Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural Entrainment Mechanism on Syntactic Structure in Second Language Learners using Neural Representational Similarity Analysis
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
TING-HSIN YEN1,2, Chun-Hsien Hsu2; 1Taiwan International Graduate Program in Interdisciplinary Neuroscience, National Central University and Academia Sinica, Taipei, Taiwan, 2Institute of Cognitive Neuroscience, National Central University, Zhongli, Taiwan
Neural entrainment is the synchronization process where the brain oscillates with the rhythm of an external stimulus. While previous studies have mainly focused on entrainment to the acoustic envelope and onsets in spoken discourse (Giraud & Poeppel, 2012; Goswami, 2018), the roles of lexical class and syntactic structure remain underexplored. The multivariate temporal response function (mTRF), a neural tracking method utilizing linear convolution models to estimate dynamic responses to continuous stimuli (Brodbeck et al., 2023), has shown that neural entrainment is modulated by second language (L2) proficiency (Ihara et al., 2021). Furthermore, representational similarity analysis (RSA) allows researchers to quantify the spatiotemporal patterns of these language representations (Wang et al., 2020). Thus, this study utilizes inter-subject representational similarity analysis (ISRSA) to explore how spatiotemporal neural entrainment to lexical and syntactic features diverges between native speakers and L2 learners. Participants included 33 native English speakers (Natives) from the Alice dataset (Bhattasali et al., 2020) and 26 native Mandarin who learn English as L2 learners. Both groups listened to the English audiobook of Alice’s Adventures in Wonderland Chapter One during EEG recording (61 and 64 scalp electrodes, respectively), followed by a comprehension questionnaire to ensure full task engagement. Additionally, L2 learners completed the Vocabulary Size Test (VST), which was utilized to rank subjects from high to low proficiency within the similarity matrices. TRFs were estimated using five linguistic predictors (Brennan & Hale, 2019): word onset, function word, content word, syntactic surprisal of a tri-gram model (n-gram), and context-free grammar (CFG). Predictor significance was evaluated via cluster-based one-sample t-tests, while spatiotemporal entrainment patterns were examined through normalized inter-subject correlation (ISC) with permutation testing. Combining TRF cluster-based t-tests and ISC reveals profound processing divergences between Native and L2 listeners. For lexical categorization, Native speakers rapidly process content words (0–150 ms, p = .019), triggering a massive, highly synchronized spatial network (median r ~ 0.26). Similarly, word onsets elicit strong, shared spatial tracking (median r ~ 0.22), indicating automated, universal recognition of physical word boundaries. Function words elicit a broader TRF (0–520 ms) and moderate synchronization (median r ~ 0.20). Regarding syntactic surprisal, Natives exhibit immediate frontal TRFs (0–550 ms, p < .001) for both n-gram and CFG, representing sequential and hierarchical structure processing, respectively. However, Natives share highly uniform spatial responses for sequential prediction (median r ~ 0.24), while relying on individualized networks for hierarchical CFG parsing (median r ~ 0.15). Conversely, L2 learners exhibit temporally smeared, delayed TRFs across predictors (e.g., function words: 300–720 ms, p < .001). Crucially, despite being ranked by VST proficiency, the L2 group exhibited near-zero ISC across all predictors. Ultimately, Native fluency relies on an automated, highly synchronized processing for lexical class and syntactic features within the first 250 ms. In contrast, regardless of language proficiency level, the absolute lack of shared spatial similarity in L2 learners indicates that non-native cognitive load forces the use of highly variable, individualized networks rather than a universal L2 processing strategy.
Topic Areas: Computational Approaches, Multilingualism