Single neurons predict grammar and meaning during natural conversation, NIH-led study finds

NIH-backed researchers reported Wednesday that recordings from individual neurons in the human brain, gathered during natural conversation, allowed artificial intelligence models to predict the grammar, meaning and context of what a person was about to say. The work, based on single-cell activity rather than broad brain-region signals, offers a more detailed view of how human speech is assembled in the brain.

The findings were announced June 17 by the National Institutes of Health under the headline, “Researchers discover single-cell brain activity that underlies human speech.” NIH said the National Institute on Deafness and Other Communication Disorders supported the work, which was published in Nature as Jing Cai et al., “Mapping the neuronal building blocks of human language with language models,” DOI: 10.1038/s41586-026-10691-5. The study was led by researchers at Massachusetts General Hospital and drew on microelectrode arrays implanted in eight patients for the separate clinical purpose of epilepsy monitoring.

Why it matters is the scale. Brain research has long linked speech and language to larger regions of the cortex, but this study aimed to show how those functions appear at the level of individual neurons in humans during everyday speech. NIH said that kind of cellular-scale map could help scientists better understand how the brain generates language and could eventually inform speech neuroprostheses — systems designed to translate neural signals into machine-generated speech for people who cannot speak.

According to NIH, the researchers recorded naturally flowing conversations in English while the patients were being monitored. They captured activity from hundreds of neurons in the frontotemporal cortex, a set of brain areas involved in language, and aligned those signals with time-stamped transcripts of the conversations. Using natural language processing and other machine-learning models, the team reported that it could predict grammatical structure, meaning and context from neuronal activity. Signals detected just before participants spoke predicted many properties of the speech that followed, NIH said.

The researchers also reported what NIH described as a “division of labor” among neurons. Some appeared to encode more basic information, including the meaning of specific words and the roles those words played. Others appeared to represent higher-order functions, such as grouping words and phrases into structured sentences. That suggests human language production may be distributed across specialized neural components that work together across multiple levels, from word choice to sentence construction.

“For the first time we’re describing processes not only at the regional but cellular scale that produce speech,” Jing Cai, a Mass General researcher and the paper’s first author, said in the NIH release. “Having identified these fundamental building blocks, we’ve set the table for us to begin answering some really interesting questions.”

The study extends a line of recent work from the same team. NIH said earlier studies in 2024 linked single neurons to speech production and to semantic encoding during comprehension. The new paper pushes that approach further, examining grammar and sentence-level context during spontaneous conversation rather than focusing only on isolated aspects of speaking or understanding. In practical terms, experimental speech-decoding brain-computer interfaces already exist, and findings like these could help move them beyond phonemes or single words toward richer language structure.

The advance also touches, briefly, on emerging questions about neural data privacy. NIH said methods like these may one day be used to infer speech-related thoughts, though the new work is presented as basic science rather than a near-term application.

The study was small, involving eight epilepsy patients with clinically implanted devices, and the recordings were collected in the context of medical monitoring rather than a purpose-built communication system. That makes the findings an important step in basic research, not an immediately deployable technology. Still, Debara Tucci, director of the National Institute on Deafness and Other Communication Disorders, said the detail matters: “This level of granularity is necessary for us to more completely understand how the brain generates speech and, ultimately, how we can develop technologies to restore it for individuals with communication disorders.”