Abstract
Working memory performance declines when items are stored over long intervals, suggesting that memories degrade over time. Here we examine the nature of this information loss from working memory. One possibility is that items gradually fade in quality, becoming less veridical over time. Alternatively, items may drop out of memory suddenly (termed sudden death). Recently, Zhang & Luck (2009) claimed to have found evidence for sudden death by showing that, over time, the number of remembered items decreases while the quality of those that remain is roughly constant. We considered an alternative model in which information is lost gradually over time, leading to failure once information about an item is depleted ("death by natural causes"). The inspiration for this model is the finding that working memories are variably precise (Fougnie, et al., 2012, van den Berg et al., 2012) and the possibility that poorly-remembered items are more likely to fail or are harder to retrieve. Our model draws on signal detection models where items are coded by independent samples that are averaged to reduce uncertainty. We propose that time-based changes in memory occur from volatility in stored information—samples are lost independently of each other over time, resulting in representations that gradually decay until they cease to exist. The moment a poorly-remembered item is forgotten, its removal improves the average quality of the remaining items, masking gradual decay and producing results that have been taken as evidence of sudden death. This model performed better than the sudden death model at describing changes in memory quantity, quality, and variability over time. Furthermore, the model can be fit at one time point and predict data at other time points by adjusting only the expected number of lost samples. Thus, we find that memories do not suddenly fail, but slowly degrade until death by natural causes.
Meeting abstract presented at VSS 2013