Is Generalization General? Differences Between Predictive and Category Learning Reflect Knowledge of “Cognitive Kinds”

Abstract

Shepard (1987) observed that generalization gradients were invariant between species and stimulus dimensions, and proposed that these universal properties reflect knowledge of “natural kinds” that exist in the world. Here, we test a natural extension of Shepard’s ideas, by examining whether generalization differs between different types of learned associations. To do so, we leverage generalization as a tool to diagnose the underlying content of learning in two domains (predictive learning and category learning), while equating stimuli and procedural details between tasks. Experiment 1 showed that empirical differences in generalization gradients between domains can be largely explained by differences in the testing procedures. When these were equated, the underlying generalization function was largely similar between predictive and category learning. The results of Experiment 2 were more complex, revealing some differences between domains which may be explained by differences in the representation of outcomes and categories. We conclude that the similarities between learning domains (underlying gradient shape and response to experimental manipulations) provide support for Shepard’s (1987) proposal that generalization exhibits universal properties, and propose that the differences between learning domains reflect knowledge about “cognitive kinds”; categories are created to be mutually exclusive while outcomes are not.