Comprehender's predictions in the face of uh speech disfluencies

Abstract

Spontaneous human speech is peppered with errors and disfluencies. Previous research has demonstrated that such errors and disfluencies are not always detrimental to comprehension. When confronted with a filled pause, listener's are able to make predictions about upcoming speech. But, does the length of the edit interval affect how far listener's back-track through given information in order to make a prediction about the repair? The present experiment examines the on-line effect of fillers on reference comprehension in order to investigate whether edit-length has a direct effect on the listener when deciding how far to back-track through the information already heard in order to make a prediction about the upcoming speech. Participants eye-movements were tracked whilst they were presented with visual and auditory stimuli. Auditory stimuli involved disfluencies with either a long edit interval (830ms silence) or a short edit interval (415ms silence) preceded by an editing expression uh. Relative to the two word reparanda, listener's predicted that a no-change repair would be made after hearing a filler followed by a short edit-interval and that a repair involving a change in the word immediately preceding the filler would be made after hearing a filler followed by a long edit-interval. The study demonstrates that edit-length does have an effect on the type of predictions made by listener's. However, regardless of edit-length, listener's do not tend to back-track when making predictions. Rather, they expect speaker's to interrupt themselves immediately after an error has occurred.