"How Was Your Weekend?" A Generative Model of Phatic Conversation

Abstract: Unspoken social rules, such as those that govern choosing a proper discussion topic and when to change discussion topics, guide conversational behaviors. We propose a computational model of conversation that can follow or break such rules, with participant agents that respond accordingly. Additionally, we demonstrate an application of the model: the Experimental Social Tutor (EST), a first step toward a social skills training tool that generates human-readable conversation and a conversational guideline at each point in the dialogue. Finally, we discuss the design and results of a pilot study evaluating the EST. Results show that our model is capable of producing conversations that follow social norms.