Chládková, K. 1 , Scharinger, M. 2 & Schröger, E. 1
1 Cognitive and Biological Psychology, University of Leipzig
2 Max Planck Institute for Empirical Aesthetics
Humans track the distributional probabilities of stimuli in their surroundings and warp, or categorize, the perceptual space accordingly. The outcomes of distributional training studies vary depending on the number of perceptual dimensions involved. To find out whether listeners use the distributional information available from multiple perceptual dimensions, we assessed unsupervised distributional learning of novel auditory categories cued by two perceptually independent dimensions. Because distributional learning is typically defined as an automatic, exposure-triggered, process, we predicted training effects for both dimensions.
The stimuli were inharmonic tone complexes varying in terms of spectral peak and duration, and clustering around a 'low and long' category vs. a 'high and short' category, which were perfectly separated in the two-dimensional space. Listeners were exposed to 400 tokens drawn from the two clusters, and labeled each token as A or B, without receiving feedback. Before and after training, we measured listeners' neural mismatch responses to spectral and durational differences. Robust learning effects (stronger mismatch response after training) were found for the durational dimension, with the effects for the spectral dimension being smaller, if existent at all. This indicates that distributional learning could be a selective, dimension-specific, mechanism and may not automatically operate over all available cues.