Disfluencies: Comprehension Processes

Judgments about the Speaker

When speakers are unsure of an answer to a question they tend to hesitate, and these hesitations may be marked by utterance-initial disfluency. Speakers are asked to rate their own confidence in their answers (their feeling of knowing or FOK), produce ratings that correlate negatively with disfluency. Listeners are equally able to make this kind of judgment: Their ratings of FOAK (feeling of another's knowing) are lower when answers are preceded by silence and lower still when the silence is replaced by a filler-silence combination of the same length. These findings provide evidence that disfluency can influence the judgments that participants are asked to make. However, they do not directly answer the question of whether disfluency routinely affects language comprehension.

Effects on the Comprehension Process

Listeners are faster to identify target words following uhs in recorded speech than when the uhs are excised and are faster to respond to instructions to press a given button when those instructions are repaired (yel-uh-orange). Disfluencies appear to have a helpful effect on the ongoing comprehension of speech, at least in carrying out specific tasks. More compelling evidence comes from studies in which listeners' eye movements are recorded as they respond to instructions to manipulate objects on a computer display. In line with other work on language comprehension, listeners appear to predict what is most likely to be mentioned at any given stage, and these predictions are reflected in anticipatory eye movements to the objects depicted. When the instructions are disfluent, participants tend to fixate earlier on objects that are hard for the speaker to describe or name (such as visually complex abstract objects). When listeners are told that the speaker is aphasic and prone to random disfluency, the effects disappear, suggesting that they can be directly attributed to the listeners' theories about what (normal) speakers find difficult to say.

[Page 253]Additional work has used event-related potentials (ERPs) to examine whether disfluencies routinely affect how listeners process speech. In these studies, there is no specific task to carry out or set of objects referred to. Participants simply listen to recorded utterances, which include some target words that are surprising (difficult to integrate semantically or acoustically deviant in context). Relative to unsurprising control words the targets give rise to clear ERP effects (an N400 for semantic and a P300 for acoustic surprise), but these effects are attenuated when the target words are immediately preceded by an uh disfluency. These ERP findings implicate prediction (not being able to predict what comes next may reduce the semantic surprise and hence the N400) and attention (the P300 reflects attentional orientation to input; because attention is oriented by the disfluency, uh results in a P300 and the subsequent P300 is diminished). In other words, disfluency indicates that the words that follow may not be predictable, which causes listeners to attend more closely to what follows.

...