, 63–100, 200

interpretational variance 53, 71, 89–90, 96

introspection 151, 179, 181, 322invoke 150, 197–202, 232, 282,

299, 382see also evoke

Kknowing–how 36

see also abilityknowing–that 36

see also conceptual knowledgeknowledge, see conceptual

knowledge, encyclopaedic knowledge, facet of knowledge, linguistic knowledge, world knowledge

knowledge element 26–27, 100, 198–199, 231, 283, 325

knowledge mode 111, 308–309, 312

knowledge representation 20, 30, 35, 54, 94, 96, 126, 146, 206, 224, 384see also frame, representation

knowledge resource 55, 138, 147, 379

knowledge types 57, 113, 132–133, 143–147, 184–186, 381

Llanguage acquisition 111, 118,

122, 241


language change 299–300, 302–304

language community 171–172, 176, 178–179, 199, 204–205, 257, 265–266, 302, 317, 339, 378

language faculty see ability language processing 56–58, 90–92, 142see also information

processing language knowledge, see

linguistic knowledgelanguage use 39–40, 121–122,

151, 204, 240, 297, 301levels of knowledge 132level see two–level semantics,

three–level semantics, multi–level semantic model

lexeme 46, 70, 118, 181, 218, 323, 354, 366

lexical unit see form–meaning pair, word

lexicography 15–18, 104, 256, 266, 322, 391

lexicology 15–18, 266life form 101linguistic knowledge

36, 38–39, 52–53, 57–58, 61–62, 65–77, 113–114, 120, 149–151, 167, 380

linguistic sign 101–102, 128–129, 135–146, 150–152, 166–167, 224, 244, 247see also form–meaning pair,

symbolic unitlocust 274–279, 316, 328–333,

339–349, 365–376 long–term memory 32, 35, 45,

137, 223–226 long–term potentiation 306

Mmacroproposition 138, 140,

335–336mapping 26, 323–327, 330–332

see also projectionmatrix frame 233, 238–240,

270–275, 278–281, 283, 318–322, 330, 354–359, 363–364

maximalist inference theory 102 see also inference theory

meaning 1–4, 50–52, 56–61, 101, 147–210, 380–385see also frame

meaning actualization 50–51, 66–67, 169, 325

meaning construction 26, 68–69, 164

meaning potential 68–69, 78, 92, 202–207, 370, 382

meaning side 155, 157–164, 205see also content side

memory 182, 186, 213, 220–227, 304–305see also long–term memory

mental space 15–16, 26–28, 32, 101, 133, 239–241, 325–328see also blending

metaphor 66–67, 322–332, 339–440, 371–377

metaphtonymy 69, 328metonymy 66–67, 116, 328–

329, 368, 399 mind 29–30, 38, 53–54, 92–93modularism 29, 38–39, 43–44,

50–53, 55, 59, 86–87, 92–94module 36, 53, 56–57, 88–98morpheme 154, 157, 162–163,

197, 214, 286–287, 350motor 50, 54, 92–95, 212–213,

298–299, 301, 306 see also sensorimotor

multi–level semantic model 39, 45, 100, 120–122, 151see also two–level semantics,

three–level semantics

Nnativist 87, 110, 168network 97–98, 134, 161–178,

180–185, 231–232, 383neuronal network 304–307nominalization 253–254,

268–269, 273, 308–309, 313non–linguistic knowledge

see conceptual knowledge, encyclopaedic knowledge

noun 31–32, 162–265, 274noun type 269–271, 279

Pparticle 133, 254–256, 307–314part–whole 34, 67, 140, 144,

293patient 80–84, 325–327pattern recognition 91pattern 180, 339–340, 348

argumentation pattern, see topos

neuronal pattern 168 see also neuronal networkstructural pattern 229–230,

236–237perception 18–20, 30–33,

92–96, 132–133, 212–216, 228–232, 248–251

perceptual constancy 125, 200perceptual modalities 230, 301performance 35, 55, 150–151,

167, 252perspectivation 68, 284–285,

289–290see also figure/ground

perspective 285, 289phenomenon of the third kind

302–303, 307phoneme 212phonological structure 56–57,

154–156, 155phonological unit 154–155, 158,

164–165, 196–201, 247–248, 251–252

phraseologism 70plan 229–230, 236–237polysemy 63, 77, 82, 89–90,

107, 116–117postulate of U–relevance

102, 122–123, 128–135, 142, 146–147, 184, 192

practice 31, 125, 183–185, 187predicate 9, 52–53, 59, 110, 246,

262–269, 288predication analysis 315, 339,

352–353predication potential 205–207,

261–263predication 177–179, 205–207,

245–247, 261–269, 283, 313, 383see also explicit predication,

implicit predication

Subject index 427

predicative attribution see predication

predicator 270, 278–279, 348, 357–361, 369predicator class 278–279,

318, 354, 366–370predicator schema 270, 319

presupposition 63, 67, 75–76, 133, 189–190, 255

primitives, see semantic primitives

principle of contextual adaptability 119

probabilistic 316, 377procedural routes 81–82procedure, see cognitive

operationprocessing depth 79, 296processing operation

105, 221, 227profile, see semantic profileprojected reality 137, 249–251projected world 44, 249–251projection 54, 137, 178, 248–249proposition 21–23, 123–127,

138–139, 184–188, 244–247, 264–265, 281–282

propositional content 138, 245–247

prototype 21–23, 69, 97–98, 173, 280, 293–294, 297

psycholinguistics 135–136, 166, 281, 377

Qquasi–explicit predication

251–257, 286, 291

RRadical Construction Grammar

41reading 52–53, 63–67, 69, 71, 104 reception aesthetics 313–313recurrence 177–179, 182, 204,

208, 225, 231–233, 294–296, 301–302, 306–307, 333see also frequency

reductionism 3–4, 77, 79–83, 101–104, 106, 121, 174, 258–260

refer 248, 250–256, 258–260

reference 31–32, 118–119, 245–251, 255, 261, 268, 282, 383

reference object 246–247, 251–252, 280–281, 282–288, 314, 337–340

reference potential 60–61, 67, 73, 78–79, 96, 254

referentialization 32, 136, 251–252, 257, 289–290, 327

representation 30–31, 137, 143, 148, 178, 232

representation format 20, 28, 52, 91, 98, 137, 175, 195, 220–221, 287, 325see also frame, mental space,

schemarule system 53, 57–58, 87, 92,

114, 122

Ssalience 173, 226, 285, 292–294,

296, 329–330, 347, 377–378scene 12, 69, 126, 188–195, 251 schema 145, 174–176, 218–241,

280, 297–298, 383 schema instantiation 34, 84,

221–222, 287–288, 292–295, 300, 301, 307

schema–instance relationship 163–176, 180–181, 287–288, 298–299, 301–302, 307–309

schematization 167, 185–186, 303, 341, 379

script 13–14, 21–22, 142, 200–201, 236

selection 59, 78, 171, 222, 226selection restriction

59, 114–115, 380semantic component

38, 45–46, 59–60, 67, 111, 116–117

semantic constant 52, 60–65, 68, 73–74, 76, 80, 111see also semantic primitives

semantic core 46–47, 99, 107–109, 148

semantic interpretation, see interpretation

semantic knowledge 49–50, 56–57, 103–104, 119, 121, 161–162

semantic primitive 57, 73–74, 77, 87, 111–112, 121, 177see also semantic constant

semantic profile 24–25, 32–33, 132, 283, 285, 370, 375see also figure/ground

semantic representation 50–54, 60–63, 67–70, 92, 109, 377, 380see also cognitive

representationsemantics of understanding

379 see also U–semantics,

postulate of U–relevance, space of understanding

semantic space 169–170semantic unit 154, 162–165,

169–172, 178–198, 201–202, 211–212, 287–288

sensorimotor 50, 53–54, 92–93, 95–96, 123, 137see also motor

sentence 13, 133–134, 162–163, 232, 245, 351

short–term memory 35, 91, 225sign, see linguistic sign situated inference theory

140–143see also inference theory

situation 123–124, 126, 141–142, 144–145, 169–170, 182–189, 196–197

situative setting 185, 187slot 4, 19–20, 201–202, 205–

209, 224–225, 227–241, 256–279, 299–301, 329–330, 354see also facet of knowledge,

sub–questionsource domain 323–325,

346–347, 356space of understanding

121–130, 133, 146, 164–165, 170–171, 185–187, 381

specification 60–62, 72, 82–83, 107–109, 171–172, 180, 218–219, 262, 285–290

standard interpretation 119–120

store 226–227, 304


structural constituent 10, 228–232, 243–247, 295, 300–301, 315, 349, 377, 379

subject 59, 113, 115, 159–160, 162–163, 245, 253

sub–question 199, 272, 278–279, 283see also slot

subslot 272, 282–284superframe 363–364symbolic principle 150–152,

158–159, 161, 204, 301, 382symbolic structures 22–23,

155–156, 175symbolic unit 96, 150–170, 179,

183–185, 196–197, 199–202, 315–316, 341, 343, 382 see also form–meaning pair

synonymy 110syntactocentrism 92, 194, 380syntax 41, 57–58, 80, 86, 92, 94,

96, 114–115, 158

Ttarget domain 323–325, 335,

346text 135, 139, 149, 198–201, 225,

243, 333–334theta grid 63, 80–81, 84three–level semantics

38, 77–86, 96–97, 107

see also two–level semantic, multi–level semantic model

token see fillertoken frequency 294–302, 311,

313–315, 343, 346–347, 375–376topos 180, 348TPA model 91–96trail see cognitive trailtripartite parallel architecture,

see TPA modeltwo–level semantics 38–39, 44,

56–61, 61–77, 96, 97, 107–121see also three–level

semantics, multi–level semantic model

type frequency 294–299, 307, 345–360, 370–373, 375, 383–384

Uunderspecified 20, 53, 60–62,

80–81, 202understanding, see postulate

of U–relevance, space of understanding

understanding semantics 2, 381 see also U–semantics,

semantics of understandingU–semantics 2, 13, 89, 379, 381

see also postulate of U–relevance, space of understanding

unit, see semantic unit, phonological unit, symbolic unit

U–relevance, see postulate of U–relevance

U–relevant knowledge 102–103, 128–134, 191–195, 257, 335, 379–381

usage–based 40, 90, 118, 122, 130, 134

usage event 168–169, 171–172, 176, 303–304, 321, 341, 376 see also conceptual meaning

Vvalency 11–13, 191–195, 257–

258, 380value, see filler, default valueverb 17–18, 59, 97, 126, 162–163,

184–194, 207, 262, 284, 319–320, 359–360

Wword 18–19, 34, 50–51, 115–116,

125–126, 162–163, 182, 197–198, 252–253, 280–281

world knowledge see encyclopaedic knowledge

Date post:	08-Aug-2015
Category:	Education
Upload:	nelli17
View:	46 times
Download:	4 times