Categorization

Background

Categorization involves mapping multiple individual items into a single conceptual representation that can be used to make predictions about new items encountered in the world. Categorization research concerns itself with understanding the nature of these representations and how they are learned from experience. Theories of categorization posit that individual items are composed of features (e.g., color, shape, size) and that categories are represented by relationships between features. Different theories posit different relationships between features and category representations.

Researchers concerned with learning often use novel, artificially constructed categories with clearly defined features to investigate how people learn category structure from features of individual category members. For example, category members might be circles of varying sizes with a central line that varies in angle. Different category structures could then be defined by the size, angle, or any of a number of possible combinations of the two.

Other research investigates natural categories to examine how knowledge is represented and used in other processes such as induction. A key feature of natural categories is typicality. Many categories appear to have graded membership in that some items are “better” members than others. For example, a robin is a more typical “bird” than is an ostrich. Typicality effects are ubiquitous in natural categories. For example, typical category members are classified more quickly than nontypical members.

Many researchers believe there are at least two distinct category-learning systems, although some [Page 319]maintain that there is a single system. In the dual-systems framework, one system relies on hypothesis-testing strategies and represents categories using explicit, verbalizable rules (e.g., “large circles are Category A; small circles are Category B”). This system is learned quickly and can be generalized widely to novel items, but it is limited to learning categories defined by explicit rules. The prefrontal cortex plays a critical role in the explicit system.

Another, more implicit, system represents items in a perceptual space of possible feature combinations and maps portions of that space to responses (category labels). This system is learned more slowly, cannot be generalized widely, but one can learn complex category structures that cannot be described by explicit rules (like most real-world concepts). The striatum plays a crucial role in this system.

The prefrontal cortex and striatum develop at different rates and decline differently with age, so the relative development and decline of the two systems may explain many age-related changes in category learning (discussed later). Exemplar theories (one type of single-system account) propose that items are categorized by assessing their featural similarity to items stored in memory from potential categories, and because of the high reliance on memory, the medial temporal lobe is thought to be critical for categorization.

The role of feedback during learning is an important issue in categorization research. One distinction is whether the task is supervised or unsupervised. In supervised learning, category labels are supplied for each item. In unsupervised learning, no category labels are given. Studies have shown that both children and adults can learn categories in unsupervised scenarios by abstracting statistical regularities among the items—inferring that items form distinct clusters of features. Dual-systems accounts predict that feedback is processed differently by each system. The implicit system relies heavily on dopamine signals in the striatum that require immediate feedback to adjust behavior. Feedback processing in the explicit system is less time dependent but relies on working memory to evaluate potential hypotheses in light of feedback.

...