WHY L3/LN ACQUISITION MUST NOT BE REGARDED AS ANOTHER INSTANCE OF L2

Despite compelling reasons to do so, many researchers have not distinguished, at least consistently, between different instances of adult nonnative learners, including L3/Ln learners in populations that are reported as L2 adult learners. Such a practice essentially operationalizes the term L2 acquisition to cover any instance of nonnative adult language acquisition. However, just like the terms native speaker and monolingual are unwittingly confused and misapplied in the literature (i.e., they are not synonymous) with serious methodological and epistemological consequences, so too has the over-encompassing term L2 learner. Failing to properly differentiate true L2 from L3/Ln acquisition can have an inadvertently damaging impact for important questions studied within a more broadly defined SLA (De Angelis, Reference De Angelis2007). Putting aside the important discussion of which variables should be factored in when determining L3/Ln status, (for a discussion, see Hammarberg, Reference Hammarberg2010) it should be relatively uncontroversial to claim for chronological reasons alone that an individual who has already acquired an L2 would be different in crucial, deterministic ways during L3 acquisition as compared to a novice L2 adult learner. For example, if there is any credence to the so-called additive effect of bilingualism on L3 learning (Cenoz, Reference Cenoz, Cenoz, Hufeisen and Jessner2003), the superior metalinguistic skills and knowledge of bilinguals are a variable that must be considered as a possible source of some knowledge in L3 research. From a generative linguistic perspective that assumes adult Universal Grammar (UG) continuity, it follows that L3 learners have access to more grammatical options (the underlying feature representations of two languages) when it comes to initial hypotheses of L3 interlanguage development.

So, why have so many researchers not differentiated L2 from L3/Ln learners in many studies of SLA? Consider one of the two main historical questions within the generative SLA paradigm, that is, the recurrent question regarding UG accessibility in adulthood (see White, Reference White2003). It is not necessarily the case that combining data of L2 and L3 learners invalidates the claims made from them.Footnote ³ However, it is possible that lumping true L2 and L3/Ln acquisition into a single group might have inadvertent complicating effects for this and other relevant questions. Let's examine a tangible example to illustrate the possible effects. To do so, we will consider possible developmental implications stemming from what Rothman and Cabrelli Amaro's (Reference Rothman and Cabrelli Amaro2010) L3 study revealed.

Rothman and Cabrelli Amaro (Reference Rothman2010) tested four groups of adult English-speaking nonnative learners of French or Italian at a point in which they all had had the same amount of exposure to the L2/L3 target languages (for specific details, see Rothman & Cabrelli Amaro). Groups 1 and 2 consisted of L2 learners of French and Italian, respectively. Groups 3 and 4 were true L3 learners of the same target languages; however, these learners started the L3 process as highly advanced learners of L2 Spanish. The properties under investigation were related to the null-subject parameter (NSP), for example, referential subjects, expletive subjects, and semantic constraints on anaphora resolution (e.g., restrictions on co-reference between main clause and embedded clause subjects that are delimited by so-called universal constraints such as the ones described by the overt pronoun constraint (OPC); Montalbetti, Reference Montalbetti1984). These languages made for an interesting scenario since despite the clear structural or typological proximity of the Romance languages, French is actually different from Spanish and Italian as it relates to these properties. Whereas Spanish and Italian are null-subject grammars—they have null and overt referential subjects; expletive subjects are obligatorily null and the restrictions on co-reference captured under the OPC obligatorily apply—English and French are nonnull subject languages.

Rothman and Cabrelli Amaro (Reference Rothman2010) hypothesized that typological proximity between the languages would determine initial stages transfer whether or not such transfer would be facilitative. In the case of the L2 groups, transfer, if relevant at all, could only come from English. For the L3 groups, the authors predicted that both groups would transfer the Spanish positive setting of the NSP given the overall structural and typological proximity of Spanish to Italian and French, which would only be facilitative for L3 Italian. The rationale of having all four groups was the following: The true L2 groups would serve as controls for what transfer effects would look like if English were the only language available for transfer. The prediction, then, was not only that both L3 groups would behave similarly reflecting Spanish transfer, but that both would differ significantly from their counterpart L2 groups (who did not have access to a typologically related null-subject grammar). Both L3 learner groups did in fact demonstrate unambiguous Spanish transfer of the NSP, differing from their L2 counterparts significantly and not differing from each other in the L3. Rothman and Cabrelli Amaro's goal was to determine the source of transfer at the initial stages of L3 acquisition. Nevertheless, their findings have implications for L3 interlanguage development and ultimate attainment, and display how the presence of an L2 and its transfer into L3 can have consequences for determining accessibility to UG if L3 learners are not differentiated from L2 learners.

Following from the predictions of the full transfer/full access (FT/FA) model of the initial state (Schwartz & Sprouse, Reference Schwartz and Sprouse1996), Judy and Rothman (Reference Judy, Rothman, Franich, Iserman and Keil2010) and Judy (Reference Judy2011) tested Spanish native speakers of highly advanced L2 English proficiency with the same NSP properties. Their prediction was that Spanish natives would not be able to reset the NSP, despite what has been robustly shown in the opposite direction (for review of the many studies that show this, see Rothman, Reference Rothman2009), not because UG is inaccessible, but rather because the Spanish setting of the NSP constitutes a proper superset to the English subset value. FT/FA maintains that L2 parameter resetting is in principle always possible, but not inevitable. One of these inevitabilities follows from the subset principle (Wexler & Manzini, Reference Wexler, Manzini, Roeper and Williams1987). Accordingly, if transfer of a superset applies, then a learnability obstacle arises in the very same way it would for child learners, since transfer of the superset value would effectively block the parsing failure needed for grammatical restructuring towards the target grammar. In other words, having transferred a superset from the L2 may prevent the L3 learner from being able to reduce that superset into the subset relevant to the L3 grammar. Judy and Rothman (Reference Judy, Rothman, Franich, Iserman and Keil2010) and Judy (Reference Judy2011) showed that the predictions of the subset principle and FT/FA were supported. Despite high proficiency in L2 English, the Spanish natives did not show consistent evidence of NSP resetting for L2 English, ultimately continuing to accept null expletives in English, especially in embedded contexts, and misapplying the OPC in anaphora resolution.

Based on the above studies, let's consider the implications this has for the L3 learners in the Rothman and Cabrelli Amaro (Reference Rothman2010) study. Recall that both L3 groups transferred Spanish into the initial stages of L3 Italian and French. As it relates to the L3 Italian group, this was facilitative, whereas for the L3 French, it was not. If we take seriously the claims of Judy and Rothman (Reference Judy, Rothman, Franich, Iserman and Keil2010) and Judy (Reference Judy2011) and the general claims of the FT/FA model, we now have a case in which the same learnability problem for native Spanish speakers of L2 English arises for the case of L1 English/L2 Spanish/L3 French. Having transferred the Spanish NSP setting into the initial stages of L3 French, this group's ability to reset the NSP should be subject to the tenets of the subset principle. We might expect, then, that if we were to track this L3 French group through development, they, like what was shown of native Spanish learners of L2 English, would not be able to reset the NSP as a consequence of the subset–superset relationship between the grammar they initially transferred (Spanish) based on typological proximity and the target L3 (French). This is predicted to be so despite the fact that their L1, like the L3, is a nonnull subject grammar.

Of course, the above scenario embodies an empirical question that can and should be investigated directly. However, if we are on the right track here, we have uncovered a situation in which collapsing L2 and L3 learners into a single group (or having a group of claimed L2 learners that in fact have at least some L3 learners) could obscure our understanding of UG accessibility. Let's imagine that we did have a mixed group of Anglophone L2 French learners where some or all are actually L3 learners with Spanish as an L2. At the level of the aggregate, we would likely find artificial levels of variation from which some might argue that UG is inaccessible. But if this variation actually stems from influence of the L2, Spanish, and this influence is predicted to not subside because of the subset–superset relationship of the grammars, then this would simply mean that the so-called L2 group was not a well-chosen one for the question of UG accessibility. At this point we might begin to ponder the extent to which some, perhaps much, of the ubiquitous individual differences in SLA morphosyntax studies could be better explained by taking into account previous experience with other languages for some of the learners functionally overlooked or not reported at all within so-called L2 groups.

L3/LN ACQUISITION FROM A FORMAL LINGUISTIC PERSPECTIVE: AN OVERVIEW

Having justified the need to distinguish between L3/Ln acquisition research and strict SLA, and moving towards the narrow focus of this article, in this section we will review the genesis of generative L3/Ln morphosyntax studies through the present day. Again, as mentioned and justified above, our main focus will be on studies of morphosyntactic transfer at the initial stages of L3 acquisition.

FORMAL MODELS FOR L3 TRANSFER (INITIAL STAGES)Footnote ⁴

More recent research on L3/Ln acquisition has produced several formal models that attempt to explain the source and selection of transfer at the L3 initial stages. Here, we outline these formal models for transfer along with some exemplary empirical studies providing evidence for each of the models.Footnote ⁵

FINAL THOUGHTS: ON THE DYNAMICS OF MULTILINGUAL TRANSFER AND WHAT IT TELLS US

In summarizing generative research with a particular emphasis on transfer at the L3/Ln initial stages, we have made reference to some latent potential that such study has beyond contributing to the understanding of the dynamics of multilingual acquisition alone. Abstracting away from L3/Ln acquisition per se, it seems reasonable to argue in the general study of linguistic transfer that only in the case of multilingualism can one begin to test and understand its real dynamic nature. In the case of L2 acquisition, where transfer has been seriously investigated for decades, complex questions of what transfer patterns might reveal cannot be as meaningfully contextualized to the extent they can be in the case of multilingualism. This is so because in L2 acquisition, transfer either materializes or it does not. When it does, there is no mystery about what the source of transfer is; it must be the L1.Footnote ⁸ However, in the case of multilingualism there exists the possibility of several sources of linguistic transfer. Modeling and predicting the patterns of this transfer relates to larger questions. These questions pertain to what L3 transfer can tell us about mental architecture, linguistic representation, and design as well as the interface of principles of cognitive economy and language in ways that cannot be explored by looking at transfer in various instances of bilingualism (cf. Rothman, Reference Rothman, Cabrelli Amaro, de Bot, Herschensohn and Young-Scholten2013).

The formal linguistic models of L3 transfer reviewed in this article exemplify this very point. The L2 status factor, the TPM, and the CEM, despite important differences, all embrace the idea that transfer in multilingual acquisition scenarios is dynamic but ultimately predictable, and, by extension, logical, based on cognitive factors. At a substrate level, it seems that all three models take seriously the notions first offered by Flynn et al. (Reference Flynn, Foley and Vinnitskaya2004): (a) that language acquisition is inherently nonredundant, (b) that nonredundancy has a cognitive economic drive, and (c) that L3/Ln transfer patterns are the nonredundant surface reflex of internal mechanisms or processes of some kind. As we see it, the models can be taken to stem from these basic tenets, yet their respective interpretations on how these underlying driving factors play out differ, resulting in their partially overlapping, yet crucially distinctive predictions (see above for details). Which of these models are ultimately correct is of no consequence to our point here, though. The argument we wish to make is that each of them proposes, albeit in different manifestations and with distinct stipulations, that cognitive economy drives selection when multiple options are present. Simply put, language acquisition is taken by all three models as nonredundant in the sense that each model maintains that L3/Ln does not start over again from a tabula rasa devoid of previous acquisition experience, although logically speaking it could, and that selection and deployment of previous linguistic knowledge is constrained. If the general tenets are on the right track, this shows how L3/Ln transfer research embodies unique and independent evidence for the connection general cognition has with acquisition processes.

On the purely linguistic side, the study of L3 acquisition allows us to investigate theoretical notions important to linguists, cognitive scientists, and developmental psychologists interested in language. L3 studies provide data that come to bear on questions of the structure and composition of linguistic representation in the mind as well as the formal complexity of specific linguistic structures. Take, for example, the CEM and TPM, both based somewhat differently on the proposal that the mind is especially attuned to linguistic function and form, forcing learners to naturally take advantage of the full array of linguistic knowledge (functional features in generative terms) they have in order to facilitate the cumulative process of language acquisition in adulthood (even if this leads to nonfacilitative results; see the TPM). As such, these models make sense in formal linguistic terms of what has been anecdotally observed previously as the additive effect of bilingualism (e.g., Cenoz, Reference Cenoz, Cenoz, Hufeisen and Jessner2003). The very fact that all three of the existing models are based on the notion that the L1 and L2 are separate systems in the mind or brain informs debates on bilingual language representation. After all, if both L1 and L2 grammars can be deterministic in L3/Ln acquisition, this must mean that each of these grammars has its own reality of mental representation in the mind that is, at least partially, unique from the other. If it can also be shown that some L2 properties are transferred in L3/Ln that embody properties predicted by some theories of L2 acquisition to be unattainable (e.g., evidence of transferred new L2 uninterpretable features and their associated functional categories), this would mean that L3/Ln transfer informs most directly debates related to formal linguistic approaches to the critical period (see Iverson, Reference Iverson2010).

Finally, we wish to point out that the descriptive facts of L3/Ln acquisition need to be captured under linguistic theory to the same extent and for the same reasons the facts of bilingual acquisition must. It seems that the human mind, both in childhood and adulthood, is capable of multilingual acquisition. These multilingual grammars have to be accounted for by the human language faculty and fit within a harmonious model of linguistic design and architecture that covers more than the idealized monolingual standard. Alternatively, linguistic theory needs to address how the empirical facts of multilingual grammars are commensurable with hypotheses of how the by-products of acquisition filtered through and delimited by the same biological linguistic endowment (i.e., UG) can result in differences related to factors outside the scope of linguistic domain specificity. In principle, this is not a hard task. Linguistic nativism and its hallmark of linguistic modularity are not at all in conflict here.

In psychological literature on mental modularity, language being one module of the mind, the claim has always been that the encapsulation of a modular subsystem entails that the specific design itself cannot be altered by external systems, but it has always maintained that modular systems themselves can provide output to other systems and use, if not need, input from other systems (e.g., Fodor, Reference Fodor1983; Jackendoff, Reference Jackendoff2002). Thus, linguistic theory must come to terms with the reality of multilingual grammars and what the facts of their development and outcomes bring to bear on larger notions.

We have highlighted just some of the reasons why we believe much can be learned from the linguistic study of multilingualism beyond questions belonging solely to the provenance of multilingual acquisitionists. We predict that future generative work examining the L3 initial stages and its developmental route will help linguists to better understand linguistic architecture and representational constitution by demonstrating how multiple languages are positioned in the mind relative to one another, how and to what degree they are independent, and, finally, where and why they interact.

Much more research is needed in the domain of formal linguistic L3/Ln acquisition of morphosyntax. We predict that in a few years’ time new foci will emerge, especially as models of L3/Ln development and ultimate attainment are articulated as they have been for L3/Ln transfer models in recent years. Generative acquisition research with adults is largely experimentally based and thus requires significant numbers of participants. However, difficulty in finding significant numbers of L3 learners, especially for language pairings beyond Germanic and Romance, is a real obstacle imposed by reality within the Western world. Moving formal linguistic L3/Ln studies forward, given this, also embodies an opportunity for collaborations where we actively engage the multilingual world outside of the traditional loci for such research, that is, the majority of the world in geographical terms, in which multilingual acquisition is not the exception but the rule. Some interesting work is already emerging in this regard, for example, Hermas's (Reference Hermas2010) work in Morocco involving Arabic-French-English L3 acquisition. We look forward to the near future when experimental methodologies are brought to these interesting language combinations, this being exactly what our present models need to be tested and refined and what will spawn the creation of the next generation of multilingual morphosyntactic models and hypotheses.