Revisiting peak shift on an artificial dimension: Effects of stimulus variability on generalization

Abstract

One of Mackintosh’s many contributions to the comparative psychology of associative learning was in developing the distinction between the mental processes responsible for learning about features and learning about relations. His research on discrimination learning and generalization served to highlight differences and commonalities in learning mechanisms across species and paradigms. In one such example, Wills and Mackintosh (1998) trained both pigeons and humans to discriminate between two categories of complex patterns comprising overlapping sets of abstract visual features. They demonstrated that pigeons and humans produced similar “peakshifted” generalization gradients when the proportion of shared features was systemically varied across a set of transfer stimuli, providing support for an elemental feature-based analysis of discrimination and generalization. Here we report a series of experiments inspired by this work, investigating the processes involved in post-discrimination generalization in human category learning. We investigate how post-discrimination generalization is affected by variability in the spatial arrangement and probability of occurrence of the visual features, and develop an associative learning model that builds on Mackintosh’s theoretical approach to elemental associative learning.