Categorisation and perceptual Learning: Why tDCS to Left DLPFC Enhances Generalisation.

Abstract

In the 27 years that have passed since the McLaren, Kaye and Mackintosh (MKM) model of perceptual learning was first proposed, it has undergone considerable theoretical development and been subject to extensive empirical test. But we would argue that the basic principles of the theory remain as valid today as they were in 1989. One of these principles was that salience modulation of stimulus representations based on prediction error was a key component of latent inhibition and perceptual learning. It was this modification of what was otherwise a fairly basic adaptation of the model for categorisation proposed by McCleland and Rumelhart (M&R) that transformed a system that would exhibit enhanced generalisation as category learning progressed, into one that would instead offer an improved capacity for discrimination between exemplars as a consequence of experience with the category. This modification has only been tested indirectly up until now, by looking at the predictions that flow from it and then comparing them to animal and human discrimination following stimulus pre-exposure. In this chapter we test this principle more directly, by using tDCS to disrupt the modulation of salience by prediction error, and show that when this is done, people exhibit the enhanced generalisation predicted by the standard M&R model. We conclude that our results provide further support for the MKM approach to stimulus representation.