Speaker：Dr. Jing Cai (蔡静), Harvard Medical School

Time： Feb 14, 10.a.m.

Venue：Room 1113，Wangkezhen Building

Host：Qian Wang

Abstract

Natural conversation relies on producing and understanding language, which involves assembling and deciphering meaningful sentences from streams of words. This is a complicated process, and we still lack fully understanding of how neuronal activity responds to individual word meanings, nor how neurons collectively respond to overall sentence meaning during comprehension or assemble words into sentences during production. To investigate this, we leveraged rare opportunities for single-neuron and intracranial sEEG recordings in humans. I will begin by discussing single-neuron and local field potential (LFP) coding that supports processing the meaning of words during language comprehension and production. We found that most single neurons exhibited selective responses to specific semantic domains, and word meaning could be decoded with high accuracy from both single-neuron activity and LFP. Further, to fully capture the compositional meaning of language in natural conversation, we used state-of-the-art large language models (LLMs) to represent the compositional meaning and linguistic features that conveyed between speakers. We found parallel processing between neural activity and the artificial model, with neural activity reflecting both production and comprehension broadly distributed across frontotemporal regions and multiple frequency bands. Together, our findings reveal a detailed neuronal computation underlying natural language communication, and begin to identify some of the basic components of single and populational neuronal processing in human cognition.

Bio

Dr. Jing Cai is currently an Instructor in Neurosurgery at Harvard Medical School. Her long-term goal is to bridge artificial intelligence models with human neural recordings to explore the neural mechanisms underlying high-level cognitive behaviors. Specifically, her research focuses on natural human communication, leveraging single-neuron recordings and intracranial EEG to gain deeper insights into these complex neural processes. Her representative works have been published in Science, Nature, and other leading journals.

2025-02-10