Web resource selection for dialogue system generating natural responses

Abstract

Using Web information in example-based dialogue systems is considered to be a good way to increase the topical relevance of system responses. However, Web content is mostly written documents, not colloquial utterances. To alleviate the discrepancy in style between written and spoken language, we suggest that the corpus should be selected subsets of the Web, viz., online bulletin boards. The naturalness provided by casual utterances is especially important when the goal of a dialogue is chatting rather than formal question answering. By appropriately selecting the information source, our text-based dialogue system can generate friendly responses to users. These characteristics were evaluated with a questionnaire. © 2011 Springer-Verlag.