Content area
Abstract
The ability to create and use relational categories—such as barrier, duty, or grandfather—is crucial to many higher-order cognitive processes, ranging from analogy and problem solving to language and perception. However, most studies of categorization have focused on how people learn and use categories defined by features. A series of empirical studies examine how relational categories are learned, used, and remembered, and how they differ from categories based on features. In Chapter 1 I present a brief review of the literature and build an argument for the utility of relational categorization. Chapter 2 describes an empirical experiment contrasting how relational and featural categories are learned. I present evidence suggesting that while featural categories can be learned with both a deterministic and family resemblance category structure, relational category learning is catastrophically impaired in the family resemblance condition. An ideal observer model is developed in Chapter 3 to account for differences in difficulty and presentation order and is applied to the findings, demonstrating that the differences between conditions were not due to incidental causes. Chapter 4 examines the interaction of relational and featural information within a single category. By manipulating the two types of information independently, it was found that participants exclusively use relations for making goodness-of-exemplar judgments, and exclusively use features when making recognition judgments. When forced to use the less preferred type of information, participants' judgments were at chance—even though they showed evidence of having differentiating information available. Chapter 5 examines typicality, one of the most robust findings in categorization studies. Typicality is almost universally equated with and measured by goodness-of-exemplar judgments. However, most previous studies have used feature-based categories. I show that when using relational categories with relational and featural information in conflict, participants do not consider typicality and goodness-of-exemplar judgments to be equivalent. Together these results suggest a fundamental difference in how relational and featural categories are learned, used, and remembered, and highlight the need for additional understanding of relational categorization both in the laboratory and the real world.
