Emergent stimulus relations depend on stimulus correlation and not on reinforcement contingencies

Abstract

We aimed to investigate whether novel stimulus relations would emerge from stimulus correlations when those relations explicitly conflicted with reinforced relations. In a symbolic matching-to-sample task using kanji characters as stimuli, we arranged class-specific incorrect comparison stimuli in each of three classes. After presenting either Ax or Cx stimuli as samples, choices of Bx were reinforced and choices of Gx or Hx were not. Tests for symmetry, and combined symmetry and transitivity, showed the emergence of three 3-member (AxBxCx) stimulus classes in 5 of 5 human participants. Subsequent tests for all possible emergent relations between Ax, Bx, Cx and the class-specific incorrect comparisons Gx and Hx showed that these relations emerged for 4 of 5 the participants after extended overtraining of the baseline relations. These emergent relations must have been based on stimulus-stimulus correlations, and were not properties of the trained discriminated operants, because they required control by relations explicitly extinguished during training. This result supports theoretical accounts of emergent relations that emphasize stimulus correlation over operant contingencies.

Keywords: emergent relations; humans; mouse-click; stimulus correlation; stimulus equivalence.

Fig 1

Schematic diagram showing the baseline A–B and C–B conditional discriminations for Class 1 (procedures were similar for Classes 2 and 3). On these trials, the correct comparison, B1, was always presented with class-specific incorrect comparisons, G1 and H1. Responding to B1 was reinforced while responding to G1 and H1 was never reinforced. The stimulus pairing of 3 s occurred at comparison presentation where the choice response was completion of a fixed interval 3-s requirement. A response made to any comparison before 3 s had elapsed had no effect.

Fig 2

Schematic diagram showing examples of the intended stimulus classes based on comparison–stimulus pairings. Relations explicitly trained with reinforcement contingencies are indicated by the solid arrows. The comparison–stimulus pairings implemented at the comparison–choice phase are indicated by the solid lines. Derived relations are indicated by broken arrows. An example of the equivalence classes predicted to emerge based on operant reinforcement contingencies is shown on the Class 2 stimuli (left diagram). The derived B–A, B–C, A–C, and C–A relations are shown with the broken arrows. Examples of the relations predicted to occur based on the comparison–stimulus pairings are shown on the Class 3 stimuli (right diagram). The derived A–G, A–H, B–G, B–H, C–G, C–H, G–A, G–B, G–C, G–H, H–A, H–B, and H–C relations are shown with the broken arrows.

Fig 3

Training stimuli for all potential stimulus classes.

Fig 4

Percent correct for each participant on Classes 1 to 3 stimulus relations for all Parts 1 and 2 training blocks in Session 1. Part 2 accuracies are shown with the rightmost data point.

Fig 5

Percent correct for each participant on Part 3 baseline trials and equivalence tests (symmetry, and combined symmetry and transitivity) for all stimulus classes.

Fig 6

Percent correct responses for each participant on Part 6 baseline and test trials (A–G, A–H, B–G, B–H, C–G, C–H, G–A, G–B, G–C, G–H, H–A, H–B, and H–C) for stimulus classes based on comparison–stimulus pairings. The 85% criterion for the emergence of these classes is shown on each graph as a dotted line.