Bramantyo Supriyatno: Computational Modeling of Language Timescales

BCCN Berlin / Technische Universität Berlin

Abstract

Half of the world's population speaks more than one language, raising important questions about how different languages are processed in the brain. While research has shown hierarchical organization across the cortical surface for English, it remains unclear whether the hierarchical processing of the native language differs from that of the second language. This study addresses this issue by analyzing functional magnetic resonance imaging (fMRI) responses from bilingual participants whose native language is Chinese and second language is English as they read the same narratives in both languages. We utilized a multilingual language model to derive language timescales, with each timescale representing a different hierarchical level—words, sentences, and paragraphs. Voxelwise encoding models were then employed to identify brain regions associated with these timescales. Our analysis reveals significant shifts in processing timescales across various brain regions, with faster timescales observed for Chinese compared to English in the same areas. These shifts tend to occur in regions associated with language processing and emotional or relational semantics. These findings suggest that processing a second language may require longer linguistic timescales than a native language, highlighting the cognitive demands of bilingualism.

 

Additional information:

Master thesis defense

 

Organized by:

Prof. Dr. Fatma Deniz & Prof. Dr. Klaus Obermayer