Graduation Year


Document Type




Degree Name

Master of Science (M.S.)

Degree Granting Department

Communication Sciences and Disorders

Major Professor

Stefan A. Frisch, Ph.D.

Committee Member

Nathan Maxfield, Ph.D., CCC-SLP

Committee Member

Kyna Betancourt, Ph.D., CCC-SLP


categorical error, gradient error, variation, curve-to-curve analysis, individual


This thesis investigates errors on speech sounds (or phonemes) produced in laboratory speech stimuli designed to generate phonological onset errors. The present study adds to the literature on phonological speech errors with an instrumental analysis of tongue posture during speech error production and an investigation of the nature of speech errors as unintended variation in articulation.

This study utilized ultrasound instrumentation to visualize speech errors made on velar and alveolar stop consonants at the point of stop closure. Two types of errors were of interest, categorical errors and gradient errors. Categorical errors are those that are heard by the listener and instrumentally appear to be a correct production of an incorrect target. Gradient errors are those that are usually heard to be the correct target, but on instrumental examination display characteristics of an incorrect production. Six participants repeated eight tongue twisters in both a baseline and an experimental condition. This study was interested in errors produced on the onset stop consonant pairs /t, d/ and /k, g/. Recordings were transcribed to determine the perceptual identity of each target. Ultrasound videos were then analyzed and an individual frame representing the articulatory posture for each closure was extracted. These frames were fit with a smoothing spline curve using Edgetrak software. A curve-to-curve analysis based on the methods of Zharkova (2009) was conducted as a means of further investigating variation in individual speakers as well as providing a quantitative measure of errors.

Results from the six speakers showed that all produced both categorical and gradient errors. The speakers showed individual variation in the stability of their productions and overall rate of errors. There was an observable trend for speakers who were more stable in their baseline productions to produce fewer errors, both gradient and categorical, in the experimental portion. Conversely, those speakers who exhibited more variation in their baseline productions had a higher rate of error under the experimental condition.