The scope of Myers-Scotton's new book is far wider than that of her earlier study (1993), which focused exclusively on code-switching. She has now applied her Matrix Language Frame (MLF) model to a much broader range of language contact phenomena, covering areas such as convergence (what others have termed “interference”), language attrition and shift, and creoles and mixed languages. There is no doubt that the model does offer many insights into the dynamics of various types of language contact, and while Myers-Scotton's remarks remain fairly tentative with regard to some of these fields, this book certainly opens up new avenues for exploration in each of them. However, the most substantial and possibly the most controversial part of the book is again that which presents the latest developments in her account of bilingual code-switching, which has already achieved worldwide renown, so for want of space we will focus only on this part of the book.
As usual, Myers-Scotton's views are clearly and meticulously argued and are supported by a wide range of examples drawn from her own data and those of other researchers. The basic tenets of the MLF model still hold. It maintains that when two languages come into contact in the speech of bilinguals, they are not equal in status; rather, one language, the Matrix Language (ML), is the main source for the morphosyntactic frame into which elements from the other, the Embedded language (EL), can be inserted, either as single elements or as longer constituents, the latter being termed “EL islands” and being well-formed according to EL requirements. Predictions as to which elements may come from which language are formulated in terms of the distinction between content and system morphemes. Myers-Scotton has now added two further supporting models: the 4-M model, which introduces, alongside the content/system morpheme contrast, three subcategories of system morpheme; and the Abstract Level model, which identifies three levels of abstract structure in lexical items – lexical-conceptual, predicate-argument, and morphological realisation patterns.
In fact, Myers-Scotton seems to be anxious to point out that the 2002 version of the MLF model is not very different from the 1993 version. She rejects others' claims that it has “changed radically,” instead referring to “provisions of the MLF model that either were not well articulated by me, or not well understood by others” (53). At several other points, too, she deplores what she considers to be other researchers' apparent misunderstanding of the model (64, 87). However, a careful comparison of the 1993 and 2002 versions leads us to conclude that the model has indeed undergone substantial changes, and we do not understand Myers-Scotton's reluctance to admit that she has, quite simply, changed her mind on certain issues.
On the one hand, we find a fairly dramatic change in the procedure for identifying the Matrix Language. In 1993 her definition was quite clearly couched at the discourse level: “The ML is the language with the higher frequency of morphemes in a discourse sample in which CS occurs” (1993:232). However, now the ML is defined only at the level of the CP, and the model has nothing to say about any units larger than this; indeed, Myers-Scotton dismisses as of no interest for a grammatical analysis of code-switching cases where a switch of language occurs between two conjoined CPs or between a main clause and an embedded clause. This shift of position has the advantage of removing at one fell swoop many of the problems posed by the earlier definition, relating to exactly what constituted a discourse sample and when the ML could be considered to have changed (see Bentahila 1995:136–37). However, reducing the ML to a property of CPs seems to weaken the ties between grammatical and psycholinguistic and sociolinguistic factors that were for many an attractive feature of the earlier version. In fact, Myers-Scotton's remarks reveal that she still envisages the ML as a constant property of whole conversations and indeed of entire varieties (62–64, 112), but the model as it now stands seems to have nothing to say about the difference between a case where the matrix language never varies within the speech of a whole community and one where it is constantly changing between one CP and the next.
Moreover, whereas in 1993 Myers-Scotton insisted that the criterion of morpheme counting was essential for empirical verification and to ensure a noncircular, independent definition (1993:66), the new definition has abandoned this: The ML is still “the language making the larger contribution,” but now “contributing more does not mean more morphemes” but rather “more abstract structure and structure of a certain type” (15). Since there is nothing in the model to stop the ML from changing between each pair of CPs in a discourse, the predictions made by the new version would seem to be considerably less constrained than those made by the 1993 version. Moreover, it is later pointed out that the ML is after all “not to be equated with an existing language,” but is rather “an abstract frame for the morphosyntax of the bilingual CP” (66). The structural requirements imposed by the ML are in fact not exactly the same as those imposed by the language it is based on, which allows for the possibility of inserting into the ML frame certain EL elements and constituents that are not “optimally morphologically integrated into the Matrix Language” (67).
Other adaptations of the model have the effect of equipping it to deal with what might earlier have been felt to constitute counterexamples to its predictions. As in earlier versions, the ML frame is claimed to impose ML word order and ML system morphemes in mixed-language constituents. However, using the distinctions drawn by the 4-M model, Myers-Scotton now maintains that in fact only one of the three categories of system morpheme, “outsider” late system morphemes, such as verbal inflections and case markers, must be from the ML in mixed constituents. Apparently, then, the model does not block EL morphemes of other types – such as determiners, demonstratives, and possessive markers – from mixed constituents, whereas in 1993 the discussion implied that these were also blocked by the system morpheme constraint (“Within ML + EL constituents, all active system morphemes are from only one of the languages participating in the CS, the ML” (1993:99, 102–9). Myers-Scotton does in fact admit that her earlier discussion was misleading (87, 91); for instance, she is at pains to point out that, contrary to what was implied in the earlier study, she now views the occurrence of Arabic djal within a French matrix as posing no problem for her model, since this is to be regarded as a “bridge” late system morpheme.
However, while now admitting that mixed constituents may contain such EL bridge morphemes, as well as EL early system morphemes (such as determiners), she repeatedly describes such occurrences as unusual (87, 91, 114). To account for the scarcity of such patterns in her data, she is then obliged to introduce the Uniform Structure Principle, which claims that all constituents must satisfy the well-formedness requirements for their language; hence ML system morphemes will be preferred in a bilingual CP because they ensure this uniformity of structure. The more abstract definition of the ML now adopted then allows her to postulate a concept of “sufficient congruence” according to which EL system morphemes may occur in mixed constituents if the structures of the two languages are similar though not identical. For instance, she suggests that the occurrence of a French Det-N string within a larger NP containing an Arabic determiner, a common feature of Moroccan Arabic-French code-switching, is possible because the French determiner complex is “nearly identical” to that of Arabic (in contrast to that of Dutch, which apparently does not exhibit sufficient congruence with the Arabic complex, so that EL Dutch nouns are not accompanied by Dutch determiners). Whether or not one agrees with this particular analysis of French NPs, one cannot help feeling slightly uneasy about the vagueness of the concept of sufficient congruence, which could lead to circularity: It is all too easy to say that if a particular ML + EL sequence is possible, there must therefore be sufficient congruence between their abstract structures.
Another loophole seems to arise in the discussion of EL islands, whose morphemes are predicted by the model to follow EL order. However, an apparent counterexample (building high-rise) is dismissed as follows: “The problem here is that these morphemes follow the word order of the Matrix Language, French, not English; therefore this is simply a sequence of two Embedded Language content morphemes in a mixed constituent under Matrix Language order” (139). The use of therefore here again seems to imply circularity, opening up the possibility that any sequence not meeting the EL order requirements might be reanalyzed as a sequence of separate EL elements rather than an island.
The changes made to the model have thus removed some problems and expanded the range of data it can account for. However, this has not been achieved without some cost. The original model dealt very neatly with the patterns of code-switching found in many communities, but the patterns found in some other data sets were less well accounted for. In seeking to expand the coverage of her model to all known code-switching varieties, including those featuring the “unusual” configurations with EL system morphemes, Myers-Scotton has perhaps reduced its fit with the original data. Moreover, while she has made a valiant attempt to account for the patterns of North African Arabic–French code-switching within the MLF model, it still remains lacking in both descriptive and explanatory adequacy. Examples such as the switches for the morpheme djal and between determiners, while now allowed, are still presented as exceptional and puzzling types of switch, whereas in Arabic–French code-switching discourse they are absolutely normal, typical and highly recurrent patterns. Moreover, while examples like (1) and (2) are now apparently allowed for by the MLF model, it is important to note that patterns like (3) and (4) also occur.
While in (1) the MLF model will identify la feuille as an internal EL island within the larger ML-framed NP containing the Arabic determiner had, a quite different analysis will be needed for the similar sequence in (3), where the ML is French and had is a lone EL early system morpheme. Moreover, while the Arabic inflections on a French verb root in (2) are in accordance with the requirement that outsider morphemes must come from the ML (here Arabic), the same pattern in (4) clearly violates this rule, since here we have outsider morphemes from the EL attached to an ML root! The fundamental problem for Myers-Scotton's model seems to be that the ML, as she defines it, is simply not relevant to the acceptability of certain types of mixed constituents in Arabic–French code-switching. These Arabic–French determiner sequences and French verbs with Arabic inflections occur not only in CPs with an Arabic-based ML, but also in ones where the ML is based on French, and what is needed is a model which can account for all these occurrences. The MLF model simply misses the point here.
Myers-Scotton's model undoubtedly provides an effective and insightful account of what happens in many types of language contact data, including many code-switching varieties. However, the quest for even greater comprehensiveness that seems to underlie the modifications of the model adopted in this book does not seem to be entirely fruitful. The model still does not allow for all attested patterns; it fails to capture some important generalizations and treats some regularly occurring patterns as puzzling exceptions. While we admire Myers-Scotton's single-mindedness in seeking to prove that her model can cope with all types of code-switching, we feel that ultimately this approach may prove counterproductive.
