Anticipation in conversational turn-taking

February 26, 2015
3:00 pm - 4:30 pm
Pearson Laboratory (room 104)
Speaker: Jan de Ruiter, Bielefeld University
Host: Departments of Computer Science, Philosophy, and Psychology

Abstract

Ordinary conversation feels like an effortless activity. Yet listeners in a conversation have several cognitive tasks to perform in parallel. Obviously, they must try to comprehend what the current speaker is saying. But they must also, while the utterance of the current speaker is unfolding over time, plan an appropriate response, and time the delivery of this response appropriately. As Sacks, Schegloff and Jefferson documented in the 1970s, conversationalists try to minimize both temporal gaps and overlaps between turns of successive speakers. In order to achieve this, a would-be speaker needs to accurately anticipate the end of a current speaker’s turn, even though it is of unpredictable length.

My research addresses some of the cognitive-psychological questions that arise from the study of conversation: A) How good are listeners in conversations at timing their turns, in practice? B) What information from the current speaker's turn do they exploit? C) Which cognitive mechanism(s) do they employ for anticipation? And D) how long before the end of the current speaker’s turn do listeners initiate anticipation processing?

In my talk, I will present a series of studies aimed at addressing these questions. Using corpus analyses, on-line and off-line behavioural tasks, and neurocognitive methods, we found empirical support for the following answers: A) Listeners are very good at anticipating the end of the current speaker’s turn, and this appears to be the case cross-culturally. B) Listeners primarily exploit symbolic information in the current speaker’s utterance, and rely more on semantic than on syntactic information. C) They estimate the duration of the current turn by predicting its content, and D) anticipation processes can be detected very early, on average more than 1000ms before the end of the current turn.

I will also discuss why it is, at once, highly desirable and extremely challenging to model this human communicative competence in artificial systems.

*Note: Different room than usual for this talk*