Disfluency as metacommunication

Research on spoken language comprehension has challenged the traditional view that disfluencies are mere performance errors that disrupt comprehension. By now, there is a range of evidence that disfluencies often facilitate language comprehension by supporting predictive inferences about upcoming speech. What is less well-understood is the exact nature of this benefit. How do listeners derive meaning from disfluency? To what extent are the benefits of hearing a disfluency dependent on potentially "signaled" elements, such as the fillers um and uh, versus more symptomatic elements, such as the length of a silent pause? Do disfluencies benefit comprehension through low-level mechanisms such as priming, or do they call upon more high-level inferences? How sensitive are listeners to different possible sources of disfluency? In this talk, I will review results from a research program investigating the nature and processing of utterance-initial disfluencies during referential communication. In these studies, the form of a speaker's disfluency was experimentally manipulated to discern its impact on the listener. The results generally support the view that the primary meaning of disfluencies is metacommunicative [1]: a disfluency signals that the speaker is experiencing trouble, and the form of the disfluency provides information about the severity of the trouble. The impact of a disfluency on listeners is to cause them to attend to the speaker and to attempt to diagnose the cause of the trouble. Moreover, listeners appear to do this in a flexible and largely speaker-specific, rather than "egocentric", manner. This model gives good coverage of the phenomena, and suggests important avenues for future investigation.