Regard et communication face-à-face entre un locuteur humain et un agent conversationnel animé. Attention mutuelle et monstration multimodale
Stephan Raidt · HAL (Le Centre pour la Communication Scientifique Directe) · 2008
In the context of synthetic generation and decoding of linguistic information, not only the audible component but also the visual component of speech conveys valuable information. We address gaze as an important modality to enhance speech and to convey additional information. Gaze is an important deictic gesture as well as it plays various roles in the organization of dialogue and social interaction. In a first experiment, we investigated how the gaze of a talking head can be used as a deictic gesture in an on-screen search and retrieval task. We found that such gestures are appropriate to reduce processing time as well as cognitive load. Multimodal gestures incorporating speech in a coherent way showed to be more efficient than only visual gestures. In a second experiment, we investigated the relations between the gaze of a target subject and different elements of conversational interaction. We defined different stages in the dialogic exchange of information and found that these are related to the variations in the measured gaze behavior. Based on the observed characteristics we propose a model to control the gaze of an embodied conversational agent in close dyadic interaction.