Semantic Dependency Analysis Method for Japanese based on Optimum Tree Search Algorithm

There is combinatorial order of ambiguities in each level of natural language analysis, such as the morphology, syntax, and semantic levels. Moreover, much of the linguistic knowledge in each level is preference knowledge and has mutual interference. Deterministic processing is usually introduced to avoid the combinatorial explosion. However, this restricts the ability of natural language processing system because of the above-mentioned features of linguistic knowledge. This paper describes a sentence analysis method which uniformly evaluates syntactic and semantic preference knowledge, and shows an algorithm (based on the branch and bounding method) to search the optimum semantic tree from a semantic dependency graph which holds syntactic and semantic ambiguities in a Japanese sentence.