High Capacity Visual Short Term Memory for Layout

Document Type

Article

Publication Date

12-2001

Digital Object Identifier (DOI)

https://doi.org/10.1167/1.3.124

Abstract

How much information is represented in a viewer's memory during scene perception? Change detection tasks probe visual short term memory (VSTM) and numerous studies document severe limits on memory, of about 3–5 items. However, with certain stimuli the capacity of VSTM can be quite high (e.g., Phillips, 1974; Rensink, 2000). These later results have been explained by assuming that a configuration of objects can be encoded as a single unit. We propose that the “single unit” can be better defined for these and other studies as a matrix of units—a matrix of locations that are filled by something. The change detection (VSTM) task was used to explore this idea with set sizes of 4 to 8 items. In all 3 experiments, performance was generally very high across set size. In Experiment 1, layouts of shapes were used and estimated capacity ranged as high as 7.0 shape-locations out of 8, as measured by a simple model in which filled locations are remembered correctly or guesses are made (see below for the model). In Experiment 2 the stimulus was a dollhouse with 36 objects divided into 12 units—12 locations that could be filled or unfilled. Performance was again high; estimated capacity ranged as high as 5.2 units or 15.6 objects. In Experiment 3, locations were filled by letters whose identity was either constant or varied between memory and test. Viewers again demonstrated high capacity memory for filled locations, even when identity changed. Capacity estimates ranged as high as 7.3 locations out of 8. Taken together, these results are consistent with the idea of a high capacity memory for a matrix of filled locations, where identity is not critical. This idea may be more definitive and more fruitful than the idea of “encoding as a single unit.” Model H = C/D+[(D-C)/(D*g)] ; H=hit rate, C=estimated capacity, D=# of items in display, and g=guessing (false alarm) rate.

Was this content written or created while at USF?

Yes

Citation / Publisher Attribution

Journal of Vision, v. 1, issue 3, art. 124

Share

COinS