ABSTRACT
Most theories of verbal self-monitoring assume that we detect speech errors through at least two channels: Overt speech (the external channel) and internal speech (the internal channel). The postulation of two channels raises questions about their relative contribution. We argue that existing proposals for determining this “division of labor” are inadequate: either they fail to take into account that monitoring the internal channel is sometimes slow or they hinge on the unjustified assumption that the two channels are equally accurate. We propose a probabilistic model that expresses a relation between the detection rates of the channels and the frequencies of disfluencies and corrected and uncorrected speech errors. By fitting the model to existing data sets with normal speech and noise-masked speech, acquired from speakers with Broca’s aphasia and control speakers, we showed that the internal channel is more effective than the external channel. In fact, the data from Broca’s aphasia were compatible with the hypothesis that the external channel is not used at all. Furthermore, the analyses suggest that the external channel is relatively unimportant in the detection of lexical errors, but important in the detection of phonological errors. We propose that the division of labor between channels is under top-down control (selective attention to the internal channel) but also depends on bottom-up influences (access to acoustical or phonetic information).
