A: An Experimental Investigation into... B: ...Split Utterances

A distinguishing feature of dialogue is that more that one person can contribute to the production of an utterance. However, until recently these 'split' utterances have received relatively little attention in models of dialogue processing or of dialogue structure. Here we report an experiment that tests the effects of artificially introduced speaker switches on groups of people engaged in a task-oriented dialogue. The results show that splits have reliable effects on response time and on the number of edits involved in formulating subsequent turns. In particular we show that if the second half of an utterance is 'misattributed' people take longer to respond to it. We also show that responses to utterances that are split across speakers involve fewer deletes. We argue that these effects provide evidence that: a) speaker switches affect processing where they interfere with expectations about who will speak next and b) that the pragmatic effect of a split is to suggest to other participants the formation of a coalition or sub-'party'.