18.2.1 Constraints in linguistic theory

As early as 1955, Chomsky had noted that the transformational component in a hybrid generative–transformational system had the disadvantage of vastly increasing the expressive power of the grammar, permitting the formulation of grammatical processes that did not seem to occur in any language. In response to the problem, Chomsky (Reference Chomsky1964, Reference Chomsky1965) and other researchers such as Ross (Reference Ross1967) posited constraints on transformational rules. Ross noticed, for instance, that a Noun Phrase (NP) could not be extracted out of a conjoined phrase, as in (2a), accounting for the ill-formedness in (2b), but could be extracted in the semantically equivalent (but syntactically divergent) example in (2c), where t denotes trace, the point of extraction of the NP, questioned as what.

While the focus of efforts to constrain the grammar was placed primarily on the transformational component, these extended to the phrase structure component as well, culminating in the formulation of X’ Theory in Chomsky (Reference Chomsky, Jacobs and Rosenbaum1970) (see Newmeyer Reference Newmeyer1986). Nonetheless, constraints were typically viewed as psychologically real restrictions on the application of transformations to phrase markers, and were therefore understood to be imposed at the level of surface structure.

The idea of a constraint in syntactic theory appealed to a number of researchers in CS, and was used to articulate the grammatical restrictions observed in CS data. While some thought of these constraints in the technical sense, as actual grammatical constructs, others used the term informally, intending to refer only to descriptive restrictions on language mixing.

18.2.2 Equivalence-based analyses

Several researchers converged simultaneously on the notion that language switching is controlled by some kind of syntactic equivalence requirement. Lipski was among the first to express the idea, hypothesizing that elements appearing in an utterance after a switch must be “syntactically equivalent” (Lipski Reference Lipski and Paradis1978:258). As Pfaff similarly suggested, “Surface structures common to both languages are favored for switches” (Reference Pfaff1979:314). Poplack (Reference Poplack1978, Reference Poplack and Durán1981) articulated this perspective in terms of her well-known Equivalence Constraint, augmented by The Free Morpheme Constraint, given in (3) and (4).

As a variationist (see Labov Reference Labov1963), Poplack argues that linguistic rules correlate with social structure and should be stated in terms of statistical frequencies, hence (3) is expressed as a tendency. The general idea is nonetheless clear: CS is allowed within constituents so long as the word order requirements of both languages are met at surface structure. Surface structures derive from the (cyclical) application of transformations to phrase markers, which originate as the output of a phrase structure grammar. The constraint in (4) defines a restriction on morphology in CS contexts, also noted in Wentz and McClure (Reference Wentz, Erica and Ingemann1977) and Pfaff (Reference Pfaff1979). To illustrate, (3) correctly predicts that the switch in (5) is disallowed, because the surface word order of English and Spanish differ with respect to object pronoun (clitic) placement; (4) correctly disallows (6), where an English stem is used with a Spanish bound morpheme without the phonological integration of the stem.

Research since Poplack’s initial proposals has found persuasive documentation that her Equivalence Constraint is empirically inadequate, that is, it does not account for the full range of relevant data (Stenson Reference Stenson and Hendrick1990; Lee Reference Lee1991; Myers-Scotton Reference Myers-Scotton1993a; Mahootian Reference Mahootian1993; MacSwan Reference MacSwan1999b; Chan Reference Chan1999; Muysken Reference Muysken2000). Note, for example, the contrast in (7) from Spanish–English CS, noted by Belaz et al. (Reference Belazi, Rubin and Toribio1994).

The surface structure of Spanish and English are alike with regard to the construction in (7), yet a switch between the auxiliary and the verb renders the sentence ill-formed, but not so in the case of a switch between the subject and the verb. However, (3) predicts that both examples should be well-formed.

Also consider the examples in (8), from MacSwan (Reference MacSwan1999b), where code switches occur between a subject pronoun and a verb, both in their correct surface structure position for both Spanish and Nahuatl, yet one example is ill-formed and the other well-formed.

The descriptive adequacy of Poplack’s Free Morpheme Constraint, on the other hand, remains controversial. While it is attested in numerous corpora (Bentahila and Davies Reference Bentahila and Davies1983; Berk-Seligson Reference Berk-Seligson1986; Clyne Reference Clyne1987; MacSwan Reference MacSwan1999b), others claim to have identified some counter-examples (Eliasson Reference Eliasson1989; Bokamba Reference Bokamba1989; Myers-Scotton Reference Myers-Scotton1993a; Nartey Reference Nartey1982; Halmari Reference Halmari1997; Chan Reference Chan1999; Hlavac Reference Hlavac2003). However, in presenting counter-examples to The Free Morpheme Constraint, researchers have often given too little attention to the specific phonological, morphological, and syntactic characteristics of the examples cited, making it difficult to determine whether they are in fact violations. For Poplack, items that are phonologically integrated into the language of the bound morpheme are regarded as borrowings rather than code-switches. This is made explicit in subsequent formulations of the constraint, as in Sankoff and Poplack (Reference Sankoff and Shana1981:5): “A switch may not occur between a bound morpheme and a lexical item unless the latter has been phonologically integrated into the language of the bound morpheme.” Thus, examples in which an other-language stem has been phonologically integrated into the language of an inflectional affix do not constitute counter-examples to The Free Morpheme Constraint.

Poplack’s constraints have been criticized as a third grammar, a term originally coined by Pfaff (Reference Pfaff1979) to designate a system designed to mediate between the two languages present in a mixed utterance, and applied specifically to Poplack’s constraints by Lederberg and Morales (Reference Lederberg and Cesáreo1985), Mahootian (Reference Mahootian1993), and MacSwan (Reference MacSwan2000), among others. As Lipski (Reference Lipski1985:83–84) noted, such mechanisms should be admitted only as a last resort:

However, although Sankoff and Poplack (Reference Sankoff and Shana1981) similarly expressed a strong preference for avoiding CS-specific mechanisms to mediate between the two languages in contact, they nonetheless concluded that something of the sort appeared to be necessary on empirical grounds. Otherwise, the authors argued, the free union of Spanish and English phrase structure grammars would yield ill-formed results. For instance, whereas English requires pre-nominal adjectives (NP → Det Adj N), Spanish requires post-nominal adjective placement (NP → Det N Adj). A speaker is free to select the Spanish rule and lexically insert an English determiner, Spanish noun and English adjective (*the casa white) or even insert English lexical items for all categories (*the house white). Therefore, in order to constrain the grammars so that they do not generate violations of (3), Sankoff and Poplack introduced a superscripting (“bilingual tagging”) mechanism that restricted lexical insertion rules so that the grammar contributing the phrase structure rule would also be the grammar from which lexical insertion rules would be drawn. Hence, under conditions of CS, the Spanish phrase structure rule would be annotated as in (9a), generating (9b). The superscripting conventions followed from heritability conditions, which essentially allowed phrase structure rules to look ahead and restrict the application of lexical insertion rules.

Sankoff and Poplack do not make explicit the mechanisms for superscript insertion; rather, they indicate that phrase structure rules are so superscripted when they are selected in the generation of a code-switched utterance, and are subsequently used to trigger language-specific lexical insertion rules (N → casa, for instance, in the case of N^sp:n). No account is presented as to how the superscript insertion mechanism is able to annotate the appropriate categories correctly – for instance, N and Adj in (9a), but not Det, where either language may be inserted without negative consequences. For these reasons, the superscripting mechanism, like The Equivalence Constraint and The Free Morpheme Constraint, appears to constitute a CS-specific mechanism, a marked disadvantage.¹

Woolford (Reference Woolford1983) similarly attempted to derive The Equivalence Constraint working within the basic assumptions of the Aspects model. Like Pfaff (Reference Pfaff1979) and others before her, Woolford emphasized that our best account of CS would avoid reference to any kind of CS-specific grammar. And like Sankoff and Poplack, Woolford recognized the basic dilemma of providing lexical items with access to the structure of the sentence in which they were inserted:

In other words, Woolford believed that lexical insertion was unconstrained in the case of phrase structure rules common to both languages; but in the case of phrase structure rules that were not shared, lexical insertion was limited to the terminal nodes associated with the phrase structure rule of the grammar to which it belonged. Woolford’s system does not seem to achieve its intended results, as it predicts that Spanish–English CS would require that a language-unique phrase structure rule (for instance, NP → Det N Adj for Spanish) could only be lexically filled by Spanish items (predicting the casa blanca to be ill-formed, contrary to the facts). In addition, while Woolford’s work is an excellent example of the articulation of the goals of CS research, she does not herself present the formal mechanism that might be responsible for achieving the results expected within her framework. No explanation as to how the unique phrase structure rules get linked to language-specific rules of lexical insertion is offered.

Woolford accounts for Poplack’s Free Morpheme Constraint by postulating that “the lexicons and word formation components of the two grammars remain separate” (Reference Woolford1983:526). While this approach seems preferable to Poplack’s, where the prohibition against word-internal switching is simply stated in descriptive terms, no rationale for the separation of the lexicons in terms of principles independent of CS itself is offered, leaving the basis for asserting that the model is free of any CS-specific mechanisms inexplicit.

18.3.1 The Government Constraint

Working within the GB framework, Di Sciullo et al. (Reference Di Sciullo, Muysken and Singh1986) proposed The Government Constraint, which posited that there is an anti-government requirement on CS boundaries. Using the standard definition of government in (10), the authors posed (11) as a condition on lexical insertion (where q indexes a category to the language-particular lexicon).

Di Sciullo et al.’s intuition was that (11) is a narrower and empirically more accurate version of (12), which they viewed as a common assumption in syntactic theory that is never made explicit.

On these provisions, the authors maintained that CS “can be seen as a rather ordinary case of language use, requiring no specific stipulation” (Di Sciullo et al. Reference Di Sciullo, Muysken and Singh1986:7). In order to permit the head carrying the language index q to percolate up to its maximal projection, they formalized the condition on CS as The Government Constraint, given in (13).

This formalism allows the language of a head to determine the syntax of its maximal projection and imposes the condition that two categories must be in the same language if the government relation holds between them.

Much like Sankoff and Poplack’s (Reference Sankoff and Shana1981) formalism, (13) (like (11)) attempts to trigger language-specific lexical insertion by identifying nodes within a phrase marker with a specific language label (termed a language index). Although the authors maintain that the mechanism underlying the language index is vacuously available to monolinguals as well, it nonetheless appears to add few advantages over Sankoff and Poplack’s version. Indeed, the constraint in (13) must be seen as a primitive in the system of grammar or it would not have the desired effect, and the very motivation for proposing it is to account for the data of CS; hence, (13) also appears to be a CS-specific constraint.

In addition, there are important counter-examples to (13), some of which were noted by Di Sciullo, et al. For instance, because government holds between a verb and its object and between a preposition and its object, (13) predicts that a verb or preposition must be in the language of its complement. This is shown to be incorrect by examples in (14), where switches occur in case-marked positions.

18.3.2 The Functional Head Constraint

Belazi et al. (Reference Belazi, Rubin and Toribio1994) proposed The Functional Head Constraint (FHC), which took advantage of a recent development in syntactic theory that distinguished between lexical and functional categories (Abney Reference Abney1987). Functional categories, or functional heads, were responsible for selecting complements with specific feature matrices. For example, for is a head (C⁰) and has a feature specifying that its complement must have the feature [-Tense]. Belazi, Rubin, and Toribio argued that the data of CS can be correctly described in terms of the generalization in (15).

The authors developed the FHC, given in (16), intended as a refinement of Abney’s (Reference Abney1987) proposal.

By language feature, the authors mean a label identifying the language from which an item was contributed, such as [+Spanish] or [+English]. If the features do not agree (e.g. a Spanish functional head with an English complement), then the code-switch is blocked. Since (16) applies only to f-selected configurations (a complement selected by a functional head), switches between lexical heads and their complements are not constrained.

Mahootian (Reference Mahootian1993) and Muysken (Reference Muysken2000) see the FHC as a further elaboration of The Government Constraint, in that it identifies an independently motivated principle of grammar but incorporates language-specific identifiers (for The Government Constraint, a language index; for the FHC, a language feature). Belazi et al., like Di Sciullo et al. (Reference Di Sciullo, Muysken and Singh1986) with respect to The Government Constraint, maintained that the FHC does not constitute a CS-specific constraint. However, although the constraints were formulated in terms of independently motivated operations, the particular language identifiers were not. Linguists take particular grammars to be derivative in nature, not primitive constructs, and hence positing a label for a particular language as a primitive in syntactic theory leads us to an ordering paradox, as MacSwan (Reference MacSwan1999b) has pointed out. In addition, (15) remains controversial as a descriptive generalization (Mahootian Reference Mahootian1993; MacSwan Reference MacSwan1999b; Muysken Reference Muysken2000).

However, MacSwan (Reference MacSwan1999b) noted that the content of Belaz et al.’s theory is greatly improved if we regard [+English] as an informal reference to a collection of formal features that define English. On this view, names for particular languages act as proxies for bundles of features which formally characterize them. The ordering paradox disappears, because language features like [+English] or [+Spanish] are no longer taken to be primitives in the theory of grammar. To evaluate the FHC, then, particular hypotheses would be needed regarding which features of a language, being distinct from features of another, result in a conflict. Such conflicts might arise in numerous configurations besides those where head-complement relations hold, leading us to move beyond the FHC to propose a wider diversity of grammatical configurations where CS might be illicit.

18.3.3 The Null Theory

Mahootian (Reference Mahootian1993) proposed the Null Theory of CS, formulated within the framework of Tree Adjoining Grammars (TAG) originally introduced by Joshi (Reference Joshi, Dowty, Karttunen and Zwicky1985b) for applications in computational linguistics and natural language processing. TAG differs from mainstream generative grammar in that the lexical items encode partial tree structures, and use operations of substitution and adjunction to assemble larger trees composed of multiple lexical items. For example, the verb build is represented in the lexicon along with its projection, and therefore the branching direction of its complement is lexically specified. A substitution operation allows a DP (e.g. a house) to integrate with build by substituting the DP (along with its category label) with the object category label of build.

Mahootian focused on the complement relation in phrase structure (see Pandit Reference Pandit and Jacobson1990; Nishimura Reference Nishimura1997), and claimed that (17) was adequate to account for the facts of CS.

Mahootian (Reference Mahootian1993) used a corpus of Farsi–English CS data that she collected in naturalistic observations. In Farsi, objects occur before the verb, contrasting with basic word order in English. She observed that in CS contexts the language of the verb determines the placement of the object, as (18) illustrates. These facts are consistent with (17).

Mahootian noticed that the TAG formalism provides an advantage for the analysis of CS data. Because structures are encoded in the lexicon, no intervening control mechanism is needed to pair up lexical insertion rules with terminal nodes in a phrase marker, as seen in previous proposals. However, like Belazi et al. (Reference Belazi, Rubin and Toribio1994), Mahootian’s analysis was restricted to head-complement configurations (MacSwan Reference MacSwan1999b; Muysken Reference Muysken2000). Not only was (17) too narrow in this regard, failing to comment on CS in other dimensions of syntax, but it also proved to be insufficiently restrictive. Note, for instance, the examples in (19), a phenomenon observed by Timm (Reference Timm1975). Although all complements are in the correct positions assigned by heads, (19a) is ill-formed but (19b) well-formed. The contrast in grammaticality appears to relate to the nature of the subject, a specifier (XP) rather than a pronominal head (van Gelderen and MacSwan 2008).

Furthermore, recall also that in (7), visto “seen,” the complement of habían/had, is in the position assigned by its head, and therefore adheres to (17), yet (7a) is well-formed and (7b) is not. In (20), Spanish and Nahuatl word order is respected with regard to the placement of the verbal complement of negation, yet (20a) is ill-formed but (20b) is not.

These examples indicate that restrictions on CS are far more pervasive than the head-complement relation alone, and appear to move well beyond issues of phrase structure alone.

18.4.1 The lexicalist advantage

The importance of constructing a theory of CS that does not appeal to CS-specific mechanisms has been emphasized throughout the history of the field. However, in essential respects, the theoretical contexts in which many influential theories were formulated did not provide the tools needed to permit the implementation of a constraint-free theory of CS. An approach to syntax which built structure from the top down, as in the Aspects and later GB models, postponed lexical insertion until well after the word order had been laid out, posing a significant problem. The structure could not be sensitive to which language contributed a specific lexical item until the end, when lexical insertion occurred, but the language contributing the lexical item appeared to have strong consequences for the syntactic structure at the onset.

The desire to avoid CS-specific mechanisms in accounts of CS goes beyond issues of elegance and economy. The more serious problem is that such mechanisms threaten to trivialize the enterprise. Rather than explaining descriptive restrictions observed in CS data, CS-specific mechanisms simply note these restrictions within the grammar itself so that no explanation is needed, and so one is left still wondering what general principles of grammar might be at work in posing the observed restrictions. Within the Minimalist Program, structures are built from a stock of lexical items, essentially beginning with lexical insertion (formalized as Select). This important development permits CS researchers to probe the structural consequences of particular lexical items from specific languages, with no need to keep track of which languages may contribute which specific lexical elements during a final stage of lexical insertion.

18.4.2 The Minimalist Program

X’ Theory effectively eliminated phrase structure grammar in favor of the view that structures are projected from lexical items; however, remnants remained, with reference to lexical insertion rules reasonably common among GB era syntacticians (Chomsky Reference Chomsky1981; Stowell Reference Stowell1981; Lasnik and Uriagereka Reference Lasnik and Uriagereka1988). Apparent redundancies among various modules of grammar within the GB framework were troubling. Subcategorization, θ-Theory, and X’ Theory all appeared to approach the same basic problems from a different angle, with none sufficient to manage the full array of issues associated with the base generation of an initial phrase marker. According to Chametzky (Reference Chametzky and Hendrick2003), the “lexical entry driven” approach to syntax was part of the general effort underlying X’ reduction, with significant contributions from Stowell (Reference Stowell1981) and Speas (Reference Speas1990), among others. With a return to its derivational roots, Minimalist syntax reduced generation to the simplest possible form – free Merge (Chomsky Reference Chomsky and Kasher1991, Reference Chomsky1994), building structures from the ground (the lexical string) up (the hierarchical phrase structure), based on the specification of lexically encoded features. Independently, Borer (Reference Borer1984) had suggested an account of language variation in which parameters were also associated with the lexicon, rather than with the system of syntactic rules. Hence, the system of rules could be seen as invariant, with all variation associated with the lexicon, the traditional repository of arbitrariness.

In the Minimalist Program there are two components of grammar: C_hl, a computational system for human language, believed to be invariant across languages; and a lexicon, to which the idiosyncratic differences observed across languages are attributed. An operation called Select picks lexical items from the lexicon and introduces them into a Numeration or Lexical Array (LA), a finite subset of the lexicon used to construct a derivation. Merge takes items from the LA and forms new, hierarchically arranged syntactic objects. Movement operations (Internal Merge) apply to syntactic objects formed by Merge to re-arrange elements within a tree (Chomsky Reference Chomsky1995, Reference Chomsky, Martin, Michaels and Uriagereka2000). Phrase structure trees are thus built derivationally by the application of the operations Select and Merge, constrained by the condition that lexically encoded features match in the course of a derivation.

Movements are driven by feature valuation, and may be of two types. A head may undergo head movement and adjoin to another head, or a maximal projection may move to the specifier position of a head. In either case, the element moves for the purpose of valuing morphological features of case and φ (number, person, and gender). In addition, its movement may be overt or covert. Overt movements are driven by strong features and are visible at PF (Phonetic Form, where they are pronounced) and LF (Logical Form, where they are interpreted). Covert movements, driven by weak features, are visible only at LF.

Principles of Economy select among convergent derivations. One such principle, Full Interpretation (FI), requires that no symbol lacking a sensorimotor interpretation be admitted at PF. Applied at LF, FI entails that “every element of the representation have a (language-independent) interpretation” (Chomsky Reference Chomsky1995:27). Thus, uninterpretable features (denoted -Interpretable) must be checked and deleted by LF. The +Interpretable features are categorial features plus φ-features of nominals; the +Interpretable features do not require valuation (checking). A derivation is said to converge at an interface level (PF or LF) if it satisfies FI at that level; it converges if FI is satisfied at both levels. A derivation that does not converge is also referred to as one that crashes. If features are not valued, the derivation crashes; if they mismatch, the derivation is canceled (that is, a different convergent derivation may not be constructed).

At some point in the derivation, an operation, Spell-Out, applies to strip away from the derivation those elements relevant only to PF; what remains is mapped to LF by a subsystem of C_hl called the covert component. The subsystem of C_hl that maps the lexicon to Spell-Out is the overt component. The phonological component is also regarded as a subsystem of C_hl. Note that the various components (overt, covert, phonological) are all part of C_hl, the computational system for human language. The model could be represented graphically as in Figure 18.1 (MacSwan Reference MacSwan1999a).

Figure 18.1 Model of the minimalist framework

It may be helpful to consider some illustrations of how feature checking drives movement in the Minimalist Program. As mentioned, movement may be of two types: (1.) a head (or X⁰) moves by adjunction to another head, forming a complex X⁰; or (2.) an XP moves to the Specifier position of another XP. Movement is driven by a need to value features, and the configuration into which the element moves for feature valuation constitutes its Checking Domain. XP movement is illustrated in (21). The tree in (21a) is formed by successive application of Merge, which uses lexically encoded categorial features (e.g. V, N, D, T) to build a classical phrase structure representation. The subject bears a case feature. T is a nominative case assigner that attracts the case feature of the subject to its Specifier position, bringing the full DP along. Because case is an uninterpretable feature, it must be valued and deleted before LF, or the derivation will crash. Hence, the DP is attracted to T, and moves to its Checking Domain, the Spec[ifier] of TP, as shown in (21b).

There are some terminological departures from earlier generative models. TP, or Tense Phrase, replaces IP, Inflection Phrase; DP, or Determiner Phrase, is headed by a determiner and dominates NP, as illustrated. The subject originates in a VP-internal position, following Koopman and Sportiche (Reference Koopman and Sportiche1991), raising to the Spec of T to check its case feature. t is the trace of movement, co-indexed by i, as in classical approaches. Chomsky (Reference Chomsky1994, Reference Chomsky, Martin, Michaels and Uriagereka2000) adopts a bare phrase structure approach, dispensing with intermediate bar levels when they are not relevant to output conditions. As is traditionally done, we have simplified aspects of the structure in (21) (and elsewhere) that are not relevant in the present context.

Now consider an example of head movement, illustrated in (22). The successive application of Merge results in the formation of a base structure. As in (21), the subject DP moves out of the VP shell to check its case feature. The resulting structure is shown in (22a). V, a head, moves to T by head adjunction in order to value and delete its φ-features, as shown in (22b).

Feature strength (weak, strong) is the primary mechanism in the MP used to account for cross-linguistic variations in word order. Notice, for instance, the contrast in (23).

In English, VP-adverbs precede verbs, but in French they follow them. We might assume, then, that in English V moves to T covertly, attracted by T’s weak φ-features. This is represented in (22b) with the use of parentheses around the verb, illustrating that the phonetic features of the V have been left behind. By contrast, in French, T has strong φ-features, resulting in overt movement. In this case, all of V’s features raise, with the result that it appears before its adverbial modifier in (23).

Feature strength can similarly be used to account for word order differences in the case of XP movement. For instance, if the case feature of T is strong, then the subject DP must move overtly out of its VP shell, bringing along its phonetic content. Overt movement of the subject DP results in preverbal subject word order (as in English, French, or Spanish). However, if the case feature is weak, then the subject DP will move covertly, resulting in postverbal word order (e.g. in Irish, Breton, or Zapotec). Let us now turn to an analysis of CS data within the Minimalist Program.

18.4.3 The analysis of code-switching in the Minimalist Program

The leading aim of the Minimalist Program is the elimination of all mechanisms that are not necessary and essential on conceptual grounds alone. Thus, only the minimal theoretical assumptions may be made to account for linguistic data, privileging more simplistic and elegant accounts over complex and cumbersome ones. These assumptions would naturally favor accounts of CS that make use of independently motivated principles of grammar over those that posit rules, principles, or other constructs specific to it. MacSwan (Reference MacSwan1999b, Reference MacSwan, Bhatia and Ritchie2004) presents this research program in the context of the Minimalist Program as in (24), where the minimal CS-specific apparatus is assumed.

Notice that (24) does not use “constrain” in a descriptive sense, to imply that there are no unacceptable code-switched sentences. Rather, constrain is used in its technical sense here, to mean that there are no statements, rules, or principles of grammar which refer to CS. In other words, (24) posits that all of the facts of CS may be explained just in terms of principles and requirements of the specific grammars used in each case. More formally, the claim is that for G_x a grammar of Lx and Gy a grammar of Ly, CS falls out of the union of the two grammars ({Gx ∪ Gy}) and nothing more (MacSwan Reference MacSwan1999a). In this respect ungrammaticality in CS is understood to relate to mechanisms motivated for the analysis of monolingual language, or which are conceptually necessary for reasons of optimal design.

Note that our conception of these conflicts is very much determined by our conception of the organization of the grammar. In classical GB theory, parametric differences were generally assumed to be properties of the computational system. For instance, noting that some subjacency violations of the English variety are acceptable in Italian, Rizzi (Reference Rizzi1982) proposed that the bounding nodes for the Subjacency Principle were parameterized (NP and IP in English, NP and CP in Italian). On this conception of parametric variation, in which the computational system itself differs across languages, it is very difficult to know how a conflict in language-specific requirements should be precisely defined. In an Italian–English mixed construction, for instance, what determines whether the sentence will be sensitive to IP or CP as a bounding node for the purposes of the Subjacency Principle? The answer depends upon which computational system is in use (Italian or English), and it is very unclear what factors might determine this in the absence of a system permitting CS-specific constraints to mediate conflicts.

In a Minimalist approach to CS which adheres to the research agenda stated in (24), lexical items may be drawn from the lexicon of either language to introduce features into the lexical array, which must then be valued (and deleted by LF, in the case of uninterpretable features) in just the same way as monolingual features must be valued, with no special mechanisms permitted. In this lexicalist approach, no CS-specific mechanism is required to mediate contradictory requirements of the systems in contact. The requirements are simply carried along with the lexical items of the respective languages.

18.4.3.2 Code-switching and conditions on interface levels

Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky2001a) and Boeckx and Stjepanovic (Reference Boeckx and Sandra2001) have recently suggested that head movement is a phonological operation. The motivation for the idea derives from a need to address a range of issues in syntactic theory. In the context of CS research, associating head movement with the phonological component suggests some common ground between the ban on CS in head movement contexts and the prohibition against word-internal CS noted by Poplack (Reference Poplack and Durán1981). In both instances, switching from one phonological representation to another within a word-like unit is disallowed. Recall Poplack’s (Reference Poplack1980) classic examples of the ban on word-internal mixing of the sort illustrated in (34).

(34)

(34)

1. a. *Juan está eat-iendo

   Juan be/1Ss eat-dur

   “Juan is eating.”

2. b. *Juan eat-ó

   Juan eat-past/3Ss

   “Juan ate.”

3. c. * Juan com-ed

   Juan eat-past

   “Juan ate.”

4. d. * Juan eat-ará

   Juan be/1Ss eat-fut/3Ss

   “Juan will eat.”

In cases such as (34), CS occurs within a single syntactic head, a structure represented in (35a). In the head movement cases, CS occurs in the context of a complex head, as in (35b).

(35)

(35)

In order to remain true to our goal of positing no CS-specific constraints, the ban on switching in word-internal and head movement contexts may not be declaratively stated as a constraint on syntax, but must be derived from independent principles. As indicated in Figure 18.1, at Spell Out a derivation is split, with features relevant only to Phonetic Form (PF) sent to the phonological component where the phonological system maps them to PF, and interpretable material is treated by further application of the syntactic component in the mapping to Logical Form (LF) in anticipation of semantic interpretation.

A common assumption in CS research is that the linguistic identity of a word (as “Spanish” or “English” or “Arabic”) is established by its morphological and phonological characteristics (Lipski Reference Lipski and Paradis1978; Pfaff Reference Pfaff1979; Woolford Reference Woolford1983; Di Scuillo et al. Reference Di Sciullo, Muysken and Singh1986; Mahootian Reference Mahootian1993; MacSwan Reference MacSwan1999a, Reference MacSwan1999b). For instance, the word taipiar may derive from English type, but taipiar is regarded as a “Spanish word” because it has phonological and morphological properties that are generally compatible with the grammar of the community of speakers known as “Spanish speakers.” The same is true of nonce or novel borrowings. In an important respect, then, CS research is concerned with interface conditions on morphophonology and syntax across discretely represented linguistic systems.

The Minimalist framework assumes that processes of word formation apply before an item is introduced into the Lexical Array when syntactic operations begin (see Figure 18.1) (Chomsky Reference Chomsky1995:20). Chomsky (Reference Chomsky1995, Reference Chomsky, Barbosa, Fox, Hagstrom, McGinnis and Pesetsky1998) stresses that the phonological system has a dramatically different character from the syntactic system. Specifically – with π indicating the PF representation, λ the LF representation, and N the initial collection of lexical items – Chomsky (Reference Chomsky1995:229) posits that:

. . . at the point of Spell-Out, the computation splits into two parts, one forming π and the other forming λ. The simplest assumptions are (1) that there is no further interaction between computations and (2) that computational procedures are uniform throughout: any operation can apply at any point. We adopt (1), and assume (2) for the computation from N to λ, though not for the computation from N to π; the latter modifies structures (including the internal structure of lexical entries) by processes very different from those that take place in the N → λ computation.

We assume that affixation interacts with phonology (at least) pre-lexically (before items are selected into the Lexical Array), and that phonology is sensitive to word boundaries or discrete syntactic heads.

What properties of the grammar might explain the prohibition against switching head-internally, banning language mixing in structures such as (34)? Current approaches in phonology posit that lexical form (input) is mapped to the surface form (output) in one step, with no intermediate representations, and hypothesize that phonological constraints are prioritized with respect to each other on a language-specific basis. Each set of internally ranked constraints is a constraint dominance hierarchy, and a language-particular phonology is a set of constraint dominance hierarchies (see McCarthy Reference McCarthy2002). Since language-particular phonologies differ with respect to their internal rankings, we might reasonably posit that bilinguals have a separately encapsulated phonological system for each language in their repertoire in order to avoid ranking paradoxes resulting from the availability of distinct constraint dominance hierarchies with conflicting priorities. This property of the bilingual language faculty emerges as a result of the design constraints imposed by the phonological system; without it, bilingualism would not be possible. It further leads us to anticipate that phonological systems may be switched between syntactic heads but not within them, since every syntactic head must be phonologically parsed at Spell Out, and the mapping of phonological structure occurs in a single step, with no intermediate representations and therefore no opportunities for switching from one phonological system to another. We state the condition as in (36) as the PF Interface Condition.

(36)

(36) The PF Interface Condition

1. (i.) Phonological input is mapped to the output in one step with no intermediate representations.

2. (ii.) Each set of internally ranked constraints is a constraint dominance hierarchy, and a language-particular phonology is a set of constraint dominance hierarchies.

3. (iii.) Bilinguals have a separately encapsulated phonological system for each language in their repertoire in order to avoid ranking paradoxes, which result from the availability of distinct constraint dominance hierarchies with conflicting priorities.

4. (iv.) Every syntactic head must be phonologically parsed at Spell Out. Therefore, the boundary between heads (words) represents the minimal opportunity for CS.

By stipulating that syntactic heads subject to phonological parsing include both simple and complex heads, as illustrated in (35), we extend (36) to both word-internal CS and CS in head movement contexts. We might alternatively posit that head movement is itself a phonological operation that first builds a complex sequence of phonological features deriving from both adjoined heads, then attempts to subject them to phonological processing as a single word-like unit.

We may now appeal to the PF Interface Condition to account for the ban on CS in head movement contexts. Because the condition follows from independently motivated properties of the grammatical system, it complies with the research agenda articulated in (24), namely, the supposition that there are no rules or principles of grammar that refer to CS. We turn to a final example, CS among languages that differ with respect to their basic word order requirements, calling upon the PF Interface Condition to help identify syntactic properties of lexically null functional categories.

19.1.1 No chaos allowed: the Uniform Structure Principle

This chapter elaborates and illustrates the research program framed by the MLF model. What is new is that it emphasizes how a principle of uniform structure drives the explanation of what does and does not occur in CS. The first goal of this chapter is to show how the Uniform Structure Principle (USP) implies a particular view of processing and production in bilingual speech, especially in CS. Bilingual speech is defined as surface level morphemes from two or more language varieties in the same clause. With the 4-M model, the USP clarifies and strengthens the Matrix Language Frame (MLF) model, as a model of CS. A succinct way of viewing the USP is the phrase, “no chaos allowed.” This may be obvious for monolingual speech; that it applies to bilingual speech is not so obvious. A priori, the ways in which languages participate in bilingual speech are unconstrained. In particular, the source of grammatical structure within a bilingual clause could be shared in any number of ways. But this does not happen. For bilingual speech, “no chaos allowed” means a particular asymmetry between the participating languages. This is formalized in the USP, as follows:

The second goal here is to make more explicit how specific morphemes are classified under the 4-M model and to show how differences in morpheme type explain their distribution in CS. These distributions will be shown to reflect the USP. In doing this, the chapter focuses on CS in general, but gives special attention to prepositions, complementizers, and pronouns.

First, the MLF model of CS and its relation to the USP is summarized. Next, the view of language production motivated by empirical CS data and the MLF model is outlined, as is the 4-M model and how it relates to the USP and the MLF model. Finally, the descriptive sections of the chapter are shown to support the theoretical goal of explaining the asymmetries that pattern CS data. With the USP as an overarching framework in which the MLF and 4-M models add specific hypotheses, a set of principled predictions emerges about what does and does not occur in CS. These predictions should have relevance to other types of contact phenomena as well.

19.2.1 Exemplifying the MLF model

Example (1) comes from a corpus of Turkish–Dutch CS, where elements from the EL appear in italics. Turkish is verb-final, and in this example the inflected (main) verb yap “do” occurs after its predicate (the Dutch infinitive), not before it as it would in Dutch. This configuration supports the MOP, which states that only one of the participating languages supplies morpheme order in such constituents. Note as well that all instances of subject–verb agreement come from Turkish. This supports the SMP, which states that only one of the participating languages supplies a certain type of system morpheme (SM), now called an outsider late SM under the 4-M model. Subject–verb agreement is such a morpheme. Based on the example’s support of the MOP and the SMP, Turkish is identified as the ML.

Example (2) supports both principles as well. Note the order of certificate and its modifiers; they follow Swahili word order, not that of English. In addition, although the main verb is from English (depend), subject-agreement, an outsider morpheme, (i-, class 9) comes from Swahili, agreeing with a subject mentioned before (saa hiyo is an introductory phrase, not the subject). These data support Swahili as the ML in this example.

19.2.2 Three premises summarizing the MLF model

Three basic premises have always structured the MLF model (Myers-Scotton Reference Myers-Scotton1993a, Reference Myers-Scotton1997):

1. (1.) Participating languages do not play equal roles in the bilingual clause.

2. (2.) In bilingual constituents within this clause, not all morpheme types can come equally from the ML and EL.

3. (3.) The SMP limits the occurrence of system morphemes that build clausal structure of the ML.

19.2.3 Relating the MLF model to other data

The implicit domain of the MLF model always has been participating varieties that are not mutually intelligible. It may well apply to other varieties, but that would be an unintended bonus. As already noted, the model applies only to what is defined above as Classic CS. This type of CS contrasts with Composite CS in which the abstract grammatical structure underlying surface configurations still comes largely from one language, but also partially from another. The Abstract Level model (see Myers-Scotton and Jake Reference Myers-Scotton, Jake and Nicol2001; Myers-Scotton Reference Myers-Scotton2002a) is especially relevant to Composite CS. More research may show Composite CS to be more common than Classic CS. The USP applies to both types of CS, of course, to the extent there is an ML, and so does the 4-M model; both are universal.

19.4.1 The Differential Access Hypothesis

In CS, and in line with what the USP would predict, the distribution of morpheme types across the ML and EL is quite different. Not only are there distribution differences between content and system morphemes, but also within the category of SMs itself. Recognizing this motivates new ways of classifying morphemes and leads to the 4-M model. In turn, how the 4-M model classifies morpheme types leads to a hypothesis that abstract differences at the production level account for surface level differences in morpheme types. The Differential Access Hypothesis (DAH) offers an explanation for the observed differences. The DAH is the following:

The hypothesis suggests the following scenario. As already noted, lemmas underlying content and early SMs send language-specific directions to the formulator to build larger linguistic units. To build these units, these directions contain information about assigning late SMs. These late SMs become salient only when they are structurally-assigned at the formulator. Separating the activation of abstract elements underlying surface morpheme types echoes Garrett’s view that “major and minor grammatical category words behave quite differently” (Reference Garrett, Blanken, Dittmann, Grimm, Marshall and Wallesch1993:81). However, he and others, such as Ullman (Reference Ullman2001), who posits that the grammar and lexicon are two separate systems, do not differentiate the distribution of different types of SMs.

This theory differs from contemporary linguistic theories that project “functional” elements as the heads of maximal projections. The following sections exemplify EL morphemes in CS in terms of SM types, showing how their distribution follows the USP and implies the DAH.

19.4.2 Early SMs

Early SMs are so-designated because they, along with their content morpheme heads, become salient in the mental lexicon as the basic building blocks of constituent structure, such as NP, VP, AP. Yet, they are still SMs because only content morphemes receive and assign thematic roles. Early SMs typically occur with the content morpheme heads that select them. Early SMs may be free or bound. For example, definite articles are early SMs but always occur with nouns in English.

Plural and derivational affixes are examples of early SMs. Unfortunately, to date, few studies include quantitative evidence on the distribution of either type of early SM. However, in one quantitative study considering determiners in bilingual NPs in a Spanish–English corpus, 151/161 (94%) of English nouns in well-formed mixed NPs occur with Spanish determiners, such as el garage (Jake et al. Reference Jake, Myers-Scotton and Gross2002). Because Spanish can be identified as the ML, the overwhelming number of these mixed NPs supports the USP because ML structure is maintained in these NPs even though the noun is from English. However, as early SMs, definite articles can come from the EL without violating the SMP, and occasionally do, as in Palestinian Arabic–English CS el pharmacy is very boring [. . .] (Okasha Reference Okasha1999:110).

Verb satellites (also called particles) that occur with what are often called phrasal verbs are also early SMs because they depend on their heads for their appearance and they add meaning to their heads. Under the MLF model, these and other derivational morphemes may come from the EL because they are not the type of morpheme that the SMP restricts. EL phrasal verbs often appear with their EL verb satellites. An example from Swahili–English CS shows this, u-na-chase after (“you are chasing after”), as does another example from Arabic–English: [an engine is] locked up. In an Ewe–English example, an Ewe object suffix -e “him” can attach to the verb, as required in Ewe, but the particle remains in the EL, English: keep-e away from Eun (“keep him away from Eun”) (Amuzu Reference Amuzu1998:53). That Ewe supplies the third person singular object suffix shows how the USP is supported as grammatical structure from the ML is maintained.

19.4.3 Late system morphemes

In contrast to early SMs, two types of late SMs are structurally-assigned. The term “late” suggests that they are not activated until a later production level. While early SMs largely build semantic structure, late SMs build syntactic structure. These late SMs are labeled “bridges” and “outsiders.” The DAH, discussed in §19.4.1, explains observable differences in data distribution by postulating a fundamental difference in how late SMs are accessed. It states that not all morphemes become salient at the same level of language production. Information about content morphemes and early SMs is available at the level of the mental lexicon; late SMs do not become salient until the level of the formulator. The role of late SMs is to construct larger constituents out of conceptually-activated morphemes; they assemble phrases and connect phrases to realize full clauses. Put simply, late SMs satisfy the requirements of the USP that constituents maintain a consistent structure. “The late system morphemes [. . .] indicate relationships within the clauses; they are the cement that holds the clause together” (Myers-Scotton Reference Myers-Scotton2006a:269).

19.5.1 Prepositions

Linguistic theory has long recognized that prepositions do not behave as a uniform class (see e.g. Abney Reference Abney1987). Under the 4-M model, prepositions can be content morphemes or any of the three types of SMs. Sometimes the same phonological form fits into more than one category. For example, in He walked across the street, across assigns a thematic role and is a content morpheme. In CS, content morpheme prepositions can come from the EL, as in (10). There are not many examples of such EL prepositions in mixed constituents; more frequently, they occur in PPs that are EL islands (e.g. before tomorrow evening).

Some prepositions can also be indirectly elected at the conceptual level, and are then early system morphemes: in he comes across as ill-prepared, across occurs with come, its content morpheme head. The discussion in §19.4.2 above includes examples in which prepositions are satellites of phrasal verbs, and suggests that the satellite comes from the same language as the verb, either EL or ML verb. However, sometimes such early SMs occur in the EL even when the verb is in the ML. In (11), the expression “change [something] around” is realized in both Spanish and English, with English supplying the preposition around.

Sometimes prepositions are late SMs that are not activated until the level of the formulator; these primarily contribute structure, and not content. For example, prepositions that are bridge SMs make a phrasal constituent well-formed. Although EL bridge prepositions can occur in mixed constituents, very few actually do. As noted above in some French–Arabic CS, Arabic djal (equivalent to “of”) occurs in associative constituents in French-framed CPs. Below are discussed some instances of EL bridges that occur with more frequency, namely, Comp bridges.

Some locative prepositions are bridges; they do not encode directionality or motion, but locate a figure with respect to a ground (see Talmy Reference Talmy2000). For example, in Joe’s in school, in adds little conceptual information to the mapping of the theme (Joe) to the ground (school). Such bridge prepositions show variation (e.g. Joe’s at school). However, in can also be a content morpheme or an early SM. In He’s all done in, in is an early SM. Further, in is a content morpheme in In this example, they illustrate the distinction; like other content morphemes, the thematic role assigned by in can be questioned, as in Where do they illustrate the distinction?

In some languages, prepositions are also outsiders. Consider Spanish a. It can be a content morpheme assigning directionality, as in va a Hamburg “he/she goes to Hamburg,” or an early SM, as in miremos al año que viene, (“we are looking forward to the coming year”) (al = a + el “to+the.m.def”). As a bridge, it connects purpose infinitives with matrix CPs, as in prepare a venir (“prepare to come”). Finally, a occurs as an outsider when it assigns case to animate direct objects, as in veo a Eva “I see Eva.” The prediction is that as ML, Spanish will supply personal a to objects in mixed constituents, as in refieres a tus coworkers (“you are talking about your coworkers”) (Jake et al. Reference Jake, Myers-Scotton and Gross2002), but that as EL, Spanish NP animate objects do not have to occur with personal a, as in the police officers have seen un ladrón (“The police officers have seen a thief”) (Belazi et al. Reference Belazi, Rubin and Toribio1994:230).

The 4-M model articulates how morphemes are classified. Even so, the fact that one prepositional form can be activated at more than one level and is thus subject to different conditions in CS demands careful analysis. The SMP requires that all outsider prepositions come from the ML in mixed constituents. The distribution of bridges in most CS also supports the USP; one language, the ML, provides most of the grammatical frame.

19.5.2 Pronouns

Pronouns are another lexical category that is not uniform because they can be members of any of the four morpheme types (see Jake Reference Jake1994). Some are content morphemes; i.e. they occur in argument position and receive thematic roles. As content morphemes, EL pronouns can occur in clauses framed by the ML. For example Klintborg (Reference Klintborg1995) reports English pronouns in Swedish-framed clauses in Swedish–American English CS, as in När vi var hemma sista gånge me and min hustru (“When we were home last time me and my wife”) and [. . .] he var smed för tyket (“[. . .] he was a blacksmith by trade”).

But even when pronouns are content morphemes, EL pronouns occur very infrequently except in EL islands. Why? First, preference is for ML elements. Also, ML counterparts play a role. When pronouns in the ML are clitics or affixes licensing null pronouns in argument position, they are outsider SMs and must come from the ML (see the Blocking Hypothesis (Myers-Scotton Reference Myers-Scotton1993a, Reference Myers-Scotton1997:120) which requires congruence with the ML).

However, EL content morpheme pronouns that establish topics or contrast occur widely in bilingual clauses. They convey both conceptual and procedural information, as noted by Wilson and Sperber (Reference Wilson and Dan1993:21). Example (12) shows an English pronoun within a Malay grammatical frame, expressing a “dual notion of fusion and contrast” (Jacobson Reference Jacobson and Jacobson2000:68).

Other examples from diverse language pairs abound. Haust (Reference Haust1995) includes examples of Mandinka emphatic pronouns occurring in English-framed CPs and Wolof-framed CPs. English contrastive topic pronouns occur in Spanish-framed CPs, as in You estás diciéndole [sic] la pregunta in the wrong person (“You are asking the question in the wrong person”) (Sankoff and Poplack Reference Sankoff and Shana1981:13).

Some researchers have commented on “pronoun doubling,” as in (13), but this is not the true doubling that occurs with early SMs. Each pronoun is activated independently and occurs in a separate position in the bilingual CP. In (13), for example, the Arabic discourse emphatic pronoun nta (“you”) is adjoined under Comp and the French tu (“you”) is an agreement clitic not in argument position. A null pronoun is assumed to occur in subject position. In (14), Arabic ?iħna (“we”) is adjoined under Comp and English we occurs in subject position.

And in the Spanish–English example cited above, the English emphatic pronoun you is a topic, and a Spanish null pronoun occurs as the subject.

In sum, the distribution of pronouns in CS reflects their classification under the 4-M model. The requirements of the MLF model foreshadow the import of this classification. Both the SMP, which requires ML outsider pronouns in mixed constituents, and the Blocking Hypothesis, which requires cross-linguistic congruence, imply how critical it is to recognize morpheme type at the abstract level of clause construction. Taken together, they maintain the integrity of the frame in line with the USP.

19.5.3 Complementizers and other clause connectors

Complementizers and complementizer-like elements are similar to prepositions and pronouns in not showing a uniform distribution in CS. In current syntactic theory, COMP is the head of any clause identified as CP, projection of Complementizer. Variation among COMP elements themselves and cross-linguistic variation in their patterning in CS complicate their discussion. Also, there is no uniform agreement regarding what elements are rightly classified under COMP. Complementizers include not just elements such as that, but also subordinating conjunctions, relative clause markers, other elements that indicate clause boundaries, and even coordinating conjunctions. These elements are discussed according to how they are elected and how they participate in the construction of a CP.

Like pronouns, most complementizers convey procedural knowledge. Many constrain the truth-conditions of propositions and participate in the discourse-thematic structure of propositions. For example, porque “because” in (15) assigns a discourse-level thematic role of Cause or Reason.

Most complementizers straddle two CPs. In this way, they are at the intersection between inter-sentential CS and intra-sentential CS. In example (15), porque is the head of the subordinate CP (hereafter CP2), yet it is within the domain of a matrix clause (hereafter CP1), and is in the language of CP1.

19.6.1 Predictable patterns

The bulk of this chapter is descriptive, but with the theoretical goal of demonstrating how the asymmetries that one finds in CS show a predictable pattern. The goal here has been to demonstrate that the contributions of morphemes of participating languages depend on the four morpheme types and to relate this observation to production. When one views morphemes in terms of these types and in terms of the Differential Access Hypothesis, a principled explanation for differences in their cross-linguistic distributions in CS is forthcoming. Three examples illustrate insights of this chapter.

First, it is predicted that double plural marking on EL nouns is possible, but that double subject–verb agreement or double case marking is not. This disparity is owed to the difference in how these morphemes are accessed in language production. Plural affixes are early SMs while subject–verb agreement and case markers are outsider SMs. EL early SMs can be accessed with their content heads because they are conceptually activated and available in the mental lexicon, but outsiders become salient only later, at the level of the formulator.

Second, a strong preference for the ML to supply “that-type” complementizers at clause boundaries is predicted. However, subordinating complementizers are less constrained and they come from either language. The reason again is differences in morpheme type and hypotheses about their production history. That-type complementizers are bridge SMs and, although not as critical in building clauses as outsiders, are still part of constituent structure and are salient at the level of the formulator. When they come from the ML, the USP is satisfied. In contrast, many subordinators are content morphemes; they are activated by speakers’ intentions (conceptually activated), and are available in the mental lexicon.

Third, prepositions fall into all four types under the 4-M model, but not all types are predicted to come from the EL, or with the same frequency. In fact, EL prepositions do not occur frequently, and this may be because of their role in structuring constituents and the requirement of uniform structure, i.e. the USP. Among prepositions, early SMs are more frequent, perhaps because they allow for an elaboration of pragmatic and semantic structure without creating syntactic structure. Prepositions that are content morphemes (i.e. that assign thematic roles) are the most unconstrained; even so, they are not frequent.

Of the three types of SMs, early SMs are the least constrained because they are conceptually activated, whereas bridges and outsiders are structurally assigned. Even so, the most frequent early SMs seem to be the most contentful ones, plural affixes and definite articles. Only a few types of EL bridge SMs occur, and no EL outsider SMs occur in mixed constituents (except for fairly rare types of EL islands). More research needs to be done on EL islands, but the overall point about prepositions holds for all lexical categories: morpheme type, as discussed at many points above, makes the difference in their distribution.

19.6.2 Cognitive support systems

Such differences in morpheme distribution across languages are systematic and follow from the MLF and 4-M models, as well as the USP; they are also empirically verifiable. However, these differences also imply some speculations about language production and the cognitive systems supporting language. First, the division of labor between languages in CS, with only one language providing the morpho-syntactic frame of the bilingual clause, seems to imply some sort of divisions within the cognitive component supporting this surface asymmetry. Certainly, any intra-clause CS implies that both languages are active during bilingual production, but that the ML has a higher level of activation.

Second, the USP implies that cognitive energy is conserved by allowing only minor-level switches to EL islands. If we look at the frequency with which EL content morphemes are integrated into an ML-framed structure and juxtapose this frequency with the USP’s injunction against changing languages, one conclusion is that accessing words from the EL requires a different type or level of activation than creating morpho-syntactic structure.

Third, the role of the EL is explicitly limited. Most obviously, the EL never structures any constituents that include ML morphemes. This means that the EL has little opportunity to supply any outsider SMs to the bilingual clause except in EL islands, but typical EL islands have few structures that would require outsiders. Further, except for occasional early SMs or even less frequent bridge SMs, the EL supplies only content morphemes within constituents structured by the ML. The dearth of outsider SMs from the EL motivates the conclusion that the cognitive component supporting this morpheme type may be independent from that which coordinates simpler syntactic constructions. Keep in mind the complex tasks that this mental architecture must accomplish: outsiders are critical in signaling thematic roles and other relationships of the semantic–syntactic interface; without them, there can be no clause.

19.6.3 Testing hypotheses

Finally, as already noted at many points above, CS data support the DAH that is derived from the 4-M model. This hypothesis suggests a language production model in which some of the elements underlying surface level morphemes are salient at one level and others are not salient until another level. Specifically, late SMs are not salient until they are called by the lemmas underlying content morphemes to construct larger constituents in the formulator. Obviously, the extent to which this hypothesis is supported has relevance beyond CS and other types of contact phenomena to both child and second language acquisition, among other topics. The results of psycholinguistic experiments testing this hypothesis will add crucial support to a language processing model that accommodates the notion of saliencies at different levels (Myers-Scotton Reference Myers-Scotton2006b). Not only does this matter for production models, but also for comprehension models.

19.6.4 Supporting the Uniform Structure Principle (USP)

In sum, this chapter has shown how the distribution of morpheme types in Classic CS, at least, is compatible with predictions of the MLF and 4-M models, and the USP. As an empirical window on divisions of labor between participating languages, CS implies intriguing hypotheses about some ways in which language is supported in our cognitive systems.