A population code for semantics in human hippocampus.

Hippocampal neurons encode semantic information with complex, context-dependent patterns, showing strong correlation with GPT-2 embeddings and semantic distances.

• Why it matters: Understanding how the brain represents semantics is crucial for bridging neural activity with language comprehension, addressing gaps in neurocomputational models of meaning.
• What they did: The study recorded hippocampal responses to narrative speech from individuals, using an encoding model to analyze semantic encoding and compare neural responses with language model embeddings.
• The result: Findings reveal that hippocampal activity reflects semantic relationships similar to language models, enabling insights into neural mechanisms of word meaning and contextualization.