Session 2D

Mandarin Chinese allows null subjects when the referents are Topics. In the case of intransitive main clauses as in (1), the embedded null subject refers to the main clause subject. In addition, Chinese null subjects can alternate with pronouns such as ta ‘he’ as in (2), and then the interpretation has to be disjoint: ta cannot refer to the main clause subject. 1. (1) Daxiang1 chang-de ?1/?2 kuqilai le. Big.Elephant sing-DE ? begin.to.cry ASP ‘Big Elephant sang, and as a result he (Big Elephant) began to cry.’ 2. (2) Daxiang1 chang-de ta?1/2 kuqilai le. Big.Elephant sing-DE he begin.to.cry ASP ‘Big Elephant sang, and as a result he (someone else) began to cry.’ English works differently in this respect, not allowing null subjects and not enforcing a disjoint interpretation on the overt subject. Research on null and pronominal subject interpretation in L2 Mandarin (Kong, 2007; Yuan, 1993; Zhao 2011, 2012, 2014) has been vigorous, but it still offers a mixed picture in terms of successful acquisition. In this presentation, we adopt Reinhart’s (2006) Reference Set Computation (RSC) analysis. It predicts that children’s pronoun interpretation accuracy would be around chance, since interpretation incurs high processing costs. Children’s processing resources are taxed and they resort to guessing. Slabakova et al. (2017) showed that the RSC analysis is supported in L2 English pronoun interpretation. In this experiment, we follow Shibata & Yashima (2014) in testing the RSC predictions in L2 Mandarin. Shibata & Yashima (2014) found that 5-year old monolingual Chinese children, evaluating sentences as in (1) and (2), were above 80% accurate on the null subject interpretation (comparable to a full noun phrase in the same position) while they were only 37% accurate on the ta pronoun. In other words, they were at best guessing who the pronoun referred to. This is surprising because at this age children already use pronouns correctly. The RSC analysis offers an elegant explanation of these findings. We hypothesize that what is difficult for Chinese children will also be difficult for learners of Chinese as a second language. We used Shibata & Yashima’s Truth Value Judgment Task, with minimal modifications for adult speakers. An example context involves two characters: Big Elephant, who likes to sing sad songs and Little Monkey, who loves listening to songs. In one version of the story, Big Elephant is so moved by his own sad song that he begins to cry, while Little Monkey is happy because he is listening to music. In the context of this story, the sentence in (1) with the null subject is the correct interpretation. In a second version, Little Monkey listens to the sad song Big Elephant is singing and begins to cry. The sentence in (2) with the overt pronoun ta is true for this story. We used six presentation lists which included 12 stories, each with two versions as exemplified above. Test sentences alternated null subjects, ta and full NPs. Each participant saw 12 story–test sentence combinations. We tested 35 learners of Mandarin with English as their native language and 19 controls. The average accuracy by condition and group is given in Table 1 overleaf. For learners and natives, there is a significant difference between the accuracy of ta and both null subjects and full NPs (p < .0001), but not between null subjects and full NPs (p = .98; p = .3). These findings are entirely in keeping with Shibata & Yashima’s child language results. Since there is no difference in the experimental conditions except the embedded subject linguistic properties, we argue that the RSC analysis is supported by evidence from Mandarin L2 acquisition.

Purpose | Background: We test the predictions of the Feature Reassembly Hypothesis[1] (FRH) as applied to the acquisition of French 3rd person accusative pronouns by L2 learners of Mexican Spanish. The FRH proposes that bundles of atomic features are transferred from L1 to L2—these must then be reassembled to meet L2 specifications, a process that can be problematic. Previous studies of L2 acquisition of French accusative clitics suggest that learners initially struggle using grammatical gender to accurately interpret French clitics when there is a mismatch in L1/L2 pronominal features (e.g. English-French interlanguage)[2], but less is known about cases where reassembly is not required. According to the FRH, acquisition should proceed smoothly. French accusative pronouns le/la encode grammatical gender and do not lexically distinguish between between [±human] referents (unlike English). Similarly to French, Spanish accusative clitics lo/la also encode grammatical gender. However, even though Spanish abstract pronominal features match French pronominal features perfectly, the pronominal paradigms share some surface-level similarities, i.e. le is 3rd person dative in Spanish but 3rd person accusative in French. Additionally, Spanish le does not encode gender and does not obligatorily encode [+human] but frequently does.[3] In its current formulation, the FRH does not consider how surface-level idiosyncrasies of lexical items might affect the mapping and reassembly processes. Procedure: Two tasks were used to investigate grammatical representations and online processing. The picture-selection task (2x2; factors: ±human, gender; n=3) introduced two potential referents as part of a short context (Fig. 1). Four picture choices were provided: ?gender-matched, *gender-mismatched, *both masc/fem referents possible, *distractor. Similarly to the first task, the self-paced reading task (2x2; factors: ±human, gender match/mismatch, n=24) also contained short contexts and test sentences including a clitic. Unlike the picture-selection task, though, only one of the two referents was discourse-prominent in the context and the clitic in the test sentence either matched or mismatched this discourse-prominent referent in gender. L2 participants also completed a c-test that served as a proficiency measure. Participants: (Mexican) Spanish-French bilinguals [n=36] tested in Mexico and French Native Speakers [n=27] tested in France. Results: Picture-selection results revealed that L2 learners were overall quite accurate and used gender to interprete clitics (Fig. 2). Further statistical analysis is under progress. Self-paced reading task results (segment: verb following the clitic) are presented in Fig. 3 and 4. Since the RT were not normally distributed, a log transformation was used to analyse the results statistically. We ran two mixed-effect models, one for masculine referents and one for feminine referents (lme4 package, R) with main fixed effects for gender match/mismatch and ±human and random intercepts for subjects and items. Both models revealed only one significant main effect for group (Masculine: ?=.4707, SE=.0087, t=5.408, p=0.000; Feminine: ?=.0.505, SE=.0849, t=5.951, p=0.000). Further statistical analysis is currenly in progress (e.g. to incorporate proficiency results into the analysis), however, it is important to point out that among L2 learners the higher reading times trend for gender mismatched clitics was reversed for masculine clitic (le) when referring to [-human] referents. Conclusion: Overall, these results suggest that the FRH might need further elaboration to account for surface-level idiosyncrasies at the lexical level since the current study suggests that even when no reassembly is expected (L1 French-L2 Spanish), surface-level similarities in forms can hinder acquisition. Works Cited [1] Lardiere, D. (2009). Some thoughts on the contrastive analysis of features in SLA. Second Language Research, 25(2), 173-227. [2] Shimanskaya, E., & Slabakova, R. (2017). Re-assembling objects: a new look at the L2 acquisition of pronominal clitics. Bilingualism: Language and Cognition, 20(3), 512-529. [3] Roldán, M. (1975). The great spanish le-lo controversy. Linguistics, 13(147), 15-30.

The distribution of post-verbal subjects (VS) in Italian is governed by lexicon-syntax (with unaccusative verbs; Belletti 1988) and syntax-discourse (information-focus on the subject; Belletti 2004) interface conditions. In L2, phenomena related to the former interface are more easily attained (Sorace 2011). Accordingly, the performance of advanced L2-learners of Italian with VS differs depending on whether the subject follows an unaccusative verb or is marked for focus (Belletti et al. 2007; Lozano 2006 on L2-Spanish). However, it remains unexplored how VS emerges at lower levels of proficiency and whether the divide between lexicon-syntax and syntax-discourse interface phenomena is reflected at the developmental level (cf. Bettoni, Di Biase & Nuzzo 2009 and Nuzzo 2015 for a similar perspective). We conducted a corpus study on the LIPS Corpus, that collects transcriptions of language assessment exams of learners with varied L1 and proficiency (A1-C2) (Vedovelli 2006). After dividing the transcriptions in units and isolating VS-structures (Table 1 PDF), we analysed them following a multi-layered annotation of verbs and subjects. We considered verb-type (distinguishing between unaccusatives – coding separately the “piacere (like)-type” verbs with ThemaNOM and ExeriencerDAT – unergatives and transitives) and semantic (agentivity) and information-structure features of the subject. To avoid any arbitrariness in coding topic and focus, we relied on the inventory of topic- and focus-associated functions reported in Riester & Baumann (2013). The former features include givenness (at both lexical and referential level), while the latter contrast with a set of alternatives and occurrence with focus-operators (accordingly, given constituents be focused, too). As for the verb-type, Figure 1 (PDF) reveals a decrease in the use of VS with V=“piacere-type” from B2 onwards and an increase in the use of VS with other types of unaccusatives from B1 to B2. There is also a steady increase in the number of VS with V=transitive. A relevant percentage of VS with V=unergatives (21%) occurs only at C1. The analysis of the information structure shows that the higher the proficiency, the greater is the tendency for postverbal subjects to be associated with focus (Figure 2 PDF, middle line), with the difference between B2 and C1 reaching significance (?(1) = 7.29, p =.005). Most of these subjects are contextually given (line on top), which is observed across all levels (with the lowest percentage – 55% - at B1). Finally, postverbal subjects are increasingly associated with agentivity across levels A1-C1, with a significant increase between B1 and B2 (?(1) = 5.31, p =.02) and with a slight decrease between C1 and C2 (bottom line). Overall, our results (as based on a developmental approach and decompositional analysis of VS-structures) comply with the hypothesis of a developmental trajectory from lexicon-syntax to syntax-discourse interface phenomena. The semantic analysis suggests that post-verbal constituents are progressively assigned subject-status, being more and more associated with agentivity. This developmental pattern is also confirmed by the error analysis concerning verb-subject agreement (20% of errors in A1&A2, 14% in B1, 6% in C1 and 3% in C2). Likewise, the increasing occurrence of focused subjects (together with the extension of VS to unaccusatives, unergatives and transitives) denotes learners’ developing sensitivity to information-structure (with the preferred interpretation being contrastive focus). When looking at lower proficiency levels, the data seem to challenge the idea that “piacere-type” verbs emerge relatively late because of non-canonical alignment between thematic roles and grammatical functions (Bettoni et al. 2009). However, the observation that in these structures, postverbal constituents are not necessarily assigned subject-status (see data on subject-verb agreement errors) suggests that piacere-verbs function as “pivots” for a syntactic schema that is later extended to other verb-types.