Mapping the neuronal building blocks of human language with language models - PubMed
4 hours ago
- #syntactic representation
- #neuronal language encoding
- #frontotemporal cortex
- The study used wide-scale single-neuron recordings and natural language processing models to map linguistic representations in the human frontotemporal cortex.
- Neurons were found to represent grammatical relationships between words, parts of speech, and higher-order syntactic structures like phrase transitions and sentence sequences.
- These neurons encoded syntactic and semantic properties combinatorially, incorporating specific sentence contexts for detailed information processing.
- Neuronal populations were locally organized, with microscale representations differing from wider field potential patterns.
- Linguistic encoding was broadly distributed across the frontotemporal cortex but left-lateralized and varied regionally, revealing cellular building blocks of language at micro, meso, and macro scales.