Kris R. Thórisson
This paper examines some of the general characteristics of real-time face-to-face interaction relevant to AI and human-computer interaction. It supports the claim that a real-time multimodal system needs to be designed in layers and include both reactive and reflective behaviors. It presents a layered feedback-loop model of face-to-face dialogue and shows how contextual analysis of the function of multimodal acts is linked with feedback generation and multimodal interpretation. A brief description is given of a modular architecture called "~'mir, based on the proposed model, for modeling psychosocial dialogue skills.