Interpreting Loosely Encoded Questions

James Fan, Bruce W. Porter

Abstract: Knowledge-based question-answering systems have become quite competent and robust at answering a wide range of questions in different domains, however in order to ask questions correctly, one needs to have intimate knowledge of the structure of the knowledge base, and typical users lack this knowledge. We address this problem by developing a system that uses the content of the knowledge base to automatically align a user's encoding of a query to the structure of the knowledge base. Our preliminary evaluation shows the system detects and corrects most misalignments, and users are able to pose most questions quickly.