Combining CCG Parsing and DL for Chinese Question Answering

Abstract

This paper presents a novel Web Question Answering(WQA) approach based on the combination of Combinatory Categorial Grammar (CCG), Description Logic (DL) ontology and Web Service (WS), in order to promote semantic-level accuracy through deep text understanding and improve the solutions for non-WH questions. Our framework transforms a question answering problem into a service planning problem for which HTN planner-SHOP2 is used. It addresses transition from natural language to semantic expression and dynamical composition of web services via planning, and execution of the web services planning. We propose to take DL based semantic modeling, i.e., translating lambda-expression encoding of question meaning into DL based semantic representations. The advantage of such approach is a seamless exploitation of existing semantic resource coded as DL ontology, which is widespread in such areas as Semantic Web and conceptual modeling. The experiments are conducted with a repository of complex Chinese questions which involves the satisfaction of some object property restrictions. The experimental results show that producing the semantic representations with the combination of CCG parsing and DL reasoning is an effective approach for question understanding at semantic level, in terms of both understanding accuracy promotion and semantic resource exploitation.