Spoken Dialogue System for Information Navigation based on Statistical Learning of Semantic and Dialogue Structure

摘要

The thesis addresses a framework of a spoken dialogue system that navigates information of text documents, such as news articles, based on statistical learning of semantic and dialogue structure. Conventional spoken dialogue systems require a clear goal of the user, but this assumption does not always hold. Moreover, domain knowledge and task flows of the systems are conventionally hand-crafted. This process is costly and hampers domain portability.