Speech categories vary in their similarity to one another, with consequences for phenomena ranging from patterns of confusability under noisy listening conditions to phonological processes described as operating on relatively abstract representations. Studies of inter-category similarity are typically conducted by studying error rates in noise, or by using metalinguistic similarity judgments. Here, we exploited the continuous and non-ballistic properties of computer mouse movements to investigate, in a fine-grained manner, the inter-category similarity of speech sounds. Participants listened to recordings of naturally produced words ("pin," "pen," and "pan") and used a computer mouse to select the matching stimulus from an array of two pictures. The same participants also performed a dissimilarity judgment task. Both tasks revealed evidence for graded effects of inter-category similarity, albeit in the context of strongly categorical classification. Multidimensional scaling of mean dissimilarity judgments revealed a very tight distribution of stimuli from the same category in perceptual space and, at a larger scale, differences between pairs of categories in their similarity to one another. Although accuracy in the word identification task was nearly perfect, arm-movement dynamics revealed attraction toward the alternative response, scaled in magnitude to the perceived similarity of the two categories.
Bibliographic reference. Zevin, Jason D. / Farmer, Thomas A. (2008): "Similarity between vowels influences response execution in word identification", In INTERSPEECH-2008, 2048-2051.