Challenges in Exploiting Conversational Memory in Human-Agent Interaction

In human interactions, language is used to project and maintain a social identity over time. The way people speak with others and revisit language across repeated interactions helps to create rapport and develop a feeling of coordination between conversational partners. Memory of past conversations is the main mechanism that allows us to exploit and explore ways of speaking, given knowledge acquired in previous encounters. As such, we introduce an agent that uses its conversational memory to revisit shared history with users to maintain a coherent social relationship over time. In this paper, we describe the dialog management mechanisms to achieve these goals when applied to a robot that engages in social chitchat. In a study lasting 14 days with 28 users, totaling 474 interactions, we find that it is difficult to leverage the shared history with individual users and to also accommodate to expected conversational coordination patterns. We discuss the implications of this finding for long-term human-agent interaction. In particular, we highlight the importance of topic modeling and signaling explicit recall of previous episodes. Moreover, the way that users contribute to interactions requires additional adaptation, indicating a significant challenge for language interaction designers.