Poplack, Shana. 2004. Code-switching. Soziolinguistik. An ...

Text-only Preview

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.

1. Linguistic manifestations of language contact

Code-switching (CS) is but one of a number of
the linguistic manifestations of language contact
and mixing, which variously include borrowing
on the lexical and syntactic levels, language
transfer, linguistic convergence, interference,
language attrition, language death, pidginization
and creolization, among others. There is little
consensus in the literature over which aspects
should be subsumed under the label code-
. In this article, CS refers to the
utterance-internal juxtaposition, in unintegrated
form, of overt linguistic elements from two or
more languages, with no necessary change of
interlocutor or topic.
Mixing may take place at any level of
linguistic structure, and a long research tradition
has grown up around questions of language
choice and language negotiation among
interlocutors in bilingual contexts (Gumperz
1976/1982; Heller 1982). But the combination of
languages within the confines of a single
sentence, constituent or even word, has proved
most intriguing to linguists. This article surveys
the treatment in the literature, linguistic and
social, of such intra-sentential CS.

2. Theories of CS

First dismissed as random and deviant (e.g.,
Weinreich 1953/1968); intra-sentential CS is now
known to be grammatically constrained. The
basis for this is the empirical observation that
bilinguals tend to switch intra-sententially at
certain (morpho)syntactic boundaries and not at
others. Early efforts to describe these tendencies
(e.g., Gumperz 1976/1982; Timm 1975) offered
taxonomies of sites in the sentence where CS
could and could not occur (e.g., between
pronominal subjects and verbs or between
conjunctions and their conjuncts), but these were
soon met with a host of counter-examples.

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.
The first general account of the distribution of
CS stemmed from the observation that CS is
favored at the kinds of syntactic boundaries
which occur in both languages. The Equivalence
Constraint (Poplack 1980) states that switched
sentences are made up of concatenated fragments
of alternating languages, each of which is
grammatical in the language of its provenance
(see also Lipski 1978; Muysken 2000; Pfaff
1979). The boundary between adjacent fragments
occurs between two constituents that are ordered
in the same way in both languages, ensuring the
linear coherence of sentence structure without
omitting or duplicating lexical content.
That general principles, rather than atomistic
constraints, govern CS is now widely accepted,
though there is little consensus as to what they are
or how they should be represented. Many theories
assume that the mechanisms for language
switching are directly predictable from general
principles of (monolingual) grammar. As
extensions of the formal linguistic theories
successively in vogue, these tend to appeal to
such abstract grammatical properties as inter-
constituent relationships (e.g., government, case
assignment) and/or language-specific features of
lexical categories (i.e., subcategorization of
grammatical arguments, inherent morphological
Di Sciullo, et al. (1986), for example,
identified the relevant relations as C-command
and government: CS cannot occur where a
government relation holds. Replacement of the
function of government in standard theory by the
notion of feature agreement led to a parallel focus
on feature matching in CS studies. The
Functional Head Constraint (Belazi, et al. 1994)
adds language choice to the features instantiated
in functional and lexical categories, prohibiting
CS where a mismatch occurs. MacSwan’s (1999)
adaptation of the Minimalist proposal restricts CS
at structural sites showing cross-language
differences in monolingual features.
The distinction between lexical and functional
categories is a hallmark of theories invoking the

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.
complement structure of individual lexical items
to characterize permissible CS sites (e.g., Joshi
1985 and its sequel, the Null Theory of CS
(Santorini and Mahootian 1995); see also
Bentahila and Davies’ Subcategorisation
(1983)). Perhaps the most detailed
model involving the contrast between lexical
properties and functional (or “system”)
morphemes is the Matrix Language Frame model
(Azuma 1993; Myers-Scotton 1993). Here,
structural constraints on CS result from a
complex interaction between a dominant matrix
language and the prohibition against embedding
“system” morphemes from the “embedded”
language in matrix language structure.
The assumption that bilingual syntax can be
explained by general principles of monolingual
grammar has not been substantiated. While such
formal theories of grammar may account well for
monolingual language structure (including that of
the monolingual fragments in CS discourse),
there is no evidence that the juxtaposition of two
languages can be explained in the same way. As
described in ensuing sections, bilingual
communities exhibit widely different patterns of
adapting monolingual resources in their code-
mixing strategies, and these are not predictable
through purely linguistic considerations. The
equivalence constraint, as formalized by Sankoff
(1998a; 1998b; Sankoff and Mainville 1986;
Sankoff and Poplack 1981), is a production-based
explanation of the facts of CS, which incorporates
the notions of structural hierarchy and linear
order, and accounts for a number of empirical
observations in addition to the equivalent word
order characterizing most actual switch sites.
These include the well-formedness of the
monolingual fragments, the conservation of
constituent structure, and the essential
unpredictability of CS at any potential CS site.

3. Fitting theory to data

Which of these competing (and often
conflicting) models offers the best account of

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.
bilingual CS? Testing the fit of theory with the
data of CS should be a straightforward matter;
however, disparate assumptions, goals and
domains of application have thus far hindered
such efforts. Assessment of the descriptive
adequacy of a theory of CS requires that at least
two methodological issues be resolved. One
involves identification and principled
classification of language mixing phenomena, the
other, confronting the predictions of the theory
with the data of actual bilingual behavior.

3.1. CS vs. borrowing
It is uncontroversial that CS differs from the
other major manifestation of language contact:
lexical borrowing. Despite etymological identity
with the donor language, established loanwords
assume the morphological, syntactic, and often,
phonological, identity of the recipient language.
They tend to be recurrent in the speech of the
individual and widespread across the community.
The stock of established loanwords is available to
monolingual speakers of the recipient language,
along with the remainder of the recipient-
language lexicon. Loanwords further differ from
CS in that there is no involvement of the
morphology, syntax or phonology of the donor

3.2. Borrowing vs. nonce borrowing
Recent research on borrowing as a synchronic
process (e.g., the papers in Poplack and Meechan
1998a; Poplack, et al. 1988) has shown it to be far
more productive than its result (established
loanwords) would imply. Crucially, the social
characteristics of recurrence and diffusion need
not be satisfied, resulting in what has been called,
after Weinreich (1953/1968), nonce borrowing
(Sankoff, et al. 1990). Like its established
counterpart, the nonce borrowing tends to involve
lone lexical items, generally major-class content
words, and to assume the morphological,
syntactic, and optionally, phonological identity of
the recipient language. Like CS, on the other
hand, particular nonce borrowings are neither

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.
recurrent nor widespread, and nonce borrowing
necessarily requires a certain level of bilingual
competence. Distinguishing a nonce borrowing
from CS of a lone lexical item is conceptually
easy but methodologically difficult, especially
when this item surfaces bare (i.e.,
morphologically uninflected, or in a syntactic slot
shared by both languages), giving no apparent
indication of language membership.
The classification of such lone other-language
items is at the heart of a fundamental
disagreement among CS researchers over 1)
whether the distinction between CS and
borrowing should be formally recognized in a
theory of CS, 2) whether these and other
manifestations of language contact can be
identified in bilingual discourse, and 3) criteria
for determining whether a given item was
switched or borrowed. Researchers who classify
lone other-language items as CS tend to posit an
asymmetrical relationship, in which one language
dominates and other-language items are inserted
(e.g., Joshi 1985; Myers-Scotton 1993). On the
other hand, for those who focus only on the class
of (unambiguous) multiword CS, both languages
are postulated to play a role (Belazi, et al. 1994;
Sankoff 1998a, 1998b; Woolford 1983).
Muysken (2000) admits the possibility of both

3.3. Identifying the results of language contact
Quantitative analyses of language mixing
phenomena in a wide variety of language pairs
have now established that such lone other-
language items are by far the most important—in
some cases, virtually the only!—component of
mixed discourse (e.g., Backus 1992; Berk-
Seligson 1986; Budzhak-Jones 1998a; Nortier
1989; Pfaff 1979; Poplack 1989; Poplack, et al.
1987; Treffers-Daller 1994). In comparison, CS
of multiword other-language fragments, other
than tags and other frozen forms, while frequent
in some communities, is in the aggregate
relatively rare.
Both CS and borrowing are based on

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.
principled combination of elements of the
monolingual vernaculars of the bilingual
community. Recent research suggests that the
structure of these source vernaculars can reveal
whether a code-mixed element is behaving like
one or the other. Focussing on the structural
variability inherent in CS qua oral phenomenon,
Poplack and Meechan (1998b) developed a
method, adumbrated in Sankoff et al. (1990), to
compare bilingual structures with the
monolingual source languages of the same
speakers. Making use of the framework of
linguistic variation theory (Labov 1969; Sankoff
1988), the variable patterning of such forms is
discovered, and used to determine their status.
The method involves cross-linguistic comparison,
on a given diagnostic criterion, of the ambiguous
lone other-language item, with its counterparts in
both the donor and recipient languages, as well as
with established loanwords and unambiguous CS.

3.3.1. Morphological measures
If the rate and distribution of morphological
marking and/or syntactic positioning of the lone
other-language items show quantitative parallels
to those of their counterparts in the recipient
language, while at the same time differing from
relevant patterns in the donor language, the lone
other-language items can be considered to have
been borrowed, since only the grammar of the
recipient language is operative. If they pattern
with their counterparts in the monolingual donor
language, while at the same time differing from
the patterns of the unmixed recipient language,
the lone other-language items must result from

3.3.2. Bare forms
Even where lone other-language items surface
bare, the comparative method can determine their
status. Bare forms have figured prominently in
the formulation of code-mixing theories, where
they are frequently cited as examples of
exceptional or ungrammatical ways of
incorporating foreign material (Jake and Myers-

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.
Scotton 1997; Picone 1994). Quantitative analysis
of actual CS discourse, in contrast, shows that
bare other-language forms occur overwhelmingly
in just those contexts where they are permitted in
the recipient language, and more strikingly, at the
same rate (Budzhak-Jones and Poplack 1997; Eze
1998; Ghafar Samar and Meechan 1998; Sankoff,
et al. 1990; Turpin 1998).
Indeed, code-mixed structures that appear
exceptional when compared with an idealized
version of the source language generally turn out
to conform closely to counterparts in the spoken
vernaculars of the bilinguals under study. Lack of
productivity in the recipient language may also
explain apparently unusual morphological
strategies for incorporating lone other-language
items (Poplack and Meechan 1998). Where the
status of bare forms is pursued systematically,
they are seen to mirror productive use in the
recipient language.
Empirical analyses of lone other-language
items, marked and bare, with their source-
language counterparts (Adalar and Tagliamonte
1998; Budzhak-Jones 1998a; Eze 1998; Ghafar
Samar and Meechan 1998; Poplack and Meechan
1998; Turpin 1998) confirm their quantitative
parallels with dictionary-attested loanwords. And
both pattern like their unmixed counterparts in the
recipient language, regardless of the typological
properties of the language pair. This is evidence
that most lone items are borrowed, if only for the
nonce. The same method shows CS, on the other
hand, to pattern like donor-language
counterparts, in terms of the same linguistic
criteria. Thus a first imperative in developing a
theory of CS capable of accounting for the data of
CS is to determine the status of the linguistic
elements involved.
Most of the voluminous literature on intra-
sentential CS, however, especially of the
“insertional” type (Muysken 2000), is based on
data which represents, properly speaking, lexical
borrowing. It follows that many of the theories
applying to both types of language mixing (e.g.,
Mahootian 1993; Myers-Scotton 1993) are more

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.
properly theories of borrowing. This in turn
explains on the one hand, why some seem to
account for many of the facts of code-mixing
(since most of the mixed items are in fact
borrowings), and on the other, why their handling
of (multiword) CS may appear unwieldy and/or
descriptively inadequate (e.g., Myers-Scotton
1993 and many others).

4. The data of CS

The data of CS are relevant both to evaluating
theories and to understanding the social role of
CS within the community. With respect to
evaluation, the literature on CS is largely
characterized by the “rule-and-exception”
paradigm. Despite the onslaught of counter-
examples provoked by successive CS theories, as
of this writing, few have been tested
systematically against the data of spontaneous
bilingual usage. Instead, both the theories and
assessments of their applicability tend to be based
on isolated examples, drawn from judgements,
informant elicitation, linguist introspection or the
published literature. The relation between such
examples and the recurrent and systematic
patterns of everyday interaction is tenuous or
In many bilingual communities, speakers
conventionally make use of both languages with
the same interlocutors, in the same domains, and
within the same conversational topic. To
understand the social role of CS in such
communities, the analyst must observe, uncover
and document those conventions, as instantiated
in everyday situations in which spontaneous CS
is a discourse norm. This requires first identifying
a community in which such situations regularly
arise, and characterizing its social structure in
terms of language knowledge and language use.
Second, samples of sustained discourse including
CS must be obtained from enough community
members in quantities sufficient to detect
recurrent patterns of speech behaviour. It is in
these steps, prior to any linguistic analysis, that

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.
social, political, historical and demographic
knowledge of the community are most pertinent.
These characteristics could then be related to its
members’ linguistic production to arrive at a
community profile, or “social meaning” of CS.
Curiously, however, although the last three or
four decades of research have produced a wealth
of data from a wide range of bilingual
interactions world-wide, relatively little is known
of the bilingual norms of the communities from
which they are drawn. Nor is it clear how the
social forces typically described in such detail
(Backus 1996; Gardner-Chloros 1991; Nortier
1989) shaped those norms, let alone the structural
form of the language mixes, beyond the fact that
two or three languages ended up being spoken.
As detailed below, in most bilingual communities
empirically studied, one or another manifestation
of language contact is (inexplicably) preferred to
the detriment of others; thus the social “meaning”
of the languages, individually or in combination,
reveals little about the differential use of
linguistic resources in the social life of a given
community. This is because the patterning of
utterances containing elements from more than
one language is not predictable from community
or language typologies. It emerges only from
systematic examination of how the languages are
used by community members.

5. Community Strategies for CS

When two languages are combined in a single
sentence, various problems of incompatibility
may arise. The most obvious derive from word-
order differences, but incompatibilities may affect
any level of linguistic structure, especially in
typologically distinct language pairs.
Nonetheless, it has been observed repeatedly in
systematic studies of bilingual communities that
speakers tend to circumvent these difficulties,
producing bilingual structures which are
felicitous for the grammars of both languages
simultaneously. This is achieved through
participation in prevailing community norms,

Poplack, Shana. 2004. Code-switching. Soziolinguistik. An international handbook of the science
of language
, 2nd edition, ed. by U. Ammon, N. Dittmar, K.J, Mattheier & P. Trudgill. Berlin:
Walter de Gruyter.
relating to both the overall rate and type of
language mixing. In what follows we detail four
empirically established community-wide
strategies for combining languages intra-
sententially: smooth code-switching at
equivalence sites, flagged code-switching,
constituent insertion and nonce borrowing.

5.1. Equivalence-based CS
The New York Puerto Rican community, with
a high degree of Spanish-English bilingualism,
favours smooth intra-sentential CS,
grammatically constrained by the equivalence
constraint (Poplack 1980). Characteristics of
smooth CS include copious occurrences, smooth
transitions between languages, and lack of
rhetorical effect. Also documented as a norm in
other Spanish-English bilingual communities
(e.g., Pfaff 1979), this pattern is sometimes
attributed to the many typological similarities
enjoyed by the Spanish-English pair. However,
the operation of the equivalence constraint has
been empirically verified in communities
featuring such typologically distinct language
pairs as Finnish-English (Poplack, et al. 1987),
Arabic-French (Naït M'Barek and Sankoff 1988),
Tamil-English (Sankoff, et al. 1990), Fongbe-
French and Wolof-French (Meechan and Poplack
1995), Igbo-English (Eze 1998), French-English
(Turpin 1998) and Ukrainian-English (Budzhak-
Jones 1998a).

5.2. Flagged CS
It is logical that typologically similar language
pairs should be particularly propitious to intra-
sentential CS, but its occurrence in them is by no
means a foregone conclusion. The French/English
situation in the bilingual Ottawa-Hull region of
Canada is a case in point. Instead of engaging in
smooth intra-sentential CS at the many available
equivalence sites, French-English bilinguals
prefer to flag CS and use them for specific
rhetorical purposes (Poplack 1985). Flagged
switches are marked at the discourse level by
repetition, metalinguistic commentary, and other