Context-based multimodal interpretation : an integrated approach to multimodal fusion and discourse processing

摘要

This thesis is concerned with the context-based interpretation of verbal and nonverbal contributions to interactions in multimodal multiparty dialogue systems. On the basis of a detailed analysis of context-dependent multimodal discourse phenomena, a comprehensive context model is developed. This context model supports the resolution of a variety of referring and elliptical expressions as well as the processing and reactive generation of turn-taking signals and the identification of the intended addressee(s) of a contribution. A major goal of this thesis is the development of a generic component for multimodal fusion and discourse processing. Based on the integration of this component into three distinct multimodal dialogue systems, the generic applicability of the approach is shown.