Constraining Multiple Grammars

Amaral and Roeper (this issue; henceforth A&R) present a theory about interlanguage representations that accommodates optionality as manifested in the parallel use of mutually conflicting rules within a single grammar. Multiple Grammars advances the claim that optionality is a constitutive characteristic of any one grammar, with interlanguage grammars being perhaps the clearest examples of a system that subsumes several different and often conflicting subgrammars made up of properties of the first language (L1), the target language and idiosyncratic rules.

In this respect, the approach accords well with recent advances in psycholinguistic research on the multilingual lexicon that suggest that multilinguals accommodate words from different languages within one lexicon, with access to the multilingual lexicon being non-selective (for review, see Kroll et al., 2012). Similar proposals have been made for syntax in the shared syntax account (e.g. Hartsuiker and Pickering, 2008) where syntactic nodes are interconnected across languages within a single system. In this sense, the Multiple Grammars approach can be understood as extending non-selective lexical approaches to grammatical representations. Unlike these psycholinguistic models, however, the Multiple Grammars approach does not commit to any assumptions about language processing and how grammatical representations are accessed and selected in real-time production or comprehension. In fact, according to A&R, such processing considerations seem to be at best secondary in accounting for optionality in second language (L2) development. ‘[A]s far as the MG theory is concerned, any type of morphosyntactic variation is motivated by the (natural) existence of multiple sub-grammars in the speaker’s mind’ (pp. 36-37).

A&R are certainly right in pointing out that where different grammatical representations are involved, a theory of interlanguage representation must give a coherent account of the nature of different coexisting grammatical representations within a single language user. However, it is far from clear whether all (or even many) cases of morphosyntactic optionality in advanced to near-native L2 acquisition necessarily involve multiple grammatical representations. Instead, what looks like morphosyntactic optionality at the grammatical level may be conditioned by grammar-external (e.g. lexical) representations or processing considerations that constrain access to a fully target-like L2 grammar.

A&R discuss Sorace and Filiaci’s (2006) finding that L1 English near-native speakers of Italian overuse overt subject pronouns to establish local anaphoric relations in contexts where native Italian speakers prefer using a null subject. In their criticism of Sorace’s (2011) Interface Hypothesis they point out that consistent optionality calls for a formal description of the underlying representations involved. However, it is an open question whether a formal description equals a grammatical description of optionality in terms of multiple grammars. The selection of an overt or a null pronominal form is the consequence of a division of labour between syntax, the interpretive system and processing considerations (e.g. Reuland, 2011). Following Carminati (2005), Sorace and Filiaci (2006) argue that interpretive preferences for null and overt pronouns follow from the application of a parsing strategy (the Position of Antecedent Strategy, PAS) that correlates pronominal forms to the syntactic position of antecedents. Given the complexity of the task, different outputs can obtain depending on whether all steps in the computation are effected, e.g. the PAS is consistently and fully applied. If language users do not fully compute the PAS, they are left without a grammatical resolution of the pronoun. In this case, they will likely make random pronoun choices, because the grammar and associated parsing strategies do not provide them with a determinate outcome. Hence, Sorace and Filiaci suggest that optionality in bilinguals follows from the inconsistent application of the PAS, which is potentially magnified in bilinguals due to the higher demands of parsing an L2. In support of this account, both native and non-native speakers of Italian show an asymmetry in performance between forward and backward anaphora that impose different processing loads in computing appropriate form-to-function mappings (Sorace and Filiaci, 2006). In a similar spirit, O’Grady (2011) suggests that the overuse of overt pronouns may be driven by a processing strategy to project an overt form in cases of non-local antecedence to avoid memory overload. On either account, speakers consistently refer to the same (target) grammar, yet they cannot consistently access it. Such performance gives rise to what looks like grammatical optionality, even though only a single grammar is involved.

In morphosyntax, similar effects of optionality can surface as the result of incomplete computations. In a series of experiments, Hopp (2010) finds that L1 English and L1 Dutch near-native L2 speakers of German demonstrate persistent optionality in the use of case marking and subject–verb agreement in comprehension, whereas L1 Russian speakers converge on the target language system. Given that L2 groups behave differently according to their L1s, the pattern of results seems to fit the bill for an explanation in terms of Multiple Grammars: Near-native L2ers seem to access different grammars, i.e. one that includes syntactic features for agreement by case and one that does not. It can be shown, however, that native speakers of German performing the task under increased processing demands approximate the inflectional variability observed in the L2 groups (see also McDonald, 2006). The fact that natives demonstrate L2-like comprehension of morphosyntax under elevated task demands suggests that processing limitations on access to target-like representations rather than access to different (sub-)representations mandate morphosyntactic variability at L2 ultimate attainment.

In similar terms, recent research on syntactic optionality in L2 gender agreement (Grüter et al., 2012; Hopp, 2013) finds that advanced to near-native speakers show residual variability in syntactic gender agreement. Again, it may be argued that these learners variably access a sub-grammar that does not encode syntactic gender features, e.g. the L1 English grammar, and a sub-grammar that does. Upon closer examination, the variability turns out to be constrained by systematic optionality in lexical, rather than grammatical, representations, with learners who have target-like lexical gender assignment and who are fast in accessing lexical gender representations performing target-like on syntactic gender agreement. These findings indicate that syntactic optionality in part reflects aspects of lexical encoding, i.e. how detailed lexical entries are and whether these entries can be fully accessed and computed in bilingual processing.

These three examples illustrate that persistent and systematic morphosyntactic optionality in L2 acquisition is not necessarily evidence of and support for access to multiple grammars; rather, optionality may index processing restrictions in how learners use a single grammar, i.e. the target-language grammar that does not contain (or access) conflicting rules. A&R state that ‘processing accounts are usually inexplicit in their relation to representations, but since representations must be involved, it may well be the representation which holds the explanation’ (p. 32). Then again, it may well not be the representation that holds the explanation. Without an explicit theory of processing that situates Multiple Grammars in a model of L2 acquisition and that defines how and under which circumstances speakers access subgrammars, there is no telling whether optionality reflects access to different subgrammars or follows from other grammar-external constraints on L2 performance.

Needless to say, any grammatical account of optionality faces this very issue, yet it seems magnified within the framework of Multiple Grammars, since the approach allows for an infinite number of subgrammars in principle. In this respect, the strictly representational approach taken by Multiple Grammars runs the risk of overgenerating grammatical explanations to optionality in L2 acquisition. Any type of systematic variability may be classified as reflecting access to multiple grammars, with the number and types of subgrammars within a speaker essentially proliferating to all feature combinations and parametric settings available within the hypothesis space delimited by Universal Grammar.

Of course, it is an empirical question whether any observed token of optionality in monolinguals or bilinguals is due to representational or computational factors, yet it stands to reason that this question can only be meaningfully addressed in a theory if its framework articulates the relation between multiple grammars and processing. In sum, the innovation and the merit in the Multiple Grammars approach lie in advancing a coherent account of the grammatical representations involved in morphosyntactic optionality. At the same time, the challenge for the model is to embed Multiple Grammars in an overall theory of L2 acquisition that constrains the use of multiple grammars.