Predictions in conversation

Abstract

In everyday interactions, natural conversations are often effortless and managing them seems to be easy. However, a closer look at the timing pattern of conversations implies a complex cognitive architecture underlying turn-exchanges. Corpus studies of recordings of natural conversations have shown that the duration of turn-transitions—the time between the end of a turn and the beginning of the new one—is most frequently around 200 ms (Heldner & Edlund, 2010; Levinson & Torreira, 2015; Stivers et al., 2009). Several studies have tried to explain how interactants manage conversations since the 1960’s (Pléh, 2012). Recent models of turn-taking consider the cognitive processes of the speakers and listeners during conversations, and most agree that listeners need to predict the end of the turn of the current speaker, in order to produce a timely response. Different models, however, suggest different predictive mechanisms underlying smooth turn-transitions. Some accounts do not agree about when listeners who speak next start to prepare their turn, i.e., towards the end of a current turn or as soon as it is possible to formulate a response. Some models also do not agree about how listeners can predict the end of a turn. In this paper, it will be discussed whether the different accounts can be reconciled.