The disambiguation strategies of semantic analysis in Chinese spoken dialogue system

Abstract

Semantic frame analysis is one of the most commonly used semantic analysis methods in Chinese spoken dialogue system research. And the two typical ambiguous structures commonly encountered in semantic analysis are relation-ambiguity and structural-ambiguity. According to the features of these two ambiguous structures, this paper puts forth the semantic PCFG (probabilistic context free grammar) model based disambiguation strategy to solve structural-ambiguity, and the expectation model (EM) based disambiguation strategy to solve relation-ambiguity. Efficient algorithms of the two methods are also provided. The experimental results show that applying these two disambiguation strategies can greatly improve the performance of language understanding in a base-line system. Especially, sentence accuracy is improved from 75.7% to 91.5%, and the three targets of semantic unit understanding rate-correction, recall, and precision are also improved by 10% on average.