Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Resting-State Functional Connectivity Reflects Bilingual Language Entropy and Speaking Performance: Evidence from Network-Based, Whole-Brain Data-Driven Predictive, and Pairwise FC Analyses
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
This poster is part of the Sandbox Series.
Yushen ZHOU1, Yingying PENG2, Hanxiang YU3, Sifan ZHANG3, Keyi KANG3, Zhuotong SUN4, Haoyun ZHANG5; 1Master student, University College London, 2PostDoctoral Fellow, The Hong Kong University of Science and Technology, 3PhD student, University of Macau, 4Master student, The University of Edinburgh, 5Assistant Professor, University of Macau
Research on bilingualism-induced brain structure and function changes has disproportionally linked resting-state fMRI (rs-fMRI) connectivity and self-reported L2 proficiency, a subjective and static index of language profile. Yet, the bilingual’s language profile is a multifaceted phenomenon heavily anchored to daily language use, which further impacts language performance. Nonetheless, the neural consequences on the brain functional connectivity of both bilingual experience and performance remain largely unexplored. We ask whether these two underexplored dimensions, indexed by language entropy (the flexibility and density of language use across everyday contexts) and speaking fluency derived from bilingual speech snippets could better predict bilingualism-induced functional connectivity reorganization. We recruited 79 university students, who were native Cantonese speakers learning L2 English. They underwent a resting-state fMRI scanning session and completed the Language History Questionnaire (LHQ3) and a three-minute English monologic narration. From behavioral tests and questionnaire, we derived self-reported speaking proficiency, language entropy, and pause-based fluency measures to index L2 English speakers’ language performance. We then conducted a three-level analysis pipeline to test which measures could better predict brain functional connectivity. First, in a hypothesis-driven ROI-to-ROI analysis, we extracted ROIs from Language (LANG) and Multiple Demand (MD) networks, subcortical (SUB), and SomatoMotor (SomMot) regions. Resting-state functional connectivity (rsFC) was computed between Language ROIs and ROIs in each of the three target networks and further correlated with proficiency, entropy, and number of pauses. Second, we applied connectome-based predictive modeling (CPM) across whole-brain parcellations to test, without a priori regional hypotheses, whether distributed FC patterns could predict the three language indices. Third, we developed a cross-subject FC-RSA to test whether pairwise individuals on either the higher- or lower-end of these three behavioral measures showed more stereotyped (or more concentrated) versus idiosyncratic pairwise FC patterns. The hypothesis-driven analysis revealed distinct network signatures. Higher speaking proficiency was primarily associated with stronger positive LANG-SUB coupling, showing greater integration. Higher entropy showed broader but negative associations across LANG–MD, LANG–SUB, and LANG–SomMot connectivity: higher the speaker’s language entropy is, lower the cross-network FCs are. This pattern suggests that more flexible and diverse language use is associated with greater functional segregation between the language network and control, motor, and subcortical systems. More pauses were negatively associated with LANG–SomMot coupling, that is, the more a speaker tends to pause, the weaker their LANG-SomMot functional coupling. CPM provided converging results that entropy and pauses could be predicted by FC among the language, control, motor, and subcortical regions. Cross-subject FC-RSA further revealed that participants with higher language entropy tends to exhibit more diversified functional connectivity patterns, confirming the earlier one-on-one hypothesis-driven results. Moreover, participant pairs with more pauses were consistently associated with more idiosyncratic LANG–MD and LANG–SomMot FC patterns. Together, these findings suggest rsFC patterns could reflect not only bilinguals’ language proficiency, but also their languages use and further language performance. Critically, language entropy emerged as the most robust language-profile marker across these three-level analyses, highlighting the importance of quantifying dynamic, context-dependent, and multifaceted language use in understanding bilingualism-induced brain changes.
Topic Areas: Multilingualism, Computational Approaches