Speaking through a noisy channel - Experiments on inducing clarification behaviour in human-human dialogue
Speaker:David Schlangen
Institution:University of Potsdam
Abstract:
In natural, unrestricted conversation, participants have at their disposal a variety of devices to ensure sufficient mutual understanding. Adressees can signal understanding by providing a relevant reply or can signal problems by posing a clarification request (CR); the original speaker then can, should a signal of understanding reveal a misunderstanding, intervene and repair the original utterance.
The models of the human conversational repair system that have been put forward so far mostly derive from analysis of recorded dialogues occurring either in 'natural' situations (Schegloff and colleagues; e.g. (Schegloff, 1987)) or in experimentally controlled sitatuations (Clark and colleagues; e.g. (Clark, 1996)). While this approach has led to impressive advances in our understanding of this conversational competence, it has certain limits:
- If there is a CR, the problem that caused it must be inferred from its form and the original speaker's reply, as it cannot be directly observed.
- As it is not obvious for the annotator whether there has been a problem or not, strategies for avoiding to ask for clarification cannot be studied straightforwardly.
- The effectiveness of the repair system can only indirectly be studied.
In this talk, we first present a classification scheme for CRs and discuss problems related to it, and then report the results of an experiment where we introduced in a controlled way understanding problems, by occasionally replacing in one channel signal with noise. This gives us control over (one type of) communication problems, and hence addresses these limitations.