Abstract

Query understanding is the foundation of information retrieval study. In this paper the authors build a low cost prototype system that can parse Chinese wh-questions posted by internet users into structures of conceptual graphs with rich semantic information. As a novel point, this paper draws an analogy between the semantic parsing process and Example-based Machine Translation (EBMT) by viewing questions as the source language and conceptual graphs as the target language, thus simplifying semantic parsing process into a matching, alignment and recombination process. Instead of depending on a syntactic parsing tree like other approaches, the suggested approach uses only some lexical and syntactic features that can easily obtained by heuristics or a POS tagger. The work balances between shallow and deep parsing and shows encouraging evaluation results.

Paper Details