Early prosodic manifestations of disfluency

Theoretical models of speech production have hypothesized a relation between different types of disfluencies and the mechanisms responsible for them. Some disfluencies, such as filled pauses (e.g. ‘um’, ‘uh’) and repetitions (i.e. ‘the the’), are argued to arise from difficulty in planning, while cutoff disfluencies (e.g. ‘horiz-[ontal]’) are argued to arise from selfmonitoring. This distinction predicts that prosodic manifestations of disfluency, i.e. durational slowing and pitch/intensity modulation, should occur earlier for planning disfluencies than for self-monitoring disfluencies. The present study examined segmental duration, pitch, and intensity in speech produced just before filled pause, repetition, and cutoff disfluencies in the Switchboard corpus. The results showed that durational slowing occurs earlier and is more extensive before filled pause disfluencies than before repetitions and cutoffs. In addition, decreases of f0 and intensity occurred earlier before filled pauses than before repetitions, and intensity decreased more gradually before cutoffs than before repetitions and filled pauses. These findings support theoretical models in which cutoffs are associated with a self-monitoring mechanism and filled pauses/repetitions are associated with planning difficulties. Furthermore, differences in effect magnitudes between filled pauses and repetitions indicate that filled pauses may be associated with more severe planning difficulties than repetitions.