A semantotopic map in human hippocampus.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Linguistics

Background:

• The hippocampus is crucial for abstract conceptual and semantic information encoding.
• The precise topographical organization of semantic encoding within the human hippocampus remains largely unknown.

Purpose of the Study:

• To investigate the spatial organization of semantic representations along the mediolateral axis of the anterior human hippocampus.
• To determine if semantic features are systematically mapped within this hippocampal region.

Main Methods:

• Utilized Neuropixels probes to record neural responses during passive language listening in three anesthetized patients undergoing neurosurgery.
• Computed semantic tuning functions by regressing neural activity (threshold crossing events, single unit responses) against semantic embeddings from GPT-2, Word2Vec, and SBERT.
• Analyzed neural responses across recording sites to identify patterns of semantic similarity and dissimilarity.

Main Results:

• Found that more distant recording sites exhibited more dissimilar tuning functions, supporting a semantotopic organization.
• Observed systematic changes in semantic features like animacy, concreteness, and familiarity along the mediolateral axis, though these effects were individual-specific.
• Detected a small but significant increase in semantic similarity at shorter spatial scales, suggesting a modest periodic organization.

Conclusions:

• Demonstrated a multiscale functional organization of semantics within the human hippocampus.
• Provided evidence for both broad semantotopic and finer periodic organizational principles in hippocampal semantic processing.
• Highlighted the potential for individual variability in the topographical mapping of semantic information.