| In nine experiments I explored the effects of naturally occurring speech disfluencies on comprehension.;I found that spontaneous repetitions, like the gift the gift department, might help comprehension by supplying information missing in the first occurrence of words in the repetition. Information from the second occurrence might help listeners to recognize the preceding words and free attention for the upcoming word. Another explanation for the effect is that repetitions might help comprehension by putting words into fluent phonological phrases after an interruption.;I found that spontaneous false starts, like and um they begin to disc-she asks him about uh keeping the ant, slow recognition, but only in those cases where they force the listener to do work to repair the utterance. When false starts begin utterances, listeners may abort the false starts with no cost to comprehension. But when false starts are in the middle of utterances, listeners have to figure out where the false start begins in order to know exactly what to abort and where to attach the restarted information. This process is a load on memory and comprehension is slowed.;I also found suggestive evidence that the speech fillers um and uh affect comprehension, and that the two fillers affect comprehension differently. Ums seem to help comprehension, perhaps by providing information about the meta-communicative process, such as directing listeners' attention to the upcoming phrase. The presence or absence of uhs, in contrast, had no effect on word recognition, perhaps because the effects were masked by pausing effects.;No existing model of speech comprehension incorporates the processing of speech disfluencies. All models predict either the opposite of the effects found, or a myriad of effects of which the actual effects are a possibility. To include disfluencies, I propose that a model should incorporate: (1) information that defines phonological phrases and (2) a repair process based on identifying the beginning of an error in the speech stream. |