How categories master variability: Insights into category learning and generalization

Variability permeates every aspect of our environment and categories help us navigate this variability. This thesis presents three projects that leverage variability to deepen our understanding of category learning and generalization. Project 1 investigates how different types of variability, learned in a prior relationally structured category learning task, affect later category generalization. The findings show that categories experienced as more diverse lead to broader generalization than homogeneous ones. Specifically, generalization widens when category exemplars exhibit heterogeneity, but not when participants encounter many different exemplars within a diverse category. In Project 2, I use variability to explore category learning processes, focusing on how the immediate context of a category—specifically its counter-category—shapes learning. By manipulating category variability in a newly developed self-regulated category learning task, I show that greater variability prompts participants to draw more samples until their category representation suffices. Interestingly, not only the category's variability but also the variability of the counter-category influences the number of samples drawn. In Project 3, I explore how category learning in the self-regulated task can be modeled within the sequential sampling framework. Our findings suggest that category variability determines the accumulation rate, while the counter-category influences the decision to stop sampling exemplars. Within this framework, I examine variability perception and the shape of the accumulation rate. I also model how between-category processes impact learning, finding that learning assimilates to the counter-category’s characteristics. In summary, this thesis provides new insights into how category variability shapes generalization, influences the category learning process, and highlights the intricate link between a category and its counter-category.