Conversational Completions: Tracing Human and Artificial Linguistic Creativity

摘要：
This talk highlights the creativity underlying an apparently trivial everyday human activity: everyday conversation. It discusses the sophistication of the linguistic mechanisms involved in this activity in relation to the current surge of artificial conversational agents, grounded in Large Language Models. The talk then focuses on “conversational completions”: situations in which one speaker terminates an utterance initiated by another speaker. This phenomenon will be used to approach traditional linguistic and cognitive units of language analysis from a radically dialogic perspective. Conversational completions are especially interesting because they question the idea that utterances in a conversation are produced and understood by isolated individuals. The talk concludes by discussing whether humans, while performing these completions, are less predictable and more variable than artificial agents. Taking a broader perspective, we ask whether this higher variability is part of what makes conversation human. It argues that human conversational creativity may lie not purely in producing novel linguistic forms, but to find creative solutions to conversational challenges in real time, under linguistic, pragmatic, and social constraints.