As people become more familiar with a stimulus or stimulus sequence, accuracy of recall improves. With sequences of stimuli, improvement over successive exposures manifests in two parallel changes: (i) a flattening of the serial position curve, and (ii) a decrease in overall error rates. We used a visuomotor sequence-learning paradigm to investigate the extent to which these two changes in visual working memory (VWM) were driven by separable familiarity-induced capacity increases and resolution improvements. Our stimuli consisted of a disk that traversed a trajectory defined by quasi-randomly directed linear motion segments. Forty-four participants (from four experiments) viewed the disk's motions multiple times, and, after each such presentation, used a graphics tablet and stylus to reproduce the disk's path from memory. Reproductions were recorded for offline analysis of the error made in reproducing each segment. The analysis confirmed that, as in other sequence-learning tasks, participants' error magnitudes showed a strong serial position curve on first seeing each sequence, which flattened with subsequent presentations; the flattening was accompanied by an overall increase in accuracy. For a finer-grained analysis, we created error distributions for each combination of segment serial position, participant and stimulus repetition. Then, for each distribution, we found the best-fitting Gaussian + uniform mixture model. The serial-position dynamics associated with the model's two parameters revealed that the flattening of the serial position curve arose primarily from improvement in VWM's resolution, which quickly reached ceiling (after two or three presentations). In contrast, the continued overall decrease in error magnitude came from a change in VWM capacity, which slowly approached 100% of the segments being in memory. The differing dynamics of representation accuracy and capacity limits suggest that learning-induced improvements in working memory depend on two separable sources of improvement.
Meeting abstract presented at VSS 2012