Abstract
A previously published computational model of the Attentional Blink (Wyble and Bowman, 2004) has given rise to predictions about the time course of the U-shaped T2 performance curve that is characteristic of the AB. Specifically, we posit that the blink curve is a hallmark of limitations in the temporal resolution of the binding of working memory tokens to types. In our model, lag-1 sparing results from a temporal window approximately 150 msec in length during which multiple items can be bound to the same token. Recovery of the blink occurs when a second token becomes available for binding.
In testing these predictions, we have performed two AB experiments. The first experiment compared the shape of AB curves for RSVP streams in the style of Chun and Potter (1995), presented at either 10 or 20 items/sec. If the AB is the result of temporal processes, the shape of the curve should be constant with respect to time. In our data, the blink curves had the same shape with respect to the time lag between T1 and T2, not the number of items. Specifically, in the faster presentation stream, we observed lag-2 sparing, maximal depth at lag 6 and recovery by lag 12.
Our second experiment involved presenting subjects with a letter pair for each of T1 and T2. If sparing involves binding T1 and T2 into a single conglomerate token, there should be a high percentage of binding errors at lag 1, in which one element of T1 is swapped with one element of T2. This prediction was confirmed, subjects made many binding errors at lag-1, and returned to baseline levels of error by lag-2.
This work details our continuing efforts to use the AB paradigm to explore the temporal structure of working memory. These data support our theoretical position that the AB stems from limitations in the consolidation of working memory tokens. We present these data in the context of a previously published connectionist model of the AB.