Effectiveness of Combining Learning Rules and Analogy in Semantic Analysis for Japanese Unknown Sentences

H. Shibuki, Y. Momouchi, K. Araki, and K. Tochinai (Japan)


Natural Language Processing, Knowledge Acquisition,Temporal Reasoning, Casebased Reasoning


In natural language analysis, it is one of the most difficult problems to analyze various sentences using limited knowledge. A Rule-based analyzer has the advantage of regular analysis using relatively fewer rules, and an Example-based analyzer has the advantage of flexible analysis using analogy. We propose a framework of hybrid analyzer that has both of the advantages. The hybrid analyzer learns rules from training data, and analogizes unknown sentences with the rules. We conducted experiments of assignment of semantic roles using three analyzers including the hybrid analyzer. In accuracy, the hybrid analyzer was the best result of them.

Important Links:

Go Back