Jing Cai, PhD

What happens inside a single brain cell the moment before you speak? Every sentence we produce is the result of neurons firing in patterns—yet how individual cells capture the building blocks of language, from grammar and meaning to sentence structure, remains unknown.

A new study from the Mass General Brigham Neuroscience Institute, led by Jing Cai, PhD, and supervised by co-corresponding authors Sydney S. Cash, MD, PhD, and Ziv M. Williams, MD, addressed this gap by studying patients with epilepsy undergoing planned surgical monitoring, allowing single-neuron recording in an awake, speaking brain.

Sydney Cash, MD, PhD

Ziv Williams, MD

Using ultra-fine electrode arrays, they recorded the activity of 579 individual neurons across the frontotemporal cortex, in parts of the brain’s language network, while participants constructed sentences.

Using AI language tools to label the grammatical and contextual properties of each spoken word, the team found that different neurons captured different features of language: some tracked parts of speech, others reflected how words group into phrases or fit within a sentence’s structure.

These signals emerged up to one second before a word was spoken, indicating the brain plans language ahead of articulation. Language-capturing neurons were broadly distributed, but the strongest responses were concentrated in the left hemisphere. The team also found that human neurons and large AI language models process language in similar ways.

These findings provide a cellular map of how the human brain produces speech and may ultimately inform efforts to restore communication in individuals with stroke or amyotrophic lateral sclerosis (ALS).

Published in Nature on June 17, 2026 | Read the paper: “Mapping the neuronal building blocks of human language with language models”

Summary reviewed by: Ziv M. Williams, MD, senior author