How do humans learn categories? In most cases, categories are learned in relation to one another. In a recent paper (Hosch, Hoffmann, & von Helversen, 2024, JEP:LMC), we used a self-regulated category learning task to investigate this in more detail. As variability is a defining feature of categories, we manipulated the variability of two categories (i.e. the focal and the counter-category) and participants sampled exemplars until they felt confident to have a representation of the categories. Briefly, we observed evidence for an increasing sample size with increasing category variability.
We here propose and explore a model that aims to capture this relational learning process with an evidence accumulation model. In the model, the variability of the focal category contributes to the rate of evidence accumulation, while the counter-category variability modulates the boundary at which the learning process ceases. The rate of evidence accumulation either remains at the same expected value or diminishes with an increasing number of sampled exemplars. Variability was mapped to parameters either linearly or using an entropy measure to capture the nature of the perceptual stimulus variability.
We hypothesize that categories are learned either discriminatively, by contrasting the focal category's variability with the counter-category's variability, or assimilatively, where the focal category's variability level aligns with that of the counter-category. Our findings reveal that an assimilation process best fits the data in this task, underscoring the relational nature of category learning.