Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural signatures of non-native video learning: proficiency-specific strategies revealed by neural coupling and encoding
Poster Session E, Sunday, September 14, 11:00 am - 12:30 pm, Field House
This poster is part of the Sandbox Series.
Yingying PENG1, Samuel A. Nastase2, Yuxi Li1, Yuhan Huang1, Ping Li1; 1The Hong Kong Polytechnic University, 2Princeton Neuroscience Institute, Princeton University
Higher education has seen exponential growth in international students learning in a non-native language. However, current neurocognitive research on bilingual processing has overwhelmingly focused on isolated words or sentences, leaving the mechanisms supporting non-native comprehension of complex, real-world content largely unexplored. In this study, we examine how learners with varying English proficiency levels process and comprehend naturalistic, college-level educational video. Participants included native English speakers (n = 16), high-proficiency (n = 17), and medium-proficiency non-native speakers (n = 19). All participants viewed a ~12-minute English video lecture during fMRI acquisition. The instructor was also scanned while delivering the lecture, serving as a neural model of content delivery. Intersubject correlation (ISC) analyses were performed (a) within the three learner groups and (b) between learners and the instructor, focusing on three networks implicated in video learning: the language network, the multiple demand (MD) network, and the default mode network (DMN). Support vector machines (SVMs) classified learners by proficiency based on ISC patterns. We also employed neural encoding models using GPT-2 embeddings as a proxy for idealized linguistic representations to examine how well learners’ brain responses aligned with the lecture content. Behaviorally, native and high-proficiency groups showed comparable performance and outperformed the medium group, who also reported more unknown words, which further negatively predicted learning. fMRI results revealed that ISC in semantic areas of the language network was highest in native learners and lowest in the medium group, with ISC negatively associated with unknown word count. The MD network showed higher ISC in non-native groups, with neural synchrony in the high-proficiency group associated with learning outcomes. In the DMN, ISC was highest in the native group and also predicted learning. Classification based on ISC features achieved 81% accuracy for both the language and MD networks, and 62% for the DMN, indicating robust group-specific neural signatures of engagement. Teacher-student ISC analyses revealed that delayed coupling in the right IFG and angular gyrus predicted learning in the native group. In contrast, for medium-proficiency learners, delayed coupling in the left IFG and posterior temporal lobe correlated negatively with unknown words and positively with learning. Neural encoding analysis indicated that, in native learners, prediction accuracy in the left IFGorb and MFG positively predicted learning, whereas in both non-native groups, accuracy in the left ATL predicted learning. Together, these findings reveal distinct neural strategies for video learning across proficiency levels. Native speakers exhibited more homogeneous semantic processing, with learning supported by teacher-student coupling in the right hemisphere and by neural tracking of linguistic structure in left frontal regions, suggesting a balanced reliance on syntactic, semantic, and integrative cues. Medium-proficiency learners showed reduced synchrony in semantic areas but relied more on left-lateralized semantic processing, particularly with limited lexical knowledge. High-proficiency non-native learners exhibited stronger MD engagement, suggesting greater cognitive effort and flexible resource allocation. These results underscore differentiated neural mechanisms that support non-native video learning and highlight the need to tailor multimedia educational strategies to individual learners’ linguistic and cognitive profiles.
Topic Areas: Multilingualism, Computational Approaches