Causes and Strategies for Requesting Clarification in Dialogue

Abstract

We do two things in this paper. First, we present a model of possible causes for request-ing clarifications in dialogue, i.e., we classify types of non-understandings that lead to clar-ifications. For this we make more precise the models of communication of (Clark, 1996) and (Allwood, 1995), relating them to an indepen-dently motivated theory of discourse seman-tics, SDRT (Asher and Lascarides, 2003). As we show, the lack of such a model is a prob-lem for extant analyses of clarification moves. Second, we combine this model with an ex-tended notion of “confidence score ” that com-bines speech recognition confidence with dif-ferent kinds of semantic and pragmatic confi-dence, and argue that the resulting processing model can produce a more natural clarification and confirmation behaviour than that of current dialogue systems. We close with a descrip-tion of an experimental implementation of the model. 1