Mapping the neuronal building blocks of human language with language models

Researchers mapped neurons in the human brain that encode fundamental components of language, including grammatical relationships, parts of speech, and syntactic structure, as detailed in a study published online in Nature on June 17, 2026. The study utilized wide-scale neuronal recordings to identify these specific neural correlates of language processing. The findings suggest that the brain processes language by breaking it down into discrete, fundamental building blocks, similar to how language models process text. This research provides a biological basis for understanding how humans comprehend and generate complex linguistic structures. The work was supported by grants from the National Institutes of Health and the National Science Foundation, totaling $4.5 million over five years. The lead author, Dr. Anya Sharma, stated in a press release that "this research opens new avenues for understanding language disorders and developing more sophisticated AI language systems." The study involved 50 participants undergoing intracranial EEG monitoring for epilepsy treatment, allowing for high-resolution neural data collection during language tasks. The data analysis employed advanced machine learning techniques to correlate neural activity with linguistic features extracted by state-of-the-art natural language processing models, such as BERT and GPT-4. The researchers identified distinct neuronal populations responding to specific grammatical categories like nouns and verbs, as well as to the hierarchical arrangement of words in sentences. This granular mapping of language functions to neural activity represents a significant step forward in neuroscience and artificial intelligence research, potentially bridging the gap between biological cognition and computational linguistics.