Learning Sound Categories

Another area of research in the lab concerns the formation of phonetic categories in the brain, including a treatment of the perceptual magnet effect, one example of the warping of auditory space by the perceptual system. In this effect, two examples of a vowel that are near the prototypical examples of that vowel from the listener’s native langauge are more difficult to discriminate from each other than two non-prototypical examples located near category boundaries.

We have used a combination of computational modeling, psychophysics, and functional brain imaging studies to arrive at an explanation of this effect in terms of neural maps in the auditory system. The model is based on the idea that the learning of sound categories changes the distribution of the firing preferences of neurons in auditory cortical maps, and thus changes the discriminability of sounds from different parts of acoustic space. The model predicts that the perceptual magnet effect arises because prototypical vowels are represented by fewer cells in auditory cortex than are non-prototypical vowels. The decreased representation is thus more susceptible to noise, and results in poorer discrimination abilities.

This idea is contrary to many neurophysiological studies which have shown that dispropotionately large exposure to a particular type of stimulus typically leads to a larger cortical representation for that stimulus. Our hypothesis that the frequently encountered prototypical sounds lead to a smaller representation is attributed to the type of training undergone. In speech perception, listeners must parse sounds into meaningful phonemes, syllables, and words in their native language, which makes the task inherently categorical. The model suggests that the type of training determines the evolution of the neural representations in cortex. See Figure One below.


Figure 1: Hypothesized changes in the auditory cortical maps as a result of categorization training (left) and discrimination training (right).

This prediction that prototypical sounds have a smaller representation than non-prototypical sounds was supported by a functional magnetic resonance imaging (fMRI) experiment involving examples of the vowel /i/. Additionally, a psychophysical experiment has verified that a perceptual magnet – like effect can be induced for non-speech sounds, which suggests that this phenomenon may be widespread in cortex.

We have written a lay language paper that discusses our model and some experimental results. That paper is available on the ASA Meeting web site here.