Affiliation: College of Arts and Sciences, Department of Psychology and Neuroscience
During conversation, listeners can use disfluency (e.g. uh, um) as a signal to expect discourse-new information. However, inferences from disfluency are attenuated once the difficulty is attributed to a characteristic of a speaker (e.g. stutterer, non-native). The current study tests whether the distribution or frequency of disfluency can change its informativity to the listener. In Experiment 1, I created a context where disfluency only occurred prior to discourse-given reference, resulting in an atypical distribution of disfluency. In Experiment 2, the stimuli were manipulated in that disfluency occurred frequently, but followed a typical distribution of disfluency (before discourse-new information). Both experiments found no effect of distribution; listeners showed a greater bias towards new information following disfluent utterances than fluent ones, regardless of whether the distribution of disfluency was novel (Exp. 1) or frequent (Exp. 2). Further research is needed to determine whether distributional learning occurs in the context of discourse processing.