Brain alignment of language models differs during speaking and listening
What Shapes Brain Alignment of Large Language Models during Dialogue? Insights from Language Production and Comprehension
Computation and LanguageMachine Learning
Summary
Understanding how large language models relate to the human brain can help improve AI communication. This study looked at brain activity when people both speak and listen during conversations. The researchers found that the way language models match brain activity changes depending on whether a person is producing or comprehending language. Different brain areas and language model layers show unique patterns linked to these two processes. They also discovered that certain linguistic and communication features affect this alignment differently in speaking versus listening.
What this means in practice
- •For dialogue system developers: Improve conversational AI by tailoring models to align with brain patterns during both speech generation and understanding.
- •For neurotechnology engineers: Design brain-computer interfaces that better interpret user intentions by considering distinct brain alignment patterns during speaking and listening.
Authors
Yuko Nakagi, Yuya Chiba, Yasue Kishino
Abstract
Brain alignment of large language models (LLMs) offers a way to investigate representational similarities between LLMs and the human brain during language processing. While previous studies have mainly focused on passive language comprehension, little is known about such alignment during dialogue, in which speakers engage in both language production and comprehension. In this study, using fMRI data collected during dialogue, we examine brain alignment during language production and comprehension across LLM layers and brain regions. We assess its association with the intrinsic dimension (ID) of LLM representations and the contributions of several interpretable linguistic and communicative features. Our results reveal distinct patterns of brain alignment between production and comprehension, as well as strong associations with ID in brain regions associated with each processing mode. Furthermore, we show that the contributions of linguistic and communicative features also vary across processing modes, LLM layers, and brain regions.