Statistical answer-type identification in open-domain question answering

One of the most critical components of a question-answering system is the identification of the type, or semantic class, of the answer sought. Systems today use widely-varying numbers of such classes, but all must map the question to one or more classes in their repertoire. In this paper, we present a statistical method of associating question terms with candidate semantic classes that has been shown to achieve a high degree of accuracy and to be applicable to different underlying semantic classifications.