Abstract
Long-term memory (LTM) can be surprisingly precise (Brady et al., 2013), even when compared to working memory (WM). However, the maximum precision of LTM remains unexplored. In order to quantify the upper bound on the precision of LTM, we looked at how precisely people can remember objects in LTM as we manipulated the amount of exposure to a particular object through repetition. We showed participants arbitrarily colored real-world objects. In the WM condition, they were shown a single object, then after a brief (1s) delay were asked to report the object's color on a color wheel. In the LTM condition, they saw sequences of 40 objects, each repeated either one or two times. Using a mixture model to extract "guess rates" and standard deviation as measures, we found that the WM condition was the most accurate (gr=1.9%; σ=15°), as expected. In LTM, participants had the least accurate memory for non-repeated objects (gr=28.3%; σ=23°), but this accuracy improved significantly for objects repeated twice (gr=18.5%; σ=19°; p< 0.005), demonstrating that repetition significantly improves the precision of LTM. In Experiment 2, we increased the number of repetitions of each object. Participants were shown blocks of objects repeated one or eight times, and reported the object's color after a distraction task. The no-repetition condition was similar to previous experiments (gr=50%; σ=21°). The eight-repetition condition's accuracy increased dramatically, to the level of a single item in WM (gr=4.5%; σ=13.6°; improvement: p< 0.001). These results show that repetition can be used to increase the accuracy of LTM. Repeating objects only eight times was sufficient to create representations of dozens of objects that were as precise as a single WM item. This demonstrates that the upper bound of the precision of LTM is still not fully understood.
Meeting abstract presented at VSS 2018