A Methodology to Design Human-Like Embodied Conversational Agents

16 pagesInternational audienceThis paper presents a bottom-up methodology to study human-human and human-agent dialogues in order to improve the design of embodied conversational agents (ECAs). The methodology uses a matrix representation of dialogues, constructed by means of a (semi-)automatic annotation process. A multidimensional dialogue pattern extraction and a clustering algorithm are applied to the coding matrices. The extracted patterns model the human behaviours