Test-enhanced learning for pairs and triplets: When and why does transfer occur?

In four experiments, we explored conditions under which learning due to retrieval practice (i.e., testing) transfers to the case in which the cue and response words are rearranged (e.g., a training test on gift, rose, ?, wherein the target is wine, and a final test on gift, ?, wine, wherein the answer is rose). In both Experiment 1 and a supplementary experiment, we observed divergent results for pairs and triplets: Relative to a restudy control condition, strong transfer was observed for pairs, but none for triplets. In Experiments 2 and 3, the theoretical basis of the specificity of learning for triplets was explored. The results rule out the possibilities that transfer is wholly absent for triplets and that transfer occurs only for the case of exact cue-response reversal on the final test. Rather, it appears that, for both pairs and triplets, transfer will occur unless both of the following conditions hold: (1) two or more independent cues are presented on the training test, and (2) the correct responses on the training and final tests are different. We show that the majority of the results can be explained by combining the dual-memory theory of the testing effect with an inclusive-OR representation that forms when two or more cues are presented on the training test. Follow-up analyses that were conditionalized on training test accuracy suggest that specificity of learning is greater on a correct than on an incorrect training test trial, although selection confounds and contradictory experimental results preclude a strong conclusion.