Reliability and Dispersion of Nasality Ratings

Abstract

Reliable average scale values of nasality have been obtained in many studies using the psychological scaling method of equal-appearing intervals. In these studies, reliability has generally been evaluated through the use of correlation procedures. Many investigators have also computed a Q value, the semi—interquartile range, to measure the dispersion of ratings assigned by the judges to each stimulus. The mean Q value has then been reported as an index of interjudge reliability. The speech stimuli presented to the judges for rating of nasality have typically been of one or more of four types: isolated vowels, CVC syl— lables, connected speech played forward, and connected speech played backward. In those studies (10, 12, I4, 15) in which more than one type of speech stimulus have been presented to the judges, there has been a tendency for the reliability coefficient and mean Q value to vary systematically with the stimulus type. There is evidence, for example, that smaller mean Q values are obtained for connected speech samples rated during forward play, than for connected speech samples rated during backward play, or for isolated vowels (10, 12, 14, 15). Differences in the dispersion of nasality ratings assigned to various types of speech stimuli suggest that the degree to which nasal voice quality can be defined reliably by judges varies as a function of the speech stimulus. In no study, however, have measures of dispersion or reliability been obtained for all four types of speech stimuli produced by the same speakers or rated by the same judges. Moreover, data regarding the reliability and dispersion of ratings of specific vowels, CVC syllables, and connected speech sam— ples within each of the four types of speech stimuli are unavailable. While mean Q values have been reported as measures of the reliability of average scale values of nasality, little interest has been shown in the Q values of subject groups or individual subjects. Thurstone and Chave (17) used Q (in this instance, the interquartile range) as a measure of the ambiguity of the sample, the sample in their case being a statement The authors are affiliated with the