Web text classification for response generation in spoken decision support dialogue systems

This paper addresses issues in generating responses by extracting sentences from the Web for spoken decisionmaking dialogue systems. Various decision criteria are usually involved when selecting an alternative from a given set of alternatives. Such a dialogue system is required to explain the alternatives in terms of each decision criterion focusing on why the alternative is recommended. Preparation of a generation template is difficult for such explanations; thus we consider extracting and then adapting sentences retrieved from the Web. In this work, we address our preliminary experiment in selecting sentences appropriate for system responses.