‹
Back
11:00 - 12:30
Parallel sessions 5
›
11:00 - 12:30
Room: HSZ - N3
Chair/s:
Rebecca Albrecht, Florian Seitz
Categorization processes are fundamental to how humans structure, interpret, and interact with the world. They shape individual perception as well as higher-level judgments and decisions, and at the societal level play a key role in stereotyping, prejudice, and group formation. Yet, research on perceptual and social categorization has largely proceeded along separate tracks. Whereas perceptual research has aimed to identify domain-general mechanisms underlying individual categorization, social research has aimed to uncover the broader implications of categorization for group dynamics. Integrative approaches that bridge perceptual and social categorization remain rare.

Differences in research goals are mirrored by differences in methodology. Perceptual research typically relies on simplified, abstract tasks that maximize internal validity and support formal modeling, often at the expense of external validity. Social research, in turn, embeds categorization in realistic contexts that reflect lived experience and intergroup relations, but at the cost of making it harder to isolate and formalize the underlying processes. Recent methodological advances—from richer behavioral paradigms to automated theory discovery using ML/AI—create novel opportunities to combine the strengths of both traditions. These developments make it increasingly possible to study categorization in contexts that are both controlled and ecologically valid, paving the way for genuine integration of perceptual and social research.

This symposium is structured around three complementary steps in building this bridge. The first examines computational models of categorization and asks what predictions they offer for understanding social categorization. The second starts from the opposite direction, considering how phenomena of social categorization and judgment can be explained within computational frameworks. The third takes a meta-perspective, highlighting how recent methodological advances—ranging from large-scale experimentation to theory-driven simulations and formal model comparison—create new opportunities for linking the two traditions. Together, these perspectives show how research on categorization can move beyond separate traditions by uniting mechanistic explanations with social consequences and aligning methodological control with ecological relevance, paving the way toward a unified theory of categorization.
Submission 680
A Unified Similarity Framework for Modeling Category Variability and Feature Correlations
SymposiumTalk-04
Presented by: Florian Seitz
Florian Seitz, Jana Jarecki, Jörg Rieskamp
University of Basel, Switzerland
Understanding how people use variability and correlations among features is essential for understanding social cognition. Social groups often differ in their variability (e.g., more diverse appearance among artists than bankers), and many features are correlated (e.g., colorfulness of clothing correlating with hairstyle distinctiveness). Such distributional information can shape social categorization and may contribute to stereotyping—for example, when group variability is underestimated or feature correlations are overweighted. This project presents a unified similarity-categorization model that models how feature variances and correlations jointly shape psychological similarity and category inference. Specifically, we use an exemplar framework to compare two ways of similarity computation, representing categorization processes that ignore or incorporate distributional information, respectively. Two experiments manipulated feature variances or correlations across categories to test whether people rely on distributional information when categorizing transfer stimuli. In the variance experiment, transfer stimuli were positioned between a high-variability and a low-variability category. In the correlation experiment, transfer stimuli aligned with the correlation structure of one category but differed less from the other category. Across both experiments (Ns = 43), people generally ignored distributional information, suggesting reliance on cognitively simple categorization strategies. Importantly, the materials used offered participants no strong prior beliefs about category variability or feature correlations, and processing distributional information was not required for category learning. We discuss how the modeling framework can be extended to real-world social categorization tasks—where people may have stronger prior distributional beliefs—providing a principled approach to study how distributional information shapes social judgments in ecologically meaningful contexts.