Unifying the structures of language in a neural population code.

Large language models encode linguistic diversity into a high-dimensional space that closely parallels neural population codes, suggesting a shared computational strategy.

• Why it matters: Understanding how language emerges from neural population activity is crucial for bridging artificial intelligence and cognitive neuroscience, addressing limitations of rule-based systems.
• What they did: The authors analyzed the structure of LLMs' embeddings and compared them to neural population codes, demonstrating that LLMs can serve as cognitively plausible models of human neural language processing.
• The result: This work implies that explaining language emergence requires new theoretical paradigms beyond algebraic-symbolic theories, advancing our understanding of neural and artificial language systems.