Analogical Mapping Method and Semantic Categorization of Japanese Compound and Complex Sentence Patterns

To overcome the limit of the conventional machine translation (MT) method based on compositional semantics, we proposed an Analogical Mapping (AM) method based on Semantic Typology and built a semantic category system for Japanese compound and complex sentences. The AM-method maps linguistic expressions into other expressions with the same meaning with semantic categorization (based on concepts called Truth Items). We also built a semantic category system composed of two sub-systems: one for classifying the meanings (222 categories) represented by the relation between two clauses and the other for classifying the meanings (740 categories) represented by clauses. These semantic codes were assigned to sentence patterns (SPs) (226,800 patterns) registered in the sentence pattern (SP) -dictionary we recently developed. We ascertained that these were useful for selecting semantically correct candidates from matched patterns in input sentences.