Features As The Basic Storage Units In Visual Working Memory

An influential theory of visual working memory (VWM) asserts that VWM stores integrated objects, rather than independent features (Luck & Vogel,1997). Other evidences supporting this popular theory that visual items are stored as integrated objects in visual working memory have been suspected, except for the object-based benefit (Xu,2002a). Xu found that features belong to an object were better retained than features from spatially separated objects. With a change detection task, in Study 1, we tested whether there existed object-based benefit or not. In Experiment 1, the to-be-memorized items were mushrooms (features belong to different parts of an object, in Exp.lA) and colored triangles (features belong to the same part of an object, in Exp.1B), respectively. The results show that the object-based benefit can be reproduced when each to-be-memorized visual item occupies a unique spatial location during memory encoding. No such effect is obtained, however, when all items are sequentially presented at fixation during memory encoding.In Experiment 2 and 3, we excluded the potential confounds come from the task difficult due to the sequential presentation and illusory conjunction error. In Experiment 2, to-be-memorized items were presented sequentially, and manipulated the location of the probe to make location relevant or irrelevant. When the location was relevant, we found the object-based benefit. However, when the location was irrelevant, object-based benefit cannot be reproduced. In Experiment 3, we made the probe consisted of the features of to-be-memorized objects to further discuss the effect of illusory conjunction error during perceptual encoding. We found that, when to-be-memorized items were presented sequentially, the conjunction errors generated by them were equivalent. In summary, by eliminating the effect of task difficult, these findings suggest that the object-based benefit reported in previous studies may actually be the result of a location-based cost, possibly arising from failures to bind feature and location in visual working memory.Based on the findings in Study 1 and previous findings (Fougnie & Alvarez,2011; Wheeler & Treisman,2002) suggested that features may be stored in visual working memory in parallel, we conducted Study 2. In Experiment 4, two colors were assigned to either two, or six, to-be-memorized objects. The identity of these objects was defined by color, location, and shape. We excluded the effect of perceptual grouping by spatial and temporal separation in Exp.4A and Exp.4B, separately. We also added a verbal load task in Exp.4C to exam whether our results were influenced by verbal encoding. Finally, we found that participants’ memory performance was found to be unaffected by the number of objects, suggesting that the storage of color in VWM is dissociable from that of objects. In Experiment 5, the effect of perceptual grouping was evaluated by manipulating different levels of perceptual grouping. The results showed that there were no differences between different perceptual grouping levels. Experiments 6-7 extended the observation of color to the spatial frequency and the orientation.In Experiment 8-when required to memorize all features of the to-be-memorized objects-participants’ overall memory performance declined dramatically when the number of objects increased from two to six. This performance decline, however, was primarily due to memory failures on the feature dimension where the number of feature values exceeded the capacity limit of VWM. Experiment 9 further showed that, when the number of feature values was fixed at six on one feature dimension whereas varied between two, four, and six on the other feature dimension, the variation in CDT performance was mainly the result of memory failures on the feature dimension with varying number of feature values. In Experiment 10, we used a mixture design to exclude the effect of attention set (which make participants always attend the feature dimension that is easy to be memorized), and the results were the same as that in Experiment 8-9.Together, these findings provide convincing evidence that visual features can be stored in parallel in VWM and be bound together to form objects.