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.