|Hiroshi Umemoto||Future Work Research Laboratory,
Corporate Research Group
This paper describes the implementation of a Japanese semantic parser based on the glue approach. The parser puts linguistically deep analyses into practice and is utilized for verifying the semantic validity of text and detecting logical or deductive relations such as redundancy and consistency within a text. It is designed as domain-independent, and produces fully scoped higher-order intensional logical expressions, coping with semantically ambiguous sentences without a storage mechanism. It is constructed from an English semantic parser on top of Lexical-Functional Grammar (LFG), and it attains broad coverage through relatively little construction effort, thanks to the parallelism of the LFG grammars. I outline the parser, and I present the analyses of Japanese idiosyncratic expressions including floating numerical quantifiers, showing the distinct readings of distributive and cumulative, as well as a double-subject construction and focus particles. I also explain the analyses of expressions that are syntactically parallel but semantically distinct, such as relative tense in subordinate clauses. In the future, I would like to explore the application of the parser for knowledge extraction from text combined with domain knowledge in fields such as sales and medical care.