Abstract
Working memory performance increases as subjects learn to chunk regularly co-occurring visual information (Brady et al. 2009), and one account of this increased visual working memory capacity is that instance-specific detail is lost about the items in favor of a “content-free” representation of the chunk as a pointer to long-term memory (Huang & Awh, 2018). To test whether detailed instance-specific information persists in these kinds of tasks, we implemented a version of the Brady et al. (2009) study in which subjects can exploit regularities in color pairings to form chunks, but with the added manipulation that the colors within each pairing vary slightly in luminance from trial to trial. On each trial, participants were asked to remember 6 colors organized into three pairs occupying concentric circles. The training phase consisted of 7 blocks where the variance in luminance of the colors was incrementally increased each block and participants were probed on one of the color locations with a 2-AFC response, with the options differing only in hue so luminance was irrelevant. In the last two blocks, the test phase, participants were given a “detailed” probe that consisted of a 4-AFC, with choices of the target hue at two luminance values and a non-target hue at two luminance values. We find that working memory performance increased during the training phase, suggesting chunk learning occurred. But in contrast to the content-free view, during the test phase people still displayed knowledge of instance-specific luminance in the detailed tests at the same rate for chunked pairs as un-chunked pairs. This suggests that detailed, instance-specific information persists even when participants use chunks to structure a display in memory. We suggest that is because chunked items are stored in a hierarchical structure (Allen et al. 2021) rather than being replaced by content-free ‘pointers’ in memory.