Natural language processing models reveal neural dynamics of human conversation

Area of Science:

• Neuroscience
• Computational Linguistics
• Cognitive Science

Background:

• Human conversation relies on complex, alternating speech production and comprehension.
• The underlying neural mechanisms for these processes are not fully understood.

Purpose of the Study:

• To identify neural signals reflecting speech production, comprehension, and transitions during natural conversation.
• To explore the role of deep learning in analyzing neural data for language processing.

Main Methods:

• Utilized pre-trained deep learning natural language processing (NLP) models.
• Combined NLP models with intracranial neuronal recordings from individuals during conversation.

Main Results:

• Neural activity for speech production and comprehension is broadly distributed in frontotemporal areas across multiple frequency bands.
• These neural activities are specific to conveyed words and sentences, depending on context and order.
• Neural patterns partially overlap between production and comprehension, with specific activity changes during listener-speaker transitions.

Conclusions:

• Neural activity during conversation exhibits a dynamic organization supporting both language production and comprehension.
• Deep learning models are effective tools for uncovering neural mechanisms of human language processing.