Session 1E

Research on bilingual language development has often focused on whether the two languages of bilingual children influence each other and under what circumstances cross-linguistic influence/CLI occurs [1-3]. Most studies have investigated CLI from the first language/L1 to the second language/L2. In this study we examined the production of multiple wh-questions/MWHs in bilingual children living in the UK who have Romanian as heritage language/L1 and English as majority language/L2. The study aimed at uncovering whether qualitative differences emerge in the production of Romanian and English MWHs in Romanian-English bilingual children when compared to their monolingual peers, and whether differences in performance stem from differences between the L1 and L2 grammatical systems. MWHs provide a good test case: they have language-specific syntactic properties which children need to acquire despite overall little evidence in the input for the use and interpretation of such structures [4]. Romanian, contrary to English, exhibits obligatory multiple wh-movement. A differential object marker (DOM) pe precedes wh-objects and which-objects are doubled by a clitic (Examples (1a-d), Appendix 1).
18 Romanian-English bilingual children, 32 Romanian monolinguals and 20 English monolinguals (age range 6-9yo) participated in an elicited production study in which they had to produce questions with two wh-phrases (1a-d). Bilingual children were assessed in both their languages. 
Romanian heritage children and Romanian monolinguals produced three main types of questions, at different rates (Fig.1, Appendix 2): MWHs with multiple wh-movement (multiple move); MWHs with one fronted wh-phrase, one in-situ (single move); simple wh-questions. A GLMER revealed a significant Group x QuestionType interaction (heritage children produced significantly more MWHs with one fronted wh-phrase, one in-situ, while monolingual children mainly produced these, albeit to a lesser extent, in questions containing which-elements (1b)). No difference appeared between Romanian-English bilinguals and English monolinguals: both groups produced MWHs with one fronted wh-phrase, one in-situ at a similar rate (Fig.2, Appendix 2). 
This indicates that, when compared to monolinguals, quantitative differences emerge only in bilingual children's heritage, but not in their dominant language: bilinguals produce less complex MWHs in Romanian involving movement of only one wh-phrase. Language production in the children's L1 seems to be affected by L2 properties, under CLI, but no CLI from Romanian to English emerged in the production of English MWHs. This asymmetry suggests that more complex structures are less likely to be transferred and that the likelihood of CLI increases when one language offers a less complex structural option, especially if this is the children's dominant language.
References: 
[1] Hulk & Müller, (2000). Bilingual first language acquisition at the interface between syntax and pragmatics. BLC, 3(3), 227–244; 
[2] Serratrice (2013). Cross-linguistic influence in bilingual development: Determinants and mechanisms. LaB, 3(1), 3–25; 
[3] van Dijk et al. (2021). Cross-linguistic influence in simultaneous and early sequential bilingual children: A meta-analysis. JCL, 1-33; [4] Grebenyova (2011). Acquisition of multiple questions in English, Russian, and Malayalam. LA, 18(3), 139–175.

Capitalizing on crosslinguistic similarity (cognateness) facilitates the acquisition of L2 and L3 vocabulary (Parasitic Word Learning Model; Ecke, 2015). However, the precise mechanisms of L3 lexical learning based on similarity are unknown. It is unclear whether the L2 has a cognitively different status than the L1 (Bardel & Falk, 2021), and whether one or all previous languages influence L3 acquisition. Bartolotti and Marian (2017) proposed two possible mechanisms of L3 cognate learning. The accumulation account posits that formal overlap across L1, L2 and L3 accumulates to enhance learning, so L3 words similar across more previously known languages will be learned better. According to the scaffolding account, any overlap of an L3 form with the L1 or L2 form suffices to boost learning. Bartolotti and Marian (2017) demonstrated the scaffolding mechanism, however for an artificial L3.
We investigated how previously known natural languages impact the L3 word knowledge in speakers of L1_Polish, L2_English, and L3_Italian. We tested the knowledge of 120 L3_Italian words: 40 noncognates (gorzki/bitter/amaro), 40 L2=L3 cognates (skromny/modest/modesto), and 40 L1=L2=L3 cognates (stabilny/stable/stabile), controlled for part of speech, frequency, concreteness, length, and orthographic overlap. Assuming that L3 lexical learning may be influenced by individual learner differences, we examined whether participants' L3 knowledge was modulated by their L2 and L3 proficiency (DIALANG; https://dialangweb.lancaster.ac.uk), learning aptitude (LLAMA_B; Meara, 2005), and working memory (Polish Reading Span Test; Biedroń & Szczepaniak, 2012).
The participants were 93 L1_Polish learners of L3_Italian (Mage=21.82, SD=3.31) with L2_English. Their proficiency ranged between pre-intermediate and advanced in L2_English (MDialangENG=668.65, SD=166.72) and L3_Italian (MDialangIT=433.12, SD=217.22). To measure productive L3 lexical knowledge, we adapted the Vocabulary Knowledge Scale (VKS; Wesche & Paribakht, 1996) to activate all three languages: L2_English (stimulus presentation), L1_Polish (language of instructions), and L3_Italian (word and sentence elicitation). Our L3_VKS study also utilized confidence ratings and 40 L2-English-like nonwords to control for guessing. The L3_VKS results were rated by two proficient Italian academic teachers (κ=.989 for words, κ=.967 for sentences). 
Generalized linear mixed-effects models were fitted separately for word-level and sentence-level knowledge (motivated by the binomial distribution of scores). The results revealed that, at both the word and sentence level, the L1=L2=L3 cognates were known better than the L2=L3 cognates, which were known better than the noncognates. Regarding learner-related variables, only L3 proficiency turned out to be a significant predictor of word knowledge. Additionally, guessing disadvantaged performance on all word types. However, for word-level knowledge, participant propensity to guess (guessed nonwords) disadvantaged the L2=L3 cognates more than the L1=L2=L3 cognates and noncognates. 
Overall, our results suggest that learning L3 words is significantly facilitated by participants' L3 proficiency, but not their L2 proficiency, and by cumulative L1=L2=L3 crosslinguistic similarity. The results agree with the accumulation account of L3 lexical learning by Bartolotti and Marian (2017), which contradicts their results obtained for an artificial L3. The results also contradict the special status of L2 knowledge (Bardel & Falk, 2021) in learning L3 words.

In null subject languages like Spanish, monolingual speakers show a preference for interpreting a null pronoun (pro) as referring to a subject antecedent (Pedro), while overt pronouns (él ) are more likely to refer to non-subject antecedents (Carlos).
 Pedroi saludó a Carlosj cuando élj/proi cruzaba la calle
 Pedro greeted Carlos when he crossed the street

Bilingual speakers of a null subject language (e.g., Spanish) whose second language is a non-null subject language (e.g., English) can show some optionality in the interpretation of subject pronouns in the first language (L1, Spanish), choosing the subject antecedent for an overt pronoun more often than Spanish monolingual speakers. However, it is still unclear how frequent patterns of pronoun use in the L2 affect pronoun resolution in the L1 of bilingual speakers, determining potential cross-linguistic interference. In the present study, we look at the interpretation of subject pronouns in Spanish heritage speakers (HSs; minority language speakers) (experiment 1) and a group of late learners of English (L1 Spanish, experiment 2), to explore the effects of second language (L2) proficiency on pronoun resolution in the native language. 
In experiment 1, forty-eight HSs of Mexican Spanish and forty-eight monolingual Spanish speakers participated in a sentence comprehension task where they had to choose the referent of an ambiguous null/explicit pronoun in anaphoric or cataphoric position (Table 1 in attached materials). The results showed that HSs chose the subject antecedent significantly more often than monolingual speakers for all pronouns (main effect of Group< .0001). The results show that pronoun interpretation in the heritage language deviates from monolingual biases, with HSs demonstrating a stronger subject antecedent preference for anaphoric and cataphoric null/explicit pronouns than Spanish monolinguals. The results further suggest that English proficiency is inversely correlated with the development of monolingual-like interpretation biases in heritage Spanish (p< .0001). 
In experiment 2, we test a group of intermediate-high proficiency L2 English speakers (L1 Spanish, N=31) compared to a group of thirty-one monolingual Spanish speakers using a sentence comprehension task. Participants chose the referent of an ambiguous Spanish null/explicit pronoun in anaphoric position (Table 2). The results show that English learners gave more subject interpretations for null and overt anaphoric pronouns than monolingual Spanish speakers (main effect of Group: p< .001). In addition, lower English proficiency was linked to more monolingual-like interpretation patterns (p< .0001), demonstrating that L1 anaphoric interpretation is susceptible to influence from a late learned L2 when L2 proficiency is intermediate-high.  
While several factors may contribute to HSs performance in experiment 1 (input quality/quantity, attrition, cross-linguistic influence), we tentatively link the performance of HSs (experiment 1) and L2 speakers (experiment 2), speculating that the pressure from the L2 system may at least partially contribute to the performance observed in the two groups of bilinguals. We conclude that exposure to pronoun uses in the L2 can contribute to shape bilinguals' interpretation biases in the L1, and cross-linguistic interference can be greater at increased levels of L2 proficiency, determining divergent comprehension patterns in the L1.