Abstract
A central question for models of visual working memory is whether the number of objects remembered depends on object complexity. Early research found that observers could remember fewer complex objects than simple objects (Alvarez & Cavanagh, 2004). However, Awh et al. (2007) showed that despite worse performance at detecting subtle changes in complex objects, observers maintained the ability to detect large changes (e.g., changing a cube to a Chinese character). This was taken as evidence that observers could remember 3-4 complex objects, but that they had lower fidelity representations of these objects. Here we show that this method greatly overestimates capacity for complex objects because it does not account for ensemble representations. 100 observers participated in a change-detection task with displays of cubes and characters. They had to detect changes both within-category and across-category. We found that the dispersion of the items within a category (how close together the cubes were) was a major predictor of success on across-category changes but not within-category changes. This suggests that an ensemble representation was responsible for the estimates of 3-4 low fidelity cubes, rather than an individual item memory. Consistent with this conclusion, a second experiment with heterogeneous displays (mixed sets of cubes, Chinese characters, polygons, and Snodgrass objects) disrupted grouping and observers remembered fewer than 2 objects even with across-category changes (M=1.79, significantly lower than displays with grouping present, p<0.001). Thus, observers are capable of remembering individual information about only 1-2 complex objects. This is in direct contrast to claims of slot models that suggest observers always represent 3-4 objects and only fidelity is affected by object complexity. Our results show the necessity of understanding the representations observers form when estimating capacity: in particular, the importance of taking ensemble representations into account given their prevalence in working memory displays.
Meeting abstract presented at VSS 2013