Configural learning theory, primarily associated with John Pearce (1987, 1994), proposes that compound stimuli (e.g., a tone+light pair) are represented as unique configural units rather than as collections of individual elements. When a compound AB is presented, the organism forms a configural representation of AB that is similar to, but distinct from, representations of A alone or B alone.
Pearce's Configural Model
S(A, AB) = (n_common / n_A) × (n_common / n_AB)
Generalization: V_A = S(A, AB) × V_AB
where V_AB = associative strength of the AB configural unit
Elemental vs. Configural
The configural approach contrasts with elemental models (like Rescorla-Wagner) where compound stimuli are represented as the sum of their elements. The key empirical test is negative patterning: training A+ and B+ (reinforced) but AB− (not reinforced). Elemental models struggle with this because the compound AB should have twice the associative strength. Configural models handle it naturally because AB has its own representation that can be associated with no-reinforcement. The debate between elemental and configural representations remains active, with evidence suggesting that both mechanisms operate depending on the similarity structure of the stimuli.