Researchers at Baylor College of Medicine, Rice University, the University of California, Berkeley and Texas Children's Hospital reported on October 11, 2026, that groups of neurons in the hippocampus represent the meaning of spoken words and how that meaning changes with context. The study, published in Nature Neuroscience, drew inspiration from large language models (LLMs), which process language by representing words numerically.

The researchers recorded activity from individual hippocampal neurons in 10 people who listened to stories. Tiny wire electrodes implanted during clinical monitoring for epilepsy captured the activity. Mathematical models connected words and their meanings in the stories to the recorded neuronal activity; the analysis found semantic encoding even after accounting for phonemics (speech sounds) and grammar. Individual neurons responded to multiple words across multiple meaning categories.

The team compared differences in groups of neurons' responses to pairs of words with differences in the words' meanings. They also compared the brain recordings with numerical word representations produced by language models processing the same stories, and examined whether LLM-based measures of polysemy (a word having multiple related meanings) tracked variation in neural activity. Neural response patterns were most closely aligned with GPT-2 embeddings, and patterns varied with LLM-derived polysemy measures. The authors also reported that distances between neural population responses correlated with semantic distances, while responses to semantically similar words showed an inverse correlation even with contextual embeddings; they attributed that pattern to contrastive coding's noise-mitigating benefits.

The authors said the findings provide a neurocomputational account of how hippocampal neurons encode word meaning during spoken-language listening. They said the work could inform understanding of how the hippocampus represents spoken language and inspire future studies using LLMs to investigate brain language processing.

Researchers at Baylor College of Medicine, Rice University, the University of California, Berkeley and Texas Children's Hospital reported on October 11, 2026, that groups of neurons in the hippocampus represent the meaning of spoken words and how that meaning changes with context. The study, published in Nature Neuroscience, drew inspiration from large language models (LLMs), which process language by representing words numerically.

The researchers recorded activity from individual hippocampal neurons in 10 people who listened to stories. Tiny wire electrodes implanted during clinical monitoring for epilepsy captured the activity. Mathematical models connected words and their meanings in the stories to the recorded neuronal activity; the analysis found semantic encoding even after accounting for phonemics (speech sounds) and grammar. Individual neurons responded to multiple words across multiple meaning categories.

The team compared differences in groups of neurons' responses to pairs of words with differences in the words' meanings. They also compared the brain recordings with numerical word representations produced by language models processing the same stories, and examined whether LLM-based measures of polysemy (a word having multiple related meanings) tracked variation in neural activity. Neural response patterns were most closely aligned with GPT-2 embeddings, and patterns varied with LLM-derived polysemy measures. The authors also reported that distances between neural population responses correlated with semantic distances, while responses to semantically similar words showed an inverse correlation even with contextual embeddings; they attributed that pattern to contrastive coding's noise-mitigating benefits.

The authors said the findings provide a neurocomputational account of how hippocampal neurons encode word meaning during spoken-language listening. They said the work could inform understanding of how the hippocampus represents spoken language and inspire future studies using LLMs to investigate brain language processing.