Refined natural language questions with context analysis

In this paper, we present an approach to refine a natural language question by analyzing contexts and transforming them into a set of queries in the form of Boolean expressions and phrases. We try to change the question sentence into answer phrases or a set of terms which are likely to appear in target documents. Our transforming strategies are based on interrogative words and structures of the question sentences. We tighten the general terms in the query with their synonym set in hypernym / hyponym (is-a) relation. After that, a set of refined queries is sent to a search engine, which we use as our knowledge base. The benefit of our system is that it can be used as an interface to help users to create effective queries over existing querying systems such as FAQs, newsgroups, web boards and search engines. From our preliminary experiment, we found that the top rank documents, which are returned from search engines, consist of more precise answers than using the natural language question alone.