Abstract
Much of what we know about visual short-term memory (VSTM) derives from change detection studies, in which observers are charged with identifying whether two successive stimulus displays, interleaved by a blank interval, are the same or different. More elaborate techniques require observers to indicate what specific element of the display changed, or where a change occurred. However, since a (typically) binary response is required, guess rate is high, the rate of data accumulation is low, and the richness of data gathered is necessarily limited. Furthermore, unless eye tracking is performed, one cannot be sure that the changed stimulus was foveated, predictably reducing detection rate as stimulus displays become more populous, obfuscating capacity estimation. A procedure akin to the game ‘Concentration’ is proposed in which observers search mosaics for matching stimuli, wherein only two stimuli are visible at any time. This procedure, unlike change detection, is conducive to modelling and simulation, enabling human performance data to be compared to spatial-memory-only models and slot simulations with specific storage capacities and recency/decay effects. In an example experiment, observers matched pairs of unfamiliar faces. Human performance exceeded that of an idealized spatial-memory-only model, being comparable to a simulated observer with a slot capacity of 1 to 2 items; furthermore, no long-term memory accumulation was found, evidenced by linear performance over time. Interestingly, increasing simulated slot capacity beyond 2–3 items had negligible impact on performance, suggesting that limited human VSTM capacity may underlie an optimally parsimonious neural implementation. The procedure proposed is engaging to participants (an inattentive approach would significantly extend trial duration, a disincentive to indolence). Rich data is produced, facilitating partial-match and time-course analyses, and enabling comparison to model and simulation results. The information unit in simulated analyses is the object-file, compatible with a dominant model of VSTM incorporating eye movements, trans-saccadic memory.