Abstract
When people are asked to remember many items in visual working memory, performance suffers. Many prominent models claim this is because some items are unrepresented, or represented extremely poorly. For example, when asked to report all 6 items from a set size 6 display, participants appear to have little or no information about the last few items they report (Adam et al. 2017). We propose that this strictly item-based approach to performance is incomplete. Rather than thinking about each item independently, we show that when there are many things to remember, people encode items relative to each other and take advantage of the structure of the display. In particular, we reanalyzed data from previous whole-report visual working memory paradigms (N=22), but rather than looking independently at performance for each item, looked at the relationship between the items people reported. In such paradigms, people have to remember 2–6 items and report all of the items in any preferred order. We find that when the set size is high (>= 4 items), people tend to report the colors that are close to the mean color of the display first, and these first responses are systematically attracted toward each other (e.g., reported closer together in color space than the items truly had been). The number of items that are grouped depends on the entropy of the display (high entropy->more grouping). Examining the last items people report on a given display also reveals they are not random responses. Instead, the later responses (including the last) are systematically repelled from the earlier ones. Thus, rather than encoding items independently, memoranda are compressed by combining and separating items in relation to each other. Overall, we show that no memorandum is an island and each response reflects how that item fits into the whole picture.
Acknowledgement: NSF CAREER BCS-1653457 to TFB, Thai Red Cross Society to CC