Abstract
Are the contents of visual memory influenced by a subject's task? Our previous work using natural tasks in virtual reality has shown strong task effects, coupled with minimal memory usage. Hollingworth (2003) has shown, however, that people can encode a great deal of information when viewing natural scenes. In this work we show that even when people encode a lot of detail there are still strong task effects. We conducted 2 experiments using a novel change-detection paradigm. In both experiments subjects completed 2 tasks; the first was to describe objects in a set of natural images, reporting a specific property of each object when a crosshair appeared above it, such as its name or dominant color. The other task was viewing a modified version of each scene, and detecting which of the previously described objects had changed. Subjects were split into separate groups, which described the same objects, but different aspects (name, or dominant color). In Experiment 1, subjects were told they were in a memory experiment and performed the change detection task for each image immediately after describing the objects in that image. In Experiment 2, subjects finished describing all images before being informed that we were studying memory and performing the change detection task. Surprisingly, subjects in both experiments showed comparable levels of overall change detection, even though in Experiment 2 the memory task was unexpected and memory contents were probed after a 1–5 minute delay. We also found task effects, but only in Experiment 2. When subjects described the color of objects, their ability to detect color changes was increased by 16%, whereas when they named the objects, their ability to detect object additions increased by 22%. We conclude that people automatically encode a surprising amount of visual details when viewing natural scenes, but the details encoded depend strongly on people's task, and their understanding of that task's overall demands.
We thank Ronald Rensink for donating the stimuli we used in this research. This work was supported by the University of California Academic Senate, San Diego Division, under grant RC142C-TRIESCH.