What if i Interrupt You

Today, with the development of science and technology, we hope to improve human-agent interaction, agents will become more 'human-like', so that they communicate autonomously with humans both verbally and non-verbally. A challenge for the Embodied Conversational Agent is then to handle and manage speaking turn exchanges, and, more particularly interruptions, these are inherent in human-human interaction. We present our ongoing work on modeling interruption management in human-agent interaction. Our research contains two main aspects: 1) when and how should the virtual agent interrupt human users, and 2) how should the virtual agent respond when being interrupted by human users. To achieve this goal, we first started by analyzing human-human interaction data.