3.1 The L2 Status Factor

The L2 Status Factor maintains that the L2 takes on a significantly stronger role than the L1 in the initial stages of L3 morphosyntactic acquisition (e.g. Bardel & Falk, Reference Bardel and Falk2007; Falk & Bardel, Reference Falk and Bardel2011). Essentially, it is argued that the L2 acts as a filter of sorts to the L1 grammar, presumably due to the way in which an L2 is represented differently and stored in the mind as compared with the L1. In a recent paper, Bardel and Falk (Reference Bardel and Falk2012) appeal to Paradis’ (Reference Paradis2004) psycholinguistic model of bilingualism to offer a neurolinguistic basis for the effect the L2 Status Factor advocates. The main data sets provided by the model's authors that seem to best support the L2 Status Factor come from Bardel and Falk (Reference Bardel and Falk2007) and Falk and Bardel (Reference Falk and Bardel2011). Bardel and Falk (Reference Bardel and Falk2007), for example, examined placement of negation in two groups with respect to V(erb)2(nd): L1 V2/L2 non-V2 and L1 non-V2/L2 V2, learning either Swedish or Dutch as an L3. Their data showed the L1 non-V2/L2 Dutch/German group outperformed the L1 V2/L2 English group in producing post-verbal negation. They maintained that only a privileged role for the L2 is corroborated by the data.

The L2 Status Factor is a particularly strong hypothesis since, like an absolute L1 transfer position, it makes very straightforward predictions that are testable irrespective of language pairings. However, just like showing L1 transfer would only be consistent with absolute transfer under certain methodologies and language pairings, demonstrating L2 transfer alone might only be consistent with the possibility of L2 transfer as opposed to falsifying alternative explanations for its selection. It might be suggested that L2 transfer even under this approach can be circumvented by structural or other factors, which Bardel and Falk do not deny in their published work; however, it seems increasingly less clear as to how this would be possible under the current explanation based on Paradis’ model. In other words, it is not clear how or why factors such as relative structural similarity could bypass the filter imposed by so-called differences in mental storage location and type of representation (e.g. declarative vs. procedural memory) of the L1 and L2.

3.2 The Cumulative Enhancement Model (CEM)

The Cumulative Enhancement Model (CEM) proposed by Flynn, Foley and Vinnitskaya (Reference Flynn, Foley and Vinnitskaya2004) is one of two models described here that hypothesize the possibility of morphosyntactic transfer from either the L1 or L2. The CEM maintains that developmental acquisition patterns are inherently not redundant, which should be increasingly observable in multilingual acquisition. In this sense, language acquisition is viewed throughout the lifespan as a collective process whereby experience with any prior language acquisition can facilitate subsequent language acquisition precisely because the mind avoids repetition. Quite distinct from the L2 Status Factor, the CEM claims that previous linguistic knowledge transfers in multilingual development from any language available to the multilingual learner irrespective of order of acquisition. However, transfer crucially only obtains when such knowledge has a facilitative effect, hence the expectation that redundancy in acquisition will not be observed. Alternatively, when transfer from either language would not be facilitative it is effectively neutralized, that is, blocked.

Flynn et al. (Reference Flynn, Foley and Vinnitskaya2004) base their claims on data from the production of restrictive relative clauses in L1 Kazakh/L2 Russian/L3English speakers. Given the patterns observed which demonstrated that the L2 can influence development of CP structures in L3 acquisition and that experience in any previously acquired language can be taken advantage of, they claim the CEM is supported. Still, there has not been much published work that supports the CEM unambigiously (but see Berkes & Flynn, Reference Berkes and Flynn2012; Jaensch, Reference Jaensch2011 for claims of support for a “weak” version of the CEM; see also Slabakova and García Mayo (published online July 24, 2013) for a discussion of the roles of cumulative enhancement and its interaction with cumulative inhibition). What does seem to be supported by a growing literature, as we will see reviewed next, is the CEM's claim that transfer is not restricted to a default L1 or default L2. It appears that there are no clear motivations for why the CEM rejects non-facilitative transfer as a possibility. From my perspective, having to avoid non-facilitative transfer a priori would place an unrealistic burden on limited cognitive resources during the course of forming the emerging L3/Ln system. At a minimum, it implies that the learner would have to have enough experience with the L3/Ln on a property-by-property basis to determine what could, for each domain, be non-facilitative.

3.3 The Typological Primacy Model (TPM)

Like the CEM, the Typological Primacy Model (TPM; Rothman, Reference Rothman2010, Reference Rothman2011, Reference Rothman, Baauw, Dirjkoningen and Pinto2013), envisions accessibility to multiple sources of transfer in multilingual syntactic acquisition and assumes that the L3 initial state provides access to both the L1 and L2. Differently from the CEM, the TPM acknowledges the possibility of non-facilitative transfer; in fact, it predicts it in specific contexts. The possibility of non-facilitative transfer, however, derives from the same general spirit underlying the original CEM, which is the belief that L3 developmental patterns, starting with transfer at the initial stages, reflect general cognitive economy reflexes that force language acquisition to be a non-redundant process. Both the CEM and the TPM agree that multilingualism is conditioned by a cumulative effect of previous linguistic acquisition; however, the TPM views selection of a language for transfer as being conditioned by factors related to structural similarity between the languages at play, as opposed to mere facilitation. Recall that for the CEM, transfer at the initial stages and beyond is predicted to be maximally facilitative or otherwise neutralized. Unlike the CEM, the TPM hypothesizes that transfer is complete (the entire L1 or L2) and early in L3 interlanguage development, constrained by what the internal parser takes to be most similar (actual or perceived structural similarity) among the three grammars. More precisely, it makes reference to structural similarities at an underlying level of linguistic competence across the three languages. Therefore, the possibility of non-facilitative transfer is taken to not only be possible, like the L2 Status Factor (albeit for different reasons), but rather predictable.

Although research by Rothman and Cabrelli Amaro (Reference Rothman and Cabrelli Amaro2010) hinted at the basic tenets of the TPM, I first introduced this label in Rothman (Reference Rothman2010). In that study, I examined the L3 acquisition of Brazilian Portuguese (BP) contrasting two sets of L3 learners: (a) L1 speakers of English with advanced L2 Spanish and (b) L1 speakers of Spanish with advanced proficiency in L2 English. Word order restrictions relating to transitive verbs and two types of intransitive verbs (unergatives and unaccusatives) in declaratives and interrogatives as well as relative clause attachment preference were tested. Despite the fact that Spanish and BP are clearly typologically similar, BP patterns much more like English than Spanish in these domains. The data unambiguously show Spanish transfer irrespective of whether it was an L1 or L2, supporting the TPM and providing evidence against the predictions of the L2 Status Factor and the CEM. In recent years, several studies have emerged showing that relative structural similarity between the L3 and one of the previously acquired systems is the most deterministic factor for multilingual transfer.

Much of the additional work finding support for the typological factor in adult multilingualism comes from language triads where two Romance languages and English are involved (e.g. Foote, Reference Foote2009; Giancaspro, Halloran & Iverson, in press; Ionin, Montrul & Santos, Reference Ionin, Montrul and Santos2011; Iverson, Reference Iverson, Pires and Rothman2009, Reference Iverson2010; Montrul, Dias & Santos, Reference Montrul, Dias and Santos2011). This fact might leave one questioning whether the TPM makes predictions beyond such obvious language pairings. Promisingly, recent research with more varied L3 language pairings has shown similar support for the TPM (e.g. Tuvan/Russian/English, Kulundary & Gabriele, Reference Kulundary and Gabriele2012; Uzbek/Russian/Turkish, Özçelik, Reference Özçelik2013; Polish/French/English, Wrembel, Reference Wrembel2012).

4.1 Why does the multilingual mind place restrictions on transfer selection?

In Rothman (Reference Rothman, Baauw, Dirjkoningen and Pinto2013), I justified two stipulative claims: (a) typological proximity assessment has to occur very early on in the L3/Ln acquisition process and (b) once typological proximity is assessed, the entire L1 or L2 is transferred in the sense of Full Transfer (Schwartz & Sprouse, Reference Schwartz and Sprouse1996). The stipulations are motivated by economic considerations. If transfer is essentially reflexive to avoid redundancy in acquisition and thereby lessen the cognitive burden of an additional grammar, it should then obtain as early and completely as possible to be maximally useful towards these fundamental goals. Although in principle possible, property-by-property transfer would be a gradual process that is inherently more costly and slower. Such transfer, although possibly guaranteeing more accurate facilitation in the sense the CEM advocates, is not maximally efficient for reducing the cognitive burden of multilingual acquisition and/or capitalizing on previous linguistic experience to bootstrap successive linguistic acquisition. For one, doing so would not allow for increased inhibitory control to develop because the two languages would need to remain easily accessible and could possibly lead to an unnecessary increase in processing differences with each language added.

In summary, the TPM's claims are based on considerations stemming from the notion of general cognitive economy and cognitive processing factors inherent to the bilingual mind. As is well known, bilingual language production requires involvement of the executive control system differentially than a monolingual mind since, among other things, inhibition is needed to suppress the effects of activation of the other grammar. In consideration of the complex processes inherent to the management of multiple grammars in a single mind, it is reasonable to claim that complete transfer based on early assessment of typological/structural factors between the L3/Ln and other systems would be the most efficient strategy. The question remains, as to how the TPM maintains this is accomplished.

4.2 How does the parser determine typology?

An essential part of this article is to explicate and rationalize an articulation of the TPM that renders universally applicable predictions (see Rothman Reference Rothman, Baauw, Dirjkoningen and Pinto2013). The idea is to leave behind impressionism of what would likely transfer in a given language combination L1 X → L2 Y → L3 Z, replacing it with a principled proposal based on information the parser can use, namely linguistic cues, to determine transfer selection. In principle, such a proposal yields predictions that coincide with impressionistic notions of what “should” be the selected candidate much of the time, as it does in the case that Romance languages are involved as we have reviewed in several studies thus far. But it is also possible that such a proposal will yield predictions that run contrary to impressionism. It is under these latter scenarios that such proposals meet their greatest empirical chance for falsification. The goal is not to be right per se, but rather to be accurate. I embrace the risk of being wrong in putting forward a strong version of the TPM to permit new ground to test a justifiably logical proposal in pursuit of increased accuracy over time. Ideally, such a proposal will be accurate under ample empirical investigations across the widest array of multilingual language pairings as the field, not one researcher, will take on over time. Ultimately, determining whether this proposal is indeed universally applicable is an empirical question. My task then is to clarify the strong position offered herein and justify what evidence leads me to the conclusion that the proposal is defensible and has a fair chance of being accurate, accepting that it will likely need various degrees of modification at a later stage after empirical facts require reconsideration. The remainder of this section details the strong, universally applicable proposal the TPM advances for how the parser determines transfer selection.

Since the TPM predicts that multilingual transfer selection is not dependent on obvious surface level (dis)similarity, it rejects the notion that conscious psychotypological assessment on the part of the learner brings anything to bear. I would take as coincidental any correlations between the learner's perceptions and independently measurable assessments of linguistic competence. In many cases, for example, when languages are genetically related, one could predict with relative certainty that the parser's unconscious and the learner's conscious assessments of comparative similarities would coincide. However, such a correlation in no way bespeaks a necessary causation loop. Prior to this latest version of the TPM, reliable predictions were not so easily tenderable when language relatedness was not relevant; however, defining or proposing linguistic cues used by the parser to make its assessment yields more reliable predictions that are universally applicable. To be clear, by conscious psychotypology I am referring to Kellerman's (Reference Kellerman, Gass and Selinker1983, Reference Kellerman, Kellerman and Smith1986) definition, which relates to the learner's conscious perception of typological relatedness. From a generative perspective, it is reasonable to reject a priori that consciousness would have anything to bear specifically on underlying linguistic representation. If there is any merit to the proposal that transfer is essentially a reflex of cognitive economy, then it stands to reason that the parser is indefinitely scanning the input to make decisions about which system is the “best choice” to transfer whether or not there is an actual genetic relationship between the target L3/Ln and one of the existing systems. In other words, the parser obviously has no preference or motivation to select what seems intuitive in accord with the learner's conscious impression of similarity, but rather is charged with determining what would be the most economic choice linguistically. It is worth noting that this proposal, in my estimation, applies to all transfer but can only be seen in the dynamic environment of multilingualism for obvious reasons. When there is only one existing system, as is the case with L2 acquisition, the default can only be the L1. In fact, this applies even to L1 acquisition where the input is directly gauged against the template of UG. In other words, what we see uniquely in L3 is simply a specific multilingual surface manifestation of the universality of language acquisition. Structural similarities, at least as I mean them here, are linguistic theory internal notions, not necessarily intuitive ones. Given the TPM's conceptualization of the grammatical parser, only linguistic information is interpretable; that is, the parser is designed specifically to use linguistic information only as it relates to constituting linguistic competence.

With the aforementioned in mind, I provided in Rothman (Reference Rothman, Baauw, Dirjkoningen and Pinto2013) a continuum of four types of linguistic cues from L3 input that are used by the parser to fulfill the remit of what the TPM seeks to model; that is, multilingual transfer selection based on early and limited input that reveals structural proximity in linguistic terms. To be sure, following the logic espoused in Schwartz and Sprouse's (Reference Schwartz and Sprouse1996) Full Transfer Hypothesis for L2 acquisition, the TPM advances the idea that one of the two systems must be transferred completely in the initial stages. These four factors are listed below in order of relative impact/influence, represented in (1):

1. (1) Lexicon→ Phonological/Phonotactic Cues →

   Functional Morphology → Syntactic StructureFootnote ⁴

Not all of these factors are as easily usable by or equally accessible to the parser at the same time, partially depending on the specific language pairings. For this reason, the above list is intended to be implicationally hierarchical. The TPM does not idealize an unrealistic situation in which these four factors are mutually exclusive to one another. Rather, there is clear mutual dependency of the levels in the hierarchy. For example, syntactic structure clearly depends on functional morphology, which in turn is determined in the lexicon and interfaces with phonology. The extent to which this influences or is deterministic for the proposed hierarchy is an important question to be fleshed out over the course of time. As I will detail below, detecting lexical similarities is much more straightforward than detecting phonological similarities, morphological similarities and finally syntactic similarities, in this order. To start with, the latter two require more experience with the L3 and a deeper level of (implicit) knowledge about the L3 than do the former.

Recall that the goal of this proposal is one of universal application. In other words, I intend to offer a strong version in which, even when seemingly counterintuitive, this hierarchy is absolute. A reviewer offered the following scenario questioning, based on intuition, whether the hierarchy would work for certain language pairings. Let us consider the example offered by the reviewer of Mongolian L1 → L2 French → L3 Turkish, where the L1 and L3 are both Altaic languages and thus share higher degrees of typological similarities in syntax, but the L2 and L3 share much greater degrees of similarity in the lexicon due recent borrowings. The reviewer states that he/she feels that the hierarchy would not hold acknowledged intuition only. It is certainly possible that his/her intuition would hold true under empirical scrutiny, however, in the absence of actual data to the contrary I use this scenario as a mere example to highlight what the TPM, nonetheless, predicts. Insofar his/her description of the relative similarities between these languages is objectively accurate under some independent measure of similarity across these domains – and this would entail that there is indeed enough lexical similarity (e.g. cognates) to warrant French as a likely candidate for transfer – then indeed the TPM would predict counter to his/her intuition, that French would be transferred over Mongolian at the initial stages. Arguing this follows from the logic that underscores the implicational nature of the hierarchy itself; parsing lexical items at the very early stages of L3 exposure is an easier task than parsing syntactic structures per se. In Rothman (Reference Rothman, Baauw, Dirjkoningen and Pinto2013), I am clear that of the four possible types of cues appealed to, it is ultimately the language combinations themselves that determine how many and which, if any, of the four factors are usable. Ultimately the TPM predicts that the established linguistic system with the most detectable/usable structural crossover, at the highest levels of the cue hierarchy, at the earliest of timing at the very initial stages will be selected for complete transfer.

Because the TPM is an acquisition model conceived of within the generative framework, it relies on formal linguistic theory to specify how the parser detects and determines what is structurally similar across the languages based on generative proposals of underlying linguistic representation. The TPM proposes that lexical similarity is the most important factor and, when relevant, this accords with some type of unambiguous typological relatedness. Given current generative assumption of how particular language syntax is acquired from a feature-based lexicon, the parser would not need to go past the lexicon to render a “best guess” decision for transfer in the case of, for example, Spanish/English bilinguals acquiring L3 Italian. It would also be the case that much of the functional morphology (inclusive of functional features) would indeed be similar, as would the syntax that is projected from it. However, not all L1/L2/L3 language pairings will be such that there is a shared lexicon of only either the L1 or L2 respective to the L3. Such would be the case if both the L1 and L2 share more or less the same degree of lexical similarity with the L3, for example a Portuguese/Italian bilingual acquiring Spanish as an L3 or when neither the L1 nor L2 share obvious lexical crossover (e.g. a Hebrew/Russian bilingual acquiring L3 Spanish). Referring back to the Mongolian L1 → L2 French → L3 Turkish example cited by the reviewer, the question becomes if there is enough lexical similarity between French and Turkish for the parser to stop “searching” and make an unconcious decision for transfer selection without moving on to higher levels of the hierarchy such as syntax, which would entail more experience with the L3. If so, and this is to be determined by linguists intimately familiar with these languages (ideally via some type of corpus, objective measure), then French is predicted to be selected. For the TPM to make predictions for situations in which there is not enough similairty at a lower level of the implication hierarchy, it has to assume that in the absence of or in addition to a “tie” in lexical similarity, other factors play a deterministic role. The implicational hierarchy in (1) above enumerates the next steps when (a) the higher cue is not relevant given the pairing or there is weak similarity between either the L1 or L2 and the L3 or (b) when the L1 and L2 apparently tie at the next highest level.

4.3 Some clarifications on structural vs. typological similarity

In this section, I seek to make one important clarification from Rothman (Reference Rothman, Baauw, Dirjkoningen and Pinto2013) relating to the terms “structural” and “typological” in relation to the TPM. Therein, I made it seem that the terms were somewhat interchangeable without explaining exactly what I meant by each as separate terms in relation to the model. At this point, it should be clear that the TPM does distinguish between typological and structural similarity on the whole. More specifically, it proposes that the parser's early perception of typological similarity is based on whatever actual structural similarities (or lack thereof) can be assessed from limited L3 input as compared to the L1 and L2. In other words, structural similarity along the continuum presented in the last section determines the “best guess” of typological similarity for the purposes of full transfer of either the L1 or L2. Under this logic of the TPM, once full L1 or L2 transfer has taken place at the L3 initial stages, structural similarity is not predicted to be deterministic at the property-by-property level.

Under the TPM it is perfectly possible that actual structural similarity can exist between the language selected for transfer and the L3 for any given property X, but crucially this should only obtain predictably when this happens to accord with the language that is predicted to be selected by the parser as the closest overall typological match. Equally predictable and, incidentally, unexpected by the CEM, should be cases when structural similarity does exist between the L3 and either the L1 or the L2, but where transfer does not obtain since the structurally similar language is not predicted to be assessed by the parser as the typologically similar language at a holistic level. The TPM maintains that typology, strictly determined in the way explained above, prevails over actually structural similarity at the individual property level. However, structural similarity as a construct is important under and appealed to by the TPM for the purposes of determining the parser's assessment of typological similarity for complete selective transfer in adult multilingualism. This distinction is seen in Özçelik (Reference Özçelik2013), who argues explicitly for and shows convincing evidence of overall typological and not property level structural transfer in line with the TPM, showing that Uzbek–Russian bilinguals of L3 Turkish transfer scopal properties of Uzbek, a Turkic language like Turkish, despite the fact that Uzbek works differently and Russian and Turkish are identical in this regard.

5.1 Should the TPM have different predictions for distinct types of bilinguals?

There are reasons to believe that there would be differences across bilingual types, for example early (child bilinguals) and adult bilinguals, concerning the predictions of the TPM. If multilingual transfer as described by the TPM is driven by cognitive economy, it might be fair to predict that early bilinguals would not need to transfer either the L1 or L2 completely as a function of more temporal bilingual experience and the growth of inhibitory control that has been shown to ensue. Bialystok (Reference Bialystok2009) summarizes much of her work and the work of others, all of which seems to indicate that having to engage two systems by mere virtue of being bilingual from an earlier age conveys various cognitive advantages. Some work on younger bilingualism as compared to later adult L2 acquisition shows that earlier bilingualism provides advantages for inhibitory control (Bialystok, Craik & Ryan, Reference Bialystok, Craik and Ryan2006; Bialystok & Shapero, Reference Bialystok and Shapero2005). If this finding is truly generalizable, it is logical to predict the possibility of finding differences between young bilingual acquirers of L3/Ln as compared to adult L2 acquirers of an L3/Ln since the former group should be better than the latter, on average, at inhibiting the effects of two other activated systems. The proposal would be that the mind of an early bilingual with more experience with linguistic suppression might be more efficient than an adult L2 learner's mind at allocating the cognitive resources needed to inhibit both an L1 and an L2 during the initial stages of the L3 acquisition process. The result could be that transfer of the L1 or L2 happens later, when the parser can get more information from the L3 input stream to more accurately assess typological proximity. If comparing early and adult bilingual L3 acquisition at the very beginning stages showed differences in timing of transfer this would be supportive. Such a result might ultimately convey an advantage for early bilinguals, who would have more linguistic information fed to the parser before determining transfer, which could avoid some of the non-facilitative transfer predicted by the TPM. The relevance of the psycholinguistic research related to, for example, inhibitory control reviewed above for the TPM and other L3 transfer models rests in what research of this type tells us about the non-linguistic, yet cognitive complexities that must be considered as deterministic factors in multilingualism acquisition.

There are also reasons to argue for a null effect of bilingual type; after all, if what the TPM describes is an economical strategy to lessen the cognitive burden on the one hand but crucially also to reduce acquisition redundancy, why would a more experienced bilingual mind not seek to be maximally indolent? What would be the motivation for a more experienced bilingual mind to wait to transfer if the parser gets reliable information that a “best guess” can be made simply because it has greater ability to allocate cognitive resources? Should economy not require the more experienced bilingual brain to avoid redundancy at all costs as well by being lazy? Moreover, the linguistic manifestations, if any, of increased inhibition and other general cognitive advantages stemming from early bilingualism is far from uncontroversial.

The question of an advantage vs. a null effect of bilingualism is, of course, an empirical one. Fortunately, there is some research that has compared early bilinguals to adult L2 learners in the context of the beginning stages of L3 acquisition (e.g. Iverson, Reference Iverson, Pires and Rothman2009, Reference Iverson2010). Although not the focus of these studies to address the question posed above, I discuss the findings here given their valuable empirical evidence.

Iverson (Reference Iverson, Pires and Rothman2009, Reference Iverson2010) asked whether transfer patterns at the initial stages of L3 can tell us anything about adults’ ability to acquire new syntactic features not instantiated in the L1 in contra the claims of the Interpretablity Hypothesis (IH; Tsimpli & Dimitrakopoulou, Reference Tsimpli and Dimitrakopoulou2007). He argued one could compare adult L2 learners and child (heritage) bilinguals at the initial stages of an L3 when all the languages are held constant across groups – thus the crucial difference being the timing (age) of acquisition of the heritage language/L2 – if the L2/heritage language is targeted to be the likely source of transfer in the L3. His logic was that if the transfer patterns of the two sets of bilinguals each reflect transfer of the L2/heritage language, such data would show that these bilinguals have the same representation for the property under investigation. Since the early bilinguals would have acquired this property as children and thus have a native mental representation, by extension, the similar patterns of transfer in the L3 initial stages, if shown, would provide indirect evidence that adult L2 acquisition of the targeted property obtained. Iverson tested his predictions by examining the initial stages of L3 BP by Spanish/English heritage speaker bilinguals and L1 English/advanced L2 Spanish speakers. The property targeted was N(oun)-drop, a restricted form of nominal ellipsis that is a licensed reflex of grammatical gender features, which Spanish has and English lacks. Iverson examined production and comprehension across three different types of N-drop that range in relative structural complexity. He found no statistical difference between the bilingual groups, both of whom unequivocally showed native-like knowledge of N-drop in BP after minimal exposure to the language. Iverson concluded that the lack of difference between the groups provides evidence in favor of both bilingual types having the same mental representation for gender, which could only come from Spanish and was attained by each group irrespective of when Spanish was acquired.

Although it was not Iverson's intention to address the question presented at the beginning of this section, his studies provide empirical evidence to suggest that the answer to whether or not different types of bilinguals would show different timing patterns of initial stages transfer in L3 acquisition is “no”. However, the actual question posed was whether it followed by the TPM's latest assertions that such a difference might be predicted. The TPM, in my view, is in fact neutral on this. My personal inclination, which hardly matters without empirical scrutiny, would have been that type of bilingualism would in fact not invoke differences in initial stages transfer patterns for the reasons I gave above. More work that designs studies specifically around this question are certainly warranted to understand the extent of generalizability one can draw from the few existent studies such as Iverson's. However, questions like the one posed are a good start towards making the necessary links between linguistics, psychology, and cognitive science more generally that multilingual acquisition requires and that models like the TPM are attempting to facilitate.

5.2 Does proficiency in the L2 matter for the TPM's predictions?

Another reasonable question in light of the claims of the TPM regards whether transfer is predicted to be the same when previous non-native languages are not fully developed at the onset of L3/Ln exposure. In other words, do developing L2 interlanguage systems transfer in the same way the TPM predicts for “end-state” bilinguals? Interestingly, this is a question I have been asked many times over the past few years and it is especially relevant given the reality of L3/Ln acquisition, which often begins at a point in L2 development prior to ultimate attainment. The answer to this question, from my perspective, follows straightforwardly from the argumentation offered throughout this article, which regards the basic tenets of the TPM that follow from considerations of cognitive economy. It seems to me that under these same considerations the answer to this question would have to be a decisive yes; L2 developmental interlanguage grammars would transfer if, in their constitution at the onset of L3 exposure, they meet the parser's criteria for determining typological proximity.

After all, interlanguage grammars as first conceived by Selinker (Reference Selinker1972) and ever since are taken to be proper natural grammars, albeit different to varying degrees from the target and source grammars. If indeed L2 grammars are natural grammars, then at any and every point in the interlanguage continuum these grammars, although different from the target and the next level of interlanguage grammar as development continues, are complete grammars. How would the mind know that an interlanguage grammar has not yet reached ultimate attainment, and why would this system not be available for transfer? The mind does not know of or much care about so-called completeness or relative proficiency level; these are shorthand descriptions and theoretical constructs that we have invented for theory internal purposes. From the mind's perspective, at the onset of exposure to L3/Ln, the bilingual mind has whatever systems it has, whether the L2 is at a fossilized state or still in the process of modification. These systems are available to fulfill the remit of what the TPM seeks to describe, which is the avoidance of acquisition redundancy.

We should be mindful that if interlanguage L2 grammars are transferred and that such a grammar has a different value for a given property than the target native version of the L2, then this must be revealed. I have previously argued that L3 transfer studies need to test knowledge of the target property in all three languages (Rothman Reference Rothman2010, Reference Rothman2011; Rothman & Cabrelli Amaro, Reference Rothman and Cabrelli Amaro2010). The possibility I refer to here highlights the reasons why doing so is crucial. By testing the L1 and L2 after testing the L3 so as to not prime the subjects, one will be able to know for each individual participant the status of her knowledge in all three languages. This will let the researcher know if the property of interest is available in the L2 for transfer, should such be predicted given the language pairing. Recall that interlanguage grammars have been shown to embody autonomous differences (not stemming from the source L1 or the target L2) and knowing whether this is the case for the property under investigation is important to avoid making conclusions about transfer that are inadvertently false. Obviously, one cannot expect transfer of something that does not form part of the properties available to the learner from previous acquisition. Furthermore, one cannot assume that each individual has property X in her L2 simply because they are regarded as highly proficient by general proficiency tests.