Reproduction of seen actions: stimulus-selective learning.

Subjects observed and reproduced abstract, irregular stimulus models generated by the steady movement of a disk across two-dimensional paths. The paths comprised 3 to 7 randomly oriented linear segments linked head-to-foot. Reproductions were expressed by moving a stylus over the surface of a graphics tablet while the disk was tracing its trajectory (concurrent reproduction), or soon after the disk had finished (delayed reproduction). For both concurrent and delayed conditions, fidelity of reproduction fell with increasing number of segments in the model. Overall quality of reproduction did not differ between the two conditions. When a few models were repeated, interspersed among non-repeated ones, performance improved but only when reproduction was delayed. This improvement was stimulus-selective, not a general improvement with practice. Two additional experiments showed that (i) memory for a seen model is well preserved for at least 6 s, with relatively modest need for rehearsal, and (ii) successful reproduction is possible with remarkably little information having been extracted from key points in the model's trajectory.