Abstract:

The traditional view of disordered speech characterizes speech errors as deletions, insertions, or transpositions of one or more phonetic symbols in sequence. Recent articulatory research, however, suggests that such a characterization of disordered speech is at best incomplete, and at worst incorrect. A consequence of abandoning the traditional view is that the value of phonetic transcription as a tool for analyzing speech errors is in question. We examine the discrepancy between the phonetic transcription and the actual articulatory events produced by an apraxic speaker whose articulation was imaged with real‐time MRI with synchronous noise‐canceled audio in a speak‐after‐me shadowing task. Results indicate that transcription was somewhat reliable between two trained transcribers, but several systematic discrepancies existed. In addition to the transcription being an incomplete record, we also found cases in which the transcription was misleading, unhelpful, or simply incorrect regarding the actual articulation produced by the speaker. Further examination of the acoustic signal revealed that the problems with the transcription cannot easily be overcome with detailed analysis of the acoustics. This study suggests that accurate characterization of disordered speech requires data of a type that cannot be provided solely by transcripts or acoustics. [Work supported by NIH]