Abstract
The capacity of visual short-term memory (vSTM) was measured via a “flicker” technique, where an original and a modified image continually alternated, with a uniform gray field briefly appearing (120 ms) at each alternation. The images were arrays of outlined rectangles, half with one value of a feature (e.g., horizontal), and the rest a different value (e.g., vertical). In half the trials, one of the items changed its value; in the others, all items remained the same. Observers searched for changes in displays of 2, 10, or 18 items, with search slopes computed from the average reaction times for each set size. As the display time in each alternation cycle increases, search slope becomes proportional to display time, with the proportionality constant an estimate of vSTM capacity (Rensink, Visual Cognition 7, 345–376). Two display times were examined here (80 ms and 800 ms), with 12 observers used for each pair. For orientation changes, slopes in the short- and long-display conditions differed considerably, indicating that performance was memory-limited. Estimates based on the proportionality constant showed vSTM to have a capacity of 6 items, a value similar to that obtained using other techniques. However, for changes in size and changes in contrast polarity, no reliable difference was found between slopes for the short- and long-display conditions, indicating that performance was not memory-limited. Estimates based on the proportionality constant for these types of change showed vSTM to have a capacity of at least 14 items. These results suggest that individual items may not be the only type of structure stored in vSTM. Rather, it appears that some (although not all) simple features allow items to be grouped into a more global structure, or constellation, and that such constellations may also be basic units of visual memory.
Supported by Nissan Technical Center North America, Inc.