7.1 The acquisition of features in L2 acquisition: Features and proficiency

The results presented in this paper indicate that second language acquisition is constrained and selective. Our results are only partially consistent with the Interpretability Hypothesis (e.g., Hawkins & Hattori, Reference Hawkins and Hattori2006; Tsimpli & Dimitrakopoulou, Reference Tsimpli and Dimitrakopoulou2007). Recall that all interpretable features involved in personal a are available in English and according to the IH only uninterpretable features are problematic, not the interpretable ones. Therefore, if the IH was on the right track, our findings would have been rather unexpected, since L2 learners showed difficulties in acquiring interpretable features in general, casting doubt on some of the IH premises; conversely, we can advance that not all features are equal regarding L2 acquisition (Guijarro-Fuentes & Larrañaga, Reference Guijarro-Fuentes and Larrañaga2011). However, how can we then explain the finding that L2 learners showed difficulties with all other conditions?

Recall that the IH claims that uninterpretable features not accessed during first language acquisition will be inaccessible during adult L2 learning. All other aspects of UG are assumed to be available. L2 learners will at some point encounter positive evidence that may conflict with their interlanguage grammar. This will force the grammar to restructure. Therefore, pressure to restructure within the constraints of UG (and not through general problem solving) means that learners may exploit only interpretable features to model properties of the L2. Crucially, there is no implication that L2 learners will use interpretable features, or must acquire interpretable features in the L2, that are different from the L1. However, restructuring of the interlanguage grammar is just as much subject to learners encountering unambiguous positive evidence as it is to some degree in the IH account.Footnote ¹⁶ When evidence is ambiguous, learners may never converge on the grammar of a native speaker, and may end up with different, but nevertheless UG-constrained grammars. Given that the conditions for use of personal a are quite subtle (see Section 3), it is possible that English speakers of L2 Spanish will never fully establish them. This does not imply, however, that they do not have access to interpretable features like animacy, specificity and telicity, only that the evidence for how they are deployed is not sufficiently clear from the input received in the case of L2 acquisition, in stark contrast to L1 learners who quickly and easily acquire this construction from the input (Rodriguez-Mondoñedo, Reference Rodriguez-Mondoñedo2008). Thus, unlike L1 acquisition, maybe in L2 it is a matter of “murky” input which selectively affects L2 learners’ knowledge of personal a uses (more on this below). Remember that the six conditions differ in terms of their complexity (e.g., C1 is [+animate, +specific], whereas C6 [+accomplishment/achievement predicate, ±human subject]), and given the fact that all of our participants were classroom-instructed learners, with no naturalistic input at the time of testing (although they have been exposed to native language via their teachers’ production), one can claim that a lack of (sufficient) exposure (at least for the more complex conditions) may have affected our results. Additionally, the differences observed between Spanish L1 monolingual acquisition and L2 acquisition may be related to the fact that input received (i.e., exposure) is a variable, at least in quantity and quality in L2. Although it was not controlled in the present study, the exposure of our L2 learners to Spanish outside of classroom had been limited, as they all were learning Spanish in the UK. The DOM in Spanish undoubtedly presents a Poverty of Stimulus problem in L2 acquisition, even though it is a mapping onto an overt morpho-phonological form whose distribution needs to be acquired on the basis of its distributional pattern in the input. Note, however, that some DOM features (e.g., examples (9b) and (10b)) may be quite rare in the input. It is clear from our experimental conditions that DOM is sometimes obligatory and other times optional, marking a specific semantic interpretation and not others. The latter cases might be more confusing and troublesome for L2 learners, if the semantic nuances that DOM involves for native speakers are not perceived in cases in which it is optional.

To offer an explanation of this finding, we adhere to, but at the same time expand, Lardiere's (Reference Lardiere, Liceras, Zobl and Goodluck2008, Reference Lardiere2009) Feature Reassembly Hypothesis. In the present paper, all of the interpretable features (including specificity and animacy) already exist in the L1 of our speakers (see Section 3), but they need to be redeployed to capture the distribution of all available readings that are mediated by the distribution of the personal a. We argue that since English lacks the mapping of the interpretable features corresponding to Spanish personal a in the DP in the object position, this redeployment may be problematic to our L2 speakers, and consequently this grammatical restructuring may take longer. Thus, the learnability problem is not completely overcome regarding L2 interpretable features, even though there is no “completely new” featural content to be acquired, but instead a reorganization of features mediated distinctly in particular grammars. Our English L2 learners indeed know from their L1 about telicity and distinct marking of different types of objects, other than a morphological equivalent to personal a in terms of feature content. Consequently, what they have to learn is how to associate these properties with Spanish DOM; for them it is then a matter of (re)assembling features. As shown earlier, L2 learners of Spanish have acquired the least complex constraint regarding the distribution of the preposition a, which involves one feature, [±animacy], but have not yet acquired the other constraints. This asymmetry could be caused by complexity constraints in terms of the number of features, the complexity of the configuration and/or intrinsic difficulty of the interpretable feature in question. Increased complexity may require a longer period of exposure until the L2 learners are capable of processing the input and identifying the features that are relevant for the personal a, and to reassemble them in the way they are conditioned and realized in their Spanish. The task of reassembling the relevant interpretable features in combinations such as the DOM in the L2 is an arduous task, although not insurmountable, which may require more intensive exposure to the language over a longer period of years. A group of near-native speakers of Spanish would be needed to corroborate this claim.

Nevertheless, one can call into question the different L2 hypotheses outlined throughout this paper. They all predict no acquisitional struggle, since all interpretable features which are needed for personal a marking in Spanish are also present in the native language of our participants. Although we recognize this as a potential hindrance, we opt to follow the FRH, albeit with caution, given the fact that the reorganization of features could be problematic: namely, all but one feature (i.e., animacy), are acquired in a straightforward way. An alternative account to the feature redeployment approach is the possibility that some of the factors listed as part of personal a may not correspond to features of individual lexical items. Animacy seems easy to be motivated as a lexical feature, but telicity does not (this may also be the case with specificity). Telicity is one of those facets of interpretation that seems to operate at a higher order of complexity, or rather at the phrasal level, than the feature level, in that it involves interpretive relations among different constituents in the clause, i.e., properties of the event, of its participants and their mapping to times. Specificity may also involve both event and argument properties. Likewise, agency is the result of compositional factors (including animacy and non-stativity of the event). If this account is true, then there may be two (or more) distinct sources of personal a: the one that can be listed in the “lexicon”, associated with a feature [+animate], and another that is something else altogether.

To that end, it may be erroneous to encapsulate all features related to personal a at the macro-level (i.e., at the level of whole language structure); by contrast, the number of features needs to be narrowed down at the micro-level (i.e., individual lexical items such as nouns, verbs and so on) in order to account for the acquisition patterns detected here.

Extending the parametric variation theory (Kayne, Reference Kayne, Cinque and Kayne2005; Manzini & Savoia, Reference Manzini and Savoia2007, among others) to second language acquisition, we postulate micro-parameters that would add predictive power. That is, the general model of language development has at least partially shifted from the notion of macro-parameters, which operate at the level of the whole language, to micro-parameters, which operate over word classes, or even individual lexical items, leading to an approach to language variation which is essentially lexicalist. Principles are narrowed down further and further in scope in order to give an explanation for the richness of natural language data, given that predictions built on supposed parametric clusters of phenomena are not borne out in L1 and L2 acquisition research (Guasti, Reference Guasti2002; Guijarro-Fuentes, to appear). Thus, a new notion of micro-parameters is of great utility to the language learner confronted with the challenge of the task of acquiring all intertwined features in personal a, for instance. There may be more fine-grained differences in terms of the semantic interpretation on certain features in the two languages and, consequently, our L2 data demonstrate that language-wide parameter settings cannot capture the full spectrum and linguistic inter-group and within-group variation observed in this domain. What the L2 learners of Spanish need to come to know when faced with the “personal a parameter” is not the simple setting of a parameter switch for the whole “personal a parameter”, but the particular lexical semantics of all individual structural domains where a may or may not appear, depending on different syntactic-semantic constraints. Thus, taking together our findings and comparing features and individual features across conditions, one could consider a lexical learning hypothesis by which the learner starts with the [–animate] = “no personal a” and expands from there to other contexts. That is, given the strikingly uniform success of all our participants across the board and across the two experimental tasks with only one feature, the bundle of features related to personal a must be narrowed down. With this in mind, we may be able to account for the diverse behaviour of our learners who seem to be unable to generalize and amalgamate all features when facing the vastness of the task of acquiring personal a. Micro independent interpretable features with possible ramifications on different language sub-domains seem to be at work here, making the learning task even more challenging as learners need to reorganize, but at the same time need to assemble information from different sub-domains. Hence, the poor rate of performance on some of the conditions may have come from the learners’ inability to reassemble all semantic properties of personal a at the same time. Thus, individual semantic features must be reassembled before learners understand how personal a may legitimately appear in different contexts.

The learnability task for interpretable features is such that, even under optimal conditions of input (e.g., Bowles & Montrul, Reference Bowles, Montrul, Bruhn de Garavito and Valenzuela2008), internal factors such as transfer from L1, non activation of all features at the same time and/or incorrect redeployment hypotheses adopted by the learners may extend the appearance of non-target-like grammars as shown here; these deviant patterns may persist until an advanced level of proficiency is achieved.

In explaining the challenging and belaboured task that learners of L2 Spanish may face in dealing with personal a, we cannot overlook the intrinsic impenetrability that some particular features may present. That is, some interpretable features may be more difficult to be redeployed than others due to their own intrinsic/internal interpretability configuration. From our findings it is difficult to weigh up such a claim, but let us for example take the telicity feature. The acquisition of telicity has been proven to be notoriously difficult by L2 learners at different proficiency levels (e.g., Montrul & Slabakova, Reference Montrul and Slabakova2003). In the same vein, some complex combinations of interpretable features such as the one responsible for personal a (in somewhat idiosyncratic, language-specific structure) might indeed not be attainable after the critical period. The effect of age, of course, can really be an issue here as has been argued for the deficit of representation around the uninterpretable syntactic features; however, we remain cautious as to any claims made in relation to the critical period effect since the good performance by our learners shows that, at least some interpretable features can indeed be acquired after the critical period.

From our data it is difficult to assess, however, whether L2 learners have indeed redeployed the L2 features, or are simply making performance-related errors in supplying the correct use of personal a. In that respect, we contend that in the reassembly process, some features will be more transparent form–function associations than others (that is, features such as animacy represent one single cognitive construct, whereas other features such as telicity represent more than one cognitive construct), and due to the fact that L2 learners are operating with L1 and L2 features simultaneously, the features that are less transparent/obscure form–function association (Anderson, Reference Andersen, Hyltenstam and Viberg1993; Bardovi-Harlig, Reference Bardovi-Harlig, VanPatten and Williams2007) would lose momentum over the more transparent ones. Thus, if the interpretable features involved in personal a were some monolithic light-switching cluster of features that were “all or nothing”, this variation would be anomalous unless it was interpreted under the micro-parameter framework as argued above. However, we can assume that learners after extended exposure to the L2 would be able to fine-tune the meaning of the L2 lexical items and successfully converge on the grammar in the new language. In this regard, it is worth mentioning Yang's (Reference Yang2002) Variational Learning Hypothesis, which claims that the degree of acquisition of parameter values depends on the proportion of sentences in the overall input to the learners that show signs of evidence for the value. That is, high frequency equals high rate of acquisition. In that respect, some of the DOM structures may be more frequent than others and hence the different rate in acquisition reported herein. It may be argued then that the acquisition of some facets of Spanish personal a is simply correlated with overall frequency of the respective environments. This would form a null hypothesis in which the entire superstructure of syntactic structures (interpretable vs. uninterpretable features), and all the theories implicating the existence and accessibility of UG during L2 acquisition would be superfluous to a certain degree. However, in order to reject this null hypothesis, a future study is needed in order to document the relative frequency of the various more subtle instances requiring personal a. Such a null hypothesis would not play down the importance of the fact that in any case learners still need to associate meaning to form and this implies syntactic structures, particularly with respect to telicity and thematic role of the subjects. Moreover, what we can tentatively claim in light of our findings is that the interpretable features dealt with in this paper may not have very clear-cut syntactic consequences and, hence, they may pose the same learnability problem that has been attributed to uninterpretable features. If this claim is on the right track, then the IH and FRH need to be redefined and a more nuanced account needs to be found. We call upon a new account, namely, the Feature (In)accessibility Hypothesis, which claims that all linguistic features regardless of their nature may posit the same learnability problems. In addition, this new hypothesis can also make predictions depending on the level of complexity of the feature itself; that is, a feature is more easily accessible when it is concentrated on a single item (i.e., animacy), whereas accessibility is reduced when the whole functional category or an entire phase is involved (i.e., vP). In that regard, the learnability problem may be related to the location of the feature rather than the nature of the feature itself.

Turning now to the proficiency levels of our participants, the results of the present study show that the interpretable features that determine the personal a affect the accuracy of the L2 learners within the different conditions we investigated; there is a clearly significant developmental change and significantly different performance of the learners from the native speakers. Nevertheless, we observe that the level of proficiency of the L2 learners did not fully affect their accuracy in the experimental conditions that required access to more than one interpretable feature. Some interpretable features cause real learnability problems even at advanced stages of acquisition and there are clear indications of selective attainment in some conditions. Even though the personal a may be very frequent in the input (either institutional or naturalistic setting), its usage could be unclear (or not robust enough to trigger its acquisition) for L2 learners because it is determined by a non-transparent set of features. The one-to-one principle (Anderson, Reference Andersen, Hyltenstam and Viberg1993, and previous work) seems to be in evidence here where the initial mapping is one morpheme one function (namely, a marks animate objects).

Albeit the present study was not set up to address the L1 transfer per se, an alternative explanation is offered. English does not allow object raising and does not mark accusative case overtly with a preposition, so this could explain the high number of target-deviant omissions (see Table 7). Recall that movement of the a DP seems to involve two separate operations (namely, assigning/checking inherent case to the object and then associating a certain set of interpretable features in that position). That is, do L2 speakers show any evidence of the position (i.e., having a marking)? Do they also show any evidence of mapping the relevant semantic features to that position? One might think that the results of our study help to provide an answer to both questions in that light. The data in Table 7 seem to show steady progress towards fewer omissions, which could in turn be interpreted as acquiring the vP object-raising operation (perhaps with a clear breakdown point among L2 high intermediate and L2 advanced learners, where learners stop “overshooting” by putting a all over the place). Meanwhile, acquisition of feature-mapping would follow a separate but parallel route. Furthermore, in the condition where L2 learners had a robust knowledge of Spanish a, the preposition was absent, whereas in the conditions where a is required, English L2 learners (particularly at the low level of proficiency) seemed to fail to supply it. If these effects were due to a “transfer strategy”, we would have expected to find good performance in all conditions in which a is not permitted (Condition 2, 3 and 5). Our findings do not support this and therefore we reject the idea that learners tend to resort to the least costly option (their L1); we are more inclined to think that these patterns of results are due to the different level of complexity in terms of the number of features and the configuration in the conditions constraining the realization of a, as mentioned above. Future research can address this issue by comparing the performance of learners of L2 Spanish whose L1 does not mark DPs as direct objects overtly with a preposition (English learners of Spanish) with L2 learners whose L1 marks the object of transitive verbs with a preposition similarly to Spanish (e.g., Romanian speakers of L2 Spanish).

Although tentatively, one can argue that even though all interpretable features are features relevant to the semantic component, some can be more relevant to a “pragmatic component”. Definiteness, for example, relates more to semantics, but specificity crucially relates to the discourse context. One only knows that an indefinite DP such as una secretaria “a secretary” is specific or not from the linguistic context. By contrast, animacy clearly involves semantics, and lexical aspect largely relates to the semantics of the predicate; however, whether the inherent telicity of a given predicate is actually realized or not in a particular case will also depend on the discourse context. Thus, the question would be whether the interpretable features related purely to the “core” semantic component of the grammar prove to be easier or harder to acquire than other types of interpretable features, which may be more peripheral to the semantic component and closer to a “pragmatic component”. Nevertheless, as pointed out by one of the reviewers, our results could also be interpreted in line with some interpretation of the Interface Hypothesis account (Tsimpli & Sorace, Reference Tsimpli, Sorace, Bamman, Magnitskaia and Zaller2006). Throughout this paper, we have discussed ideas regarding different levels of application of different features (specifically, the difficulty of acquiring features associated with the pragmatic component). Therefore, in line with such an account our results might also be interpreted as showing complete acquisition of the core semantic features, but difficulty with those associated with pragmatics. We remain cautious to any strong attachment to such claims and leave this issue for future research.

Finally, a comparison of our results with results from L1 acquisition (Rodriguez-Mondoñedo, Reference Rodriguez-Mondoñedo2008) shows an asymmetry between child L1 and adult L2 acquisition. L1 acquisition of DOM entails the assembly of semantic features drawn from a universal inventory and it appears to be part of the child's linguistic knowledge from the earliest stages of production. Unlike L1 Spanish-speaking children who seem to acquire the preposition a easily and quickly, our L2 English speakers show incomplete acquisition of the preposition a even at the advanced proficiency level.

In sum, this paper has looked at the controversy regarding the acquisition of interpretable features in L2 acquisition by investigating the acquisition of the Spanish personal a by English L2 learners. Our main findings partly support the claim that interpretable features are acquirable, and suggest that some interpretable features are intrinsically more difficult than others causing developmental delays, and may be vulnerable to fossilization. More importantly, our results do not seem to be fully congruent with any of the current feature acquisition accounts, and we propose an alternative account based on the saliency and configuration complexity of some features. Future research on the acquisition of the same or different features (i.e., interpretable and uninterpretable) with different populations will help shed light on some of our initial conclusions.