Modeling unsupervised perceptual category learning

During the learning of speech sounds and other perceptual categories, category labels are not provided, the number of categories is unknown, and the stimuli are encountered sequentially. These constraints provide a challenge for models, but they have been recently addressed in the Online Mixture Estimation model of unsupervised vowel category learning. The model treats categories as Gaussian distributions, proposing both the number and parameters of the categories. While the model has been shown to successfully learn vowel categories, it has not been evaluated as a model of the learning process. We account for three results regarding the learning process: infantspsila discrimination of speech sounds is better after exposure to a bimodal rather than unimodal distribution, infantspsila discrimination of vowels is affected by acoustic distance, and subjects place category centers near frequent stimuli in an unsupervised visual classification task.