The relationship between perceptual anchors, nasality ratings and nasalance scores
AuthorGalek, Kristine E.
StatisticsView Usage Statistics
BACKGROUND AND PURPOSE: This investigation explored the relationship between perceptual anchors, nasality ratings and nasalance scores when perceptual anchors were placed at different points along the 7-point scale used to rate nasality.METHOD: One hundred speech samples and corresponding nasalance scores were obtained from 95 children followed by a cleft palate team and 5 non-patient control speakers. The speech samples were randomized and duplicated to make six different sets of 120 samples each (100 experimental + 20 repeat samples). Listeners (N=129) were assigned at random to one of six listening groups and each group rated nasality on a seven-point scale that ranged from "1" normal nasality to "7" severe hypernasality. The anchors were examples of a "1", "3", "4", "5", and/or "7" on the rating scale. These anchors were played selectively to Group two ("4"), Group three ("1 and "7"), Group four ("3" and "5"), Group five ("1", "4", "7") and Group six ("7"). RESULTS: Percent agreement, Spearman-Rho Rank Correlation Coefficient and Cohen's Weighted Kappa for the first and second ratings of 20 repeated speech samples revealed that 73%-78% of the listeners from the anchored groups had "good" intra-rater agreement (Kw = 0.61). Inter-rater agreement for listener ratings of nasality was measured by Fleiss' generalized kappa and the Q value (semi-interquartile range). Findings revealed the strength of the overall generalized kappas and the corresponding kappas at each scale value (category kappa) to be low for all scale values across all the listening conditions. By contrast, the Q value revealed that the listeners had relatively good reliability across the listening conditions. Group 5 produced higher inter-rater agreement at each of the individual scale values especially at the extremes than the other five listening conditions. The inter-rater reliability and agreement data suggest that placement of perceptual anchors on a 7 point scale did affect listener judgments and inter-rater agreement and reliability was better with anchors than without. The dispersion of listener ratings of nasality at each scale value across listening conditions was generated and produced Q values notably smaller at the extreme ends of the scale ("1" and "7") and larger at each of the mid-range scale values. Group mean Q values were calculated for all groups and revealed Group 5 ("1", "4" and "7") to have the lowest (best) group mean Q value of 0.78.The effect of perceptual anchors on the dispersion of listener nasality ratings was examined through the application of psychophysical theories. Several findings were revealed from the five anchored conditions. First, the listeners in three of the five anchor groups consistently rated the first sample of the rating task as a scale value "2" even though the first sample was a control sample with a nasalance score of 11%. Second, when an anchor was played in the middle of the scale ("4") the listeners rated more samples in the middle of the scale than at either end of the scale. One anchor played at scale value "3" and scale value "5" shifted the ratings toward the anchor sites (assimilation effect) and one anchor at each of the extremes ("1" and 7") resulted in more ratings being made away from the anchor site than at the anchor site (contrast effect). A single anchor played at scale value "7" resulted in a regression of ratings away from the anchor site where more ratings were made at the opposite end of the scale than at the anchor site. Group 5 heard one anchor at each extreme and one equidistant between the extremes and had the most "even" distribution of ratings across the entire 7-point scale. Spearman Rho Correlation Coefficients were obtained between perceptual ratings of nasality and nasalance scores for each of the six different listening conditions. Correspondence between median nasality ratings and mean nasalance scores differed for the different listening conditions. Group 4 heard anchor stimuli representing "3", and "5" on the rating scale and this resulted in the best correspondence between nasality ratings and nasalance scores (rs=0.48). The poorest correspondence was for Group 1 which did not hear any anchor stimuli (rs=0.41). Sensitivity and Specificity were calculated for each of the six listening conditions and sensitivity ranged between 0.71 and 0.75. A median rating was computed across listeners for each of the 100 samples in order to assign a single rating to each sample and these were compared to the mean nasalance scores. Mean nasalance scores were essentially the same for all stimuli rated "1" (26.4%), "2" (26.9%), or "3" (26.7%). In other words, it appeared that listeners made distinctions on the mild-moderate end of the perceptual scale that the Nasometer did not. Mean nasalance scores increased only about 3% from scale value "3" (26.76%) to scale value "4" (29.72%) and increased again about 6% from "4" to "5" (36.41%). Nasalance increased substantially for scale values "6" (44.05%) and "7" (41.28%) but there was little difference in nasalance for speech samples at the high end of the scale. CONCLUSIONS: This study investigated the relationship between perceptual anchors, nasality ratings and nasalance scores. Perceptual anchors and anchor effects did appear to affect listener ratings of nasality. Placement of perceptual anchors at different points along the interval scale did affect the reliability and agreement of listener ratings of nasality, the dispersion of nasality ratings, and the relationship between nasality ratings and nasalance scores.