ABSTRACT. The bilingual lexicon of heritage speakers (HSs) in the heritage language (HL) is distinct from that of monolinguals: quantitatively, it is smaller, as the speaker must balance multiple language systems, and qualitatively, it evolves as the result of contact with the dominant societal language (SL), leading to creative lexical innovations stemming from language interaction. For instance, Polinsky (2006) found that adult HSs in the US compensate for major gaps in their HL vocabulary by code-switching and borrowing from their SL. In fact, as compared to the more-commonly studied discipline of morphosyntax, vocabulary is more vulnerable to different input factors. While researchers agree that HL input diminishes in quantity and quality as HSs age, there is no consensus regarding which input factors are most significant in predicting HL performance, and how these factors might interact with each other. Over the last few decades, input and the bilingual experience have been indexed in multiple ways, including the age of SL onset, family type, HL input over different stages of life, motivations for maintaining the HL, and current HL use in different settings. However, only recently has the field begun using network models to study the interplay between these factors, as such models provide the most accurate reflection of the bilingual experience. In the present study, we apply this new methodology to 40 adult HSs of Hebrew in the US (16M, 24F). Participants’ lexical abilities were assessed through quantitative and qualitative analyses of performance on the MINT and on a narrative elicitation task. We considered target accuracy, types of non-target responses, and cross-linguistic influence from English. Our results showed that adult Hebrew HSs fit the classic paradigm for HSs overall, performing at near ceiling-level accuracy on noun-naming in their dominant English with little variability, and reaching an average of 56% in Hebrew (SD= 16%, Range: 15%-82%). Code-switching and calquing patterns in the narratives suggested that English is the framing language from which speakers draw resources directly or indirectly. Participants also completed a background questionnaire, detailing language use across contexts and age ranges, sense of identity, and motivations and strategies for maintaining the HL. Using these data, we conducted a network analysis using the R packages bootnet and qgraph, confirming a complex set of relationships between input, identity factors, and objective lexical performance (indexed by MINT accuracy and the number of unique target lexical units produced in the narrative). Our follow-up centrality analyses found that “HL Use at Work” was the strongest node in the network model, and “Cultural Identity”- the weakest. Our study presents evidence on a highly understudied HL, further elucidating the HL lexicon by comparing our findings to more-frequently investigated HLs (Russian, Spanish, etc). Additionally, our network model shows that the HL lexicon is influenced by multiple input and background factors, which form complex bidirectional relationships. In our presentation, we will discuss specific examples of linguistic blending in Hebrew HSs’ production data and review the implications of our network analysis for future assessments of HL bilingualism.

ABSTRACT. Russian heritage language (RHL) has a prominent place in the empirical landscape of heritage languages (Polinsky, 2008; Schwartz et al., 2015; Laleko, 2019; Brehmer & Kurbangulova, 2017; Meir & Janssen, 2021; Otwinowska et al., 2021; Klassert et al., 2014, Makarova & Terekhova, 2021; among others). Yet, large-scale investigations studying RHL development across a wide age range, a large number of participants, and different national contexts are scarce. To fill this gap, we investigate HL development in a large sample of 143 pre- and primary-school children between the ages of 3 and 10 acquiring Russian in Germany, Norway, and the UK (Table 1). Our main focus here is on a less-studied domain – lexical development.

We assessed lexical production patterns in both of the bilinguals’ languages with an ecological language procedure – narrative storytelling, which taps into the ability to use vocabulary in real-life situations, in contrast to vocabulary tests. The oral language samples were obtained using MAIN (Gagarina et al., 2012) and analyzed with two widely used measures: total number of words (TNW) and number of different words (NDW), to address the following research questions:

RQ1: How does lexical development proceed in RHL of pre- and primary-school children? RQ2: How will the shift in language input and use during school age affect lexical development, if at all? Are there signs of stabilization or stagnation of vocabulary development? RQ3: Which individual background factors can explain the variance in lexical knowledge?

Analyses reveal a clear and steady increase with age in narrative length and lexical diversity for all bilingual groups in both of their languages (Figure 1). The individual variation in lexical productivity as well as the differences between the bilingual groups and between bilinguals and monolinguals are attributed to input factors, with language exposure in the home and age of starting preschool as the major predictors (Figure 2). Interestingly, when the data are separated by family type, we see that lexical diversity development in the HL slows down considerably compared to the lexical development of the majority language, but only for the children who grow up with two Russian-speaking parents (since children from mixed families are dominant in the societal language from early on and throughout development). That is, lexical development in the majority language has a much steeper slope than the slope for the HL, which only shows a slight increase over time. In line with previous research (Klassert et al., 2014), we also see that the two lines cross around age 5, indicating that the dominance shift occurs already at this young age in this group of RHL children. Overall, we conclude that our results support the view that having longer exclusive or uninterrupted exposure to a HL in early childhood is beneficial for HL development. The present study contributes further to the validation of the TNW and NDW measures in assessing bilingual children’s lexical development (Hržica & Roch, 2021). Finally, to obtain a comprehensive picture of language development in RHL, we compare these results to the evidence from grammatical (gender) and phonological (global accent) acquisition in RHL obtained in previous investigations.

ABSTRACT. The current study examines how the bilingual experience of Mandarin Heritage Speakers (HSs) impacts the computation of English doubly-quantified sentences such as (1), which are ambiguous in English (the ‘dominant’ L2) and unambiguous in Mandarin (the heritage L1). Inverse scope readings (1b) are possible for native speakers and advanced learners of English [3, 4], but are argued to be unavailable in Mandarin [5].

(1) A cat is chasing every mouse. a. SURFACE SCOPE ∃x [cat(x) & ∀y [mouse(y) -> chasing(x,y)]] There is a single cat that is chasing every mouse. b. INVERSE SCOPE ∀x [mouse(x) -> ∃y [cat(y) & chasing(y,x)]] For every mouse, there is a (different) cat that is chasing each mouse.

The inverse scope (1b) is more costly to process than surface scope (1a), possibly due to a Processing Scope Economy (PSE) principle whereby individuals prefer to parse a construction with the fewest syntactic operations [6]. The PSE predicts that HSs should be able to compute a more complex configuration (inverse scope), but they will incur a greater processing cost. In contrast, the Avoidance of Ambiguity (AA) principle [1] predicts that HSs disprefer ambiguous constructions and thus will disallow inverse scope in their L2 English [2].

The current study probes 1) whether HSs pattern more in line with the well-documented PSE or the emergent AA strategy which is specific to the HS experience, and 2) if group differences will arise based on bilingual experience between HSs and their late second-language acquiring Mandarin-English counterparts (L2s). We predicted that HSs would align with the AA strategy, but would select the inverse image at a higher rate and faster than L2s.

Method. Forty-three highly proficient bilingual participants completed a language background questionnaire and a forced picture choice task using PC Ibex software, which remotely collected three measured variables of Response Time, Picture Selection, and Goodness Rating. The design of this study is 2x3, with Sentence Type (1, a > every) vs. (2, every > a) crossed with Picture Set (Surface+Inverse; Surface+Distractor; Inverse+Distractor, Fig. 1, with extra objects) (12 items).

Results and Discussion. When a surface image is available, HSs seem to align with the AA strategy as they showed almost no differences between the Surface+Inverse and Surface+Distractor conditions (exception: a significant main effect of Picture Set on Goodness Ratings). However, in the Inverse+Distractor condition, HSs did select inverse scope at a higher rate with high Goodness Ratings which supports the PSE (Table 1). The forced choice task between two dis-preferred images revealed more nuanced evidence to inverse scope activation that supports the PSE over the AA strategy, which to our knowledge has not been documented previously. Finally, HSs overall showed higher evidence of inverse scope activation than L2s which further supports the PSE. We found that the language experience of HSs impacts doubly-quantified scope computation in the dominant L2 (particularly in comparison to late-acquired L2s), but slightly contrary to predictions of the AA hypothesis which is an emergent HL theoretical hypothesis with little experimental evidence that warrants further exploration.

ABSTRACT. BACKGROUND. Most deaf people are born to hearing families and raised in the hearing society. As a result, finding monolingual adult signers in a country like the US is nearly impossible. Three categories of signers exist (broadly speaking): Deaf-of-Deaf (children of deaf parents, who typically receive non-native linguistic input themselves), Deaf-of-Hearing (most of whom receive natural linguistic input well after the early years of life), and Hearing-of-Deaf (Codas)[1]. The latter group consists of hearing individuals raised in Deaf families who often sign for daily communication. Codas are HeritageLanguageUsers (HLUs) [2]. This suggests that (a) since they experience SignLanguage from birth, they, like the Deaf-of-Deaf, possess requisite knowledge of the language to be considered native; (b) due to typical bilingualism, their languages may be affected by the presence of one another. Thus, if Codas/HLUs children mirror monolinguals regarding a linguistic phenomenon, we can take this as evidence of a particular linguistic behavior in the language itself. In essence, we argue, examination of any linguistic phenomena, especially in a minoritized language as above, ought to include bilinguals/HLUs as evidence. The phenomenon under investigation is gradability in AmericanSignLanguage.

STUDY. Despite signer variability, the literature describes ASL, like English, as a [+degree] language [3][4]—its adjectives are inherently comparative (1). ASL strategies for comparison are in (2)-(4): comparative morphology (synthetic and analytic (2)), non-manual intensification (3), and zero marking (conjoined/paraphrase (4)). We examine gradability and emergence of comparatives in two child spontaneous production corpora in which ASL-acquiring children interact with 2-3 adult caregivers. Populations: 4 Deaf-of-Deaf/monolingual children and 3 Hearing-children-of-deaf-adults (Coda/Heritage) (Tables 1-2 [4]). We predict the presence of informative structures in adult production, the syntactic information from which children take as evidence [5]. Simultaneous search was conducted in ASL and English translation. A variety of predicates were tagged, differing in iconic potential, boundedness, frequency of occurrence [6]. Predicates were observed in canonical and contextual comparative use by adults and children, both in ASL and English. A sample of FRU of predicates in (5) by Deaf/monolingual and Coda/Heritage children is presented in (6). In ASL, gradable predicates are observed early: ~16-35ms. Most of the time comparative readings were expressed with 1-word utterances: the search yielded over 8,500 tokens of MORE but only a fraction in multi-word utterances (Fig.2). The earliest to emerge and most frequent items were SAME(-AS) and MORE (Tab.3). Zero overt comparatives co-occurred with adjectives in the Deaf/monolingual corpus; 12 tokens total were observed in the Coda/Heritage corpus, including one by a child. In English, early use of predicates and comparative structures with adjectives is evident throughout [7]. Zero instances paralleling (1b) were observed, produced by adults or children in any language. The same predicates are observed at different times in both languages.

CONCLUSION. The data support (a) the [–degree] status of ASL, (b) the lack of effect of English on ASL in terms of word-learning by HeritageLanguageUsers, and (c) Codas’/HeritageLanguageUsers’ production patterns consistently with Deaf-of-Deaf/monolinguals’

ABSTRACT. Categorising speakers as monolingual or bilingual is rather arbitrary because it is difficult to identify a clear line between these two categories (Mackey 1962). The linguistic experience of being “bilingual” is in fact, multifaceted, comprising different profiles (simultaneous, sequential, L2 learner, attriter, to name a few, Surrain & Luk, 2019) and with no strict boundaries between them (e.g., the so-called returnees). Despite this, most research on bilingualism still adopts a categorical approach to language experience (bilinguals with monolingual controls), which is supported by evidence of cognitive characteristics found only in bilinguals (Wagner, Bialystok & Grundy, 2022). Given this background, it is necessary to understand to what extent bilingual profiles can be generalisable. In this study, we demonstrate that is possible to accurately predict the linguistic profile of speakers using a machine learning model trained on the typology of their linguistic productions. For this purpose, we used data from 86 adult speakers of Italian from three different linguistic profiles captured via the LeapQ questionnaire (Marian, Blumenfeld & Kaushanskaya, 2007): monolinguals, attriters, and heritage speakers, who completed a series of picture description elicitation tasks targeting the production of Italian clitic pronouns. Clitic pronouns are vulnerable in Italian and are progressively lost across generations of bilingual speakers immersed in English-speaking settings (Smith et al. 2022). A support vector machine (SVM) was trained on ~3700 utterances to predict the linguistic profiles from the type of answer they provided. The classifier was trained on 90% of the data, randomly selected, and tested on the remaining 10% and this procedure was iterated 1,000 times to make sure that results were generalised. Results showed that the linguistic profile was correctly predicted with an accuracy above chance with monolinguals most accurately classified (~72%), followed by heritage speakers (~61%) and finally attriters (~49%). Further analyses of the classifier errors (i.e., a confusion matrix) showed that attriters were more frequently misclassified as monolinguals, whereas heritage speakers were misclassified more often as attriters. Our results suggest that linguistic performances can well differentiate speakers across linguistic profiles, which may justify treating each category of the speaker as a separate profile. However, the analyses of errors make also it clear that mono- and bilingualism do not really have sharp categorical boundaries, but rather distribute on a continuum of linguistic performances. References Mackey,W. F. (1962).The description of bilingualism. Canadian Journal of Linguistics, 7, 51–85. Marian, V., Blumenfeld, H. K., & Kaushanskaya, M. (2007). The Language Experience and Proficiency Questionnaire (LEAP-Q): Assessing language profiles in bilinguals and multilinguals. Journal of Speech, Language, and Hearing Research 50, 240-267. Smith G., Spelorzi R., Sorace A. & Garraffa, M. (2022). Language Competence in Italian Heritage Speakers: The Contribution of Clitic Pronouns and Nonword Repetition. Languages 7(3), 180. Surrain, S. & Luk, G. (2017). Describing bilinguals: A systematic review of labels and descriptions used in the literature between 2005–2015. Bilingualism: Language and Cognition, 1–15. Wagner, D., Bialystok, E. & Grundy, J. G. (2022). What Is a Language? Who Is Bilingual? Perceptions Underlying Self-Assessment in Studies of Bilingualism. Frontiers of Psychology 13:863991

ABSTRACT. Greek, Russian and Turkish (GRT) belong to different types of null subjectlanguages [1]. G is a strict pro-drop language. R is either a partial or a non-pro-drop language with abundant subject ellipsis. T is a discourse prominentlanguage which allows topic drop, i.e., subjects are dropped in highly specifiedcontexts. It has been shown that despite these typological differences, the heritagevarieties show similarities in pro-drop realization. Previous studies on GRT incontact with Germanic languages (i.e., non-pro-drop English and topic-drop German) found that overt pronominal subjects are more frequent in heritage varietiesthan in monolingual ones, suggesting transfer effects from the majority languages[2, 3, 4]. For other languages, animacy is claimed to have an impact on subject re-alization, e.g., Mandarin [5, 6], Vera’a and Teop [7], and Tamambo [8]. However,the effect of animacy related to pro-drop in GRT is understudied [9, 10].This study aims to fill this gap by investigating whether animacy, along withother extra- and intra-linguistic factors, impacts pro-drop use in GRT. Based onprevious research, we formulate the following research questions and hypotheses:RQ1 How does the realization of pro-drop align within and across different vari-eties of GRT?H1.1 We expect divergent patterns across GRT due to typological differences.H1.2 Heritage speakers (HSs) will show transfer effects from majority Germanand English increasing the overt realization of pro: Greek < Russian <Turkish. Transfer is claimed to occur in restricted domains such as thepragmatics-syntax interface [11]. With respect to the Interface Hypothesis[12], internal (core grammatical system, agreement of phi-features) and ex-ternal (information structure and discourse) interfaces seem to be involvedin the realization of pro-drop.RQ2 Are there effects of intra-linguistic factors, e.g., animacy, and extra-linguisticfactors, country of elicitation (Greece, Russia, Turkey, Germany and USA),formality (formal / informal) and modality (written / spoken), on the pro-duction of (pro)nominal reference in different varieties of GRT?H2.1 Based on the animacy hierarchy [13], we predict inanimate subjects to berealized more frequently than animate ones.H2.2 We expect pro-drop realization in monolinguals to be sensitive to formalityand modality whereas HSs lack these features that are transmitted via formaleducation. Thus, we predict a weakening in HSs’ pro-drop production dueto register-levelling and overexplicitness [14].

We conducted a study on the RUEG corpus [15] containing manually anno-tated data of 548 speakers. For each language, we ran binomial generalized linearmixed-effects models. Our results indicate moderate effects for country of elici-tation, formality, modality and animacy which confirms our hypotheses. Due totypology, we found effects on pro-drop in HSs of less strict languages, T and R,unlike in G HSs. As predicted, for GRT animacy affects pro-drop use accordingto the hierarchy features and the inherent salience of animates [16]. Specifically,inanimacy prevents pro-drop in GRT. We will extend the analysis including num-ber, person and part of speech as intra-linguistic factors.