A Formal Model of the Mechanism of Semantic Analysis in the Brain

We propose a formal model of the mechanism of semantic analysis in the language areas of the cerebral cortex. The framework of Combinatory Categorial Grammar, a framework of grammar description in theoretical linguistics, is modified so that it does not use lambda calculus to represent semantic rules. This model uses a novel form of semantic representation named hierarchical address representation, and uses only fixed-length data structures. The knowledge of syntax and the knowledge of semantics are clearly separated in this model. Therefore, it is possible to reproduce disorders specific to syntax (utterance similar to Broca's aphasia) and disorders specific to semantics (utterance similar to Wernicke's aphasia) by disabling different modules in the model. We estimate that the model can be implemented using the Bayesian network model of the cerebral cortex that we have proposed earlier. We believe that this research will connect computational neuroscience and theoretical linguistics, and greatly evolve both of them.