Rule-based category learning occurs when categories can be described by simple, verbalizable rules (e.g., "respond A if the stimulus is large"). The RULEX model (Nosofsky, Palmeri, & McKinley, 1994) formalizes this as a hypothesis-testing process: the learner sequentially tests candidate rules, starting with simple single-dimension rules and progressing to more complex conjunctive rules and exception memorization as needed.
The RULEX Process
Level 2: Conjunctive rules (e.g., "if x₁ > c₁ AND x₂ > c₂ then A")
Level 3: Rules + exceptions (store misclassified items as exceptions)
Search: test rules at each level, accept if error rate is low enough
Rule-Based vs. Information-Integration
Ashby and colleagues' COVIS model proposes that rule-based learning is mediated by an explicit verbal system (involving prefrontal cortex and working memory) that is distinct from the implicit system that handles information-integration categories. Evidence for this distinction comes from dissociations: verbal working memory load impairs rule-based but not information-integration learning; feedback delay impairs information-integration but not rule-based learning; and different category structures are optimal for each system.