Word‐based versus syllable‐based models of speech production planning: Evidence from elicited phonological errors

Abstract

Earlier error‐elicitation experiments have shown that single‐segment errors are more likely to be elicited when the target segments share position in the word onset, and less likely when the target segments cross word position (e.g., /p/ and /f/ are more likely to interact in “parade fad” and less likely in “repeat fad,” even though both target segments are in prestress position in the latter case). A separate series of experiments using CVC stimuli showed a difference in error patterns when the stimulus words are spoken in lists versus embedded in short phrases; stimuli like “leap note nap lute” elicit about the same proportion of errors in initial and final position, while the phrasal version (“From the leap of the note to the nap of the lute”) elicits far fewer errors in final position. If this difference between word lists and phrases extends to other aspects of error patterns, it might affect the word‐position/stress‐position results for mixed one‐ and two‐syllable stimuli. However, results show that, when these stimuli are embedded in phrases, the error patterns remain remarkably similar. Moreover, when speakers are asked to generate sentences using the words of the stimulus, instead of reading or reciting the phrases provided by the experimenter, results again support the claim that word‐onset consonants are more likely to interact in an error than are prestress consonants. These findings suggest that word structure plays a role in the production processing representation that is in force when single‐segment errors occur.