Abstract
W. Schulz proposed that prediction errors can facilitate the learning of sequential behaviors. We assessed Schulz's proposal, asking how prediction errors alter the fidelity with which a remembered motion sequence was reproduced. Subjects viewed a sequence five times, each time reproducing what they had just seen. Each sequence comprised six quasi-randomly directed motions. Because eye movements are influenced by cognitive factors such as learning and expectation, we supplemented measures of reproduction fidelity with measures of eye movements made while subjects viewed each sequence of motions. Beginning with the second viewing of a sequence, tracking eye movements showed clear anticipation of the upcoming motions. Thus, eye movements provide a sensitive indicator of subjects' knowledge of and expectations for complex, quasi-random sequences of motions. To determine the influence of prediction errors, a novel motion direction was occasionally injected into a well-learned sequence. This unexpected, deviant motion transiently disrupted eye movements, requiring large, corrective saccades for catch-up. However, reproduction of this motion sequence was equivalent to that of well-learned, non-deviant sequences, and was greatly improved over reproduction of entirely novel sequences. These results undermine claims that eye movements provide a substrate crucial to visuomotor learning. Immediately after a perturbed sequence, a final presentation either reinstated the original, well-learned sequence, or preserved the deviant motion. On this final presentation, eye movements anticipated the reappearance of the deviant motion component; when this anticipation was correct, catch-up saccades were smaller and velocity of smooth pursuit was higher. Motion sequence reproduction was more accurate when subjects' expectation was violated and the original sequence appeared than when this expectation was confirmed. So, one presentation of an unexpected, deviant motion produces strong learning for that component, and violating expectations improves subjects' ability to reproduce a motion sequence.
Supported in part by CELEST, an NSF Science of Learning Center (SBE-0354378), and by NIH Training Grant T32GM084907.