ABSTRACT. The co-occurrence of two languages in the same speaker inevitably raises the issue of how two linguistic systems that exist independently of each other in the input are represented and processed by the same individual. Psycholinguistic models of sentence comprehension and production have addressed this crucial question to some degree in bilingual adults – specifically adult second language learners. Less attention has been paid to the mechanisms of sentence processing in bilingual children who are exposed to two languages in infancy or early childhood.

 In this talk I will focus on cross-linguistic influence in sentence-level production and comprehension. Specifically, I will examine the production and interpretation of constructions that are either ungrammatical or semantically/pragmatically sub-optimal. Learning a language involves learning which constructions to use in which contexts, and the conditioning factors that apply. In the case of bilingualism this is complicated by the fact that the same conceptual space needs to map onto language-specific constructions which may - or may not - overlap across languages. Unlike adult L2 learners – who already have an established L1 before the introduction of the L2 – bilingual children need to carve out the conceptual space while simultaneously dealing with input whose quantity and quality may differ between languages. I will draw on psycholinguistic mechanisms of language co-activation and inhibition, construction entrenchment, and syntactic priming to propose an account of cross-linguistic influence that brings together representation and processing and that factors in individual variation in the children’s bilingual experience.

ABSTRACT. The present study used a visual world eye-tracking paradigm to investigate online processing of grammatical gender and number in 25 Mandarin-Italian bilingual children, in comparison to 32 monolingual Italian children. We examined how children anticipated upcoming nouns on the basis of grammatical gender and number information on the preceding article. While monolingual speakers are able to employ such predictive mechanisms from a very young age, to our knowledge,this is the first study that compares gender and number processing in bilingual children.The results show that, overall, participants made linguistic predictions on the basis of articles, although a post-hoc analysis focusing on a subset of our bilingual participants did not confirm the prediction effect in Mandarin-Italian bilingual children. We found a greater difference in the bilingual group than in the monolingual group, in that bilinguals tended to be slower when processing gender than number. Finally, we found that L2 proficiency had a significant effect on gender processing in the bilingual group. One interpretation of these findings is that the discrepancy between gender and number may be due to transfer, since Mandarin does not have grammatical gender while it does have a conceptual notion of number. Another factor may be L2 proficiency, and especially lexical knowledge, since gender is an arbitrary property stored in the lexicon, while number is concretely linked to the referential context.

ABSTRACT. Children need to acquire the specific properties of multiple wh-questions (MWHs) despite limited exposure to such structures in the input [1-3]. In this novel study, we compared the processing and production of Romanian MWHs in thirty-two monolingual and fourteen heritage children with English as an additional language. Romanian, contrary to English, exhibits obligatory multiple wh-movement. We aimed to examine whether (i) heritage children show similar processing patterns to monolinguals for MWHs in Romanian; (ii) who and which MWHs are equally difficult for comprehension; (iii) quantitative and/or qualitative differences appear between the two groups in production.

In a self-paced listening task, children (6yo to 9yo) listened to forty embedded questions with two extracted wh-phrases, varying WhType (cine/‘who’; care/‘which’) and WhOrder (Subject-Object; Object-Subject). In an elicited production task, children were prompted to produce twenty-four MWH questions containing who- and which-phrases.

Our findings reveal no difference in processing between monolingual and heritage children, and a speed-accuracy trade-off for comprehension. Children were more accurate with who-MWH than which-MWH, but slowed down when processing who- as compared to which-phrases. Differences emerge however in production, as the heritage children, contrary to the monolingual children, produced very few MWHs with multiple wh-movement and significantly more MWHs with a fronted wh-subject and a wh-object in-situ for all the types of elicited questions. Monolingual children mainly produced such questions, albeit to a lesser extent, in the conditions with two which-elements. This indicates that language production in Romanian heritage children is affected by L2 properties, under cross-linguistic influence [4-7].

ABSTRACT. An important question about bilingual acquisition is the extent to which bilingual children’s languages influence each other. This question has led to several accounts on this cross-linguistic influence (CLI), in which factors such as surface overlap, language dominance and age have been found to affect its presence and strength (e.g., Argyri & Sorace, 2007; Hulk & Müller, 2000; Serratrice, 2013). Results, however, are mixed, and methodological differences between studies make comparisons difficult. In this study we therefore assess the magnitude of CLI in bilingual children by means of a systematic review and meta-analysis, to our knowledge the first on this topic. Using predefined criteria (e.g., monolingual controls, experimental design) we selected studies containing sufficient data to perform a meta-analysis: 27 in total (198 data-points). Our systematic review of study designs and definitions of surface overlap and language dominance yielded two important insights: (1) authors defined and operationalized surface overlap and dominance in many different ways; and (2) study methods varied considerably, supporting the necessity to generalize over studies to explore the underlying effect of CLI. Our meta-analysis revealed a significant effect of CLI (Hedges’ g=0.47; SE=0.12, 95% CI[0.24,0.71], p<.001). Furthermore, surface overlap significantly predicted CLI, whereas age did not. These results show that, abstracting away from methodological differences, there is a small to moderate effect of CLI in bilingual children. Variation can partly be accounted for by surface overlap, but not by age, suggesting that CLI is not solely a developmental phenomenon, but rather part and parcel of being bilingual.

ABSTRACT. Evidence for crosslinguistic influence (CLI) in syntactic processing has been conflicting, both for L1 influence in L2 acquisition (Hopp2017, RankinEtAl2019) and L2 influence in L1 attrition (BergmannEtal2015, Kasparian&Steinhauer2017). Here we compare the two directly.

We present findings from two groups of highly proficient German-English sequential bilinguals (21 L1-German expats in the U.S., 22 L1-English expats in Germany) who participated in a Visual-World eye-tracking experiment containing ambiguous subject- and object-wh questions in German, examining whether their interpretive preferences were influenced by CLI from the unambiguous English word-by-word translation. Such influence would manifest in an effect of Tense, with bias for subject-question interpretations in present (WasSUBJ-OBJ/whatSUBJ leckt/licks die/the KatzeOBJ-SUBJ/catOBJ?) and object-question in perfect tense (WasSUBJ-OBJ/whatOBJ hat/has die/the KatzeOBJ-SUBJ/catSUBJ geleckt/licked?).

Sentence-final interpretations (Fig1) show no difference in interpretive bias by Tense among L1-German controls in Germany (n=22). L1-German expats did not differ significantly from German controls in final interpretation or eye-gaze patterns; proficiency, exposure and usage did not modulate performance. By contrast, L1-English expats differed from both controls and L1-German expats, with significantly more subject-question interpretations in present than perfect tense (Mpresent: 82%; Mperfect: 26%), and significantly more fixations on agent referents in the present tense than the L1-German groups. Although not modulated by L1-proficiency and L2-exposure, the Tense effect correlated positively with length-of-residence in an L1-English environment (r=.58, p=.005).

These findings indicate CLI is asymmetric: Entrenchment of L1 processing routines can lead to persistent CLI in L2 processing, while it protects from L2 influence in L1 processing, even after prolonged L2 immersion.

ABSTRACT. The aim of this study was to examine under which circumstances the Foreign-Language effect, an increase in bilinguals’ rate of utilitarian (rational) decisions in their foreign versus their native language, holds. More specifically, the influence of modality (reading versus listening) and type of dilemma (personal versus impersonal) on the Foreign-Language effect was examined. Personal dilemmas involve the self and are therefore assumed as more emotional. 154 highly proficient Dutch-English bilinguals were asked to read and listen to personal and impersonal moral dilemmas (within-subjects factors) in Dutch or in English (between-subjects factor). Importantly, the reading task had the character of a self-paced reading task in order to be more comparable to the nature of the listening task (i.e., differences in timing and ability to review dilemmas). In both modalities, participants’ task was to indicate whether the proposed action was appropriate or not. A mixed effects logistic regression model revealed two main findings. First, the Foreign-Language effect was found and modulated by the amount of emotion involved. That is, more utilitarian decisions were made in the foreign language than the native language on personal dilemmas. This pattern was not found on the impersonal dilemmas. Secondly, an effect of modality was found, that is, participants responded overall less utilitarian on the reading than the listening task. Overall, these findings give insight in the interplay between language, emotion and task demands, revealing that moral decision making is context-dependent.

ABSTRACT. Background: Investigating children’s interpretation of sentences with novel verbs can determine whether children have acquired syntactic representations, (Noble, Rowland & Pine, 2011). However, most research has been conducted with monolingual children. Low-socioeconomic status (SES) bilingual children have poorer vocabulary and grammar than low-SES monolingual children, (Hoff, 2013). The current study therefore examines the relationship between SES and monolingual and bilingual children’s developing sentence interpretation. We report preliminary data from low-SES children here; testing of low-SES and high-SES children is ongoing. Method: English-speaking monolingual and Gujarati-English-speaking bilingual three-year-olds (N=25) and five-year-olds (N=26) from low-income backgrounds (determined by parental questionnaire) completed a sentence comprehension task (Noble et al., 2011). Children viewed animations in a forced-choice paradigm whilst hearing transitive and intransitive sentences with novel verbs; their understanding of these syntactic structures and transitive role assignment was measured by their points to the corresponding video. Results: Five-year-olds outperformed three-year-olds, but monolinguals did not outperform bilinguals overall, though results differed across sentence structures. There was no reliable difference between monolingual and bilingual three-year-olds’ comprehension of transitive and intransitive structures, but monolingual three-year-olds more accurately assigned semantic roles to transitive structures. Five-year-old monolinguals only interpreted intransitives more accurately than bilinguals. Conclusion: Low-SES bilingual children showed lower performance on some measures, however they were not always outperformed by low-SES monolingual children, suggesting that SES may go some way to explaining these results rather than language background. Comparisons to high-SES children will help explore the impact of SES on developing sentence interpretation.

ABSTRACT. High-Variability Phonetic Training (HVPT) (Logan et al., 1991) has been shown to be effective in improving the perception of even the hardest second-language (L2) sounds. Most previous work with training paradigms focused on recognizing individual sounds (prelexical perception). Truly successful training, however, should also enhance the recognition of words (lexical perception), an issue we address here.

Twenty-four adult intermediate French learners of English completed eight online sessions of HVPT on the perception of /h/. This sound does not exist in French, and French listeners confuse it with silence both on the prelexical (Mah, Goad & Steinhauer, 2016) and the lexical level (White et al., 2017; Melnik & Peperkamp, 2019). In pretest and posttest participants completed an identification task (presence vs. absence of initial /h/ in non-words) and a lexical decision task (real word vs. non-word). Test items in the latter consisted of words and nonwords, with or without an initial /h/ (e.g. husband, officer, *usband, *hofficer). The training sessions used the same identification task as in pretest, but with different stimuli and with immediate corrective feedback.

In identification, there was an effect of Session (pretest vs. posttest) (p < .001), with a mean improvement of 13%. Crucially, in lexical decision there was also an effect of Session for test items (mean improvement: 16%). Thus, online prelexical training completed in the home environment improved not only the perception of /h/ but also the recognition of words containing this sound. We will discuss theoretical and practical consequences.

ABSTRACT. This study sought to compare the linguistic performance of Persian bilinguals and multilinguals, and their underlying cognitive processes represented in narration. Both groups were aged between 19 to 22 years, female and sequential in their languages (bilinguals with Persian as their L1, and English as their L2, multilinguals with Persian, English, and Spanish as their L1, L2, and L3 respectively). They all had learned English in the EFL educational context, and the multilinguals were studying Spanish at Allameh Tabataba’e University, Iran. At first, the Oxford Placement Test was applied to a group of bilinguals and multilinguals in order to check their L2 homogeneity and then, two groups of 15 participants in each one, having upper-intermediate English level, were selected based on the availability sampling. Then, a narrative task with four steps was applied with a certain response time for each step. The variables are compensatory strategies, cohesive devices, discourse markers, self-repair mechanisms, and mental state terms. All the variables were assessed in terms of their frequency in narration. The results were analyzed based on Halliday & Hassan’s Model of Cohesion (1976). The analysis of variance indicated multilinguals’ advantage in their working memory relative to bilinguals’, and also in their selective attention, a crucial element of the brain’s executive function. They were better at semiotics and were more creative when facing lexical access difficulties. Generally, this group outperformed the bilinguals in linguistic variables relying heavily on developed cognitive skills including the ability to resist inter-linguistic interference.

ABSTRACT. We tested 4-to 6-year-old L2 learners of English enrolled in a bilingual German-English kindergarten (N = 30; mean age = 5.7 years, SD = 0.7). In a visual word task using eye tracking, children had to decide which of two pictures displayed on the screen matched the previously heard auditory stimulus word. The words were (1) either German-English cognates (e.g., Baby/baby) or noncognates (e.g., Baum/tree), and (2) were presented with either a semantically-related or unrelated competitor picture. We predicted that the children would process (1) cognates faster than noncognates (i.e., show a cognate-facilitation effect; Poarch & Van Hell, 2012; Von Holzen et al., 2019), and (2) semantic-related competitors slower than unrelated competitors (i.e., show a semantic interference effect; Poarch, Van Hell, & Kroll, 2015; Vales & Fisher, 2019). Both behavioral (reaction times and accuracies) and eye-tracking data (percentage looks to target) yielded a significant cognate-facilitation effect, indicating that the children’s performance was boosted by cognates. Furthermore, while the semantic manipulation of the competitors was not found to influence processing behaviorally, the children’s eye movement data revealed semantic interference of the related items. The results suggest that cognate status exerts a stronger impact on word recognition and processing in these young L2 learners than semantic relatedness. Finally, a subsequent correlational analysis on the children’s cognate and noncognate performance and their cognitive control assessed through the Flanker task (Eriksen & Eriksen, 1974) showed a significant positive correlation between noncognate performance and the Flanker effect, indicating that enhanced cognitive control facilitated noncognate processing.

ABSTRACT. Although L1 syntactic transfer robustly surfaces in off-line comprehension (e.g. Grüter & Conradie, 2005/06; Rankin, 2014) and cross-linguistic syntactic priming (e.g. Nitschke et al., 2014), it remains unclear whether and when L1 effects emerge during L2 sentence processing (Clahsen & Felser, 2018; Rankin et al., 2018).

To examine the temporal dynamics of L1 activation in L2 sentence processing, we conducted a visual-world eye-tracking experiment with cross-linguistic syntactic priming. 28 L1-German and 28 L1-Japanese intermediate-to-advanced learners of English listened to English subject and object wh-questions (targets), each preceded by wh-questions in the respective L1 (primes), and selected the target picture. Critically, English subject-questions (Which animalSUBJ chases the catOBJ?) are ambiguous between subject and object-questions when mapped onto German verb-second syntax (Welches TierSUBJ/OBJ jagt die KatzeOBJ/SUBJ?), yet cannot be accommodated by Japanese SOV syntax.

Participants' final interpretations of English questions revealed a Group-by-Question-Type interaction, yet no cross-linguistic priming effects. L1-German learners had lower accuracy on subject questions, consistent with L1 influence, while L1-Japanese learners scored lower on object questions. By contrast, eye-movement patterns were similar across groups, but showed cross-linguistic priming effects, albeit only for object-questions.

In sum, processing of wh-questions seems cross-linguistically governed by parsing principles favouring subject-questions, and can be affected by cross-linguistic priming of the dispreferred object-questions. L1 transfer only surfaces late in off-line interpretation, and is not modulated by cross-linguistic priming. We argue that the L1 grammar only gets selected when participants assign a global interpretation to the parse, yet does not guide incremental parsing.

ABSTRACT. This study investigated the role of syntactic constraints, discourse prominence, and antecedent proximity in how Turkish learners of English process English reflexives. Two eye-tracking experiments (Experiments 1&2) and one pen-and-paper antecedent identification task were conducted. In Experiment 1, the binding-accessible antecedent was linearly closer to the reflexive and both antecedents (accessible&inaccessible) c-commanded it. In Experiment 2, the binding-inaccessible antecedent was linearly closer to the reflexive, but it did not c-command it. In both experiments, the inaccessible antecedent received discourse focus. Gender congruence was manipulated to create a match/mismatch between the reflexive and the (in)accessible antecedents. In the antecedent identification task, the sentences were presented either with or without discourse context. Data were collected from 95 advanced Turkish learners of English. The results showed that structural constraints affected the antecedent retrieval behavior of Turkish learners of English, but the integration of this information was observed in later measures (e.g., regression path duration) compared to the early measures (e.g., first fixation duration) reported for native speakers in Sturt (2003). There was no evidence for initial or later sensitivity to discourse prominence/linear proximity of antecedents in L2 learners’ online processing, but discourse-related information was used in their final interpretations. Results suggest that L2 learners, specifically Turkish learners of English, can be native-like in their antecedent retrieval behavior for reflexives although their integration of linguistic information (both structural and non-structural) in this process could be delayed, due to their slower processing speed.

References: Sturt, P. (2003). Journal of Memory and Language, 48(3), 542-562.

ABSTRACT. In this study we investigated parafoveal processing by L1 (n=51) and late L2 (n=51) speakers of English (L2ers had L1 German) while reading in English. We hypothesized that L2ers would make use of semantic and orthographic information parafoveally. Using the gaze contingent boundary paradigm (Rayner, 1975), we manipulated 6 parafoveal masks in a sentence (Julia removed th*e skin from the salmon): identical word mask (skin), English orthographic mask (skix), English string mask (nhte), German mask (haut), German orthographic mask (haui), and German string mask (losi). We found an orthographic benefit for L1ers and L2ers when the mask was orthographically related to the target word (skin vs. skix) in line with previous L1 research (Schotter et al., 2012). English L2ers did not derive a benefit (rather an interference) when a non-cognate translation mask was used (skin vs. haut), but did derive a benefit from a German orthographic mask (skin vs. haui). While unexpected, it may be that L2ers incur a switching cost when the complete German word is presented parafoveally, and derive a benefit by keeping both lexicons active when a partial German word is presented parafoveally (narrowing down lexical candidates). To the authors’ knowledge there is no mention of parafoveal processing in any model of L2 processing/reading, and the current study provides the first evidence for a parafoveal non-cognate orthographic benefit in sentence reading for L2ers. We will discuss how these findings fit into the framework of bilingual word recognition theories, and L1 and L2 reading models.

ABSTRACT. Developmental changes in the L2 Grammar show in the speakers’ ability to readjust their non-native parse following L2-specific linguistic prompts. This experiment tested adult Armenian-Russian bilinguals in a self-paced reading task. This study used structural ambiguities to check (1) whether sentence processing depended on syntactic or non-syntactic information; (2) whether the participants implemented structural parsing in their L2-Russian. The target sentences are full RCs in Russian and reduced RCs in Armenian.

(1a) Russian: Bill saw / arrested mother-ACC man-GEN [RC that was speaking about cosmetics in the yard]. (1b) Armenian: Bill saw / arrested [reduced RC yard-in cosmetics-about talking] man-GEN mother-ACC. Bill saw / arrested the mother of the man that was talking about cosmetics / soccer in the yard.

Who was talking in the yard? (a) the mother (b) the man

First, the task used social conventions to assigned certain activities to be performed by women, like talking about cosmetics, or by men, talking about soccer by men. Social biases could favor a certain answer in (1). Second, the study investigated mental structure building in both languages. Notice, that in Russian [the mother] is a higher noun within the complex head NP. In is it the lower one. To select [the mother] in (1) means building different mental structures in Russian and Armenian. The results show no effect of social biases on the interpretation of ambiguous clauses. L2 speakers perform mental structure building and parse Russian-like in Russian and Armenian-like in Armenian.

ABSTRACT. It has been reported that listeners process speech on the basis of a unit of timing of their first language. The aim of this study is to determine whether native Japanese English learners segment L2 speech on the basis of mora and whether their segmentation strategies change as L2 proficiency develops. As stimuli of a monitoring task, ten pairs of monosyllabic words sharing the same initial two phonemes were selected such that the first two phonemes (CV: pea) made up the first member of the pair and the initial three phonemes (CVC: peak) formed the second member of the pair. The first members of the pair (CV) are legal syllable structures in Japanese, while the second members are not. Non-words were prepared by adding monosyllables to the target pairs, which formed clear word boundaries (peaksome) and ambiguous boundaries (peaklef). Sixteen native Japanese speakers with a high level of proficiency in English (JH), sixteen native Japanese speakers with a relatively low level of English proficiency (JL), and sixteen native English speakers (ES) were asked to respond to the target in a sequence of non-words. The results show that all the groups showed slower responses and higher miss rate in detecting the CV targets than the CVC targets. JL and ES were affected by the word boundaries indicated by English syllable structures. The native Japanese speakers did not use a mora-based segmentation strategy for L2 speech. Both the English group and the Japanese groups syllabified the materials in segmenting English words.

ABSTRACT. The present study bridges three research fields: psycholinguistics, sociophonetics, and second language (L2) perception research. Psycholinguistic studies suggest that pitch accent in Tokyo Japanese constrains lexical access such that minimal pitch-accent pairs (/kaki/HL ‘oyster’, /kaki/LH ‘persimmon’) do not activate each other (Cutler&Otake1999), and people interact with native and non-native speech differently (Lev-AriEtAl2018). Sociophonetics and L2 perception literatures suggest that the manipulation of social information about speakers affects real-time word recognition (KoopsEtAl2008), the perception of the strength of foreign accent and comprehension (Kang&Rubin2009).

The present study uses the visual world paradigm to investigate lexical access in Tokyo Japanese minimal pitch-accent pairs embedded in sentential contexts. In the center of the screen, a photograph of a Japanese (~ native) or a Caucasian (~ non-native) speaker was presented to manipulate the perceived nativeness of the speaker (between subjects).

Previous studies investigated processing of pitch accent in isolated words only, leaving a research gap with regard to the role of sentential context. In this study, critical nouns (e.g., /kaki/LH/HL) were paired with verbs compatible with the meaning of both items in the pair (e.g., ‘eat’) or only one of them (e.g., ‘open/plant’).

Preliminary results from 25 Tokyo Japanese speakers show an effect of the verb manipulation, but not of speaker-nativeness manipulation. This suggests that while pitch accent constrains lexical access, word recognition is facilitated by a constraining sentential context. The null effect of speaker-nativeness suggests that perceived nativeness of speaker might affect offline perception and comprehension but not real-time lexical access.

ABSTRACT. Sentence processing in bilinguals is shaped by cross-language activation (e.g., Kidd et al., 2015). A key phenomenon in which this cross-language activation is reflected is cross-language structural priming, the finding that bilinguals’ production and comprehension of sentences is influenced by previous exposure to related sentences in another language (e.g., Kootstra & Doedens, 2016; Kidd et al., 2015; Serratrice, 2013, 2016). Whereas this phenomenon has been studied extensively in adults, it is still relatively underexplored in bilingual children. For example, it is not yet clear whether priming takes place in bilingual children’s comprehension, and to what extent it has not only short-term but also long-term effects. To fill this research gap, we performed an experiment in which Dutch-English and Dutch-German bilingual children listened to Dutch ambiguous sentences with a prepositional phrase (e.g., woman chasing man with a broom) and then selected one out of two pictures that they thought matched the sentence. Using a between-subjects pretest-priming-posttest design, target sentences were preceded by Dutch or English resp. German items that primed a specific interpretation of the sentence. Both in the within- and between-language condition, we found evidence of priming of sentence interpretation – both as a short-term and long-term effect. We further found that lexical overlap between primes and targets boosted priming effects – another finding often observed in adults. The results indicate that cross-language structural priming is an important mechanism underlying cross-language interaction in bilingual children (cf., Serratrice, 2013, 2016), and that this mechanism works similarly in bilingual children and in bilingual adults.

ABSTRACT. Previous research suggests that the ability to anticipate spoken language is related to literacy, as shown by differences between literates versus illiterates (Mishra et al., 2012) and by the effect of reading skills in monolingual children (Mani & Huettig, 2014). It has been argued that reading experience trains the core processes of anticipation, but it remains unclear to what extent this effect is modulated by secondary factors, such as vocabulary knowledge (Huettig & Pickering, 2019). The present study investigated the possible underpinnings of the effect, by focusing on Italian-German bilingual children formally educated in both languages. We tested 39 bilingual third-graders (M Age= 8;8) in a visual world eye-tracking paradigm. Participants were presented with two pictures that either matched or mismatched in grammatical gender, accompanied by the Italian sentence Dov’è la/il…? ‘Where is the…?’, so that the target noun could be anticipated during the gender-marked determiner in the mismatch condition but not in the match condition. Standardized reading and vocabulary tests were administered in both languages. A principal components analysis of various Italian and German literacy measures extracted a first principal component composed of both Italian and German decoding skills. Predictive processing was found to be positively related to children’s L1/L2 dedoding skills (i.e., the first principle component) and to vocabulary knowledge in both languages. This supports the view that both primary and secondary reading abilities support prediction in spoken language processing, and that cognitive benefits of literacy may transfer across languages in biliterate children.

ABSTRACT. Grammatical gender congruency (GC) effect in L2 learners has been demonstrated in multiple tasks (Sá-Leite et al., 2019), however, the nature of this interference remains understudied. The prevailing explanation is the interaction between L1 and L2 gender systems during lexical access: during L2 word retrieval, the L1 equivalent and its gender are co-activated creating interference when the genders are incongruent. The alternative, representational, explanation is the inaccurate encoding of the gender information in the L2 lemma or the underspecified gender representation in the L2 (Lemhöfer et al., 2008). The present study tested the representational hypothesis and evaluated whether the GC effect emerges early on in L2 word learning due to incorrect gender encoding when an L2 entry is first registered in the mental lexicon. 60 native speakers of Russian learned 36 words of an artificial language (AL) with a phonologically transparent gender system. Half of the words were gender-incongruent. Participants completed several tasks testing their memory of the AL words (free recall of AL words, semantic and grammatical translation recognition tasks, a word ending fill-in task) immediately after training, one day and one month later. A semantic interference effect emerged right after training indicating that lexical-semantic representations are established even after a brief exposure to a new language. However, no substantial evidence of grammatical gender interference was observed, suggesting that the GC effect reported in the literature cannot be explained by the encoding problem. We discuss the implications for models of bilingual lexical access and adult L2 word acquisition.

ABSTRACT. A large body of research on L2 comprehension has addressed the question how bilinguals process words in their L2, and specifically to what extent meaning access is dependent on activation of the L1 form prior to semantic activation. Several studies have shown that comprehension is conceptually driven at higher levels of proficiency, but predominantly shows effects of form-level processing at lower levels of proficiency, in line with the Revised Hierarchical Model (RHM; Kroll & Stewart, 1994). Most of this work, however, has solely focused on noun stimuli, meaning that information about the representativeness of the model is only available for this particular word class. Compared to nouns, verbs are more complex in terms of morphology and semantics, which may well affect their underlying processing and learning. To test the predictions of the RHM, we examined noun and verb processing of Dutch and German L2 learners of English by means of a L2-L1 translation recognition task. Test items included form-related distractors and semantic-related distractors as false translations (e.g., for “we send”, correct translation: “wij versturen”; form distractor: “wij verstoren”/we disturb; semantic distractor: “wij leveren”/we deliver) to test whether verb processing is affected differently by form and semantic interference than nouns, and whether this is modulated by proficiency. Preliminary results indicate few effects of L2 proficiency and little form interference, but do suggest more semantic inference for verbs in comparison to nouns. This adds to the notion that verbs are more difficult to process, even at later stages of L2 proficiency.

ABSTRACT. An essential yet often overlooked step towards a full mastery of difficult second language (L2) contrasts, such as the distinction between English /ɛ/ and /æ/ for Dutch and German speakers, is the reliable encoding of the sounds in the contrast into the (phono)lexical representations of L2 words that should contain them (e.g., /ɛ/: bet and lemon; /æ/: bat and dragon). The present study investigated the contribution to this encoding of phonetic categorization ability and L2 vocabulary size at two different stages of the L2 learning process. Two groups of German learners of English differing in L2 proficiency (30 advanced vs. 49 intermediate learners) participated in an English lexical decision task including nonwords with /ɛ/ and /æ/ (e.g., *l[æ]mon, *dr[ɛ]gon), an /ɛ/-/æ/ phonetic categorization task and an English vocabulary test. Results showed that the effects of categorization and vocabulary on lexical decision performance were modulated by proficiency: categorization predicted /ɛ/-/æ/ nonword rejection accuracy for the intermediate learner group only, whereas vocabulary predicted nonword rejection accuracy for the advanced learners, who outperformed the intermediate group in all three tasks. This suggests that sufficient phonetic identification ability is key for an accurate representation of difficult L2 sounds in the lexicon, but, for learners for whom robust phonetic identification is already in place, their ultimate success is tightly linked to their vocabulary size in the L2.

ABSTRACT. The interconnectedness of phonological categories between the two languages of early bilinguals has previously been explored using single-probe data [1-3]. We tap into bilingual phonological representations differently, namely via monitoring phonetic drifts due to changes in language exposure [4]. Data revealing phonetic drift in simultaneous or early-sequential bilinguals proficient in both languages is not available to date. We report a case study of two teenage English-Czech simultaneous bilinguals who live in Canada and spend summers in Czechia. Voice onset time (VOT) of word-initial voiced and voiceless stops was measured upon the bilinguals’ arrival to and before their departure from a two-month stay in Czechia. Each bilingual read the same set of 71 Czech and 58 English stop-initial target words at each time of measurement. Linear mixed-effects models assessed the effects of target language, measurement time, and underlying voicing. After immersion in the Czech-speaking environment, for both speakers the count of voiced stops realized with negative VOT increased and the measured VOT of voiced stops (appearing different for English and Czech initially) drifted towards more negative (Czech-like) values in both languages, while no change was detected for the voiceless stops of either English (aspirated) or Czech (unaspirated). The results suggest the bilinguals maintain three-way VOT distinctions, differentiating voiceless aspirated (English), voiceless unaspirated (Czech), and voiced (English~Czech) stops, with connected bilingual representations of the voiced categories. We show that observing phonetic shifts due to changes in the ambient linguistic environment is revealing about the organization of mental phonological space in simultaneous bilinguals.

ABSTRACT. When speaking naturally, native speakers shorten and merge words. For instance, German native speakers may say “hama” for “haben wir” (‘have we’). Such reductions challenge second-language learners (e.g., Dutch learners of German), who often do not recognize or mis-recognize words (e.g., Hammer ‘hammer’). We examined to what extent reductions are problematic for adolescent learners of a second language, after four years of high-school training; and whether the problems can be related to inadequate bottom-up and top-down processing.

For this, 39 Dutch and 38 German adolescents heard either reduced or unreduced German phrases (phrase-intelligibility-task) and words (lexical decision), representing well-known vocabulary. Learners presented with reduced speech had dramatically lower accuracies (both tasks) and intelligibility judgments, and made more errors of several types (phrase-intelligibility-task). The results indicate that (1) Reduced pronunciations in a second language still impair high-school students’ perception near the end of high-school training; (2) This seems due to at least inadequate bottom-up processing of acoustic-phonetic cues; (3) Experience with natural speech varying in reductions is key for strengthening linguistic representations (theory) and improving comprehension of real-life speech (practice). Outcomes will be discussed in the context of practices in classrooms, as observed in a concurrent, nation-wide questionnaire among Dutch teachers of German.

ABSTRACT. Background. More than a million people asked for asylum in Europe in 2015, half of them children. Learning the Host Country Language (HCL) is critical, and even though children benefit from school immersion in the HCL, a subset of them suffer from mental health vulnerability [1], with consequences on cognition, including language [2]. It is therefore crucial to develop diagnostic tools teasing apart language difficulties due to mental health issues from those due to a Developmental Language Disorder or to the lack of exposition to the HCL [3,4]. Although reliable language tasks have been designed for sequential bilinguals (Litmus [5,6,7]), their validity for refugee children is unknown. This study explores the link between performance to those tasks, emotional well-being and executive functions.

Methods. Data from 77 children aged 5 to 8 were collected (20 MOnolingual, 46 BIlingual and 11 BIlingual REFugees) on the following tests: Litmus tasks, Standardized language tasks, Executive functions, Emotional well-being.

Results. Better performance is found for MO compared to BI as well as for BI compared to BI-REF on all standardized language tasks and the Litmus sentence repetition task. BI-REF show higher emotional vulnerability, and marginally weaker short-term memory. Emotional well-being and Short-term memory predict performance to the sentence repetition task.

Conclusion. Results suggest that Litmus tasks are sensitive to variables on which refugee children show vulnerability, i.e., emotional vulnerability and short-term memory. It is therefore crucial to conduct further research to build appropriate tools to assess language in that specific population to avoid DLD over-diagnostic.

ABSTRACT. Translation equivalents that share word forms across languages (e.g., “water” in English and Dutch), so-called cognates, are known to be more easily processed than those with different word forms (i.e., non-cognates). However, since typical cognates in alphabetic scripts share forms at both orthographic and phonological levels, it remains open whether such processing advantage for cognates arises from orthographic or phonological stages of bilingual word recognition. In languages using logographic scripts, the orthography is not necessarily associated with the phonology. Therefore, we addressed the question using a cross-language priming paradigm in Japanese with Chinese characters. Twenty-six Chinese-Japanese bilinguals made semantic judgments about target words in L2 preceded by masked prime words in L1 (33 ms or 50 ms in duration). The primes and the targets were Identical Cognates (IC), orthographically Non-Identical Cognates (NIC; e.g., “价值”/“価値”), Non-Cognates (NC), or Un-Related words (UR). Results showed that for 33 ms primes, though the reaction time (RT) of IC was equal to NIC, participants responded more quickly to IC/NIC than to NC/UR, indicating that similar features between languages could lead to a robust cognate priming effect. For 50 ms primes, RTIC was shorter than RTNIC, RTNC, and RTUR. RTNC was shorter than RTUR but equal to RTNIC. These RT results showed two different effects of priming, both independent of phonology. First, the observed difference between IC and NIC indicates a robust orthographic priming effect. Second, the difference between NIC/NC and UR suggests a distinct component of lexico-semantic priming.

ABSTRACT. Adult bilingualism research provides considerable evidence for language co-activation in the form of between-language lexical priming (e.g., Dong et al., 2005; Kroll et al., 2006), whereas research on such effects in bilingual children is limited and largely restricted to toddlers (Von Holzen & Mani, 2012; Singh, 2014). Individual-level factors, known to influence co-activation in adults (e.g., Van Hell & Tanner, 2012), have not yet been systematically investigated in children. The current study aims to uncover to what extent and at which levels of lexical representation between-language priming takes place in school-aged Greek-Dutch simultaneous bilinguals, in interaction with individual differences in proficiency, exposure and use.

Children matched pictures to auditorily presented Greek and Dutch prime and target words, while their eye movements were recorded. Following Von Holzen & Mani (2012) and Singh (2014), these words overlapped phonologically (e.g., roda(GR) ‘wheel’–rok(NL) ‘skirt’), semantically (fousta(GR) ‘skirt’–rok(NL)), or phonologically mediated by semantics (vrachos(GR) ‘rock’–rots(NL) ‘rock’–rok(NL)). Greek and Dutch proficiency, exposure and use were assessed through offline tasks and parental questionnaires. We predicted priming in all conditions, with stronger effects for highly proficient bilinguals.

Reaction time analyses reveal between-language phonological and semantic priming, and interactions with proficiency, exposure and use of both languages, most notably in the mediated condition. For eye-tracking, we expect to find effects in all conditions, given the direct, sensitive nature of this method. These results provide new evidence for language co-activation in bilingual children, and demonstrate which sources of variation between bilingual children constrain their lexical processing.

ABSTRACT. Research has long debated the role of L1 vs. L2 input exposure on the linguistic outcome of bilingual children [1,2]. While several studies suggest that L2 input directly affects language and literacy skills in L2 [3], according to other studies exposure to mothers’ “higher level” conversation in L1 significantly predicted L2 vocabulary knowledge [4]. In the current study, we investigated the role that L1 use plays on one aspect of L2 literacy development on a sample of L2 Italian primary students. Thirty-two L2 children born in Italy (Mean age = 9.33, sd = 1.11), whose L1 was Albanian or Spanish, were tested on a series of reading tasks (words, nonwords and passage reading). The time of home exposure to L1 was calculated by means of an ad-hoc created round clock that visually represented the hours of the day and that children had to fill-in accordingly. Three different clocks were administered to each student to quantify: i) time spent speaking L1 with their mother; ii) time spent speaking L1 with their father; iii) general passive listening of L1. Results showed a significant interaction effect between age and L1 spoken with both parents in lexical reading speed: the more the time spent speaking L1, the lower the reading speed of both word-lists and passage reading. No effects emerged for general passive listening of L1. In conclusion, results suggest that the amount of time spent by children in speaking the L1 affects reading speed in L2.

ABSTRACT. We present two projects on talker-specific adaptation and cross-talker generalization in the perception of L2 speech. Both projects highlight that findings in this line of research depend on the specific exposure & test talkers. Project 1 (Xie, Liu, & Jaeger, 2021-JEP:G) addresses a design confound in Bradlow & Bent’s seminal 2008 study (BB08). BB08 found that even brief exposure to L2 speech can elicit improved recognition that generalizes to an unfamiliar talker of the same accent (N=70). Cross-talker generalization was, however, only observed after exposure to multiple talkers of the accent, not after exposure to a single accented talker. This contrast between single- and multi-talker exposure has been highly influential, suggesting a critical role of exposure variability in learning and generalization. Our two replications (N=320 each) revisit this finding for 20 unique combinations of moderately proficient exposure and test talkers. Like BB08, we find robust evidence for cross-talker generalization after multi-talker exposure. Unlike BB08, we also find evidence for generalization after single-talker exposure. The degree of generalization depends on the specific combination of exposure and test talkers. Project 2 (Xie & Jaeger, in prep) compared talker-specific adaptation across two different moderately proficient L2 talkers and two different paradigms (transcription and 2AFC). We find no difference across tasks, but exposure effects differed across talkers. Both projects suggest that some conflicts in the literature are the result of the specific speech materials used, rather than meaningful differences between studies.

ABSTRACT. Previous research showed that native (L1) and non-native (L2) listeners frequently generate false memories (they may “remember” information that was never studied, such as unstudied words in word lists or unstudied facts in narratives or videos). However, little is known about how listeners process different types of speech, such as how processing of perceptually clear child speech differs from processing of adult speech. This study compared online comprehension and memory for adult and child speech in L1and L2. L1 and L2 listeners studied pictures of events presented with simple sentences (e.g., The dog chased the postman); half of the sentences were recorded by an L1 adult and half by an L1 6-year old child. Participants then received a recognition test that included sentences with original wording and sentences with changes in meaning (unstudied inferences, i.e., false memories: The dog bit the postman). False alarm rates to sentences expressing unstudied inferences (i.e., false memories) were similar in L1 and L2, but the two participant groups differed in their hit rates for studied sentences. L1 listeners showed poorer memory for child speech than L2 listeners, while L2 memory was not influenced by the source of the input and was comparable to L1 memory for adult speech. Thus, L1 listeners may pay less attention to and may represent child speech in less detail than L2 listeners, who may instead consistently devote high levels of resources to processing cognitively demanding L2 input irrespective of its source.

ABSTRACT. Numerous studies have indicated that exposure to two or more languages from early on in life positively affects the metalinguistic abilities in bilingual and multilingual children (Bialystok 2001; Ricciardelli 1992) and some studies even suggest that bilingual immersion education can enhance metalinguistic development in second language learners, i.e. monolingual children who attend bilingual immersion schools (Bialystok, Peets & Moreno, 2012). In these studies, it is proposed that the early confrontation with and contrast between two or more formal representation systems in these individuals growing up with more than one language leads to these results (Hofer & Jessner 2019). In many of these studies, the focus has been on languages from two different language families (e.g. English-French, German-Italian) and most of the studies conducted in school contexts have, so far, not investigated the role that explicit teaching in language awareness plays a role. The proposed research project that I would like to discuss in this presentation, intends to fill these gaps by looking at early and late Dutch-German bilinguals, whose languages are closely related and who receive explicit instruction in contrastive language awareness in the first two year of schooling at a dual bilingual immersion school in the Dutch-German border-zone. I would like to present and discuss an experimental battery that is sensitive to the typological proximity of the two languages and combines online and offline-tasks to shed light on the effects of explicit instruction on their implicit awareness.

ABSTRACT. Are internal or external factors more important to young English language learners’ (YELLs’) learning of English as a foreign language? According to previous studies internal factors are more important than external ones for YELLs’ English vocabulary knowledge (Paradis, 2011; Sun, Steinkrauss, Tendeiro & de Bot, 2016). These studies have measured vocabulary knowledge at one moment in time, but did not investigate whether the contribution of internal and external factors to English vocabulary knowledge changes over time. In our longitudinal study we assessed Dutch YELLs (N = 250) receptive vocabulary knowledge with the Peabody Picture Vocabulary Test in grade 4, grade 5 and grade 6 (from age 9 to age 12). The measured internal factors were age, gender, abstract reasoning, English phonological awareness, English phonological short-term and verbal working memory, Dutch vocabulary knowledge, motivation and linguistic self-confidence. The external factors were experience with formal English education, home language(s) and extramural English exposure. Latent growth curve models will be computed to analyze to longitudinal contribution of the internal and external factors to the students receptive English vocabulary knowledge. The results will be presented and will show whether the role of internal and external factors changes over time for YELLs learning English as a foreign language.

ABSTRACT. Speakers of Russian Sign language (RSL) can read and write in Russian, but use RSL for everyday communication, which makes them cross-modal bilinguals. Unlike other bilinguals, their reading skills are not supported by phonological encoding, which leads to poor reading in many deaf signers (Bélanger, 2013). At the same time, deaf readers have certain advantages that can compensate for the lack of phonological support: highly developed peripheral vision allows them to discern 18 characters to the right of the current fixation while hearing readers discern only 14 characters (Bélanger, 2015). To further investigate how this unique combination of advantages and disadvantages affects reading, we compared eye movements of 18 native speakers of Russian Sign Language and 96 hearing Russian speakers reading 144 sentences of Russian Sentence Corpus (Laurinavichyute et al., 2019). In line with their unique status, RSL speakers resemble both proficient readers (first fixation durations, skipping rates, and question response accuracies similar to those of the control group) and poor readers (saccade landing position closer to the beginning of the word, longer total reading times, greater number of fixations on a word, greater sensitivity to word frequency, length, and predictability). In addition, RSL speakers speed up more than the control group closer to the end of the sentence. This speedup, as well as control-like first fixation durations and skipping rates, is likely to stem from greater parafoveal preview. At the same time, the characteristics that resemble those of poor readers are consistent with general L2 reading patterns.

ABSTRACT. This study tests whether (1)exposure to native pronunciation and (2)feedback on the speaker’s pronunciation lead to the learning of L2 sounds in interaction. Acoustic analyses[1] have shown that these factors lead to an improvement on production of the English /æ-ɛ/ contrast, which has been reported to be difficult for Dutch speakers[2]. Here we assess whether these improvements are noticeable to the native ear. Forty-eight Dutch speakers engaged in a puzzle-solving game with a confederate, during which exposure (presence/absence) to and feedback (presence/absence) on the critical sounds were fully crossed. Exposure was controlled by covertly implementing prerecorded speech for the confederate with the Ventriloquist paradigm[3]. Feedback was negative and implicit: when participants instructed the confederate to click on an /æ/-word, e.g. bad, they saw the confederate systematically clicking on the paired /ɛ/-word, bed. Immediately before and after the interaction they performed a production task involving the two sounds. Native speakers of English (n=107), first, identified the recorded words (e.g. auditory presentation of bad, visual response options: bad-bed). In a second task, they rated the nativelikeness of the words using a five-point scale. Listeners’ accuracy at identifying /æ/-words improved significantly in the post-test only when the Dutch speakers had received exposure and/or feedback. The nativelikeness ratings show an overall improvement from pre-test to post-test but no effect of exposure or feedback. In conclusion, exposure and feedback in interaction benefit L2 sound learning, especially in combination. The improvement on production, which was found with acoustic measurements[1] thus also impacts native listeners’ identification accuracy.

ABSTRACT. Orthography is omnipresent in instructed L2 learning, and an increasing research pool investigates its effect on speech production and perception [1]. While orthography facilitates word learning and phonological accuracy [2], it may be detrimental on phonetic aspects of speech [3-4]. This study investigates if the presence and type of orthography affect speech sound learning in production and perception. Three groups of L1-Spanish speakers (N=70) were exposed to the unfamiliar vowels /y/ and /ɛ/ over three days. The Conflicting-Orthography group was exposed to the vowels in presence of the L1-graphemes <u> (=L1 /u/) and <e> (= L1 /e/). The Novel-Orthography group was exposed to the vowels combined with grapheme-like symbols. The Auditory-only group did not receive any visual input. The Conflicting-Orthography group produced lower F2 for /y/ than the Auditory-only group (p=0.0255), with the Novel-Orthography group falling in between. In perception, the Conflicting-Orthography group categorized tokens with lower F2 as /y/ than the other two groups (p=0.002). The groups neither differed on their production nor perception of /ɛ/. The results suggest that the presence of <u>, corresponding to L1-/u/, interfered with participants’ learning of /y/ in production and perception, while orthography did not influence learning outcomes for /ɛ/. Inconsistent grapheme-phoneme mappings between L1 and L2 can therefore modulate acquisition outcomes in the earliest stages of L2-learning. Moreover, there is no one-fits-all effect of orthography on speech production and perception, as only /y/, but not /ɛ/, was affected. These findings have important implications for models of L2 speech production and perception.

ABSTRACT. Robust speech communication requires that listeners learn linguistically-relevant representations for stable language regularities, such as the speech sounds (phonemes) that convey meaning. Yet, second language (L2) acquisition is especially challenging among adult language learners, for whom learning typically involves explicit classroom instruction. In this study we explore whether incidental category learning of L2 speech sounds, in which learners’ attention is directed away from to-be-learned categories by an engaging videogame, taps into category learning systems distinct from those engaged in more explicit learning. We test whether incidental learning of L2 speech categories may provide a ‘back door’ through which to influence adult L2 learners’ speech acquisition in terms of two difficult to acquire L2 speech sounds: Mandarin Chinese tones and Tokyo Japanese pitch accents. Preliminary results from adults engaged in structured L2 classroom learning (Chinese), as well as monolingual English adults (Japanese), demonstrate that auditory categories sharing common perceptual dimensions with second language categories scaffold L2 acquisition in ways beneficial to learning. The L2 Chinese learners demonstrated robust tone discrimination, categorization, and production improvements relative to typical classroom learners who did not undergo incidental learning. The monolingual English participants showed an average increase in pitch accent categorization but not a significant improvement, suggesting that additional training time and/or structured L2 learning may be necessary to see a larger change in behavior. Taken together, our approach demonstrates an effective and engaging alternative to explicit L2 speech learning of difficult to acquire speech sounds.

ABSTRACT. Late L2 speakers’ difficulties in L2 speech sound production, known as accents, can be remediated by training, where participants receive feedback on their production. Although this type of training appears to be effective in a short term (averaged 20% improvement after training), it remains unknown whether improvements remain in a long term. The current study addressed the durability of the production training-related gains and examined the role of speaker-variability in modulating the effect. Forty-three Spanish speakers with no experience in French were trained to produce a French vowel contrast /e-ɛ/ that Spanish listeners tend to assimilate to the Spanish /e/. On each trial, participants heard one target, repeated it and received immediate visual feedback representing their production, in F1/F2 space, along with the target. Participants were randomly assigned to a multiple-speaker (n=22) or a single-speaker (n=21) training condition. There were 1050 trials per vowel, distributed over three training days. Improvements in vowel production were assessed in a vowel repetition task, after the training and two weeks later, using two talker sets: familiar and unfamiliar; the latter assessed generalization of learning. Our results revealed that both training conditions improved non-native production and led to generalization. Yet, two weeks later, only multiple-speaker group maintained production gains in both talker sets; single-speaker group showed a deterioration in the production after an unfamiliar speaker. Our results suggest that training-related improvements remain two weeks after training, when tested with a familiar speaker; yet, only speaker variability leads to an establishment of long-lasting abstract sound representations.

ABSTRACT. Native listeners often experience processing difficulty when listening to foreign-accented speech. With increasing exposure, this difficulty typically diminishes, sometimes within minutes (Bradlow & Bent, 2008; Clarke & Garrett, 2004). One possible mechanism underlying such accent adaptation is distributional learning of talker- or accent-specific mappings from phonetic cues to phonological categories.

Distributional learning has been shown to provide a good model of potentially related, but less complex, adaptive changes in native speech, such as perceptual recalibration (Kleinschmidt & Jaeger, 2015) and related paradigms (Clayards et al., 2008; Munson, 2011). However, these phenomena involve unexpected speech with comparatively simple one-dimensional cue shifts. We demonstrate that the same Bayesian ideal observer (IO) applied to native speech can predict adaptation to foreign-accented speech that differs from native speech along multiple cue dimensions.

We analyze two experiments on native listener perception of Mandarin-accented English and Flemish-accented Swedish. Whereas the L2s have syllable-final stop voicing (e.g., bid vs. bit), neither of the L1s do, leading to non-native realization of syllable-final voiced stops. The two experiments used the same exposure-test paradigm previously found to yield accent adaptation (e.g., Eisner et al., 2013). Facilitating effects of exposure were, however, only found for Mandarin-accented speech.

We show that these seemingly conflicting outcomes are, in fact, predicted by the IO model, which provides a zero degrees-of-freedom link between the cue distributions of categories in (1) native speech (reflecting listeners’ a priori expectations about cue-to-category mappings) and (2) non-native speech during exposure to predict listeners’ categorization decisions during test.

ABSTRACT. Linguistic and non-linguistic prediction have been investigated separately, following neuropsychological evidence that language is primarily processed in the left hemisphere [1]. Recent neuroimaging research shows co-activation of domain-specific regions and domain-general regions (e.g., associated with attention, non-linguistic auditory perception, and memory) during language processing [2]. Yet, many aspects of this relationship remain vague, particularly in bilinguals. To further investigate the relationship between language and other cognitive skills, Spanish monolinguals and advanced English learners of Spanish completed four auditory tasks. To assess linguistic prediction, participants used lexical stress cues to anticipate word endings, in a visual world paradigm eye-tracking task [3]. To measure non-linguistic prediction, participants used rhythmic [4] and tonal [5] regularities to form rhythmic and tonal pattern expectations. To test working memory span, participants completed the Operation Span task [6]. Results show that all participants used stress to predict word suffixes, but natives were faster than learners, independently of the learners’ working memory, in line with [2]. Also, natives were better at predicting tones and learners at predicting rhythm, because Hispanophones rely more on suprasegmental cues (pitch, duration, intensity) whereas Anglophones rely more on segmental cues (vowel quality). Finally, higher performance on the tonal task correlated with better use of lexical stress to predict word endings, showing that prediction abilities may transfer between linguistic and non-linguistic domains, and expanding prior research showing transfer of prediction skills within two linguistic domains [7]. These findings support domain-general learning theories and suggest that linguistic and non-linguistic prediction develop concurrently.

ABSTRACT. Bilingualism has been shown to modulate the neural mechanisms of selective attention, with differences between monolingual and bilingual adults observed despite equivalent behavioural performance. This suggests that bilingualism triggers neuroplastic adaptation that cannot be captured by behavioural tests alone. However, it is unclear if bilingual children’s attention mechanisms are similarly affected. One possibility is that bilingualism reconfigures the patterns of attentional processes right from the onset, such that the effects are discernible in childhood. Alternatively, these modifications might have a protracted maturation, and be more evident in adults than in children.

To address this question, we tracked the neural encoding of attended continuous speech in monolingual and bilingual children aged 7-12. Participants attended to a narrative in English while different types of interference were presented to the unattended ear: English, Latin, non-linguistic interference, and no interference. The EEG data was cross-correlated with the attended and unattended speech envelopes. Results showed more robust neural encoding for the attended envelopes than ignored ones for both groups. The type of interference significantly modulated the encoding of attended speech for monolinguals. In contrast, bilingual children displayed equivalent encoding of the attended stream across all conditions, directly replicating the results previously observed in adults. These findings indicate that, even when behavioural responses are equivalent, monolingual and bilingual children exhibit different patterns of neural entrainment to attended speech. This suggests that the neural mechanisms of selective attention in bilinguals are already reconfigured by the time children can reliably respond to selective attention tasks.

ABSTRACT. The challenge of L2 speech communication has a strongly temporal “flavor.” This presentation will examine the temporal domain of L2 speech at various levels. First, a salient difference between L2 and L1 speech is tempo. L2 speaking rate (syllables/second) is slower across L1 and L2 talkers of a given language as well as across languages within bilingual individuals [e.g. 1, 2]. Moreover, slower L2 than L1 speaking rates have been demonstrated across languages, including Spanish, French, Dutch, and German [e.g. 3-5]. In a corpus of read and spontaneous speech in both the L1 and L2 of bilingual individuals (n=86) from various language backgrounds (n=10) we found that while L2 speech was always slower than L1 speech, L1 speaking rate significantly predicted L2 speaking rate. That is, relatively fast or slow L2 talkers were also relatively fast or slow L1 talkers, respectively. Thus, the influence of L1 versus L2 mode on speech tempo is modulated by a language-general, talker-specific “trait” characteristic [6, see also 4 and 7].

Next, to gain insight into the communicative impact of the slow tempo of L2 speech, we examined both speaking rate (syll/sec) and information density (# syllables for a given text) across L1 and L2 recordings of a standard passage in three languages, English, French, and Spanish. Across languages, L2 speech involved syllables that were both longer in duration and greater in number for a given text. Examination of syllable reduction rates showed that the greater number of syllables in L2 versus L1 speech was due to less syllable reduction resulting in lower information density (less information conveyed/syll). Thus, slow L2 speech rate combines with low L2 information density to yield a very low L2 information transmission rate for L2 speech (i.e., less information conveyed/sec). Overall, this cross-language comparison establishes low information transmission rate as a language-general, distinguishing feature of L2 speech.

Finally, to assess the impact of L1 versus L2 mode in a task with communicative intent, we compared task-completion time across various combinations of L1 and L2 interlocutors in a cooperative, picture-matching task, the diapix task. While all pairs successfully completed the task, L2 pairs were less efficient than L1 pairs in both task-completion time and number of word repetitions (type-to-token ratio). However, this difference was mitigated when L2 partners shared L1, and the communicative efficiency differences between L1 and L2 pairs diminished slightly across successive diapix trials [8].

Together, these studies demonstrate accumulation of time-related influences of the L2 speech mode. At the phonetic level, L2 speech is distinguished from L1 speech by a slow tempo combined with less syllable reduction yielding a speech signal with a low information rate (few bits of information conveyed/sec for a given text/meaning). While a direct link between information rate at the phonetic level and communicative efficiency at the discourse level remains elusive and is likely compounded by lexical, syntactic, and other aspects of L2 speech production and perception, it seems clear that the L2 speech mode influences speech communication at multiple interacting time scales.

1. S. Guion, J. Flege, S. H. Liu, & G. H. Yeni-Komshian. (2000). Age of learning effects on the duration of sentences produced in a second language. Appl. Psycholing., 21:2, 205-228.

2. M. Baese-Berk & T. Morrill. (2015). Speaking rate consistency in native and non-native speakers of English. JASA, 138:3, EL223-EL228.

3. M. L. García Lecumberri, M. Cooke, & M. Wester. (2017). A bi-directional task-based corpus of learners’ conversational speech. International Journal of Learner Corpus Research, 3:2, 175-195.

4. N. H. De Jong, R. Groenhout, R. Schoonen, & J. H. Hulstijn. (2013). Second language fluency: speaking style or proficiency? Correcting measures of second language fluency for first language behavior. Appl. Psycholing., 34.

5. J. Trouvain & B. Möbius. (2014). Sources of variation of articulation rate in native and non-native speech: comparisons of French and German.” Proc. Speech Prosody (SP7), 275-279.

6. A. R. Bradlow, M. Kim, & M. Blasingame. (2017). Language-independent talker-specificity in first-language and second-language speech production by bilingual talkers: L1 speaking rate predicts L2 speaking rate,” JASA, 141:2, 886–899.

7. T. M. Derwing, M. J. Munro, R. I. Thomson, & M. J. Rossiter. (2009). The relationship between L1 fluency and L2 fluency development.” Studies in Second Language Acquisition, 31, 533–557.

8. K. J. Van Engen, M. Baese-Berk, R. E. Baker, A. Choi, M. Kim, & A. R. Bradlow. (2010). The Wildcat Corpus of Native- and Foreign-Accented English: Communicative efficiency across conversational dyads with varying language alignment profiles,” Language & Speech, 53:4, 510-540.