Intonation in Dutch

7/30/2019 Intonation in Dutch

1/36

http://las.sagepub.com

Language and Speech

DOI: 10.1177/002383090004300201012000; 43; 127Language and Speech

Johanneke CaspersIntonation Patterns in Dutch

Experiments on the Meaning of Four Types of Single-Accent

http://las.sagepub.com/cgi/content/abstract/43/2/127The online version of this article can be found at:

Published by:

http://www.sagepublications.com

can be found at:Language and SpeechAdditional services and information for

http://las.sagepub.com/cgi/alertsEmail Alerts:

http://las.sagepub.com/subscriptionsSubscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.co.uk/journalsPermissions.navPermissions:

http://las.sagepub.com/cgi/content/refs/43/2/127Citations

by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from
http://las.sagepub.com/cgi/alertshttp://las.sagepub.com/cgi/alertshttp://las.sagepub.com/subscriptionshttp://las.sagepub.com/subscriptionshttp://www.sagepub.com/journalsReprints.navhttp://www.sagepub.com/journalsReprints.navhttp://www.sagepub.co.uk/journalsPermissions.navhttp://www.sagepub.co.uk/journalsPermissions.navhttp://las.sagepub.com/cgi/content/refs/43/2/127http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/cgi/content/refs/43/2/127http://www.sagepub.co.uk/journalsPermissions.navhttp://www.sagepub.com/journalsReprints.navhttp://las.sagepub.com/subscriptionshttp://las.sagepub.com/cgi/alerts


2/36

127

Experiments on the Meaning of

Four Types of Single-Accent

Intonation Patterns in Dutch*

JOHANNEKE CASPERS

Leiden University

INTRODUCTION

Intonation plays a role on several levels of linguistic organization; it has been reported to,

for example, mark the position of important information and the location of boundaries,

to express attitude and emotion, to signal whether a speaker is asking or telling something(sentence type), to mark information status and topic structure, and to play a role in the

turn-taking system. For comprehensive surveys of the various functions related to intonation,

* A preliminary version of the present paper, entitled Testing the meaning of four Dutch pitch

accent types, appeared in the Proceedings of the 1997 Eurospeech Conference (pp.863866).

Acknowledgments: This research was funded by the Netherlands Organization for Scientific

Research (NWO) under project #300 75001. Tina Cambier-Langeveld, Carlos Gussenhoven,

Vincent van Heuven, Nel Keijsper, Bob Kirsner, Cecilia Od, Helmer Strik, Jacques Terken, and

an anonymous referee are thanked for their valuable comments. I am grateful to Cecilia Od for

transcribing and judging the used materials and to Jos Pacilly for creating the environment for

remote participation of subjects through the internet.

Address for correspondence: Johanneke Caspers, Phonetics Laboratory/Holland Institute of

Generative Linguistics, Leiden University, PO Box 9515, 2300 RA Leiden, The Netherlands;

e-mail:

ABSTRACT

The Grammar of Dutch Intonation (GDI) provides a description of the possible

intonation contours of Dutch. The GDI distinguishes accent-lending and

nonaccent -lending pitch configurations, but refrains from further functionalstatements.This paper describes an experimental attempt to verify meaning

hypotheses for four Dutch single-accent pitch patterns as postulated in the

linguistic literature. The four pitch accent types were realized on proper names;

the abstract meanings, in terms of the manipulation of an element of the

background shared between speaker and listener, were incorporated in

situational contexts, distinguishing between a default and a vocative use of

the proper name (orientation). Listeners ranked the four melodic shapes from

most to least appropriate in their specific context. After revision of part of the

materials a second perception experiment was conducted, in which subjects had to rank four contexts

from most to least appropriate for a specific pitch accent type. Results show a distinct effect oforientation on the appropriateness of two of the investigated pitch accent types in the various context

types; the other two pitch accent types are associated with the predicted context types (and vice versa)

well above chance, indicating the viability of at least two of the linguistic proposals.

KEY WORDS

Dutch

intonationalmeaning

perceptionexperiments

pitch accents

prosody

LANGUAGE AND SPEECH, 2000, 43 (2), 127 161

http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/


3/36

128 The meaning of single-accent intonation patterns

see for example Bolinger (1986, 1989); Cruttenden (1986); Cutler, Dahan, and van

Donselaar (1997); Ladd (1996); and Tench (1996). The present investigation is concerned

with functional differences among accent-lending melodic configurations; it is an attempt

to experimentally verify abstract linguistic proposals concerning the meaning of four

pitch accent types in Dutch.

Two important semantic notions that have been associated with intonation are

prominence andphrasing (among others, Bolinger, 1975; Cutler, 1991; Ladd, 1992; Tench,

1996): speakers use melodic means to direct the listeners attention to the semantically

central aspects of the message (prominence) and to aid the listener in segmenting the speech

stream at several linguistic levels (phrasing). For the signaling of semantically central or

focused constituents, pitch accents are exploited, and the location of boundaries may be

indicated by specific boundary-marking pitch configurations. It is fairly commonly acknow-

ledged that speech melody carries another semantic aspect in addition to the signaling of

prominence and phrasing: the signaling ofinformation status with respect to the discourse(e.g., Gussenhoven, 1984; Pierrehumbert & Hirschberg, 1990; Tench, 1996). For example,

all accent-lending pitch configurations carry the meaning ofprominence ( this is important

information), but the type of pitch accent adds a further meaning component, intention

(cf. Grosz & Sidner, 1986; Pierrehumbert & Hirschberg, 1990), which pertains to the status

of the focused information with respect to a background shared between speaker and

listener (e.g., a H* pitch accent signals that the focused information should be treated as

new in the discourse, cf. Pierrehumbert & Hirschberg, 1990). It is this additional meaning

aspect that is under investigation here, which means that the present research is about

intention rather than attention (focus).

Note that the meaning differences between the various pitch accent types are expected

to be extremely subtle, not only because they share the prominence meaning aspect, but

also because it is quite clear from the experimental literature that even under contextually

constrained circumstances there are large within -speaker and between- speaker differences

in the choice of intonation contour (e.g., Caspers, 1994; Kraayeveld, 1997). Apparently,

associations between specific pitch accent types and certain communicative circum-

stances are relatively loose. Therefore, the effects of pitch accent type in experimental

investigations are expected to be moderate.The limited number of available experimental investigations of the meaning of melodic

shapes employ semantic scales as experimental tools, looking for systematic differences in

the interpretation of utterances, depending on intonational distinctions (e.g., Grabe,

Gussenhoven, Haan, Marsi, & Post, 1998; Kirsner & Van Heuven, 1996; Kirsner, Van Heuven,

& Van Bezooijen, 1994; Sanders, 1996; Uldall, 1972). Most of these studies investigate

emotions and attitudes associated directly with certain melodic shapes. These emotions and

attitudes the things that semantic scales representcan be viewed as interpretations of

an underlying abstract meaning in a specific context (e.g., Cruttenden, 1986; Gussenhoven,1984; Ladd, 1980; Ladd, Scherer, & Silverman, 1986; Liberman, 1979; Pierrehumbert &

Hirschberg, 1990). However, the available investigations do not present a comprehensive

account of how the semantic scales are to be related to more abstract intrinsic meanings.

Instead of attempting to capture abstract meanings in sets of semantic scales, the

present investigation tries to translate abstract meaning proposals into different types of

background. The semantic proposals adopted concern the information status of focusedby Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from
http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/


4/36

129J. Caspers

constituents, and an attempt is made to create situational contexts which define (part of)

the background shared between speaker and listener, creating a certain setting which should

be more or less appropriate for the pitch accent types under investigation. Contexts and

short utterances carrying the various pitch accent types were presented to listeners, who

had to indicate the best match between context and contour. This way, a number of factors

that may be relevant to the interpretation of pitch accent types is controlled, but many factorsremain unknown and therefore uncontrolled and these factors will introduce noise into the

data. However, it is expected that the basic semantic regularities will surface through this

(inevitable) noise.

Choice of intonation model

There are several different intonation models available, and a choice has to be made for a

specific model when investigating melodic phenomena, because the large variability in

natural intonation contours has to be structured in some way. However, this choice is notwithout consequences, since the model dictates which melodic differences are to be regarded

as linguistic and which as paralinguistic (i.e., which aspects form the basic parts of the

model, and which are mere variations in realization, cf. Ladd, 1996).

There are two well- known models of Dutch intonation: the phonetically oriented

model developed at the Institute for Perception Research (IPO) in Eindhoven ( The

Grammar of Dutch Intonation or GDI, cf. t Hart & Collier, 1975; t Hart, Collier & Cohen,

1990), and the more abstract, phonologically inspired autosegmental model formulated by

Gussenhoven (1988). There seems to be some controversy about the magnitude of thedifferences between the GDI and Gussenhovens model (cf. Collier, 1989; Gussenhoven,

1988; Ladd, 1983, 1996), but they do differ with respect to the definition of the basic melodic

shapes, and therefore it is of some relevance that the GDI is taken as point of departure in

the present investigation.

The GDI is based on the concept that intonation can be described in terms of sequences

of discrete pitch movements corresponding to voluntary actions on the part of the speaker.

Within this framework there is a direct relationship between accent and intonation: certain

pitch movements cause the syllable within which they are positioned to become accented,

whereas the autosegmental approach works with the notion ofassociation between a

morphological level at which tones are represented and a separate phonetic level at which

accents are represented. Furthermore, the GDI regards pitch movements (not pitch levels)

as the basic building blocks. Ten pitch movements are distinguished: five rises (1 5)

and five falls (AE), defined by perceptual features along four dimensions: direction,

timing with regard to the segmental structure, rate of change, and size. The grammar defines

the combinations of pitch movements that are possible, which result in a limited number

of pitch configurations containing one or two pitch accents, or a boundary marker (t Hart

et al., 1990). For example, a fast early full rise (1), directly followed by a fast late fullfall (A) produces the so-called pointed hat contour (1&A), the most common accent-

lending pitch configuration in Dutch. However, the grammar does not indicate how to choose

from the various types of melodic configurations, which means that there is no semantic

component other than the distinction between accent-lending and boundary-marking

configurations, and therefore ideas about meaning differences between Dutch pitch accent

types have to be sought elsewhere. by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from
http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/http://las.sagepub.com/


5/36


Analyses of the meaning of Dutch intonation

There are two theoretical analyses of the meaning of Dutch melodic shapes that are

formulated in terms of the manipulation of the information status of focused constituents:

the structuralist analysis by Keijsper (1984) and the autosegmental analysis by Gussenhoven

(1984). The latter analysis was developed for English, but since the intonation of Dutchcan be described in terms of the same phonological and morphological units as that of

English (cf. Gussenhoven, 1988, p. 96), this semantic analysis can be applied to Dutch

data as well.

The analysis presented by Keijsperin which there is a direct link between form

and meaning is based on the GDI and consists of eight meaningful melodic shapes

( type I type VIII), in which levels as well as movements are relevant. Three of the

configurations are involved in the marking of boundaries (types VI, VII, and VIII) and are

therefore not relevant here. The remaining five configurations have the meaning ofaccent

(the speaker communicates what he is and is not thinking about), and the type of accent

reveals whether or not what he thinks of exists (Keijsper, 1984, p.31). For example: a

type II pitch accent indicates that the speaker thinks of the focused information; at the

same time the next thought is announced, and this next thought is necessary for deciding

whether the focused information exists or not. A type III pitch accent, on the other hand,

indicates that the speaker knew that the focused information existed before the moment of

speaking.

Gussenhoven (1984) distinguishes three tonal morphemes or basic tones, described

as sequences of high (H) and low (L) tone segments: H*L (the fall), H*LH (the fall- rise)

and L*H (the rise). One of these tones has to be chosen at every accented position in the

segmental structure (asterisks specify the tonal segment that is to be associated with the

accented syllable on the segmental tier). [P]art of what intonation is used for is to mark

the status of the information conveyed by the sentence with respect to a background

(starting point) hypothesized by the speaker (Gussenhoven, 1984, p.200). Each tone

morpheme manipulates the background in a different way, that is, represents a different

meaning. For example: a H*L tone marks information as new, whereas a L*H tone leaves

open whether the information is new or not. All other melodic shapes are derived from thesethree basic tonal elements by rules and modifications, which add their own semantic features

to the meaning of the basic tones.

Gussenhoven further distinguishes different types of orientation of the tone

morphemes. In the default orientation the speaker refers directly to the referent of the

focused information, whereas he may also refer to some aspect of the speech situation (1984,

p.205). For example, a speaker can utter a proper name to point to a specific person I

think Marina is the best candidatean example of a default orientation, but he may also

utter the name to open a conversation Marina, can you spare a moment? which is anexample of what will be called a vocative orientation.1 In the present investigation,

addressing a person (vocative orientation) is contrasted with the default orientation because

1 Gussenhovens originaland potentially confusing term literal orientation is replaced by

default orientation; the vocative orientation is the first type of what Gussenhoven calls

metaphorical orientation (1984, p.205). by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


6/36

131J. Caspers

preliminary investigations indicated that there may be differences in suitability of pitch

accent types, depending on the orientation chosen. For instance, it was difficult to picture

situations where accent 1&A could be used to address a person, whereas it was easy to

picture a default orientation for 1&A.

MEANING HYPOTHESES

FOR FOUR DUTCH PITCH ACCENT TYPES

Melodic shapes

Four melodic shapes were chosen for experimental investigation, each containing one

pitch accent (which means that all pitch accents are utterance final):

(1) an accent-lending rise (followed by a boundary-marking rise)

(2) an accent-lending fall

(3) an accent-lending rise and fall on one syllable, and

(4) an accent-lending rise and a half fall on one syllable.

In Figure 1 the four shapes are illustrated, superimposed on a three-syllable word with

an accented middle syllable (capitalized). The letters and numbers correspond to the percep-

tually relevant pitch movements distinguished in the GDI. The GDI symbols represent:

low 0 and high 0/ declination, accent-lending rise 1 (steep, early in the syllable),

accent-lending fall A (steep, late in the syllable), accent- lending half fall E (steep, not

Figure 1

Stylized examples of the four pitch accent types on the proper name Marina.



7/36


reaching the low declination line), and & indicates that movements are combined on one

syllable.

The accent-lending rise 1 and the accent-lending fall A are the most frequently

used pitch movements in Dutch; they form the basic elements of the so-called hat pattern

contours (t Hart et al., 1990). The combination of the rise and fall on one syllable, 1&A

or pointed hat, is the standard pitch accent in Dutch, and a sequence of rise 1 and fall

A on separate syllables forms the so -called flat hat contour, a very common contour

with two accented syllables. Since the present investigation works with one-accent contours,

the flat hat is not included in the design, but the parts it is built of are. As a fourth contour

1&E is added: an accent-lending rise and half fall combined on one syllable. The inventory

of target shapes should at least be plausible to adherents of (Keijsper, 1984, and) the GDI,

while it may appear less likely to adherents of competing models (see also Table 1).

The single accent- lending rise (1) is followed by a boundary- marking rise on the

final syllable (GDI notation: 2) because a single accent-lending rise without such a

final rise seems unsuitable as a (turn) final contour. Selting (1996) presents empirical

evidence for a melodic turn -keeping cue in German, namely a contour ending in a level

pitch accent; furthermore, Caspers (1998) presents experimental evidence that the accent-

lending rise followed by level sustained pitch (10/ ) cannot readily be interpreted as

turn- final in Dutch. Since the experimental set-up requires that all contours are complete,

rise 2 was added to the incomplete rise 1 as default marker of f inality.

Hypothetical meanings suggested by Keijsper and GussenhovenKeijspers semantic analyses of the four shapes are presented in Table 1. An approximation

of the meanings of the four shapes in terms of Gussenhovens model is given also (for

some GDI shapes there is no straightforward translation into Gussenhovens melodic shapes;

see also Caspers, 1998). The GDI notation (leaving out the symbols representing high and

low declination), as well as the notations used by Keijsper and Gussenhoven are in bold

face. Only the intentional side of both analyses is presented (i.e., the attentional meaningthis is important information is taken for granted).

According to Keijsper the accent-lending rise (1) signifies that the speaker thinksof a certain referent (viz., the one that is focused by placing the pitch accent), but that the

following thought will give more information about the existence of that referent. The

precise meaning of the final rise 2 (type VII) is not clear, but it implies a boundary. In

Gussenhovens terminology the accent-lending rise (L*H) communicates that the speaker

is not sure whether the focused information is part of the knowledge shared between speaker

and listener, which seems analogous to Keijspers definition of the meaning of 1. In

more recent versions of Gussenhovens model (Gussenhoven, 1991; Van den Berg,

Gussenhoven, & Rietveld, 1992) L*H is by default followed by a high boundary tone(H%) as marker of finality. This can be taken as theoretical support for the addition of the

final rise 2 to 1 in the present investigation.

In Keijspers model the use of the accent-lending fall (A) indicates that the speaker

takes for granted that the referent exists; this means that the focused information is presented

as given. In contrast, 1&A introduces the existence of the referent, that is, adds a new

thought. 1&A equals H*L in Gussenhovens model, with the abstract meaning ofby Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


8/36

133J. Caspers

ADDITION, which parallels Keijspers definition of the meaning of 1&A. The accent-lending fall (A), however, poses a problem in Gussenhovens analysis, because an initial

stretch of high declination, followed by an early fall in the accented syllable, followed by

low declination in the final syllable (cf. Figure 1, shape 2) cannot easily be described in

his framework. A can only be accurately represented as part of a downstepped contour

(cf. Gussenhoven, 1991, p.149; Rietveld & Gussenhoven, 1995, pp.37677): !H*L. This

means that the meaning of a high onset as well as the meaning of downstep not given in

Gussenhoven (1984)should be added to the meaning of H*L (i.e., ADDITION).2

Finally, the meanings of 1&E given by Gussenhoven (inspired by Ladd, 1978) and

TABLE 1

Semantic analyses of the four accent- lending pitch configurations, as formulated by Keijsper

(1984) and Gussenhoven (1984)

Notation and Meaning

GDI Keijsper (1984) Gussenhoven (1984) Agreement?

notation

12 type IIplus type VII L*H H% yes

the existence of the referent is not TESTING: I choose not to commit

communicated and the next thought myself as to whether this Variable is

is announced (may be interpreted background (may mean: I ask you if

as a question) plus a boundary this is background) plus finality

A type III %H! H*L nothe existence of the referent was ADDITION: I add this Variable to the

projected before the moment of background (may mean: I tell you this

speaking (i.e., speaker introduces a is background) plus the meaning of a

known referent) high onset (%H) plus the meaning of

downstep (!)

1&A type I in final position H*L yes

the referent exists (i.e., speaker ADDITION: I add this Variable to the

realizes this at the moment of speak- background (may mean: I tell you this

ing, suitable when the referent is new) is background)

1&E type V H*Lplus stylization yes

the referent exists plus the infor- ADDITION: I add this Variable to the

mation contained in this accent is really background (may mean: I tell you this

superfluous (modification of type I) is background) plus This is a matter

of everyday occurrence/routine

2 Grabe et al. (1998) show that a high onset combined with a high pitch accent (e.g., %H H*L) is

judged as less favorable than the combination of a low onset with a high pitch accent (e.g.,

%L H*L): less friendly, less polite, more irritated, and more aloof. According to Ladd (1996)

downstep adds a nuance like finality or completeness (p.76), which is confirmed by recent

experimental results (Caspers, 1999). by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


9/36


Keijsper are basically the same. Note that 1&E is a special case in both analyses, namely

a modification of the prototypical rise-fall accent 1&A. Contour 1&E can be described

as a calling contour (cf., for example, Gibbon, 1976; Gussenhoven, 1993; Ladd, 1978,

1996), butas observed by Gibbon (1976) and Ladd (1978)it does not have to be used

to call someone, and calls do not have to have this contour.

Both authors essentially agree on the meaning of three of the four shapes: 1, 1&A,

and 1&E. However, they do not agree on the meaning of the single accent-lending fall

(A). According to Gussenhoven this shape has the same basic meaning as 1&A, namely,

introducing a new variable (plus the meaning of a high onset and downstep), whereas

Keijsper sees the single fall as indicating givenness. Acceptability ratings of sentences

containing explicitly new information (e.g.,Mijn huis staat in brand, My house is on

fire, an example from Keijsper, 1984, p.28) with contours 1&A or A reveal that the

former combination is acceptable, while the latter is unacceptable (Caspers, Van Heuven,

& Van Zwol, 1998). Therefore, Keijspers meaning hypothesis for A is adopted.

Form-meaning hypotheses

In this subsection a meaning hypothesis for each of the four types of accent-lending pitch

configuration is presented, adopting the terminology used by Gussenhoven (1984).

Speaker (S) and hearer (H) share a background, that is, they have some idea of

what they both know and what they do not know at the moment of speaking, on the basis

of explicit and implicit information. S may manipulate the shared background by putting

a Variable (V) into focus. The abstract manipulations TESTING and ADDITIONare used as in Gussenhovens model. For the background manipulation involving given

information, Gussenhovens notion of SELECTION is borrowed.3 Examples are inspired

by Gussenhoven (1984).

(1) 12 = testing: S leaves it up to H to determine whether V belongs to the background

or not (is V part of the background?); for example,Its a UNicorn?

(2) A = selection: S selects a V from the background (which means that V was present

in the background at the moment of speaking); for example, Yes, its a UNicorn

(3) 1&A = addition: S adds a V to the background (which means that V was not present

in the background at the moment of speaking); for example, See? Its a UNicorn!

(4) 1&E = addition plus it was predictable: V is added to the background, but that is

a matter of everyday routine; for example,I repeat: its a UNicorn

EXPERIMENTAL APPROACH

The goal of the present investigation is to test the relevance of the form- meaning- relation-ships presented above in an experimental setting. It is notoriously difficult to test the

contribution of intonation to the meaning of an utterance: [] our own melodic approach

3 In Gussenhovens model, the meaning SELECTION is attached to tone H*LH. His notions

ADDITION and SELECTION correspond withBrazils (1975)proclaimingand referringtones.by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


10/36

135J. Caspers

to intonation has produced an inventory of basic intonation patterns in Dutch and English,

but we still cannot answer the question how a speaker makes a selection among these basic

alternatives []. We speculate that these choices are influenced by the attitudinal meaning

that a speaker wants to add to the literal meaning of his utterances. But the actual encoding

of this attitudinal meaning into an individual pitch contour is evidently governed by so many

pragmatic and situational factors that we are still looking for a manageable experimentalparadigm in which to tackle this complicated issue (t Hart et al., 1990, p.114). According

to some people it is even impossible: [] even the non-specific, non- referential effects

exercised by intonation contours can be shown to be context-dependent to such a degree

that the attempt to extract from them an element of commonality valid in all contexts must

be reckoned a futile endeavor (Cutler, 1977, p. 106). Others believe that intonation does

make an independent and (experimentally) verifiable contribution to utterance interpre-

tation, which resulted in a number of concrete proposals regarding intonational function

(among others: Bolinger, 1986, 1989; Brazil, Coulthard, & Johns, 1980; Cruttenden,1986; Gussenhoven, 1984; Kowtko, 1996; Ladd, 1980; Pier rehumbert & Hirschberg,

1990; Pierrehumbert & Steele, 1989; Ward & Hirschberg, 1985, all for English; Keijsper,

1984 for Dutch; Grice, 1995 for Palermo Italian). Furthermore, there are experimental

investigations of intonational function (some recent examples are: Caspers & Van Heuven,

1995; Grabe, Gussenhoven, Haan, Marsi, & Post, 1998; Gussenhoven & Rietveld, 1991,

1997; Haan, Van Heuven, Pacilly, & Van Bezooijen, 1997a,b; Kirsner & Van Heuven,

1996; Kirsner, Van Heuven, & Van Bezooijen, 1994; Kohler, 1991; Sanders, 1996; Terken,

1993).

In the majority of these cases the experimental design involved isolated utterances

which had to be judged on some sort of (semantic) scale. The present investigation attempts

to verify the abstract meaning hypotheses by presenting listeners with target utterances

carrying different melodic shapes in specific contexts, asking for appropriateness judgments.

The situational contexts are supposed to reflect the abstract meanings hypothesized for the

various pitch accent types, by defining (part of) the background shared between speaker

and listener, creating a specific setting for the manipulation of this background that will be

signaled by the type of pitch accent realized on the target utterance. Formal semantics does

not offer a useful framework for attacking the problem of how to create these contexts,since it is recognized that the contribution of intonation to utterance interpretation is hardly

ever propositional and thus cannot readily be expressed in truth -conditional terms (Ladd,

1996; Ward & Hirschberg, 1985). Therefore, the contexts had to be constructed on the (not

very explicit) basis of the information status of a referent with respect to a background.

In addition to the four types of background manipulations presented in the previous

section, the experimental design incorporated two types of orientation of the manipulation

( default vs. vocative), which means that eight different types of context had to be

created. Proper names were chosen as target utterances, since these make a default as wellas a vocative orientation possible. The proper names, two male ( Cornelis and Jan -

Willem) and two female ones (Jolanda and Marina), consist mainly of voiced segments

to guarantee an optimally audible F0 contour, and have three syllables, of which the

middle one is stressed.

Table 2 contains a summary of the background characteristics that were to be captured

in the eight basic context categories. by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


11/36


According to Keijsper and Gussenhoven the manipulation testing expresses

indecision about the presence of the focused information in the background; when the focused

information is a person (default orientation), the background must incorporate this

uncertainty, so that testing becomes a relevant manipulation (Is person X present in the

background?). When the orientation is vocative, the relevance of addressing person X istested, which means that the description of the background must make clear that it is uncertain

whether addressing X is relevant or not. The manipulation selection involves selecting a

person from the background shared between speaker and listener; in the default orientation

thisperson should therefore be part of the background, whereas in the vocative orientation

the addressing of this person should be part of the background. In contrast, the manipu-

lation addition presupposes the absence of person X from the background in the default

orientation, and the absence of addressing this person in the vocative orientation. Finally,

the manipulation addition plus requires the same background descriptions as for addition,

but with the connotation that either the presence or the addressing of person X is predictable.

For each of the eight context types, three different situational descriptions were created.

The resulting 24 contexts are presented inAppendix 1. All contexts are situated in a school

and the speaker (S) is the teacher. Four examples are presented here, translated into English:

(a) You want to speak to a colleague about something important; this colleague, however,

is in the staff room talking to others. You join them and try:Marina.

In context (a) it is not clear whether it is possible to address Marina (vocative

orientation), and the speaker is expected to express this by using contour 12, testingthe relevance of his initiation.

(b) On the agenda are a number of classroom presentations; after a few introductory

remarks about the first subject you address the pupil concerned:Marina.

The addressing of the pupil (vocative orientation) is present in the background, since

everybody knows that Marina will be giving a presentation, so the prediction for context

TABLE 2

Relevant aspects of the background for the four types of intentional meaning, broken down by

orientation (default vs. vocative)

Orientation

Intentional meaning Default Vocative

(1) testing presence of person X in the no communicative situation with H,

background is unclear unclear whether addressing H is relevant

(2) selection person X is present in the communicative situation with H is part

background of the background

(3) addition person X is not present in the no communicative situation with H,

background addressing is new

(4) addition plus person X is not present in the no communicative situation with H,

background, but is predictable addressing is predictable



12/36

137J. Caspers

(b) is that contour A, selection, will be the most appropriate melodic shape.

(c) You are in a meeting with colleagues about the problem of ongoing thefts from the

cafeteria cash register. A number of aspects of the thefts are being discussed and

suddenly everything is clear to you; it has to be her:Marina.

In context (c) the speaker is referring to a person when uttering the proper name(default orientation). This person is considered not to be present in the shared background

(speaker and hearers were not thinking of Marina before the moment of speaking), which

leads to the prediction that contour 1&A, with the abstract meaning of addition, will

be the best fitting melodic shape in this context.

(d) A colleague has just asked you which pupil was absent that morning; you mentioned

the name Marina, but your colleague apparently has not listened well and asks: Who?

You reply, somewhat annoyed:Marina.

Apparently, the pupil referred to is not present in the shared background, whereas the

speaker thinks it should be, and expresses this by using contour 1&E: I add this

information to your background, but it was to be expected.

There are some difficulties with this particular way of structuring the background,

with the purpose of creating a setting which is suitable for one abstract manipulation only.

In principle, a speaker is free to present focused information as for example, new or

given, irrespective of the precise situational context. A closely related problem is the

fact that the notion shared background cannot be defined in exact terms, which meansthat the interpretation of what is shared background and what is not may differ from

person to person. For example, it can be argued that in example (c)Marina is given instead

of new information, because the names of all pupils are always present in the minds of

the teachers. Undoubtedly there are more contextual aspects which may influence the

responses of the subjects in an unpredictable way, which means that every context will

probably allow more than one, or maybe even all four contours.

To have some independent evidence for the association between context type and

abstract intentional meaning, a paper-and -pencil experiment was carried out. The purpose

of this experiment was to obtain a rough check of the relationship between the abstract

semantic notions of testing, selection, addition, and addition plus predictability

and the four groups of short texts.

Paper-and-pencil experiment

A short description of the four abstract meaning categories, as well as the 24 contexts

constructed, were presented on paper to a number of professional linguists, who were

requested to identify the abstract meaning best fitting each context (with forced choice from

the four alternative meanings). A positive association between predicted and responded

meaning categories was expected.

Method. Because of the abstract nature of the metalinguistic task to be performed, only

professional linguists participated in the paper-and-pencil experiment. They were presented

with a short written explanation of the theory presented above (including Table 2). Subjects

were then instructed to identify with the speaker and to choose the abstract meaning categoryby Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


13/36


which reflects best what the speaker does by pronouncing the proper name in this

particular situation. Note that the experiment was entirely mental, that is, no speech was

recorded or presented. Subjects were asked to indicate the orientation as well (addressing

someone vs. speaking about someone), which means that they had to decide for each context

which cell of Table 2 was being exemplified. Twenty- nine linguists, all native speakers of

Dutch, and not familiar with the meaning analyses of Keijsper and Gussenhoven (other than

what they had just learned from the instructions) participated in the experiment. The contexts

were presented in a single fixed quasi- random order.

Results. Results are presented in bar charts, using ANOVA as the statistical analysistechnique (in all analyses the significance level was set at .05). The data from four of the

29 subjects were excluded from the analysis, because they made two or more mistakes in

the indication of the orientation, which was taken as a sign that the task was not completely

understood, or was performed rather loosely (of the remaining 25 subjects four made one

mistake in indicating the orientation). The results of the experiment are presented in

Figure 2 (confusion matrices with frequency counts can be found in Appendix 2).

Figure 2

Paper-and-pencil experiment: Percentage of responded abstract meaning category (collapsed

across orientations), broken down by intentional meaning expressed by context; chance = 25%

(shaded).



14/36

139J. Caspers

For each of the four intentional meaning categories the relative frequency of responses

in the four abstract meaning categories is given. In 60% of the total number of cases the

predicted meaning category was chosen. In all categories the number of correct responses

is at least two times above chance (=25%). An analysis of variance was carried out on the

number of correct responses per subject, involving two crossed within-subject factors:

meaning expressed by context (four levels) and orientation (two levels). There is no

statistical difference in number of correct responses between the four types of meaning,

F(3,72)=1.14, p=.340, n.s., Huynh-Feldt Epsilon=1.00. In each of the meaning categories

the average number of correct responses is a little higher for the default orientation than

for the vocative orientation (F(1,24)= 7.00, p= .014; cf. Appendix 2), but there is no

significant interaction between meaning and orientation (F(3,72)= 1.69, p =.176, n.s.,

Huynh-Feldt Epsilon= .86). A possible explanation for the fact that the type of orientation

influences the number of correct responses can be found in the abstractness of the task

performed. It is probably quite normal for a linguist to think in terms of the presence orabsence of X in a background shared between S and H, but in the vocative orientation

subjects had to think in terms of manipulations of the background referring to the initiation

of a conversation, which is unusual and somewhat more complicated.

Conclusion. A number of subjects complained about the difficulty of the task; some of

them explicitly mentioned problems with identifying the background (what is shared

information, what information is not shared, which part of the background is relevant). Still,

the data present a reasonably clear picture, which is taken as evidence that the four context

types do reflect the abstract meaning they were designed to convey. When the results are

broken down by individual contexts, however (cf. Appendix 3), it becomes clear that some

contexts were more suitable than others: contexts D4a, V2a, V2c, and V4c elicited the

intended abstract meaning category from less than half of the subjects (note that three of

these problematic contexts have a vocative orientation). Since there was no obvious way

to improve the intention of these contexts they were in agreement with the abstract

hypotheses it was decided to maintain all 24 contexts in the perception experiment.4

PERCEPTION EXPERIMENT 1

The 24 contexts used in the paper-and-pencil experiment were presented again in a

perception experiment. Both naive and linguistically sophisticated subjects were asked to

project themselves into the speaker role in each situation, and to rank the four different

melodic versions of the target utterance from best to worst fit in their particular context. A

positive association between the contour type corresponding to the meaning captured in a

specific context and the judged appropriateness of that contour type was expected.

4 Also, a different design for the paper-and- pencil experiment did not lead to other results: when

the task was made less abstract, as suggested by an anonymous referee (for instance by presenting

a context and asking whether this person is new information for the hearer, instead of asking

the subject to select one of the four abstract notions that were introduced earlier), the results are

very close to that of the original paper-and- pencil experiment (this was established in a number

of pilot tests, which are not reported upon in this paper).by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


15/36


Method

Stimulus material. A female and a male Dutch intonologist read aloud the 24 contexts.

They also realized the four proper names with the four intonation contours; the resulting

32 melodic contours are presented in Appendix 4. For each presentation of a context plus

target utterances, the speaker of the former was not the same as the speaker of the latter,that is, when the context was produced by the female speaker, the corresponding four

versions of the proper name were produced by the male speaker, and vice versa. To enlarge

the number of data, each context was presented twice to the subjects, each time with a

different speaker and with a different accompanying proper name. There were four different

combinations of contexts and proper names available, and these were blocked over subjects.

The order of the contexts and the order of presentation of the four contours was varied

over subjects to obviate possible biases and counterbalance fatigue effects. For each

individual subject, however, the presentation order of the four different F0 contours was

fixed throughout the experiment (i.e., for every context the four different contours werepresented in the same order).

Subjects. Sixty- four native Dutch subjects participated in the experiment (27 males and

37 females). Their ages varied between 21 and 55; no hearing difficulties were reported.

Nine of the subjects were linguistically naive whilst the rest were students and professionals

30 phoneticians and 25 linguists. Of the 29 subjects who participated in the earlier paper-

and-pencil experiment, 23 participated in the perception experiment as well (the results in

the paper-and-pencil experiment of three of these were rejected, which means that there

are data for both experiments from 20 subjects).

Procedure. The experiment was organized such that it could be presented via the internet.5

The majority of the subjects participated via the internet, in response to an appeal issued

by e-mail. The rest did the experiment in the Leiden phonetics laboratory.

The experiment started with an extensive introduction, in which the subjects task was

explained, urging the reader to perform the experiment with utmost concentration, because

of the subtlety of the differences between the various contexts. Subjects were instructed to

assign each of the four target utterances a different rank (1 for best fit to 4 for worstfit; forced choice, no ties allowed). The contexts were presented one by one, switching to

the next context only when all four versions of the proper name were uniquely rated. The

program reported any mistake to the subjects, obliging them to adjust their scores (e.g.,

when two target utterances received the same number).

The situations were presented as text on the screen, and subjects could summon the

read-out version of the text as often as they felt necessary. Four icons were displayed

below the text, and clicking such an icon resulted in the auditory presentation of one of the

four melodic versions of the proper name. Four response buttons were presented next toeach icon, numbered from 1 to 4. Clicking such a button resulted in a change in color,

5 The experiment can be accessed at . The

presentation was programmed by Jos J.A. Pacilly of the Leiden University Phonetics Laboratory,

and runs on any Silicon Graphics Indy/Indigo workstation. Audio file conversion is necessary

when the experiment is to be run on other computers.by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


16/36

141J. Caspers

indicating that that particular melodic version was assigned that particular rank. Subjects

could correct their rankings by clicking on another button.

The experiment started with one practice - situation, and when this first screen

was correctly completed, the subject was asked a number of questions regarding age,

occupation/field of study, sex, native language, and hearing problems, after which the

experiment began. Generally, subjects needed between 30 and 60mins to complete the task.

Since a large number of subjects participated via the Internet, there was no strict

control over the circumstances under which the experiment was performed (such as environ-

mental noise, type of headphones, type of loudspeaker, etc.).

Results

The results are presented in bar charts, using ANOVA as the statistical test. Due to problems

with the network connection, a small number of cases (18 of the 3072) are missing from

the dataset. Confusion matrices are presented in Appendix 2.

Figure 3 presents for each context type the percentage responses favoring the contour

type predicted by the semantic analysis (further referred to as correct responses), broken

Figure 3

Perception Experiment 1: Percentage of correct responses for the two types of orientation, broken

down by intentional meaning expressed by context; chance=25% (shaded).



17/36


down by the two types of orientation. In 55% of the total number of cases the predicted

contour type was selected as the most appropriate one. An analysis of variance involving

two crossed within- subject factors ( meaning expressed by context and orientation)reveals that there is an effect of meaning (F(3,189)= 109.49, p < .001, Huynh-Feldt

Epsilon= .95), no main effect of orientation (F(1,63)=1.86, p=.177, n.s.), and interaction

between the factors meaning and orientation (F(3,189)= 130.22, p < .001, Huynh-

Feldt Epsilon= .98). This means that the percentages of correct responses depend on the

intentional meaning category (there are more correct responses in the testing category

than in, for instance, the addition plus category) and on the combination of meaning

category and the type of orientation (the testing contexts work very well in both orien-

tations, whereas the other three meaning categories receive a large number of correct

responses in one orientation, but not in the other).

In Figures 4 and 5 the relative frequency is given of the contour types judged as most

appropriate (i.e., ranked as number one), broken down by the four categories of meaning,

separately for both types of orientation.

Figure 4 reveals that in the default orientation contour 1&A is preferred in all

context types, except the testing category. In the vocative orientation (Figure 5) the

Figure 4

Perception Experiment 1: Percentage of preferred contour type in the default orientation, broken

down by intentional meaning expressed by context; predicted contour type indicated by arrow;

chance= 25% (shaded).



18/36

143J. Caspers

responses in the testing, selection, and addition plus contexts are as predicted,

whereas the preferences in the addition contexts are randomly dispersed over the four

contour types.Post hoc analyses reveal no significant effects of background of the subjects (t(1630)=

0.86, p= .390 for naive vs. linguistically trained subjects; t(1870)= 0.12, p= .906 for naive

vs. phonetically trained subjects; and t(2638)= 1.40, p =.162 for phonetically vs. linguis-

tically trained subjects), participation in the paper-and -pencil test, t(3070)= 1.62, p=.106,

or speaker of the stimulus utterances, t(3070)= 1.12, p=.261. Excluding those contexts

that did not perform as expected in the paper-and-pencil experiment does not substan-

tially influence the results of the perception experiment (cf. Appendix 5). The reason for

this is not entirely clear, but see the next subsection for further discussion.

Conclusion and Discussion

In contrast with the paper-and-pencil experiment subjects reported that they enjoyed the

test. This may be due to the more everyday-like situation in the perception experiment,

which resembles a normal speech situation more closely than the genuinely abstract task

that had to be performed in the paper-and-pencil experiment. There was no effect of subject

Figure 5

Perception Experiment 1: Percentage of preferred contour type in the vocative orientation, broken

down by intentional meaning expressed by context; predicted contour type indicated by arrow;

chance= 25% (shaded).



19/36


background (naive, phonetically trained, or linguistically trained) on the number of correct

responses, which means that for all subjects the task was equally difficult. Furthermore,

there was no effect of earlier participation in the paper-and-pencil test, indicating that

familiarity with the situational contexts did not facilitate the task to be performed.

For five out of the eight types of contexts there is a positive association with the corre-

sponding contour type. However, in the default orientation the selection, addition, and

addition plus contexts all elicited 1&A as the most appropriate contour, indicating

that contour types A and 1&E are not particularly acceptable in their specific contexts.

Nevertheless, the largest number of preferences for contour type A can be found in the

selection contexts and there are (virtually) no preferences for contour types 12 and

1&E. Furthermore, the only context type resulting in preference judgments for contour

type 1&E is addition plus. These findings indicate that under the noise, that is, the

preference for contour 1&A, there is some support for the hypotheses. In the vocative

orientation the context types behave as predicted, except for the addition contexts:responses are scattered over the four contour types, which means that the vocative addition

contexts do not discriminate between the four types of accent-lending contours.

There are a number of possible explanations for the unpredicted effects. Firstly, the

realization of the contours may be incorrect, that is, one or more contours resemble other

contours too closely. This possibility was checked in a small post -hoc experiment. A

single trained Dutch phonetician who was not aware of the purpose of the materials

transcribed the auditorily presented target utterances twice in the GDI notation, also judging

the acceptability of the melodic realization of each contour on a ten point scale. All 32 targetutterances were judged to be good realizations of the intended accent types, except for one

1&A contour, which was twice transcribed as 1&A2, a pointed hat followed by a

small final rise. This contour did not lead to deviating results, however. Furthermore, all

contours that were intended to contain a single accent-lending fall were transcribed as

5A0 by the judge, indicating that a small rise was perceived on the initial (unaccented)

syllable. However, since the A-contours do not have the characteristics of 5&A (which

can be taken as a variant of 1&A, cf. Keijsper, 1985, p.183; Vismans, 1994, p. 176)

inspection of the data confirms that the A-contours indeed contain an early fall, not a

late fall (cf. Appendix 4) it is concluded that the melodic realization of the target

utterances is satisfactory.

A number of intertwined aspects remain as possible explanations for the unpredicted

findings: the abstract meaning hypotheses themselves may not be correct, the contexts

may be inadequate, and there may be pragmatic problems.

In the default orientation all contexts, except the testing ones, lead to a preference

for contour 1&A, the default accent (preferred in approximately 50% of the total

number of cases, cf. Appendix 2, Table C). On the assumption that this accent type is themost neutral means of assigning prominence (cf. Keijsper, 1984), it would constitute a

convenient escape hatch in case of doubt on the part of the listener. This offers an explanation

for the choice for contour 1&A, but not for the fact that the contour types A and

1&E are shunned even where predicted.

The contexts designed to evoke contour A (selection) bear an element of fore-

knowledge: the pupil spoken about is present in the background (cf. Appendix 1: D2a-c).by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


20/36

145J. Caspers

By pronouncing the pupils name, the speaker thus focuses on an element which is given

information. However, the fact that it is this pupil and not another one may be viewed as

new information (i.e., the target utterance may be interpreted as elliptic), which means

that the speaker is free to present the proper name as either new or given. It may then be

the case that using contour A, that is, implying that the focused information is not new,

is felt to be a trifle impolite. The fact that the predicted A contour is judged as second

best supports this assumption: when both the first and second preference are taken into

account, the number of correct responses rises from 31% to 94% (cf. Appendix 3).6

The default addition plus contexts did not evoke contour type 1&E as the most

appropriate, which may be a result of the fact that it was assumed that 1&E expresses

an aspect ofpredictability, based on Gussenhovens analysis. Keijsper, however, chooses

a stronger wording: the information is really superfluous (cf. Table 1). It may be the case

that this is the better meaning analysis and that the contexts do not sufficiently express

this superfluousness, causing the subjects to divert to other contours, because choosingthe 1&E contour would be impolite (since it exposes the speakers feeling that what he

says is superfluous, i.e., suggesting that the listener is stupid). This assumption is supported

by the fact that the one context containing the term annoyed performed far better than

the other two (cf. context D4a, Appendices 1 and 3). The same pattern is present in the

vocative orientation (where the addition plus contexts generally performed rather well,

which may be explained by the fact that mainly pupils are addressed there, whereas in the

default orientation colleagues are spoken to): here the results are also better for the contexts

expressing annoyance (V4a and c). This means that the abstract meaning of the 1&Econtour and, subsequently, each relevant contexthas to be revised. The overly general

(i.e., predictable rather than superfluous) meaning of 1&E may explain the fact

that excluding those contexts from the dataset that did not lead to the predicted responses

in the paper-and-pencil experiment, did not substantially influence the results of the

perception experiment. The addition -plus -predictability def inition used in the paper

test made it plausible for subjects to choose this abstract meaning category in some of the

cases where selection was intended (i.e., V2a and c), whereas 1&E was not suitable

for these contexts in the perception experiment; furthermore, the stronger addition -

plus-predictability contexts did not lead to the predicted responses in the paper test, but

they performed as expected in the perception test, whereas the weaker addition -plus-

predictabili ty contexts led to the predicted responses in the paper test but not in the

perception experiment.

In the vocative orientation the results are more in line with the predictions, except for

the addition category. Surprisingly, the pointed hat contour (1&A) is preferred in only

12% of the total number of vocatively oriented contexts (cf. Appendix 2, Table D), which

indicates that the default pitch accent is not very suitable to initiate a conversation. It may

be the case that it is felt to be rather impolite to bluntly add the initiation of a conversation

6 As can be inferred from the data presented in Appendix 3, such a drastic change in number of

correct responses, only occurs for the selection contexts in the default orientation; for all other

context types including the second preference only results in a moderate increase in the total number

of correct responses. by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


21/36


to the background. However, it may also be the case that addition is an illogical manipu-

lation when addressing someone, because normally there is already some form of (nonverbal)

communication between speaker and hearer.

Summarizing, there is one form- meaning hypothesis that does not seem to be correct:

contour 1&E versus addition plus predictability. For the other three form- meaning

hypotheses no such conclusion could be drawn. Therefore the abstract meaning of 1&E

was strengthened to addition plus superfluousness and tested again in a second perception

experiment, which also served as verification of the results found for the other three form-

meaning units.

PERCEPTION EXPERIMENT 2

The same materials were used, except for the addition plus contexts, which were altered

to fit the adjusted meaning hypothesis for 1&E. This time the subjects were presentedwith a proper name bearing one of the four contours under investigation, as well as four

different situational contexts (one for each type of intentional meaning), which they had

to rank from the most to the least appropriate environment for the presented stimulus

utterance. It was hypothesized that the results would resemble those of Perception

Experiment 1, except for the addition plus contexts: the match between the adjusted

contexts and contour type 1&E was expected to be better than in Perception Experiment 1.

MethodStimulus material. The stimulus materials from Perception Experiment 1 were used, except

for the addition plus contexts, which were altered when lacking an element of annoyance

(cf. Appendix 6). All combinations of contour type, proper name, and orientation were

presented once to the subjects, blocking the order of presentation, the speaker, and the

combination of contexts per stimulus over subjects.

Subjects. Thirty- six subjects participated in the experiment, aged between 18 and 43,

fourteen males and 22 females, with no self- reported hearing deficiencies. Eight subjects

were linguistically naive, 10 were phoneticians, and the remaining 18 were linguists. Nonetook part in the previous experiments.

Procedure. The organization of Perception Experiment 2 resembled the procedure of the

previous experiment as closely as possible.7

Results and Conclusion

Results are presented in Figures 6 and 7. Again a small number of cases (7 of the 1152) is

missing from the dataset, due to a computer crash. In 59% of the total number of cases the

predicted context type was selected as the most appropriate one. Confusion matrices are

presented in Appendix 2. An analysis of variance involving two crossed within -subject factors

(contour and orientation) reveals effects of contour (F(3,105)=19.06, p


22/36


23/36


vs. phonetically trained subjects and t(1207)= 0.84,p=.403 for phonetically vs. linguis-

tically trained subjects) or speaker of the stimulus utterances (t(1463)= 1.78,p=.075).

In conclusion, the results of Experiment 2 support the revised meaning hypothesis

for 1&E; otherwise, the results obtained in Experiment 1 are confirmed.

GENERAL DISCUSSION

The experiments presented in this paper were intended to test abstract meaning hypotheses

for four Dutch pitch accent types, stemming from the theoretical analyses formulated by

Gussenhoven (1984) and Keijsper (1984). The results of the experiments support part of

the hypotheses.

The association between the 12 and 1&E contours and the corresponding(adjusted) situational contexts is fairly strong, whereas the results in the selection and

addition categories are less clear. In the default orientation the selection and addition

contexts and the A and 1&A contours behave as one group (cf. Figures 4 and 6),

indicating that subjects perceive no difference between these stimuli, at least not in terms

of the semantic categories at issue here. When addressing a person (vocative orientation),

however, there is a clear and consistent difference between these groups, in both experiments.

Figure 7

Perception Experiment 2: Percentage of preferred context type in the vocative orientation, broken

down by contour type; predicted context type indicated by arrow; chance= 25% (shaded).



24/36

149J. Caspers

The highest scores in the present investigation are found in the testing category,

suggesting that testing is an adequate description of the meaning of contour 12. Further

investigations of both the accent-lending rise 1 and the boundary-marking rise 2 in

combination with other melodic shapes (e.g., 1A, 1&A2) are necessary to shed more

light on the precise meaning of individual 1 (and 2).

The finding that the 1&A contour was preferred in both the addition and

selection contexts in the default orientation (Experiment 1) was related to possible

politeness effects. However, the fact that in Experiment 2 there is, again, no difference

between the two categories indicates that there may be a problem with either the abstract

meaning hypotheses or the situational contexts. Since the A contour and the selection

contexts perform well in the vocative orientation in both experiments (cf. Figures 5 and 7)

a finding that cannot be explained by the alternative hypothesis that 1&A and A

share the abstract meaning of addition (Gussenhoven, 1984, cf. Table 1)the contexts

seem to be the crucial factor. Maybe all target utterances in the default selection andaddition contexts could be interpreted as carrying new information as well as given

information (cf. the discussion of the results of Experiment 1). Assuming that this is the

case, that is, that the difference between the two types of contexts is too small, the findings

in Experiment 1 could be explained by a general preference for the, possibly more neutral,

1&A contour; in the second perception experiment no preference is visible anymore.

In the vocative orientation the behavior of the selection data is perfectly in line

with the predictions in both tests, which indicates the adequacy of the description of the

intention expressed by contour A. However, the random responses in the additioncategory seem to weaken the tenability of the meaning hypothesis proposed for 1&A.

When presented with a vocative addition context, subjects do not have a clear preference

for either of the four contour types, which could mean that the combination of such a context

and a proper name is odd in itself (i.e., irrespective of pitch accent type). The second

perception test indeed shows that the addition context type is not very popular overall.

The precise cause of the problematic relationship between contour 1&A and a vocative

orientation is not entirely clear, but it seems related to the somewhat paradoxal relationship

between use of a proper name (presupposing a rather intimate relationship between speakerand listener) and creating a new (i.e., unpredictable) communicative situation. As mentioned

earlier, it proved rather difficult to envisage situations where using a proper name to address

a person constitutes the addition of new information to the background shared between

speaker and listener. Further research in this field is necessary.

The revised abstract meaning for 1&E seems to be supported by the results of

Experiment 2: in both orientations the addition plus superfluousness contexts are matched

with the 1&E contours in more than two-thirds of the cases.

Results reveal rather large effects of (and interaction with) the distinction between adefault and a vocative orientation. Generally speaking, the appropriateness of contour

types A and 1&A depends on the orientation of the situational context. Adding new

information (contour 1&A) seems a prototypical default manipulation, whereas selecting

information from a background (contour A) seems a prototypical vocative manipulation.

These findings suggest that the orientation of the manipulation should be carefully considered

when investigating the contribution of intonation to utterance interpretation.by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


25/36


As explained in the introduction, testing nonlexical meaning hypotheses is a difficult

task. The results emerging from such experiments will never be clear- cut. For all the inherent

fuzziness of the present data, however, a relatively clear association emerges between the

information status of a referent as indicated by a description of the background shared

between speaker and listener, and the use of a specific pitch accent type. This indicates

that, in addition to marking a certain part of the utterance as important, different types of

pitch accent indeed express different intentions, as asserted by nonexperimental analyses

of the meaning of intonation (Gussenhoven, 1984; Keijsper, 1984; Pierrehumbert &

Hirschberg, 1990). Abstract meaning propositions as formulated by Gussenhoven and

Keijsper may explain aspects of the use of certain Dutch pitch accent types. As for now,

Keijspers (1984) meaning analysis seems to lead to slightly better predictions, but further

perception and production research on the meaning of Dutch melodic shapes is essential,

especially on 1&A and A, employing semantic scales in addition to the current experi-

mental design.

Received: February 17, 1997; revised manuscript received: June 26, 1998;

accepted: September 8, 1999

REFERENCES

BOLINGER, D. L. (1975).Aspects of language. New York: Harcourt Brace Jovanovich.

BOLINGER, D. L. (1986).Intonation and its parts: Melody in spoken English. London: EdwardArnold.

BOLINGER, D. L. (1989).Intonation and its uses: Melody in grammar and discourse. London: Edward

Arnold.

BRAZIL, D. (1975).Discourse intonation I. English Language Research, Birmingham University.

BRAZIL, D., COULTHARD, M., & JOHNS, C. (1980).Discourse intonation and language teaching.

London: Longman.

CASPERS, J. (1994). Pitch movements under time pressure. The Hague: Holland Academic Graphics.

CASPERS, J. (1997). Testing the meaning of four Dutch pitch accent types. In G. Kokkinakis, N.

Fakotakis, & E. Dermatas (Eds.), Proceedings of the 5th European Conference on Speech

Communication and Technology (pp.863 866). Patras: University of Patras.CASPERS, J. (1998). Whos next? The melodic marking of question versus continuation in Dutch.

Language and Speech, 41, 371 394.

CASPERS, J. (1999). An experimental investigation of meaning differences between the early and

the late accent-lending fall in Dutch. In R. van Bezooijen & R. Kager (Eds.), Linguistics in the

Netherlands, 1999 (pp.2739). Amsterdam: John Benjamins.

CASPERS, J., & Van HEUVEN, V. J. (1995). Effects of time pressure on the choice of accent-lending

and boundary-marking pitch configurations in Dutch. In J. M. Pardo, E. Enrquez, J. Ortega, J.

Ferreiros, J. Macas, & F. J. Falverde (Eds.), Proceedings of the 4th European Conference on Speech

Communication and Technology (pp.10011004), Madrid.CASPERS, J., Van HEUVEN, V. J., & Van ZWOL, N. (1998). An experimental investigation of meaning

differences between the pointed hat contour and the accent-lending fall in Dutch. In R. van

Bezooijen & R. Kager (Eds.), Linguistics in the Netherlands, 1998(pp.65 79). Amsterdam:

John Benjamins.

COLLIER, R. (1989). On the phonology of Dutch intonation. In F. J. Heyvaert & F. Steurs (Eds.),

Worlds behind words (pp.245259). Leuven: Leuven University Press.

CRUTTENDEN, A. (1986).Intonation. Cambridge, U.K.: Cambridge University Press.by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


26/36

151J. Caspers

CUTLER, A. (1977). The context-dependence of intonational meanings. In W. A. Beach, S. E. Fox,

& S. Philosoph (Eds.), Papers from the Thirteenth Regional Meeting, Chicago Linguistic Society

(pp.104115). Chicago: Chicago Linguistic Society.

CUTLER, A. (1991). Prosody in situations of communication: Salience and segmentation.Proceedings

of the XIIthInternational Congress of Phonetic Sciences, Vol. 1 (pp.264270). Aix-en-Provence,

Universit de Provence.CUTLER, A., DAHAN, D., & Van DONSELAAR, W. (1997). Prosody in the comprehension of spoken

language: A literature review.Language and Speech, 40, 141 201.

GIBBON, D. (1976). Perspectives of intonation analysis. Bern: Lang.

GRABE, E., GUSSENHOVEN, C., HAAN, J., MARSI, E., & POST, B. (1998). Preaccentual pitch and

speaker-attitude in Dutch.Language and Speech, 41, 6385.

GRICE, M. L. (1995). The intonation of interrogation in Palermo Italian: Implications for intonation

theory. Tbingen: Niemeyer.

GROSZ, B. J., & SIDNER, C. L. (1986). Attention, intentions, and the structure of discourse.

Computational Linguistics, 12, 175 204.

GUSSENHOVEN, C. (1984).On the grammar and semantics of sentence accents. Dordrecht: Foris.GUSSENHOVEN, C. (1988). Adequacy in intonation analysis: The case of Dutch. In H. van der Hulst

& N. Smith (Eds.),Autosegmental studies on pitch accent(pp.95 121). Dordrecht: Foris.

GUSSENHOVEN, C. (1991). Tone segments in the intonation of Dutch. In T. F. Shannon & J. P. Snapper

(Eds.), The Berkely Conference on Dutch Linguistics, 1989 (pp.139155). Lanham MD: University

Press of America.

GUSSENHOVEN, C. (1993). The Dutch foot and the chanted call.Journal of Linguistics, 29, 3763.

GUSSENHOVEN, C., & RIETVELD, A. C. M. (1991). An experimental evaluation of two nuclear-

tone taxonomies.Linguistics, 29, 423 449.

GUSSENHOVEN, C., & RIETVELD, A. C. M. (1997). Empirical evidence for the contrast between L*and H* in Dutch rising contours. In A. Botinis, G. Kouroupetroglou, & G. Carayiannis (Eds.),

Intonation: Theory, models, and applications (Proceedings of an ESCA Workshop, pp.169

172). Athens: ESCA and University of Athens.

HAAN, J., Van HEUVEN, V. J., PACILLY, J. J. A., & Van BEZOOIJEN, R. (1997a). An anatomy of

Dutch question intonation. In J. Coerts & H. de Hoop (Eds.),Linguistics in the Netherlands, 1997

(pp.97 108). Amsterdam: John Benjamins.

HAAN, J., Van HEUVEN, V. J., PACILLY, J. J. A., & Van BEZOOIJEN, R. (1997b). Intonational

characteristics of declarativity and interrogativity in Dutch: A comparison. In A. Botinis, G.

Kouroupetroglou,& G. Carayiannis (Eds.), Intonation:Theory, models, and applications(Proceed-

ings of an ESCA Workshop, pp.173176). Athens: ESCA and University of Athens.t HART, J., & COLLIER, R. (1975). Integrating different levels of intonation analysis.Journal of

Phonetics, 3, 235 255.t HART, J., COLLIER, R., & COHEN, A. (1990).A perceptual study of intonation. Cambridge, U.K.:

Cambridge University Press.

KEIJSPER, C. E. (1984). Vorm en betekenis in Nederlandse toonhoogtecontouren[Form and meaning

in Dutch pitch contours]. Forum der Letteren, 25, 2037, 113126.

KEIJSPER, C. E. (1985).Information structure. Amsterdam: Rodopi.

KIRSNER, R. S., & Van HEUVEN, V. J. (1996). Boundary tones and the semantics of the Dutch final

particles h, hoor,zeg and joh. In M. den Dikken & C. Cremers (Eds.),Linguistics in theNetherlands, 1996(pp.133 146). Amsterdam: John Benjamins.

KIRSNER, R. S., Van HEUVEN, V. J., & Van BEZOOIJEN, R. (1994). Interaction of particle and prosody

in the interpretation of factual Dutch sentences. In R. Bok -Bennema & C. Cremers (Eds.),

Linguistics in the Netherlands, 1994 (pp.107 118). Amsterdam: John Benjamins.

KOHLER, K. J. (1991). Terminal intonation patterns in single-accent utterances of German: Phonetics,

phonology and semantics.Arbeitsberichte Institut fr Phonetik und digitale Sprachverarbeitung

Universitt Kiel, 25, 115 185. by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


27/36


KOWTKO, J. C. (1996). The function of intonation in task-oriented dialog. Unpublished doctoral

dissertation, University of Edinburgh.

KRAAYEVELD, J. (1997).Idiosyncrasy in prosody. Nijmegen: University of Nijmegen.

LADD, D. R. (1978). Stylized intonation.Language, 54, 517 540.

LADD, D. R. (1980). The structure of intonational meaning: Evidence from English. Bloomington, IN:

Indiana University Press.

LADD, D. R. (1983). Phonological features of intonational peaks.Language, 59, 721759.

LADD, D. R. (1992). An introduction to intonational phonology. In G. Docherty & D. R. Ladd (Eds.),

Papers in laboratory phonology II: Gesture, segment, prosody (pp.321334). Cambridge, U.K.:

Cambridge University Press.

LADD, D. R. (1996).Intonational phonology. Cambridge, U.K.: Cambridge University Press.

LADD, D. R., SCHERER, K. R., & SILVERMAN, K. (1986). An integrated approach to studying

intonation and attitude. In C. Johns- Lewis (Ed.),Intonation in discourse(pp.125 138). London:

Croom Helm.

LIBERMAN, M. (1979). The intonational system of English. New York: Garland.

PIERREHUMBERT, J. B., & HIRSCHBERG, J. (1990). The meaning of intonational contours in theinterpretation of discourse. In P. R. Cohen, J. Morgan, & M. E. Pollack (Eds.),Intentions in commu-

nication (pp.271 311). Cambridge, MA: MIT Press.

PIERREHUMBERT, J. B., & STEELE, S. A. (1989). Categories in tonal alignment in English.Phonetica,

46, 181 196.

RIETVELD, T., & GUSSENHOVEN, C. (1995). Aligning pitch targets in speech synthesis: Effects of

syllable structure.Journal of Phonetics, 23, 375 385.

SANDERS, M. J. (1996).Intonation contour choice in English. Utrecht: Research Institute for Language

and Speech.

SELTING, M. (1996). On the interplay of syntax and prosody in the constitution of turn-constructional

units and turns in conversation.Pragmatics, 6, 357 388.

TENCH, P. (1996). The intonation systems of English. London: Cassell.

TERKEN, J. M. B. (1993). Human and synthetic intonation. In V. J. van Heuven & L. C. W. Pols (Eds.),

Analysis and synthesis of speech (pp.241259). Berlin: Mouton de Gruyter.

ULDALL, E. (1972). Dimensions of meaning in intonation. In D. Bolinger (Ed.),Intonation (pp.250

259). Harmondsworth: Penguin.

Van Den BERG, R., GUSSENHOVEN, C., & RIETVELD, T. (1992). Downstep in Dutch: Implications

for a model. In G. J. Docherty & D. R. Ladd (Eds.), Papers in laboratory phonology II: Gesture,

segment, prosody (pp.335359). Cambridge, U.K.: Cambridge University Press.

VISMANS, R. M. (1994).Modal particles in Dutch directives: A study in functional grammar. Un-published doctoral dissertation, University of Amsterdam.

WARD, G., & HIRSCHBERG, J. (1985). Implicating uncertainty.Language, 61, 747 776.

APPENDIX 1

Contexts

In this appendix the 24 Dutch contexts used in Experiment 1 are presented, with a translation

in English, and without the accompanying proper name (since these were varied). D standsfor default orientation, V for vocative orientation, the numbers represent the intended

meaning category (1: testing, 2: selection, 3: addition, and 4: addition plus

predictability ), and a to c indicate the various versions of each context type.

D1a Je praat met een collega over een mogelijke schuldige aan diefstal; je denkt dat Marina

de schuldige is, maar je bent er helemaal niet zeker van; je wilt verifiren of die collega

je vermoeden deelt: by Sorin Ciutacu on October 31, 2009http://las.sagepub.comDownloaded from


28/36

153J. Caspers

[You are talking to a colleague about a possible party guilty of theft; you think that

Marina is the offender, but you are not very sure; you want to verify whether your

colleague shares your suspicion:]

D1b Je leerlingen vertellen je dat Marina niet zal komen vandaag omdat ze naar het

Binnenhof moet voor een vergadering van de vaste-kamercommissie voor onderwijs.Je bent uiterst verbaasd: is dit een grapje?:

[Your pupils tell you that Marina will not attend classes today because she has to go

to the Houses of Parliament for a meeting of the permanent education committee. You

are utterly astonished: is this a joke?:]

D1c Je neemt deel aan een docentenvergadering. Er moet een leerling worden benoemd in

het schoolbestuur. Een aantal kandidaten wordt geopperd door je collegas en je hebt

zelf iemand in gedachten waarvan je absoluut niet weet hoe die persoon zal vallen bij

de rest; je doet een voorzichtige suggestie:[You are attending a staff meeting. A pupil has to be appointed to the school adminis-

tration. A number of candidates are put forward by your colleagues and you yourself

have someone in mind of whom you are absolutely unsure whether that person will be

acceptable to the others; you offer a tentative suggestion:]

D2a Je bent tekenleraar en je geeft antwoord op een vraag van een collega wie van twee

leerlingen beter kan tekenen, Marina of Cornelis:

[You are an art- teacher and you answer a question of a colleague which of two pupils

draws better, Marina or Cornelis:]

D2b Je neemt deel aan een rapportvergadering over een specifieke klas; er is sprake van

het toekennen van een boekenbeurs aan de leerling met de beste cijfers. Als je aan de

beurt bent geef je de naam van de leerling met de beste resultaten in jouw vak:

[You are attending a staff meeting about a specific class; there is talk of awarding a

book grant to the pupil with the best marks. When it is your turn you mention the

name of the pupil with the best results in your subject:]

D2c Je neemt deel aan een docentenvergadering. Er moet een leerling worden aangewezenals klassevertegenwoordiger. Er is een lijstje kandidaten beschikbaar, en als je aan de

beurt bent noem je jouw voorkeur:

[You are attending a staff meeting. A pupil has to be appointed class representative.

There is a list of candidates available, and when it is your turn you mention your

preference:]

D3a Je bespreekt met een collega welke leerlingen geschikt zouden zijn om te betrekken bij

het opzetten van een schoolkrant; er schiet je eentje te binnen (die journalist wil worden):

[You are s

Date post:	14-Apr-2018
Category:	Documents
Upload:	sorin-ciutacu
View:	223 times
Download:	0 times

Intonation in Dutch

Documents