Abstract
In 2012, Brady, Konkle, Alvarez, and Oliva showed that long-term memories of real-world objects are not stored as monolithic units. Different features of studied objects (e.g., an exemplar and its state) are forgotten independently, suggesting these features are not stored as a single bound unit. Here we test another implication of this unboundedness: If different object features are stored independently, they might "swap", producing false combinations of true features, even in real-world objects. In Experiment 1 (N=20), observers studied 120 categories of objects, each represented by two exemplars. These exemplars could be shown in either the same state (two full mugs) or different states (one full mug, one empty mug). After studying the items, participants were given a forced-choice recognition test where they had to choose the correct states of both exemplars. We found that participants were sensitive to whether the objects were in the same state: they chose the same states for both exemplars if the studied exemplars were shown in the same states (mean: 67%), but not if the exemplars have been shown in different states (mean: 36%). Importantly, while in the former condition the probability of choosing both correct states was above chance (57% vs. 25% chance), the probability of correctly choosing two different states in the latter study condition was at chance (29%). Thus, people remember the states presented in exemplar pairs but often fail to bind them correctly. In Experiment 2 (N=20), participants studied only one exemplar from each category and had to recognize this exemplar later, either when it was presented in a new state or its original studied state. Performance was same regardless of test state (80% vs. 76%). Together, our findings of swap errors and state-independent exemplar recognition support the idea of independent feature storage for real-world objects.
Meeting abstract presented at VSS 2017