Visual reinforcement infant speech discrimination: Developing a method of performance analysis

Abstract

While measures of speech perception are an important aspect of audiological assessment and validation of amplification fitting in older children and adults, no clinical method of speech perception assessment exists for infants and toddlers. Visual Reinforcement Infant Speech Discrimination (VRISD) is a tool that has been used to assess infant speech perception in studies for over 30 years and has been deemed to have potential for clinical use. Unfortunately, the foundational work to provide an appropriate protocol for VRISD's clinical use does not exist; the reliability and validity of VRISD have not been studied. In its current research form, VRISD consists of 30 trials and is designed not as an assessment of an individual child's ability to discriminate specific phoneme or consonant vowel comparisons, but as a technique to compare the mean performance of a group of children with another - by age, by native language, or to determine whether one comparison is more difficult than another. This dissertation aimed to begin investigation into the foundational work necessary to help VRISD become a clinical tool. VRISD was used to assess 15 normal-hearing infants' abilities to discriminate three speech sound contrasts (a/i, ba/pa, and da/ga). Results of infants. performance on these three contrasts indicated a hierarchy of difficulty, with the a/i contrast being the easiest and the da/ga contrast being the most difficult, as hypothesized. Results from each of the contrasts tested were analyzed in six different methods - five of which have been used in previously published VRISD studies - so that conclusions made from the different methods could be compared. The different methods of analysis sometimes led to differing conclusions as to the ability of an infant to discriminate a contrast. It was determined that the use of a criterion based on binomial probability was the best way to analyze performance in a manner that would yield reliable results as well as provide construct validity. The current study revealed that individual children have performance profiles that indicate that they do not perform consistently across 30 trials, particularly when they demonstrate mastery or correct performance consistently over the first 10-12 trials. This inconsistent performance has previously resulted in a significant number of children who have been unable to complete the task or whose performance is correct in the beginning of testing and then becomes incorrect, possibly because of habituation or boredom. There has previously been no universally accepted and well-defined criterion for establishing whether an individual child has mastered a specific discrimination task. This dissertation compares six different methods of determining VRISD performance. A criterion-based performance measure using binomial probability provides the best technique for construct validity.