FBS Colloquia No.391Perceptual and Cognitive Neuroscience Laboratory

Exploring neural representation of conversational content using large language models

Conversation is a fundamental aspect of daily human behavior, underpinned by multidimensional information processing encompassing motor control, perception, and both linguistic and social cognition. Despite its ubiquity, the neural mechanisms underlying conversational behavior remain poorly understood. In recent years, large language models (LLMs) have garnered attention not only as tools for information processing but also as promising frameworks for probing the neural representation of language. In this study, we recorded brain activity using functional magnetic resonance imaging (fMRI) during natural conversation, and utilized LLMs to analyze how produced speech content and perceived speech content is represented in the brain. I will present our findings, as well as relevant prior work.