Towards a psycholinguistics of dialogue: defining reaction time and error rate in a dialogue corpus.

Abstract

This study uses the multi-level coding of a designed corpus of unscripted task-oriented dialogues to demonstrate that time to respond (Inter-Move Interval, IMI) and rate of disfluency behave like psycholinguistic measures, reaction time and error rate, in reflecting the speakers’ cognitive burdens. Multiple-regression analyses show that IMI is sensitive to social distance between interlocutors, to the difficulty of the task which the dialogue serves, and to comprehension of the prior utterance and production of the current one. Rate of simple overt disfluency, in contrast, shows social and task effects, with most of the uniquely explained variance associated with planning and producing the current utterance. The results suggest that coded corpora may be useful in developing models of human interlocutors.