Topic Selection in Conversation

Keywords: conversation, communication, natural language processing

Abstract: Conversation among humans is ubiquitous, and often enjoyable. During every turn of every conversation, individuals decide: Should we stay on this topic or switch to a different one? Descriptively, how do people navigate this decision? Prescriptively, how should they? Across thousands of synchronous and asynchronous conversations in face-to-face and online settings with close others and strangers, we observe how people select topics in cooperative conversation. Compared to natural language processing algorithms, humans fall short in detecting others’ interest in topics, they are overly reluctant to switch topics—with straightforward improvements (e.g., greater enjoyment of the conversation) that can be achieved by empowering people to switch topics more readily together.