The distribution of speech errors in multi-word prosodic units

Abstract

Sequencing errors in natural and elicited speech have long been used to inform models of phonological encoding and to understand the process by which serial ordering is achieved in speech. The present study focused on the distribution of sequential speech errors within multi-word prosodic units to determine whether such units are relevant to speech planning, and, if so, how. Forty native English-speaking undergraduate students were asked to produce sentences that varied in length and in the extent to which certain phonological features were repeated (tongue twisters or not). Participants prepared their utterances in advance of speaking and were coached to be as fluent as possible once they started speaking. The goal was to ensure the production of well-structured utterances, while maximizing the number of errors produced, and minimizing the effects that excessive self-correction might have on prosodic structure. Speech errors were perceptually identified in the recorded speech and categorized. Strong and weak prosodic boundaries were prosodically-transcribed in sentences with sequencing errors. Speech error patterns were found to correspond well with the boundaries of the multi-word prosodic units defined by the strong and weak prosodic boundaries. In particular, the number of sequencing errors was found to vary as a function of position within a unit such that the fewest errors were found in initial position, more occurred in early-mid position, and even more occurred in late-mid position. This pattern of increasing errors across the multi-word prosodic unit was referred to as the cumulative error pattern. The analyses also revealed a final position effect. When multi-word prosodic units occurred in utterance-initial or utterance-medial position, a disproportionate number of errors occurred in the final position of the unit. However, when the units occurred in utterance-final position, more errors occurred in late-mid position than in final position. The cumulative error pattern and final position effect are interpreted to suggest the serial activation and decay in activation of multi-word planning domains during phonological encoding.