Transfer of dimensional associability in human contingency learning.

Several studies have demonstrated processing advantages for stimuli that were experienced to be reliable predictors of an outcome relative to other stimuli. The present study tested whether such increases in associability apply at the level of entire stimulus dimensions (as suggested by Sutherland & Mackintosh, 1971). In 4 experiments, participants had to learn associations between Gabor gratings and particular responses. In a first experiment, some gratings were more predictive of the response than other gratings, whereas in 3 subsequent experiments, one stimulus dimension (i.e., either the orientation or spatial frequency of the grating) was more predictive than the other dimension. In contrast to the learned predictiveness of individual gratings (Experiment 1), dimensional predictiveness did not affect the subsequent rate of learning (Experiments 2 and 3), suggesting changes in the associability of specific stimuli, but not of stimulus dimensions. Moreover, greater transfer of predictiveness was found in all experiments when particular stimulus values of the test discrimination did not lie between the previously relevant stimuli. In Experiment 4, an increased learning rate was found for discriminations along the previously predictive dimension compared with a dimension that was indicative of uncertainty, but again the transfer was more pronounced for specific stimuli that were compatible with the previously learned discrimination. Taken together, the results imply that a transfer of associability typically applies to individual stimuli and depends on how the transfer stimuli relate to those stimuli that individuals previously learned to attend.