Abstract
The object-based view of working memory (Luck and Vogel, 1997) postulates that features of the same multi-feature object are remembered together. This would imply that irrelevant features are automatically encoded. Here, we tested this hypothesis directly. Stimuli were colored, oriented ellipses. Either color or orientation was designated to be the relevant feature throughout the experiment. Subjects viewed four stimuli and, after a delay, another four, where one item had changed in its relevant feature. Independently, one item (the same or a different one) had also changed in its irrelevant feature. In the first 30 trials, subjects were asked to locate the change in the relevant feature, but on the 31st trial, they were asked to locate the change in the irrelevant feature. Since the 31st trial suddenly makes the irrelevant feature relevant, there is no point in continuing the experiment thereafter. Thus, we get only one “irrelevant” trial per subject. To solve this problem, we crowd-sourced data from subjects on Amazon Mechanical Turk. Online subjects were able to recall the relevant feature, although not as well as lab subjects. Performance on “irrelevant” trials pooled across subjects was at chance, suggesting that the irrelevant memory was not retrieved. However, subjects might have discarded this memory right after seeing the second display. To address this, we used a delayed-estimation paradigm (Wilken and Ma, 2004), which does not have a second display. Online subjects viewed a single colored, oriented ellipse, and reported their estimate of its relevant feature on a continuum. Again, the 31st (last) trial was an “irrelevant” trial. Subjects’ performance on “irrelevant” trials was well above chance, but not as high as on “relevant” trials. This supports a weak form of the object-based view of visual working memory, namely one in which memory quality is weighed by relevance.
Meeting abstract presented at VSS 2015