2. L3/Ln acquisition and the Interface Hypothesis

In recent research on language acquisition, there have been different proposals trying to provide a principled explanation for the attested difficulty in the acquisition of several target language structures such as the ones just presented. On the basis of work in linguistic theory (Chomsky Reference Chomsky and Kenstowicz2001; Jackendoff, Reference Jackendoff2002; Rizzi, Reference Rizzi and Haegeman1997), the concept of interface has gained importance as a construct that could help explain discrepancy in difficulty levels. Interfaces are understood as spaces where mapping occurs between the various components of grammar or between grammar and other cognitive domains. The term interface has been used in work on child language acquisition (Reinhart, Reference Reinhart2006) but it is probably work by Sorace and colleagues that has made it more prominent within the realm of L2 acquisition (Sorace & Filiaci, Reference Sorace and Filiaci2006; Tsimpli & Sorace, Reference Tsimpli, Sorace, Bamman, Magnitskaia and Zaller2006). When analyzing overt pronoun use by highly proficient L2 Italian learners and by monolingual native speakers of Italian, Sorace and Filiaci (Reference Sorace and Filiaci2006) discovered that the former provided significantly different answers from the latter regarding their interpretation of overt subject pronouns (no contrast was found in their interpretation of null subject pronouns). As mentioned in the introduction, similar asymmetries were uncovered in near-native L2 performance (Belletti et al., Reference Belletti, Bennati and Sorace2007), in first language (L1) attrition (Tsimpli et al., Reference Tsimpli, Sorace, Heycock and Filiaci2004) and in bilingual L1 acquisition (Sorace, Seratrice, Filiaci & Baldo, Reference Sorace and Seratrice2009). Thus, a strong version of the IH (Sorace & Filiaci, Reference Sorace and Filiaci2006) proposed to account for similar findings in the three bilingual domains predicts that “structures involving an interface between syntax and other cognitive domains present residual optionality (in L2 acquisition), emerging optionality (in L1 attrition) and protracted indeterminacy (in bilingual L1 acquisition) but structures that require only syntactic computations are completely acquired in L2, remain stable in L1 and are acquired early in bilingual L1 acquisition.” (Sorace, Reference Sorace2011, p. 5). In other words, structures that appear at the syntax–discourse/pragmatics for example, are predicted to be difficult to acquire even for near-native speakers of the language.

However, the dichotomy between purely syntactic domains and those interfacing with other cognitive domains was subject to criticism as research showed that not all interfaces were equally vulnerable (see Slabakova, Reference Slabakova2006, Reference Slabakova2008, arguing for successful acquisition of syntax–semantic interface properties). More recently, Sorace (Reference Sorace2011) established a distinction between internal interfaces, those between narrow syntax and the other linguistic modules (phonology, morphology, semantics) and external interfaces, those between syntax and other cognitive modules. As mentioned above, her explanation for property delays rests on processing reasons: integrating context and grammar taxes the language processor. One reason why bilingual speakers may be less efficient at processing structures at the syntax–pragmatics interface could be that syntactic processing is less automatic for them (as argued by, for example, Clahsen & Felser, Reference Clahsen and Felser2006). This effect may either be due to less developed knowledge representations or to less efficient access to these representations, or both (Sorace, Reference Sorace2011).

Sorace (Reference Sorace2012, p. 213) states that the IH is only an account of patterns of optionality found in very advanced L2 attainment but not a developmental account of L2 acquisition. However, we agree with Lardiere (Reference Lardiere2011) and White (Reference White2011) that learners who are in the process of acquiring an L2 could also experience problems with structures at the interfaces. White (Reference White2011) comments on the need to extend and test the IH in domains where it was not originally intended to apply. To the best of our knowledge, interface properties have not received much attention in L3 acquisition. However, in the spirit of the IH, we would argue that L3 learners constitute another vulnerable population with respect to acquisition of discourse-related properties, as the effects of bilingualism can be expected to be even stronger in this population.

As reported in García Mayo and Rothman (Reference García Mayo, Rothman, Cabrelli Amparo, Flynn and Rothman2012) and in Rothman, Cabrelli Amaro and de Bot (Reference Rothman, Cabrelli Amaro, de Bot, Herschensohn and Young-Scholten2013), most work within the domain of L3 syntax has focused attention on the initial state of learner interlanguage and the extent to which previous learning experiences might influence the corresponding L3. As in the case of L2 acquisition, a logical position one may take regarding L3 acquisition is that absolute transfer from the native language would obtain, and consequently that the L1 would act as a filter that would block access to acquired L2 properties. This possibility, called the L1 Transfer Factor (shortened to L1 factor in the tables below), was advanced by Na Ranong and Leung (Reference Na Ranong, Leung and Leung2009) in their study of the acquisition and interpretation of null objects by L1 Thai/L2 English/L3 Mandarin Chinese participants. However, the authors themselves warned against considering their findings as definitive, due to different shortcomings in the methodology used and the small-scale nature of their study. Other models have received more attention lately: the L2 Status Factor (Bardel & Falk, Reference Bardel and Falk2007; Falk & Bardel, Reference Falk and Bardel2011), the Cumulative Enhancement Model (CEM) (Flynn, Foley & Vinnitskaya, Reference Flynn, Foley and Vinnitskaya2004) and the Typological Primacy Model (TPM) (Rothman, Reference Rothman2010, Reference Rothman2011). As the name itself indicates, the L2 Status Factor model maintains that the L2 takes on a significantly stronger role than the L1 in the initial stages of L3 morphosyntax. On the basis of data focusing on the placement of negation in Swedish or Dutch as an L3, Bardel and Falk (Reference Bardel and Falk2007) concluded that their data could only be explained if the L2 had a privileged role.

The other two models, the CEM and the TPM, hypothesize that all previously acquired languages might be a source of transfer. The CEM is a model both of the initial state as well as a theory of developmental sequence and ultimate attainment. It states that multilingualism is conditioned by a cumulative effect of previous linguistic acquisition. That is, any other previously acquired properties are potentially available to the L3 learner to use in parsing, comprehending and producing language. What is important in this model is that these prior language experiences can either enhance subsequent language acquisition or remain neutral. The TPM is similar to the CEM in that neither of them predicts categorical transfer but the TPM views transfer as being conditioned by factors related to either actual typological similarities or perceived proximity (known as psycho-typological proximity) between the grammars of the languages involved. Thus, non-facilitative transfer is a possibility under this model, but only if based on psycho-typological proximity.

Against this backdrop, the aim of the present article is to test the IH against data from Basque–Spanish bilinguals learning English as an L3. Below, we first present the learning tasks and then spell out the concrete predictions of the IH and the three currently available L3 acquisition models as to the acquisition of these properties.

3. Syntactic analyses of Topicalization, Focus Fronting and Left Dislocation

In this experimental study, we investigate the acquisition of discourse-related dislocation constructions (Topicalization, FF and LD) in the L3 English of Basque and Spanish bilinguals. In this section, we summarize the linguistic facts and analyses in the three languages.

The left-periphery phrases exemplified in (1)–(3) above display different “connectivity” effects with their respective host sentences. Such effects include case-matching and syntactic-category–matching between dislocated constituent (henceforth dislocate) and resumptive element, variable binding, island effects, and others. These effects have been taken to signal how much the phrases are dependent on some element within the sentence. We adopt the widely held view that the dislocate in English Topicalization and FF form an A-bar dependency with an argument position inside the Tense Phrase (TP).Footnote ³ The literature explains this syntactic dependency by either postulating movement of the dislocated element from an argument to a left-peripheral position, or by proposing base-generation of the dislocated phrase and chain formation with the argument position.

Dislocate movement involves an operation that copies the original element in a left-peripheral position and deletes the lower copy, or pronounces parts of it (e.g., Boeckx, Reference Boeckx2003; Grewendorf, Reference Grewendorf, Mauck and Mittelstaedt2002; Grohmann, Reference Grohmann2003). On the other hand, according to the base-generation analysis, (e.g., Anagnostopoulou, Reference Anagnostopoulou1997a; Cinque, Reference Cinque1990; Frey, Reference Frey, Lohnstein and Trissler2004), the dislocate is merged in the Compelementizer Phrase (CP) domain, an empty operator is merged in the argument position, the latter has to move at Logical Form (LF) for interpretation. Thus, the relation between the dislocate and the operator with its copy is created by chain formation at LF. The connectivity effects of Topicalization (namely the fact that though the fronted phrase is in a non-canonical position, it still retains close relations with the core TP and its arguments) are adequately explained by both movement and base-generation approaches.

In the case of English LD, the phrase in the left periphery does not show connectivity with the rest of the sentence. This is because the pronoun or epithet coindexed with the dislocated phrase is in an argument position, thus the theta-requirements of the predicate are resolved without movement. An indication of this non-integration is the non-observance of syntactic islands (Cinque, Reference Cinque, Ehlich and van Riemsdijk1983; Vat, Reference Vat1981).Footnote ⁴ Therefore LD has mostly been analyzed as base-generation of the left-peripheral phrase and a discourse semantic connection with the sentence-internal argument. As Cinque (Reference Cinque, Ehlich and van Riemsdijk1983) points out, the connection between the hanging topic and the pronoun is the same as the relationship between any other NP and a pronoun/epithet referring to it, in adjacent sentences.Footnote ⁵

Both the movement and base-generation analyses of Topicalization handle information structure notions with syntactic means, that is, feature-assignment and movement to a functional projection for checking purposes. This forceful syntactocentricity can be considered a weakness of such approaches (see López, Reference López2009). A qualitatively different approach to (dislocated) focus and topic was proposed in Reinhart (Reference Reinhart1981) and developed by Zubizarreta (Reference Zubizarreta1998), Reinhart (Reference Reinhart2006), Neeleman and van de Koot (Reference Neeleman and van de Koot2008), and Horvath (Reference Horvath2010), among others. This approach places discourse rules and acceptability at the heart of linguistic explanation. It argues that topic and focus do not instantiate formal syntactic features and do not correspond to functional projections. Instead, topic and focus are involved exclusively in what is called “discourse grammar” by the mapping of syntactic structures to information structure representations. While topic and focus play no role in the syntactic-semantic computations, they interact with syntax phenomena indirectly, via the syntax–discourse interface.

Finally, López (Reference López2009) is a recent proposal that maintains elements from both feature-based treatments and interface-based treatments by arguing for a pragmatics module of computation. The pragmatics module yields an intermediate level of representation between the syntactic structure and the discourse structure where the features [±antecedent] and [±contrast] are assigned. The derivation proceeds as proposed in Chomsky (Reference Chomsky and Kenstowicz2001). First, individual lexical items enter the syntactic derivation by merging into phrase markers. At the end of each phase – verb Phrase (vP) and CP – a syntactic object Σ is formed. This syntactic object is then inspected by the pragmatics module, which assigns the relevant discourse features to constituents located in syntactic positions at the phase edge, yielding the information structure Σ[p]. These pragmatically enriched structures are subsequently assembled into longer units of text to form discourse representation structures.Footnote ⁶ Note that López's discourse feature [±contrast] captures the idea that the fronting operation itself (Topicalization, Clitic Left Dislocation (CLLD) or Focus Fronting) simultaneously opens up a variable position and resolves it, unlike the variable position in a wh-question, which is resolved only by the following sentence.

Spanish and English FF, Spanish Hanging Topic Left Dislocation (HTLD) and English LD have been analyzed as fundamentally parallel constructions (Kempchinsky, Reference Kempchinsky2008; Zagona, Reference Zagona2002). On the other hand, English Topicalization and Spanish CLLD as in (4) have received contrasting analyses.Footnote ⁷

1. (4)

   

Various syntactic differences between these constructions have been pointed out (Cinque, Reference Cinque, Ehlich and van Riemsdijk1983): a specificity restriction on the moved constituent in Spanish but not in English, multiple possible dislocates in Spanish versus only one in English, occurrence in root and embedded clauses in Spanish while root only in English (excepting bridge verbs), and sensitivity to strong islands only in Spanish but all islands in English. We will review here Anagnostopoulou's (Reference Anagnostopoulou1997a) analysis, which follows Demirdache (Reference Demirdache1991) and Vat (Reference Vat1981). In trying to account for syntactic differences, Anagnostopoulou compares what she calls English Contrastive Left Dislocation (CLD, the same as our Topicalization) to Romance CLLD. She proposes that both constructions involve an adjoined left-dislocated phrase coupled with an operator–variable chain. Apart from the obvious morpho-phonological difference, clitics and null operators share operator properties. The syntactic differences between the two constructions fall out of the different adjunction sites of the left dislocated constituent: CP-adjunction for CLD, TP-adjunction for CLLD. Not all analyses consider Spanish and English Topicalizations to be syntactically divergent. For example, Rizzi's (Reference Rizzi and Haegeman1997) feature-based account treats both English and Spanish fronted topics as targeting the same functional projection, Top(ic)P(hrase). The Romance–English parametric difference rests simply in the presence of a null anaphoric operator versus a clitic establishing the topic–comment connection.

Let us now turn to Basque. This is an SOV ergative-case language. The structure of finite verb forms in Basque is rather complex. Finite forms, besides being marked for tense and mood, carry agreement features with up to three arguments: subject, direct object and indirect object. All of this results in very rich verbal morphology (see (5) below). Furthermore, traditionally, Basque is also described as a three-way pro-drop language (see Goenaga, Reference Goenaga1980; Laka, Reference Laka1988; Ortiz de Urbina, Reference Ortiz de Urbina1989; Salaburu, Reference Salaburu and Salaburu1987): as illustrated in (6), the subject, direct object and indirect object (marked ergative, absolutive and dative, respectively) can be phonologically empty. Like English, this language does not allow clitic pronouns of the Romance type (Hualde & Ortiz de Urbina, Reference Hualde and Ortiz de Urbina2003).

1. (5)

   

1. (6)

   

In Basque, a focus constituent has two obligatory properties (Elordieta, Reference Elordieta2001, pp. 111ff.). It has to be identified prosodically with the focalized phrase receiving the most prominent stress of the sentence. Additionally, the focus phrase must appear in immediate preverbal position (the so-called focus–verb adjacency), a characteristic also found in other languages, such as Hungarian (Horvath, Reference Horvath1995), Italian (Rizzi, Reference Rizzi and Haegeman1997) and Greek (Tsimpli, Reference Tsimpli1995). Both the prosodic and the syntactic conditions must be met in order to identify a constituent as focus in Basque. Consider the following examples from Elordieta (Reference Elordieta2001, p.113) illustrating possible and impossible answers to a question like “Who has bought a new car?” (focused material in capitals):

1. (7)

   

Both (7a) and (7b) are well-formed answers because the focused constituent (nire nebak “my brother”) appears in an immediately preverbal position. The answer in (7c) would be ruled out as an answer to the question mentioned above precisely because it does not meet the focus–verb adjacency requirement as the object phrase intervenes between the focus and the verb.

Topic constructions in Basque resemble English ones superficially; they do not feature any clitic as Spanish CLLD does. A topicalized answer to the question in (8a), would be as in (8b):

1. (8)

   

The LD construction in Basque, unlike the English one, does not have to employ a pronoun in the main clause. Basque object agreement licences and identifies the null object pronoun. Consider the Basque sentence in (9b), corresponding to the English LD sentence illustrated in (9a):

1. (9)

   

Thus, it becomes very difficult to formally distinguish Topicalization from LD in Basque, in a marked contrast with English. Table 1 summarizes the surface differences between the three languages involved in this experimental study.

We can speculate on the perceived superficial differences between the L1s and the L3 from the point of view of the learners. In principle, a language can require the use of a pronoun, allow it optionally, or not allow a pronoun in a certain construction, or in the grammar. When we say “no pronoun” in Table 1, we mean that Basque does not allow third person pronouns at all, while English does not allow a personal pronoun in that particular construction. Both languages do not allow clitic pronouns. On the surface, Topicalization and FF seem to work similarly in Basque and English. While LD relies on null object pro in Basque, it needs an overt pronoun in the English main clause. Although Basque and English (lack of) pronoun stem from different underlying reasons, this is still a regular difference working across all constructions in the two grammars. If we assume that Romance clitics can be perceived by learners as roughly the equivalent of English strong pronouns (as suggested in Parodi & Tsimpli, Reference Parodi and Tsimpli2005), then Topicalization and CLLD are the only constructions different in Spanish and English morpho-phonetically (null operator versus clitic).Footnote ⁸

4. Predictions for second and third language acquisition

In this experimental study, we set out to test the Interface Hypothesis in L3 acquisition. This hypothesis would predict that learners will have difficulty incorporating discourse information in calculating the acceptability of three left dislocated constructions in English, even at very advanced proficiency levels. However, there are other factors that could have an effect on L2/L3 acquisition.

Adding to the difficulty of the syntax–discourse interface is the fact that all English dislocations are rare in the input to learners. Generally speaking, left dislocations in English are rare even in spoken corpora. Gregory and Michaelis (Reference Gregory and Michaelis2001), working with the Switchboard Telephone Speech Corpus, uncovered 44 Topicalizations (that is, 0.00134% of all sentences) and 187 LDs (0.0057% of all sentences). Postolache (Reference Postolache2005), examining the Wall Street Journal Corpus of 1.1 million words, reported only 24 left dislocations, including focus and topic. In comparison, the NOCANDO corpus of spontaneous Spanish speech production (Brunetti, Bott, Costa & Vallduví, Reference Brunetti, Bott, Costa, Vallduví, Konopka, Kubczak, Mair, Sticha and Wassner2011) contained 39 CLLDs and left dislocations (together) in roughly 3800 segments, which amounted to 1.35% of all finite clauses. Furthermore, there were 61 clitic doubling constructions, 2.1% of all finite clauses. Although the English and Spanish corpora are difficult to compare directly, these two frequencies differ by a factor of roughly 1000, which certainly amounts to a significant difference in frequency. To sum up, CLLD may be 1000 times more frequent in Spanish than Topicalization is in English.

Nevertheless, native speaker judgments (including the ones in this study) confirm that English dislocations are not at all marginal but completely acceptable. Augmenting the difficulty of learning these constructions is the fact that they tend to be options rather than absolutes, in the sense that there are other ways of encoding a truth-semantically identical message. Keeping this situation in mind, we will explore predictions of the IH and the effect of frequency in the input on L2/L3 acquisition. In concord with the IH, we predict that these constructions will be difficult for learners, but for different reasons (lack of enough evidence in the input).

In addition to the two factors conspiring to make acquisition difficult (discourse interface, rarity in the input), there are unique differences between the L1–L3 language pairs that are further complicated by the L2. Looking at the facts in Table 1 in another way, one can say that there are no unambiguous syntactic diagnostics in English for Topicalization vs. FF, while in Spanish the presence vs. lack of the clitic clearly distinguishes the two. On the other hand, all three left-dislocation constructions in Basque may employ null object pro, which may blur the line between Topicalization and LD in the L3 English of Basque native speakers. We submit that these inter-linguistic differences will also present learning challenges because they necessitate morpho-syntactic restructuring of the grammar.

In this article, we also endeavor to test the predictions of three L3 acquisition models discussed in Section 2: L1 transfer, L2 Status Factor (L2 factor for short in tables) and CEM. Table 2 presents the three experimental groups and outlines the predictions each of the models makes for them.Footnote ⁹

The criteria for the predictions in Table 2 are the (surface) morpho-syntactic differences in the resumptive elements in the three languages (see Table 1). Complete L1 transfer would predict that neither the L1 Spanish – L2 Basque – L3 English (Sp → B → E) nor the L1 Spanish – L2 English (Sp → E) groups would have problems with L3 FF and LD, since Spanish and English work (superficially) similarly: FF has no resumptive element while LD has a resumptive prounoun or clitic. However, this model predicts that learners would find Topicalization constructions more difficult because morpho-syntactic restructuring is necessary. The participants in the L1 Basque – L2 Spanish – L3 English (B → Sp → E) group will have problems only with LD but not with the other two constructions, again due to superficial similarity, hence no need for restructuring. Along the same lines, the L2 Status model predicts that the Sp → B → E participants will have problems only with the LD construction, which is the only one that differs between Basque and English. Recall that for that model it is the second language that will have a decisive influence over the third. The B → Sp → E group is expected to have difficulties with dislocated topics. As for the CEM, this model predicts that none of the constructions will present a challenge to the Sp → B → E and the B → Sp → E groups, since either the L1 or the L2 will freely provide beneficial transfer for successful L3 acquisition.

5. The experiment