Date post: | 01-Jan-2017 |
Category: |
Documents |
Upload: | trannguyet |
View: | 212 times |
Download: | 0 times |
1
Title: Some properties of light verbs in code-switching*
Authors: Kay González-Vilbazo
University of Illinois at Chicago
Dept. Spanish, French, Italian & Portuguese
#1708 UH/MC 315
601 S. Morgan St.
Chicago, IL 60607-7117
USA
Luis López (University of Illinois at Chicago)
University of Illinois at Chicago
Dept. Spanish, French, Italian & Portuguese
#1703 UH/MC 315
601 S. Morgan St.
Chicago, IL 60607-7117
USA
2
Abstract:
This article addresses some puzzles regarding the properties of light verb
constructions that have been amply documented in many code-switching pairs , with
particular focus on some well-known asymmetries. This article should be viewed as
an argument that generative analyses of code-switching are conceptually and
empirically superior to the mainstream approach, referred to as the Matrix Language
Frame Model.
Keywords:
Code-switching, Light verb, little v, Minimalist Program, Spanish, German, Matrix
Language Frame Model
3
1. Code-switching and light verbs
1.1 Introduction
Many code-switching varieties include a light verb as part of their grammatical
repertoire. Here we focus on a German/Spanish code-switching variety in which a
light verb taken from the Spanish lexicon hacer (�‘do�’) selects a German lexical verb
in the infinitive form.
(1) Juan hace nähen das Hemd.1 Juan does sew the shirt �‘Juan sews the shirt.�’
The hacer + V construction presents some interesting puzzles that appear in other
code-switching varieties and at least one of them also appears in monolingual light
verb constructions. This pervasiveness suggests that they reflect properties of the
human language rather than idiosyncratic features of a particular variety and therefore
demand an account. The puzzles are the following:
1. Asymmetry: the light verb can only be realized in one language (Spanish in our
case) and not in the other (German in our case). This type of asymmetry has been
described also for the following code-switching pairs: Den Dikken & Rao (2003):
Telugu/English, Joshi (1985): Marathi/English, Ritchie & Bhatia (1996):
Hindi/English, Boeschoten & Verhoeven (1985): Turkish/Dutch.
2. The light verb construction cannot be used in passive voice. As described by
Karimi-Doostan (2005), this is also the case in light verb constructions in
1 All code-switching examples are Spanish/German unless otherwise noted. Following standard conventions, lexical elements of one language (Spanish) are italicized. Many of the examples in this article involve the Spanish verb hacer. Please note that hacer is an irregular verb with a stem hiz- for preterite tenses and hech- for the past participle.
4
monolingual grammars. We do not know if it is possible in other code-switching
pairs.
3. Feature spreading: the (German) VP selected by the (Spanish) light verb has
features typical of Spanish with regards to at least three properties: word order,
prosody and information structure. Feature spreading for word order has been
described by the following authors and code-switching pairs: Myers-Scotton & Jake
(2009): Swahili/English, Den Dikken & Rao (2003): Telugu/English, Joshi (1985):
Marathi/English, Ritchie & Bhatia (1996): Hindi/English, Boeschoten & Verhoeven
(1985): Turkish/Dutch, Canfield (1980): Navajo/English, Bavin & Shopen (1985):
Warlpiri/English, Annamalai (1978): Tamil/English, Stanlaw (1982):
Japanese/English
4. Absence in the input grammars: hacer + V cannot be used as a light verb
construction in monolingual Spanish. There is no equivalent construction in
monolingual German either.
We propose an account for all four puzzles using as tool-kit a set of assumptions
that are standard in generative syntax as developed in the Minimalist Program
(Chomsky 1995 et seq.). In particular, we do not use any special mechanisms that
regulate code-switching, as in, among others, Myers-Scotton�’s Matrix Language
Frame Model (Myers-Scotton 1993, 2001, 2002, Myers-Scotton and Jake 2009). In
fact, we argue that the latter model is unable to make any predictions with respect to
the data at hand.
The article is organized as follows. In the remainder of section 1 we introduce
the phenomenon of code-switching and the light verb construction. In section 2 we
present the four puzzling properties of light verbs in code-switching. Sections 3 and 4
describe our data and methodology as well as our theoretical assumptions. In sections
5
5 �– 8 we discuss the puzzles and propose an account for each of them. Section 9
presents the alternative Matrix Language Frame Model account of the puzzles and
compares it to our own account. Finally, section 10 summarizes our conclusions.
1.2 Code-Switching
Code-switching is very common in bilingual communities. Code-switching may even
take place within the boundaries of a sentence, apparently without effort or any
unusual pauses at the switching sites. The following example was uttered by a
Spanish/German bilingual code-switcher at the German school of Barcelona
(González-Vilbazo 2005):
(2) Wir utilisieren spanische Wörter, die dann alemanisiert werden y We use Spanish words that then Germanized are and
hacen klingen un poco raro.2 do sound a bit strange �’We use Spanish words, that are then Germanized and sound a bit strange.�’
We take code-switching to be an I-language phenomenon, i.e. an expression of a type
of linguistic competence. Code-switchers are able to produce consistent
grammaticality judgments on sentences such as (1), which reveal an underlying
linguistic system. Our language consultants are highly competent bilinguals, by which
we mean that their grammaticality judgments in each language do not differ from
those of monolingual native speakers (for more details see below in section 3).
Our approach to code-switching can be aligned with other generative studies on
the subject, such as Woolford (1983), Di Sciullo et al. (1986), Belazi et al. (1995),
Mahootian and Santorini (1996), MacSwan (1999) and González-Vilbazo (2005). As
2 This example was provided by a student of the German School of Barcelona when asked to describe how his code-switching works.
6
opposed to some well known work like Joshi (1985) and Myers-Scotton (1993 et
seq.), we assume that there is no third grammar involved in code-switching, i.e. there
are no specific rules, structures, mechanisms or operations built into the language
faculty in order to regulate code-switching. This does not necessarily entail that the I-
language of code-switchers will be identical to the union of the two grammatical
systems: code-switchers may include features drawn directly from Universal
Grammar which are absent in the component grammars.
Within a minimalist framework, one is led to assume that the account of the I-
language of a bilingual speaker should not lie in the structure of the computational
system, which is expected to be universal (see Mahootian & Santorini 1996 and
MacSwan 1999 for a clear argumentation). Instead, what is distinctive of bilingual
speakers is that they have functional and lexical items belonging to two different
lexica.
Code-switching is often regarded linearly, as a performance phenomenon. Thus,
code-switching would be a literal back and forth between two languages. This view is
best represented by Poplack�’s (1980) otherwise groundbreaking article �“Sometimes
I�’ll start a sentence in Spanish Y TERMINO EN ESPAÑOL�” (sic).
(3)(a) Linear representation: performance approach (�‘>�’ sequencing operator) L1>L1>L1>L1>L1/ L1>L1>L1>L1/ L2>L2/ L2>L2.
Our own view is best represented in (3)(b) where we can see that items from the two
languages contribute to the building of a linguistic structure.
7
(3)(b) Structural representation: competence
A common problem when analyzing code-switching is the difficulty of distinguishing
it from borrowing. Conceptually the difference seems clear: the borrowed string is
grammatically (phonologically, morphologically and syntactically at least) integrated
into the host language. In code-switching, on the contrary, all language strings remain
unintegrated and retain their original grammatical properties. Empirically, however, it
is often very hard to tell the difference. Since this is a relevant issue for our discussion
of the asymmetry puzzle, we address it in detail in section 5.2.
1.3 Light verbs
The students of grammatical phenomena in contact situations have pointed out
that code-switched utterances oftentimes split the clausal predicate into two distinct
verb complexes. The first one is a variant of �‘do�’ and it appears fully conjugated. The
CP
Spec C'
C°
L1
TP
Spec
L2
T'
T°
L1
vP
Spec
L2
v'
v°
L1
VP
Spec
DO L1
V'
V°
L2
XP
IO L2
8
second one is a verbal infinitive or a bare root (Butt 2003, González-Vilbazo 2005).
Following a long-standing tradition, we refer to the conjugated word as light verb.3
Sentence (4) is an example of Spanish/German code-switching. The word hizo
(from hacer �“to do�”) is drawn from the Spanish lexicon. In combination with a
German infinitive verb as a complement hacer is re-lexified as a light verb (from
González-Vilbazo & López, forthcoming).
(4) Juan hace nähen das Hemd Juan does.3rd sew.inf the shirt �’Juan sews the shirt.�’ The following is a Turkish/Dutch example taken from Myers-Scotton and Jake
(2009), who attribute it to Backus (1992):
(5) O diyor ben uitmaken yaptm diyordu kznam. he says 1st.sg finish.inf did.1st say.3rd.sg girl.with He says �‘I broke up with a girl.�’ Light verb constructions are not a prerogative of code switching varieties. Some
monolingual grammars also have them in their repertoire. The following is a Persian
example taken from Karimi-Doostan (2005):
(6) John narmesh kard. John exercise did John exercised. Interestingly, we are going to see that the monolingual and the code-switched
instances of light verbs have at least one peculiar characteristic in common. We take
this to mean that the ingredients that make up (4), (5) and (6) are probably the same.
3 The term light verb is often used as a more encompassing concept than the one used
here. For instance, the English verb �‘take�’ in �“take a walk�” or �“take a chance�” is
referred to as a light verb because it is bleached of most of its meaning. For analysis
and references on this phenomenon, see Butt (2003). In this article we focus on the
canonical light verb �‘do�’.
9
2. The puzzle
These light verb constructions exhibit some puzzling properties that bear on
current theoretical linguistic and bi-linguistic debates:
2.1 Asymmetry
The construction is asymmetric. The light verb is taken from one language L( ) and
the lexical verb from the other L( ) (the object could come for either one). Take for
instance our example (4). In this example, the light verb is Spanish while the lexical
verb phrase is German. Interestingly, the same community of bilingual
Spanish/German speakers that create and accept examples like (4) reject examples
with a German light verb and a Spanish lexical verb:
(7) *Juan tut coser una camisa. Juan did.3rd sew.inf a shirt �‘Juan sewed the shirt.�’ This asymmetry has been noted often (see Den Dikken and Rao, to appear, and
references therein). Additionally, there is a mirror asymmetry. German/Spanish
bilinguals accept (and produce) nonce words created by joining together a Spanish
root and a German verbal inflection. However, these same bilinguals reject a word
made up of a German root and a Spanish verbal inflection. In the following example,
the Spanish roots utiliz- �“use�” and cos- �“sew�” are attached to the German infinitive
suffix �–ieren. The German roots benutz- �“use�” and näh- �“sew�” are attached to the
Spanish infinitive suffix -ear:4
(8) a. [w Rsp + Infldt] utilisieren �‘use�’, cosieren �‘sew�’�… 4 The subindices �“sp�” and �“dt�” stand for �“Spanish�” and �“German�” respectively.
10
b. [w Rdt + Inflsp] *benutzear �‘use�’, *nähear �‘sew�’�… Examples such as these impinge on the debate concerning the possibility of
code-switching word-internally (for argumentation that it is not possible see Poplack
1980 and MacSwan 1999, among many others). In this article, we show that it is
possible, but it is sharply limited (see Den Dikken and Rao, to appear, for a similar
conclusion). As we show below, the facts surrounding these data are complex. In
order to analyze them properly we will have to sort out instances of borrowing from
code-switching proper.
2.2 Passive voice
Passives cannot be used in the light verb construction. (9) is a German/Spanish
example again:
(9) *El libro fue hecho lesen. the book was done read Interestingly, the light verbs of monolingual grammars obey the same restriction, as
shown in Karimi-Doostan (2005).
This common restriction cannot be a coincidence �– instead, it strongly suggests
the working of Universal Grammar. Our analysis does in fact invoke universal
mechanisms of syntactic computation.
2.3. Feature spreading
Consider examples (4) and (5) once again, focusing on the structure of the verbal
phrase. In (4), the verbal phrase exhibits a VO order. Both the verb and the object are
German and, appearing after an auxiliary or modal, one would expect an OV order.
Likewise, in (5), the word order is OV as in Turkish, although the lexical verb is
Dutch and thus one would expect the opposite order as a result of the well-known
11
Germanic V2 effects. Thus, the grammar of the lexical verb phrase is the grammar of
the light verb, not the grammar of its constituent elements. This is what we refer to as
feature spreading.
In previous work (González-Vilbazo & López, forthcoming) we discuss the
phenomenon of feature spreading that we find in (4) in depth and argue that
Chomsky�’s (2000 et seq) concept of phase provides an account rich in empirical
predictions. On the other hand, Myers-Scotton and Jake (2009) mention (5) and claim
that it provides evidence for Myers-Scotton�’s theory of code switching called Matrix
Language Frame Model (MLFM). A few years ago, the foremost minimalist and
MLFMist engaged in a debate as to what is the best path to approach the study of
code switching (Jake et al. 2002, MacSwan 2005a, b). We believe that the light verb
construction provides a very concrete data-base to foreground the inadequacies of the
MLFM approach.
2.4 Absence in the input grammars
Let�’s consider example (4) once again. This sentence has been constructed with
lexical elements taken from the Spanish and German lexica. However, the output is a
construction that does not exist in either Spanish or German. Spanish does allow
using the verb hacer �“do�” followed by an infinitive, but the end result is a causative
construction, a complex predicate, not a plain verbal phrase as in (4). Colloquial
German allows for the usage of tun �“do�” followed by a verbal phrase, but tun is
demonstrably a pure inflectional element and not a light verb.5
5 For instance, tun cannot appear in a periphrastic tense form. For more details on the
tun-periphrasis see Erb (1995, 2001).
12
Linguists working on code-switching are divided on its nature. There are
numerous articles that argue for restrictions on code switching that are specific of this
modality and have no counterpart on monolingual grammars (Poplack 1980, Joshi
1985, Myers-Scotton, 1993, 2001; Myers-Scotton & Jake 2009 among others). On the
opposite side, there are those who argue that code-switching is nothing more than the
expression of a form of grammatical competence and therefore any restrictions on
code-switching must be derivable from the same principles that are active in
monolingual grammars (Woolford, 1983; Di Sciullo et al., 1986; Belazi et al., 1995;
Mahootian and Santorini, 1996; MacSwan, 1999 and González-Vilbazo, 2005).
The fact that neither German nor Spanish has light verbs while German/Spanish
code switching does would seem to suggest that the first camp is right and code-
switching incorporates sui generis ingredients. However, we argue below that this
conclusion is unwarranted and that a proper understanding of the mechanics of
language acquisition suffices to provide an account for this apparent puzzle.
3. Data and methodology
Most of our data are taken from students of the German School of Barcelona. The
school�’s student population ranges between 1000 and 1400 students. Our informants
belong to a socially homogeneous socio-economic class: their parents are members of
the middle class with college degrees. Most of the students are Spanish/German
bilinguals, some of whom are trilingual (Catalan). The typical student�’s parents are
either Spanish/German mixed couples or both German. There is a small population
(about 10%) of students whose parents are both Spanish. Thus, the students have a
high exposure to both languages from an early age. The consultants whose
13
grammaticality judgments we use spoke mainly German in class and at home and
Spanish and/or Catalan in any other context. As a result of this multilingual
environment, the students of the school code-switch often when talking to one
another. The students are proud of their code-switching and have a positive attitude
towards it as a badge of identity as many of them confirmed in individual interviews.
They even have given their code-switching variety a name, �“Esplugish�”, because the
school is located in Esplugues del Llobregat, a suburb close to Barcelona.6
Some of the data presented in this article were gathered at the school in 1996 by
Kay González-Vilbazo and in 2003 with the collaboration of Susanne Müller. In
1996, 27 students of the school from 10th to 12th grade were our language
consultants. In 2003 Susanne Müller gathered additional data from another 55
students from 10th to 12th grade. In both occasions the data collection consisted of
four parts:
1. All subjects filled out a social background questionnaire to identify possible
confounding variables. This questionnaire allowed the researcher to identify speakers
that were actually not bilingual (to the same highly proficient extent as the rest) or
were not usual code-switchers.
2. A natural conversation in Esplugish between two subjects was recorded. The
researcher was present but did not participate in the conversation, busying
himself/herself in an ostensible manner with a different task. The subjects seemed to
feel comfortable as some actually asked to be able to finish the conversation after the
recording time (about 15 minutes) was over.
3. The subjects filled written grammaticality judgment questionnaires individually.
6 Kay González-Vilbazo attended this school and speaks Esplugish.
14
4. The researcher interviewed the subjects (individually) starting with a debriefing
about the questionnaires and followed by general questions about attitudes towards
code-switching and language use.
Additional data were collected on an ongoing basis over the years by email. In
order to contrast the results with other code-switching data we collected random
samples through short email questionnaires from the German schools of Madrid,
Bilbao, Málaga, Tenerife, Santiago de Chile and Buenos Aires. Where code-switching
was in use, the collected data were consistent with the data from the German School
of Barcelona.
As we are only interested in competent bilinguals (rather than advanced second
language speakers, which would be the subject of another study entirely), we
statistically filtered out non competent bilinguals. We used two criteria to filter out
non-competent bilinguals. First we checked the answers for each token and marked
outlier answers (with respect to at least 90%of consistent answers from the other
subjects) and then discarded speakers that gave more than 5% of outlier answers. This
percentage is stricter than what is commonly used as a criterion in statistical analyses,
but we preferred excluding too many subjects rather than including potentially non
competent bilinguals in the data used for the analysis. The second criterion to define a
competent bilingual was reciprocal identification. After the data were collected, the
subjects were presented with anonymous recordings of other speakers and were asked
to rate the proficiency in Esplugish of the recorded speakers. The two criteria yielded
solidly coherent results (for a detail explanation of methodological issues and
statistical techniques, see González-Vilbazo (2005)).
The data base obtained in this manner yielded numerous examples of the hacer
+ V construction and an initial analysis of its properties was sketched. These data
15
were then complemented with additional questionnaires administered to 3 additional
subjects (former students of the German school of Barcelona) via telephone, skype or
email. These subjects were presented with oral or written data and were asked for
their grammaticality judgments on a l-5 Likert scale. These language consultants
allowed us to probe the hacer + V structure in depth and are the source of the
judgments in section 7. The findings confirmed the intuitions of the Esplugish-
speaking author.
4. The light verb and the minimalist framework.
The minimalist framework as pursued by Chomsky and his associates is a derivational
approach to the study of syntax, with an operation called Merge as the basic unit of
syntax. Merge takes two terms x and y and combines them forming the set {x,y},
which itself may be the input to another instance of Merge. This is the operation
needed to build syntactic structures. Following the tenets of Distributional
Morphology (Halle and Marantz 1993) we view the lexicon that feeds syntax as a set
of abstract features. Syntax manipulates these abstract features in various ways and
they only become fleshed-out in a separate Morphology module that takes syntactic
structures as input.
Current theoretical investigations in the minimalist framework posit the
following clause structure (see Chomsky 1995, 2000 et seq, among many others):
(11) [CP C [TP T [vP v [VP V �…]]]] Starting from the bottom up, we have a lexical verb (or possibly a root without a
specification for category, as in Marantz 1997) that may take an internal argument
and one or more adjuncts, forming the lexical verb phrase (VP). The VP is itself
16
selected by little v, a verbal category that additionally may introduce the external
argument. The resulting projection is referred to as a vP. vP is selected by T and TP
selected by C, in the manner standard since Chomsky (1986).
In monolingual grammars, it is very common for V to incorporate into little v,
forming one morphological word. We take it that little v is the base for the verbal
inflectional morphology (Oltra-Massuet and Arregi 2005). This is visible in German:
the suffix �–ier- attaches to a bare root and the verbal morphology attaches to the
resulting base. Thus, -ier- can be taken to be a spell-out of little v. In languages such
as English, v has no phonological representation.
González-Vilbazo (2005) argues that the light verb construction exemplified in
(1) is an instance in which the lexical verb does not incorporate into little v. Instead,
hacer spells-out little v as an independent word. The infinitive morphology that
attaches to the German verb should be regarded as default morphology because
German does not allow bare roots to spell-out:
(46) [CP C [TP Juan T [vP t(Juan) hizo [VP nähen das Hemd]]]] did sew the shirt �‘John sewed the shirt.�’
Probably �–although we do not try to make the case in these pages - the light verb
constructions in monolingual grammars should also be regarded as instances of
unincorporated little v.7
7 Karimi-Doostan (2005) doubts that the light verbs in Korean, Kurdish and Persian
can be regarded as little v under the assumption that the latter always introduces an
external argument. Recent work has successfully argued that there is an instance of
little v that does not introduce the external argument (Legate 2003). Thus, there is no
obstacle to maintaining the assumption that light verbs spell out little v.
17
Combinatorial systems as the one outlined in the Minimalist Program require
inherent devices to prevent computational explosion. One such device is the phase. A
phase is a chunk of structure that forms a computational unit such that its complement
can be transferred to the interfaces and, so to speak, be forgotten, so that the
computational system itself does not have to carry too much working memory.
Intensive research has provided empirical evidence that phases play a role in natural
language syntax, being the triggers of all syntactic dependencies within the clause. In
every analysis, CP and transitive vP have been taken to be phases:8
(12) [phase CP C [TP T [phase vP v [VP V �…]]]]
Additionally, much current work has taken DPs �– the structure formed by the noun
and the functional categories associated with it �– to be a phase as well (see Svenonius
2004).
Syntactic dependencies are set up by phase heads (see Chomsky 2008) by means
of a probe-goal mechanism that binds the unvalued features of a term with the valued
features of another. For instance, v has unvalued -features that turn it into a probe,
which will look for valued -features in its c-command domain. Typically, the
internal argument provides a value for the features of v. The operation in which the
features of a probe are valued against those of a goal is called Agree:
(13)
8 For more detailed introductions see Adger (2003) and Hornstein, Grohmann and
Nunes (2005).
18
Agree
Two other principles that restrict the computational power of syntactic
derivations are relevant in this article. One is Relativized Minimality/Minimal Link
Condition (RM/MLC) (Rizzi 1990, Chomsky 1995). RM/MLC claims that a
dependency between positions and cannot be established if there is another
position that has the features relevant to establish a dependency with and c-
commands :
(10) * i �… �… i The second one is Last Resort, according to which operations only take place if
they have to and the computational system will always choose the least effort path at
any point.
These �– admittedly brief �– words should suffice to introduce our framework of
analysis. In the following sections we use this framework to approach the puzzles
presented in the introductory section.
5. Asymmetry
5.1. Three problems
In this section we address the problem we dubbed the asymmetry puzzle. As a matter
of fact, this puzzle can be broken down into three separate questions.
(i) Why is it the case, that hacer is the only light verb in Esplugish, while an
equivalent German verb like tun or machen cannot be a free-standing light verb?
(14) *Hans tut coser la camisa. Hans does sew the shirt
19
�‘Hans sews the shirt.�’ (ii) Why is it possible for Esplugish speakers in full-fledged code-switch mode to
deploy words that involve a Spanish root with German inflection (utilisieren �‘use�’,
alemanisieren �‘germanize�’) while rejecting verbs constructed from a German root and
Spanish inflection (*laufear �‘walk�’, *anmeldear �‘register�’)?
(15) Wir utilisieren palabras alemanas. We use words german �‘We use German words.�’ (16) *Juan se ha anmeldeado. Juan refl has registered �‘Juan has registered.�’ The relevant facts in (i) and (ii) can be represented by means of a table. The
German root benutz- and the Spanish one utiliz- both mean �“use�”:
(17) Light verb vsp [Vdt]
hacer benutzen * vdt [Vsp] * tun utilizar
v-to-V incorporation * vsp + Vdt * benutzear
vdt + Vsp utilisieren
[the symbol + indicates incorporation] The table shows the complementary distribution of the synthetic and the
analytical forms. It is this complementary distribution that needs an account.
(iii) hacer is a light verb only when the lexical verb is German. If the lexical verb is
Spanish, hacer must read as a causative verb. Why can�’t hacer be a separate light
verb when the lexical verb is Spanish?
(18) Juan hizo utilizar palabras alemanas. Juan did use words german =Juan had German words used. Juan used German words.
5.2 Code-switching vs. Borrowing
20
Before we start addressing the puzzles in 5.1 we need to approach another
vexing problem in code-switching: the distinction, if it is possible to draw, between
code switching and borrowing. Consider again the table above. Although the form
*benutzear is starred, instances of borrowing in the languages of the world happen all
the time and forms similar to *benutzear pop up apparently without restriction:
(19) Tienes que cliquear en este sitio. Have-to.(2nd.person.Sg.) click in this place �‘You have to click in this place.�’ In this example, the Spanish word cliquear �“click�” is taken directly from English and
fully adapted to Spanish morpho-phonology (it is conjugated as an �–ar class verb
because this is the most productive class in Spanish). Cliquear is, in every respect, a
member of the lexicon of many Spanish speakers. This fact makes the problem raised
by *benutzear more acute: why isn�’t it possible for an Esplugish speaker to simply
say *benutzear? Is *benutzear built differently than cliquear?
A substantial portion of the code-switching literature denies that it is possible
for code-switching to take place within the word (from Poplack 1980 to MacSwan
1999). That is, according to the above references, a structure like the following is
impossible (where M is a morpheme and W is a morphological word):
(20) *[w ML1 + ML2] In fact, grammaticality judgments like the following are widely attested:
(21) *Estoy eatiendo. I�’m eating. Poplack 1980: 586 These authors suggest that borrowing can be distinguished from code-switching
because a borrowed word is fully adapted to the phonology of the host language while
code-switched structures maintain features of the donor language. For MacSwan, the
impossibility of code-switching within the word is a corollary of his PF disjunction
21
theorem (MacSwan 2000: 45). Let us summarize MacSwan�’s argument. His idea is
that two languages L( and L( will have different rankings of at least some of the
universal constraints that make up human language phonology. Let�’s say that L(
has the ranking C1 >> C2 while L( has the ranking C2 >> C1. Code switching within
a word leads to contradictory rankings of C1 and C2 and therefore to an
unpronounceable phonetic representation. Obviously, this problem does not arise if a
morpheme from L( has been borrowed by L( The process of borrowing entails
full adaptation of the phonological features of the borrowed morpheme �– hence,
cliquear.
We are not certain that MacSwan�’s reasoning logically leads to the conclusion
that code-switching within the word is impossible. Incorporation of a root into a
suffix gives rise to an endocentric structure in which all and only the features of the
head project to the newly created term. Take the unit of phonological rules/constraints
to be the phonological word W. Take M to be a morpheme. It is at least theoretically
possible (and, we believe, empirically real) for L to incorporate into ML , the
output W subject only to the phonological rules of ML , the head of the word, while
ML fully adapts to the rules of L1.
When analyzing the mixing of languages within a word, at the current stage of
understanding, we do not have a formal means to tease borrowing from word-level
code-switching apart. However, we believe that we can tell that cliquear in (19) is an
instance of borrowing while utilisieren is code-switching. But our evidence revolves
around the context in which the data is elicited rather than any formal properties. As
mentioned, when our Esplugish consultants were interviewed and purposefully placed
in a code-switching mood, they readily accepted and produced forms like utilisieren
and alemanisieren; however, they never produced anything like *benutzear. When
22
asked, they rejected such forms. Esplugish speakers only borrow when talking
monolingually to an outsider. The following sentence is attested:
(22) Mañana te tienes que anmeldear. Tomorrow you.dat must.2sg that register �‘Tomorrow you have to register.�’ This conversation took place in Germany. The Esplugish speaker was talking to a
Spanish speaker with an intermediate knowledge of German. The Esplugish speaker
was referring to the peculiar German regulation that all foreign residents, must visit a
police station and fill in a form that includes their address. The speaker could not find
an appropriate word in Spanish so he resorted to the word in German, knowing the
other person would understand it.
Thus, although formally borrowing and word-internal code switching are very
difficult to tease apart, they are different linguistic phenomena. The sharpness of the
grammaticality judgments provided by our Esplugish speakers and the contrast in our
data between the frequency of forms such as utilisieren and the total absence of forms
like *benutzear tell us that the difference between word-internal code-switching and
borrowing is indeed part of the I-language of bilingual speakers.
Within the tradition of generative grammar it is possible to provide an elegant
description of the difference between word-internal code switching and borrowing.
Borrowing is the copying of a lexical item from one lexicon to another. In the case of
bilinguals, they are in possession of two lexica and therefore borrowing may take
place between the two lexica. Therefore, borrowing takes place with total
independence of the computational system of human language:
(23) {Lexicon 1 �… w�…} {Lexicon 2 �…w�…}
23
The item copied from Lexicon 2 to Lexicon 1 will be completely adapted to the
morphological and phonological properties of Lexicon 1. Thus, anmeldear can exist
as a borrowing phenomenon: the German root anmeld- is copied into the Spanish
lexicon. Notice that the new verb automatically becomes a member of a Spanish
conjugation class (the �–ar class).
In code switching, the item from Lexicon 2 is copied directly onto the
computational system, together with other items from Lexicon 1 and Lexicon 2. We
can use the device of a lexical array (as in Chomsky 1995, 2000) for ease of
exposition. A lexical array is the set of lexical items that have been selected to initiate
a derivation. Monolingual speakers choose their lexical array from their single
lexicon. Bilingual speakers are able to choose lexical items from both lexica:
(24) {Lexical Array w1, w2, w3, w4,�…} {Lexicon 1 w1, w3 �…} {Lexicon 2 w2, w4 �…} This simple model is going to prove useful, as will become clear very soon.
5.3 Explaining the asymmetry
Now we are ready to tackle questions (i) �– (iii), rephrased here more concisely:
(i) Why is the light verb always hacer and not tun?
(ii) Why do we find in Esplugish verbs formed out of a Spanish root and German
inflection but never a German root and Spanish inflection?
(iii) Why is there no light verb hacer in Spanish as in Esplugish?
Our analysis starts with question (i), by noticing a morphological difference
between Spanish and German verbs. Spanish verbs belong in three conjugation
classes (the �–ar, �–er and �–ir classes). These conjugation classes are features of the
24
verbal root and determine the phonological form of the inflection in all tenses. Let us
then take �“conjugation class�” to be a property of Spanish verbal roots:
(25) a. cant-�“sing�” -ar class cantar (ex: cantas = you sing) b. beb- �“drink�” -er class beber (ex: bebes = you drink) c. viv- �“live�” -ir class vivir (ex: vives = you live) Additionally, we hypothesize that the Spanish little v has an unvalued feature
for conjugation class:9
(26) vsp[uConj] In order to satisfy this unvalued feature, the Spanish little v incorporates a verbal
root with a specification for conjugation class:
(27) vP v[uConj] VP cant-[-ar] �… Incorporation
As a result of incorporation, vsp acquires a conjugation class feature and becomes a
morphological base to which the Tense, Aspect and Mood morphology are attached.
The German verb is not classified into any conjugation class. Although German
grammarians do discuss the verbal Flexionsklassen, what is referred to with this word
is the changes in the vowel of the root that some verbs undergo in some tenses (i.e.:
denken �“think�” becomes dachte �“thought�” instead of the regular *denkte). However,
the actual suffixes attached to the root do not change. This is shown in the following
9 Oltra-Massuet and Arregi (2005) also associate the conjugation class feature to little
v.
25
example, where the conjugation in the simple past of denken is displayed next to that
of the regular verb leben �“live�”:
(28) singular plural singular plural 1st dach-te dach-ten leb-te leb-ten 2nd dach-test dach-tet leb-test leb-tet 3rd dach-te dach-ten leb-te leb-ten We conclude that Flexionsklassen are not conjugation classes. We take it that
the best approach to this class of verbal words is as a form of suppletion, as English
irregular verbs: certain verbal roots, in combination with inflectional features, are
substituted for a suppletive form in the module Morphology (see section 4 on
Distributed Morphology). Thus, we conclude that German verbal roots do not bear a
conjugation class feature and consequently that vdt does not have an unvalued
conjugation class feature.
Further, we assimilate the merge of hacer as a light verb to other �“last resort�”
insertions of lexical material such as the famous �‘do�’-insertion of English or �–kuwa
insertion in Kiswahili. Roughly put, �‘do�’-insertion in English takes place when a
morpheme intervenes between the lexical verb (in v) and T, preventing the
establishment of a dependency between them. As a result of this intervention, the T
morpheme ends up as an unattached suffix, a situation rescued by inserting the
dummy root �‘do�’ (Chomsky 1955). Similarly, -kuwa in Kiswahili provides support for
tense or mood morphology and agreement morphology that would otherwise be left
hanging:
(29) Juma a-li-kuwa a-me-pika chakula cha asubuhi Juma agr.past.kuwa agr.asp.cook food of morning �‘Juma makes breakfast.�’ Likewise, the presence of light verb hacer in Esplugish responds to a similar last
resort motivation. The Spanish little v requires a valued conjugation class feature,
which is absent from the German verbal root:
26
(30) *vP vsp[uConj] VP lauf- �… In this example, the German root lauf-(�‘run�’) has been brought into the
computational system as a German lexical item. Crucially, it has not been borrowed,
and therefore it has no conjugation class. Since it does not have a conjugation class, it
cannot satisfy the [uConj] feature of vsp. Hence the impossibility of being suffixed a
Spanish verbal morphology in code switching and the ungrammaticality of forms such
as *benutzear. Only insertion of a light verb like hacer, bearing a conjugation class
feature, can make this configuration grammatical.
(31) *vP [hac-[-er] +vsp[uConj]] VP lauf- �… However a German little v would never have to be realized as last resort by a
light verb (like tun). Such a German little v has no unvalued [Conj] feature. Thus,
both a German V and a Spanish V can incorporate into a German v. This answers
question (i).10
10 We gladly acknowledge that a Last Resort solution to the asymmetry of light verbs
in code-switching is already sketched in Bhatia and Ritchie (1996) for the
Hindi/English pair and Den Dikken and Rao (to appear) for English/Telugu. The
details of their analyses vary from ours, since the grammatical properties of the
languages involved are also different.
27
Let�’s now address question (ii). The German root has no conjugation class and
vdt has no [uConj]:
(32) vP vdt VP utiliz-[-ar] �… There is nothing in the configuration in (32) that prevents incorporation of the
verbal root into v. The conjugation feature in the root is valued and therefore does not
need checking or valuation. The Spanish root utiliz- can incorporate into German little
v, which is realized as the German stem affix �–ier-. The complex head v+V merges
with T forming utilisieren etc. Since light verb insertion is a last resort, the
incorporation strategy always wins and this rules out periphrastic forms like *tun
utilizar as we just saw.
Finally, we are left with question (iii): why can�’t Esplugish speakers use hacer
as a light verb with a Spanish lexical verb? Again, we appeal to the last resort
character of light verbs. Since incorporating a Spanish verbal root into vsp is always
possible, the periphrastic form is ruled out.
5.4 Conclusion
We have accounted for the asymmetry puzzles using only the different morphological
features of vsp and vdt as well as the principle of Last Resort. Along the way, we have
introduced a discussion of the distinction between code-switching and borrowing and
have argued that, although the distinction is substantial from the point of view of the
mental grammars of bilingual speakers, on the surface they outcome of both processes
28
can be similar enough that we need to take contextual variables into consideration to
tease them apart.
6. Light verbs and passive voice
In our data we have found that it is possible to have the light verb select for
transitive (33), unergative (34) and unaccusative (35) predicates:
(33) Juan hace lesen ein Buch. Juan does read a book �‘Juan reads a book.�’ (34) Juan hace schlafen. Juan does sleep �‘Juan sleeps.�’ (35) La Vase se hizo zerbrechen. The vase CL. did broke �‘The vase broke.�’ However, the light verb cannot be used in a passive construction:
(36) *Das Buch ha sido hecho verkaufen. The book has been done sold �‘The book has been sold.�’ Interestingly, it seems to be a more general phenomenon: Karimi-Doostan
(2005)�’s study of light verb constructions of the monolingual varieties Kurdish,
Persian and Korean yields the same restriction. This suggests a restriction imposed on
the computational system rather than a language-specific �– or code-switching specific
�– type of restriction. In the following, we suggest an account.
Following Legate (2003) it is commonly assumed that unaccusative and passive
predicates are headed by a little v without an external argument:
(37) [vP v [VP V DP] It is well known that there is an important difference between passives and
unaccusatives: in the former, the �“absorbed�” external argument makes its presence
29
strong enough to control a PRO in an infinitival resultative clause (see Baker et al
1989, among many others). Unaccusatives, on the other hand, have no remnant of an
external argument:
(38) a. The boat was sunk to collect the insurance. b. *The boat sunk to collect the insurance. We take it then that there is a remnant of an external argument in passive predicates,
which we represent as PRO:
(39) a. The boati was PRO sunk ti. b. The boati sank ti. For reasons that remain mysterious to us, this PRO seems to be unaffected by the
Case filter, which allows, or forces, the internal argument of a passive structure to
establish a dependency with T. Hence the well-known morpho-syntactic similarities
between unaccusatives and passives (Burzio 1986) that obscure the argument
structure difference.
The next question is how the internal argument can jump over the external
argument PRO in a passive structure. The principle of Relativized Minimality
(RM)/Minimal Link Condition (MLC) introduced in section 2 should ban the
operation represented in (40) (where IA=Internal Argument):
(40) 1. [vP PRO v [VP V IA ]]] Move IA to Spec,v 2. [vP IA [v�’ PRO v [VP V t(IA) ]]] However, the work of Holmberg (1986), Chomsky (1993) and, more recently,
Gallego (2007) and Ku erova (2007a, 2007b), has made syntacticians aware of the
role of verb movement in minimality effects. To put it in Chomsky�’s terms,
movement of a head X into position p renders the positions and equidistant
from :
(41) [pP p�’ p [ X [�… ]]]
30
We do not need to concern ourselves with position . Take X=V and p=v. For our
purposes, what is crucial is that raising the lexical verb to adjoin to the little v
creates the space for a constituent in to move to a Spec,v position higher than .
As a result, the internal argument (IA) of a passive predicate can move into a new
Spec,v:
(42) [vP IA [v�’ PRO v+V [VP t(V) t(IA) ]]] Now we are ready to account for the ungrammaticality of passives with overt
light verbs. In this construction, the lexical verb does not adjoin to the little v as little
v is realized as hacer. As a result, the positions and are not equidistant from a
lower position. Raising of IA to Spec,v gives rise to a violation of RM/MLC.
This problem does not arise if the predicate is unaccusative. Since there is no
trace of an external argument in Spec,v, nothing prevents the internal argument to
move to it, regardless of V-movement.
7. Feature spreading
Consider the patterns in (43), (44) and (45):
(43) a. Hizo nähen das Hemd. did sew the shirt �‘He/she sewed the shirt.�’ b. *Hizo das Hemd nähen. Did the shirt sew �‘He/she sewed the shirt.�’ (44) a. Er soll das Hemd nähen. he should the shirt sew �‘He should sew the shirt.�’ b. *Er soll nähen das Hemd. He should sew the shirt �‘He should sew the shirt.�’ (45) a. Er hat das Hemd genäht. he has the shirt sown �‘He has sown the shirt.�’ b. *Er hat genäht das Hemd.
31
He has sewn the shirt �‘He has sewn the shirt.�’ (43) repeats (4) above, and it shows that after the light verb the word order is
VO even when the VP is realized with German lexical material. (44) and (45) show
that in a regular German VP the word order after a modal or auxiliary is OV. Why
then does the VP with German lexical material follow the Spanish word order
pattern?
The analysis of feature spreading in the Esplugish light verb construction
proposed in González-Vilbazo & López (forthcoming)11 builds on two assumptions:
(i) the head of a phase �– little v or C �– define the grammatical properties of the whole
phase and (ii) the light verb is the head of the phase (see example ()). Taken together,
these two assumptions provide an account of the complex web of data that surround
the grammar of light verbs in Esplugish.
Since the head of the vP phase is drawn from the Spanish lexicon, its
complement VP will have to have Spanish grammatical characteristics, most notably
word order. Thus, feature spreading takes place as predicted by phase theory: it goes
from the head of a phase to its complement.
The complement of a lexical verb does not show features of the lexical verb. In
(47), the complement of the Spanish verb dijo is a German sentence. In (48), the
complement of the Spanish verb busca is a German DP:
(47) Sabine dijo, dass die Frau klug ist. Sabine said that the woman smart is Sabine said that the woman is smart.
11 In the present article we only present word order data to illustrate the puzzle of
feature spreading. In González-Vilbazo & López (forthcoming) we expand the
consequences of our analysis to prosody and information structure.
32
(48) Sabine busca ein kluges Mädchen. Sabine seeks a clever girl This follows again from phase theory: lexical verbs are not phase heads. The CP
complement of dijo in (47) is structured as a regular German subordinate sentence
with OV order. Likewise the complement of busca in (48) is organized as a regular
German DP, with Adjective+Noun order, as opposed to the regular Noun+Adjective
order in Spanish. The suffix �‘-es�’ in the adjective kluges expresses concord with the
noun that involves the neuter gender, absent in Spanish. This follows naturally under
the commonly accepted assumption that the DP is itself a phase and thus not
accessible to the little v into which the lexical verb busca incorporates.
But, as shown in González-Vilbazo & López (forthcoming) the predictions of
the phase theory with respect to word order are even more wide-ranging. Consider the
following two causative constructions in Esplugish:
(49) Juan hizo bauen ein Haus. / *Juan hizo ein Haus bauen. Juan made build a house Juan had a house built. (50) Juan hizo a Pedro ein Haus bauen. / *Juan hizo a Pedro bauen ein Haus. Juan made dat Pedro a house build Juan made Pedro build a house. Causative constructions in Esplugish are constructed around the Spanish verb
hacer, itself a causative verb in Spanish (the fact that hacer in Esplugish can be a
causative or a light verb gives rise to a number of interesting structural ambiguities
that we explore in detail in González-Vilbazo & López, (forthcoming).
Example (49) exhibits obligatory VO order (bauen ein Haus) while example
(50) exhibits obligatory OV (ein Haus bauen). Other than that, the only surface
difference between the two causative sentences is that the latter includes a causee (a
Pedro), while the former does not.
33
The phase theory, coupled with independently motivated analyses of Romance
causatives (Guasti 1992, López 2001, Folli and Harley 2009) provides a
straightforward account of the difference between (49) and (50). Causative verbs may
select for a VP or for a vP, which includes an external argument. This external
argument is what surfaces as the causee (see López, 2001):
(51) vsp hacer [VP V�…] (52) vsp hacer [vP EA/causee vdt [VP V...]] Consider first the example without a causee. The complement of hacer includes
no little v. The grammatical properties of the VP are those of the closest phase head.
The closest phase head happens to be the little v that selects for hacer. This little v is
Spanish: in the absence of an independent morpheme that lexicalizes v, hacer
(causative) must incorporate into v. Since the morphology exhibited by the verb is
Spanish (hacer not *hasieren), we conclude that this v must be drawn from the
Spanish lexicon.
The example with the causee includes a little v. The same reasoning as in the
previous paragraph leads us to conclude that the little v that introduces the causee
must be German. The properties of the VP are going to be those of vdt, hence the OV
order.
8. Absence of light verbs in the input grammars
Recall that in the introduction we subscribed to the view, put forward by
Woolford (1983), Di Sciullo et al. (1986); Belazi et al. (1995), Mahootian and
Santorini (1996), MacSwan (1999) and González-Vilbazo (2005) that code switching
should not be accounted for by using dedicated mechanisms, special features or
34
restrictions. The monolingual grammars of neither German nor Spanish include a
free-standing light verb. Is the presence of such a light verb in Esplugish an example
of a �“third grammar�” dedicated to regulating code-switching?
Our approach to the third grammar issue is slightly different from our
predecessors. Linguists who assume that there should not be a third grammar usually
propose that code-switching consists uniquely of features found in the input languages
(especially MacSwan 1999). We think this view does not fully consider the role of
UG in language acquisition.
Children construct their own I-languages as an outcome of the interaction of
Universal Grammar with their input (together with what Chomsky 2005 calls �“third
factor conditions�”, which we can ignore here). There is no reason to suppose that the
resulting I-language will be exactly like the input. As a matter of fact, there is
abundant evidence suggesting that children resort to ingredients provided by
Universal Grammar that are absent from the input. The famous case of Creole
languages immediately comes to mind (Bickerton 1983). Likewise, the sign languages
of the deaf often develop numerous features absent from their input languages (see
Newport 1999, Goldin-Meadow 2005, among many others). Even the language of
monolingual children who acquire their first language in normal circumstances also
shows features that are both rule-governed and absent from the input (Crain and
Pietroski 2001). In normal circumstances, these forms eventually disappear due to the
pressure from the input. In unusual circumstances �– such as those surrounding the
development of creole and sign languages �– forms absent from the input remain and
become the mature I-language of a community of speakers.
The I-language of Esplugish speakers is also the outcome of the interaction of
Universal Grammar with their environment. The resulting I-language may contain a
35
feature �– a free-standing light verb �– that is not present in the input grammars but is
available in the universal pool. Unlike canonical monolingual learners, Esplugish
speakers do not see any reason to strip themselves of this option as they grow up
because the input they receive exerts no pressure on the Esplugish modality, which
grows virtually unnoticed by adults and even by the speakers themselves. We
conclude that the free standing light verb remains in their grammar �– as in the
grammars of monolingual Persian, Hindi, Kurdish speakers, among many others �–
subject to the same constraints (ungrammatical in passive voice) as in monolingual
grammars. Thus, the Esplugish light verb gives us no reason to adopt the view that
there are features exclusive of code-switching. Instead, it provides us with an example
of the role that input-independent internal mechanisms �– UG, if you will �– play in
child language development.
9. The Matrix Language Frame Model
The changes in word order that take place after light verbs in code-switching
contexts are sometimes used as evidence for the Matrix Language Frame Model
(MLFM) (see Myers-Scotton and Jake 2009: 338-9, among others): In this view, the
language L( of the light verb is the matrix language and therefore its complement
has to adjust to the rules of L( . In this section, we present a brief summary of the
MLFM and show that the design of this particular approach to code switching does
not give us any predictions for the structure at hand.
For our presentation, we use the most recent formulation of the MLFM (Myers-
Scotton and Jake 2009). The leading idea of the MLFM is that the participating
languages in a code-switching event are utilized asymmetrically. One language is the
36
matrix language, providing the grammatical skeleton of the clause and determining
restrictions on e.g. word order, agreement etc. The other language is the embedded
language, which provides phrases whose insertion is acceptable to the extent that it
does not violate restrictions of the matrix language. A language is defined as matrix or
embedded for a certain corpus. That is, a bilingual speaker may choose L( as matrix
and L( as embedded in some corpus while the same speaker may be using L( as
matrix and L( as embedded in another corpus. Myers-Scotton and Jake even
acknowledge that the matrix language may occasionally vary from clause to clause
(2009: 337-8).
As Myers-Scotton and Jake (2009) assert, the MLFM is not a competence
model. Instead it �“provides a major linguistic theory of language contact dedicated to
bilingual speech and processing [�…] [the theory] accounts for a variety of bilingual
behaviors.�” (p336). Thus, it is explicitly a theory concerned with the performance
aspects of bilingualism and not with the underlying knowledge. Correspondingly, the
data base utilized to develop the MLFM theory is a form of E-language (corpora). It is
also worth pointing out that, since the notions of matrix and embedded language are
not ingredients of monolingual grammars (for obvious reasons!) it follows that the
MLFM is the type of theory that argues that code-switching is ruled by special
mechanisms absent from monolingual grammars.
Let us see how the MLFM model fares with respect to our Esplugish data.
Recall that after a Spanish light verb, the complement appears in a VO order even
though the components of the VP are German:
(53) Juan hizo flicken die Vespa. Juan did repair the Vespa �‘Juan repaired the Vespa.�’
37
One could claim that in sentences like (53) the matrix language is Spanish. It
would follow that the rules and restrictions of Spanish should apply, hence the VO
order �– we believe we follow the logic of Myers-Scotton and Jake�’s (2008) analysis
of the Dutch/Turkish example (5). The asymmetry that makes *tut comprar
ungrammatical �– discussed in detail in section 5 - could be accounted for as an
asymmetry in matrix language: Spanish is the matrix language, German is not. In the
following we point out two major flaws in the MLFM account.
First, the MLFM model misses an important datum that we have pointed out in
this article from the outset. This important datum is precisely that this VO order
happens only when the lexical VP is governed by a light verb. Code switching
between a Spanish lexical verb and a German complement clause, or between a
Spanish lexical verb and a German complement DP does not affect the internal
structure of these constituents. In (54), the complement of the Spanish verb dijo has
the verb final syntax of German clauses. In (55), the complement of the Spanish verb
vio has the adjective+noun order of German DPs, not the noun+adjetive order of
Spanish DPs:
(54) a. Juan dijo dass Johannes klug ist . Juan said that Johannes clever is �‘Juan said that Johannes is clever.�’ b. *Juan dijo dass Johannes ist klug. Juan said that Johannes is clever �‘Juan said that Johannes is clever.�’ (55) a. Juan vio a la kluge Frau. Juan saw acc the clever woman �‘Juan saw the clever woman.�’ b. *Juan vio a la Frau kluge. Juan saw acc the woman clever �‘Juan saw the clever woman.�’ In our framework, we have linked the difference to phase boundaries. The light
verb is the head of its phase, so everything within that phase has to obey its
38
requirements. The lexical verb and its complement are in different phases and
therefore the lexical verb can exert no influence on its complement.
We do not think that the MLFM model can provide a non-stipulatory account of
this difference. In particular, one could plausibly argue that the matrix language in
(54) and (55) is Spanish, since the matrix verb is Spanish. If the matrix language in
(54) and (55) is Spanish, one should expect the complement of the verb to follow a
Spanish pattern. Since this expectation is not fulfilled, one would have to claim that
the matrix language changes at the CP and DP borders - but then we are left without a
clear understanding of what a matrix language is. In fact, we are not aware of any
algorithmic way of choosing a matrix language that would yield the results that the
empirical facts demand.
The empirical problem becomes even more pronounced in causative sentences.
Recall that causative sentences with a causee yield an OV order whereas causative
sentences without a causee yield a VO order:
(56) Juan hizo bauen ein Haus. Juan did build a house �‘Juan had a house built.�’ (57) Juan hizo a Pedro ein Haus bauen. Juan did dat Pedro a house build �‘Juan made Pedro build a house.�’ In our approach, this is easily accounted for because the presence of the causee entails
the presence of a lower vP phase which, given the presence of a German lexical verb
(and no light verb), should be German. The MLFM is completely unsuited to account
for this distinction. In effect, it would have to make Spanish the matrix language in
(56) while German would be the matrix language in (57), unaccountably.
Second, recall that a matrix language is defined with respect to a corpus. Thus,
we would expect that in some Esplugish corpora the matrix language would be
39
German while in some other corpora the matrix language would be Spanish. Indeed,
the following two sentences are possible in Esplugish:
(58) a. Er hat muchos libros verkauft. he has many books sold �‘He has sold many books.�’ b. El ha vendido viele Bücher. He has sold many books �‘He has sold many books.�’ (58)(a) would presumably have German as matrix language while (b) would have
Spanish.
If German can be the matrix language in some corpora, the MLFM predicts that
sentences like the following should be just as common as the sentences we have been
discussing in this paper:
(59) *Juan tut die Vespa reparar. Juan did the Vespa repair However, this sentence is unacceptable to Esplugish speakers. The only way that the
MLFM could account for (59) is to claim that in Esplugish Spanish is always the
matrix language. But this would fly in the face of examples such as (58)(a), in which
the matrix language has to be German. If German can be the matrix language some of
the time, then the ungrammaticality of (59) is unaccounted for. In fact, the entire
paradigm shown in the table in (17) is outside the empirical scope of the MLFM. We
claim that the property of having hacer as a free-standing light verb is not a property
of a corpus but a feature of the I-language of Esplugish speakers.
10. Conclusions
We have explored the properties of light verbs in code-switching and shown that the
puzzles that they present can be satisfactorily accounted for using general principles
of the faculty of language as laid out in recent minimalist work (most especially
40
Chomsky 1995, 2000). These include the feature checking/valuation mechanism and
the centrality of a phase head in carrying it out as well as principles that derive from
the computational nature of the faculty of language such as RM/MLC and Last
Resort. The puzzle of asymmetry is accounted for by analyzing the light verb
insertion as a last resort mechanism to prevent feature mismatch in the feature
valuation operation. The impossibility of passive voice in the light verb construction
can be accounted for by the absence of incorporation of V into v which precludes the
internal argument from raising over PRO. The surprising fact of features spreading
(word order) from the light verb to its complement can be accounted for by assuming
that the light verb is little v and as a phase head it controls the grammatical properties
of its phase. Finally, the problem of the absence of the hacer light verb construction in
the input grammars (Spanish and German) can be explained as being the result of
regular language acquisition. The option of a light verb is provided by UG and only
disappears under external pressure from the input, but that is not the case with
Esplugish. Hence, the light verb construction remains available to competent
Esplugish speakers.
The MLFM, on the other hand, consist of a series of stipulations meant to play a
role only in code-switching situations. Given the more principled grounding of our
approach and the empirical advantages that it provides, it should be preferred.
41
Acknowledgments: * This research project has been carried out under the auspices of the UIC Bilingualism Research Laboratory. We want to thank the members of the laboratory for their feedback and support with this project We would also like to thank Karlos Arregi, Jeff MacSwan, Marcel den Dikken, Jürgen Lenerz, Robert Kemp, Karen Zagona, Natascha Müller, Pascual Masullo and Volker Struckmeier for feedback on previous versions or this article and the audiences that heard various incarnations of this project: In/Between conference at UIC (Chicago, April 2010), Purdue University (April 2010), Languages in Contact conference (Wuppertal, May 2010). Our gratitude goes as well to our language consultants and to Susanne Müller for providing us with valuable code-switching data. All errors are ours.
References
Arregi, K. & Oltra-Massuet, I. 2005. Stress-by-Structure in Spanish. Linguistic
Inquiry 36.1, pp. 43-84.
Backus, A. 1992. Patterns of language mixing: A study of Turkish-Dutch
bilingualism. Wiesbaden: Harrassowitz.
Baker, M., Johnson, K. & Roberts, I. 1989. Passive Arguments Raised. Linguistic
Inquiry 20, 219-251.
Belazi, H. M., Rubin, E.J. & Toribio, A.J. 1994. Code Switching and X-Bar Theory:
The Functional Head Constraint. Linguistic Inquiry 25, pp. 221-37.
Bickerton, D. 1983. The Bioprogram Hypothesis. Behavioral and Brain Sciences 7:2,
pp. 173-188.
Burzio, L. 1986. Italian Syntax. Dordrecht: Reidel.
Butt, M. 2003. The light verb jungle. Harvard Working Papers in Linguistics, vol. 9,
1-49.
Chomsky, N. 1955. Logical Structure of Linguistic Theory. Thesis Harvard/MIT.
(Published in part: New York: Plenum. 1975)
42
Chomsky, N. 1986. Knowledge of language: its nature, origins and use. New York:
Praeger.
Chomsky, N. 1993. A minimalist program for linguistic theory. In K. Hale & S. J.
Keyser (eds.), The view from Building 20, pp. 1-52. Cambridge, MA: MIT
Press.
Chomsky, N. 1995. The Minimalist Program. Current studies in linguistics 28,
Cambridge, MA: MIT Press.
Chomsky, N. 2000. Minimalist Inquiries: the Framework. In Roger Martins, David
Michaels and Juan Uriagereka (eds.), Step by step: Essays on Minimalist Syntax
in Honor of Howard Lasnik, pp. 89-155. Cambridge, Mass: MIT Press.
Chomsky, N. 2001. Derivation by phase. In Michael Kenstowicz (ed.). Ken Hale: a
Life in Language, pp. 1-52. Cambridge, MA: MIT Press.
Chomsky, N. 2005. Three factors in language design. Linguistic Inquiry 36, pp. 1-22.
Chomsky, N. 2006. Approaching UG from below. Unpublished manuscript. MIT.
Chomsky, N. 2008. On Phases. In Freidin, Robert, Carlos P. Otero and Maria Luisa
Zubizarreta (eds.), Foundational Issues in Linguistic Theory, pp. 133-166.
Cambridge, MA: MIT Press.
Crain, S. & P. Pietroski. 2001. Nature, Nurture, and Universal Grammar. Linguistics
and Philosophy 24, pp. 139-186.
Den Dikken, M. & Rao, S.B. 2003. Light switches. Ms., to appear in MIT Press
volume ed. by Jeff MacSwan.
Di Sciullo, A. M., Muysken, P. & Singh, R. 1986. Government and code-mixing.
Journal of linguistics 22, pp. 1-24.
Erb, Marie Christine. 1995. Eine Theorie expletiver Verben: Die tun-Periphrase im
Deutschen. MA Thesis, Universität Frankfurt am Main.
43
Erb, Marie Christine. 2001. Finite auxiliaries in German. Ph.D. thesis. Katholieke
Universiteit Brabant.
Folli, R. & Harley, H. 2007. Causation, Obligation, and Argument Structure: On the
Nature of Little v. Linguistic Inquiry 38, pp. 197-238.
Gallego, A. 2007. Phase Theory and Parametric Variation. Ph.D. thesis. Universitat
Autonoma de Barcelona.
Goldin-Meadow, S. 2005. What language creation in the manual modality tells us
about the foundations of language. The Linguistic Review 22: 199-226.
González-Vilbazo, K.E.. 2005. Die Syntax des Code-Switching. Esplugisch:
Sprachwechsel an der Deutschen Schule Barcelona. Unpublished PhD
dissertation, Universität zu Köln.
González-Vilbazo, K.E. Forthcoming. Little v and parametric variation. Natural
Language and Linguistic Theory.
Guasti M-T. 1992. Causative and Perception Verbs: A Comparative Study. Torino,
Italy: Rosenberg and Sollier.
Halle, M. and Marantz, A. 1993. Distributed Morphology and the Pieces of Inflection.
In K. Hale and S. J. Keyser (eds.), The View from Building 20, pp. 111-176.
Cambridge, MA: MIT Press.
Holmberg, A. 1986. Word order and syntactic features in the Scandinavian languages
and English. Ph.D. thesis. University of Stockholm.
Jake, J., Myers-Scotton, C. & Gross, S. (2002). Making a minimalist approach to
code-switching work: Adding the Matrix Language. Bilingualism: Language
and Cognition, 5 (1), 69�–91.
Joshi, A. K. 1985. Processing of sentences with intrasentential code switching. In
David Dowty, Laurie Kattunen & Arnold Zwicky (eds), Natural Language
44
Parsing. Psychological, Computational, and Theoretical Perspectives.
Cambridge: University Press.
Karimi-Doostan, G. 2005. Light Verbas and Structural Case. Lingua 115, pp. 1737-
1756.
Kucerova, I. 2007a. Agreement in Icelandic: An Argument for Derivational Theory of
Intervention Effects. In Erin Bainbridge and Brian Agbayani (eds.), Proceedings
of the 34th WECOL 2006, pp. 272-284.
Kucerova, I. 2007b. An Anti-Intervention Effect in Czech Splits: An Argument for
Late Merge. In M. Goledzinowska R. Compton and U. Savchenko (eds.),
Proceedings of Formal approaches to Slavic languages 15, pp. 161-179. Ann
Arbor: Michigan Slavic Publications.
Legate, Julie Anne. 2003. Some interface properties of the phase. Linguistic Inquiry
34, pp. 506�–516.
López, Luis. 2001. On the (non)complementarity of theta theory and checking theory.
Linguistic Inquiry, 32,4: 694-716.
MacSwan, J. 1999. A Minimalist Approach to Intrasentential Code Switching. New
York: Garland.MacSwan, J. 2000.
MacSwan, J. 2000. The architecture of the bilingual language faculty: Evidence from
codeswitching. Bilingualism: Language and Cognition, 3(1), pp. 37-54.
MacSwan, J. 2005a. Codeswitching and generative grammar: A critique of the
MLFM model and some remarks on �“modified minimalism.�” Bilingualism:
Language and Congition, 8(1), pp. 1-22.
MacSwan, J. 2005b. Comments on Jake, Myers-Scotton and Gross�’s response: There
is no �“matrix language.�” Bilingualism: Language and Cognition, 8(2), pp. 277-
284.
45
Mahootian, S. & Santorini, B. 1996. Code switching and the complement/adjunct
distinction. Linguistic Inquiry 27, pp. 464-479.
Marantz, A. 1997. No escape from syntax: Don't try morphological analysis in the
privacy of your own Lexicon. In Alexis Dimitriadis et.al. (eds.), Proceedings of
the 21st Annual Penn Linguistics Colloquium. Penn Working Papers in
Linguistics 4: 2, pp. 201-225.
Myers-Scotton, C. & Jake, J. L. 2001. Explaining aspects of code-switching and their
implications. In J. Nicol (ed.), One mind, two languages: Bilingual language
processing, pp. 84�–116. Oxford: Blackwell.
Myers-Scotton, C. 1993. Dueling languages: Grammatical structure in code
switching. Oxford: Clarendon Press.
Myers-Scotton, C. 2001. The matrix language frame model: Developments and
responses. In R. Jacobson (ed.), Codeswitching worldwide II, pp. 23�–58. Berlin:
Mouton de Gruyter.
Myers-Scotton, C. 2002. Contact linguistics: Bilingual encounters and grammatical
outcomes. Oxford: Oxford University Press.
Myers-Scotton, Carol & Janice Jake 2009. A universal model of codeswitching and
bilingual language processing and production. In The Cambridge Handbook of
Linguistic Code-switching, ed. Barbara Bullock and Jacqueline A. Toribio, pp.
336-357. Cambridge: Cambridge University Press.
Newport, M. 1999. Reduced Input in the Acquisition of Signed Languages:
Contributions to the Study of Creolization. In M. DeGraf (ed) Language
Creation and Language Change: Creolization, Diachrony and Development, pp
161-178. Cambridge, Mass: MIT Press.
46
Poplack, S. 1980. Sometimes I'll start a conversation in spanish Y TERMINO EN
ESPAÑOL: toward a typology of code switching. Linguistics 18, pp. 581-616.
Ritchie, W. C. & Bhatia, T. K. 1996. Codeswitching, Grammar, and Sentence
Production: The Problem of Dummy Verbs.
(http://www.eric.ed.gov/ERICDocs/data/ericdocs2sql/content_storage_01/00000
19b/80/14/e1/ba.pdf).
Rizzi, L. 1990. Relativized Minimality. Cambridge, MA: MIT Press.
Woolford, E. 1983. Bilingual Code-Switching and Syntactic Theory. Linguistic
Inquiry 14, pp. 520-535.