Brain cells represent word meanings through distributed hippocampal activity

Researchers drew on large language models to examine how groups of neurons encode spoken word meaning. Recordings from hippocampal neurons in patients listening to stories suggested that meaning is represented by activity patterns across multiple cells. The peer-reviewed study was published in Nature Neuroscience.

The study involved ten epilepsy patients whose hippocampal neurons were monitored with implanted microwire electrodes while they heard narratives. Researchers used computational encoding models to test whether semantic information predicted firing patterns, while accounting for sound-level and grammatical features. This suggests word meaning may be distributed across cell populations rather than tied to single neurons.

The team also compared neural responses to numerical representations from language models processing the same stories, and examined polysemy. Findings published in Nature Neuroscience indicate that meaning-related information was reliably captured, with individual cells responding to multiple words and categories. This links the memory-related hippocampus to language comprehension.