The benefit of generating errors during learning: What is the locus of the effect?

Guessing translations of foreign words (hodei?), before viewing corrective feedback (hodei-cloud), leads to better subsequent memory for correct translations than studying intact pairs (hodei-cloud), even when guesses are always incorrect (Potts & Shanks, 2014), but the mechanism underlying this effect is unknown. Possible explanations fall into two broad classes. One puts the locus of the effect at retrieval: Items studied through a generation process have more potential retrieval cues associated with them, or a more distinctive context, and are therefore more accessible at final test. The other puts the locus at encoding and involves enhanced encoding of the correct answer following the generation of an error compared with passive studying (Potts & Shanks, 2014). In 6 experiments we found support for the proposal that generating errors benefits memory through stimulating curiosity to learn correct answers following an incorrect guess, leading to enhanced processing of targets following generation. In Experiment 1, generating possible translations after seeing correct answers did not produce better memory than studying without generating, suggesting that an element of surprise or anticipation is necessary for generating to benefit memory. Experiments 2a-2c found enhanced recognition memory for targets following generating, suggesting increased focus on targets following a guess. In Experiments 3 and 4, participants rated their curiosity to learn correct answers higher when ratings were given after generating than before, suggesting that the act of generation increases curiosity to learn the answers. These findings imply that enhanced processing of feedback is a key consequence of errorful generation.