Abstract

Current computational linguistics offers a wide range of methods for assessing semantic similarity among words, basing conceptual meaning either on word co-occurrence vectors or on abstract hierarchical relationships between concepts. To what extent do these different approaches capture brain information-encoding of lexical meaning? Putting in parallel computational and neural language processing, my recent studies provide evidence that both co-occurrence and hierarchical relationships affect brain activity patterns, but in distinct neuronal networks. Modality-preferential sensory and motor cortex emphasises hierarchical links among concepts, whilst modality-independent semantic regions distributional statistics.