On the Prediction of Confusion Matrices from Similarity Judgments.
BOLT BERANEK AND NEWMAN INC CAMBRIDGE MASS
Pagination or Media Count:
This study of human identification of complex visual stimuli is a step toward an integrated description of the perceptual process whereby stimuli are represented psychologically and the decision process that makes use of the stimulus representation in selecting an identification response. A multidimensional scaling procedure was applied to judgments of stimulus similarity to derive the dimensions of a perceptual space and the relative loci of stimuli in that space, and a probabilistic decision model based on weighted interstimulus distances was developed to predict the confusion matrices in various identification tasks. The high accuracy of the predictions supplies a strong validation of the use of multidimensional scaling procedures to reveal perceptual structure, in demonstrating the ability of that structure to account for behavior in an independent task. From the other point of view, the empirical success of this approach suggests a relatively simple and practical means of predicting, and possibly enhancing, identification performance for a given ensemble of visual or auditory stimuli. Author