A convenient unit of analysis that has been used extensively in the discourse literature is the turn unit. While the notion of a turn has some intuitive appeal, there is little to support it in any objective sense, making it difficult to justify its use in an empirical study. The delimitation of a turn unit is somewhat subjective, and generally assumes that dialogue is structured in an orderly A-B-A-B-like pattern, where each speaker takes it in turn to hold the conversational floor. It does not make use of evidence that a great deal of dialogue involves two (or more) speakers speaking at once, making the concept of orderly turn-taking problematic. In this talk I outline the use of the move unit as an objective alternative to the turn unit, and its uses in segmenting speech into utterances.
Once utterances have been isolated, a problem arises in deciding whether an utterance could count as a valid response to another, and if so, which other utterance it is a response to. Again, real dialogue is often sufficiently complex in the way utterances overlap to make this decision non-trivial. I therefore outline several different problems that were encountered in deciding whether an utterance was a response to an utterance by another speaker. It was supposed that these problems could be solved using data on the durations of intervals between pairs of utterances, where cut-off points could be set to divide utterances into those which were responses and those which were not.
To download this paper, please return to Proceedings of the 1997 Postgraduate Conference