Abstract
When a person attends to an object, a temporary working memory representation is created for it, specific to that individual “token” object. This differs from more general accumulated memories for “types” of objects. Kahneman, Treisman, and Gibbs (1992) conducted a series of “object-reviewing” studies showing that these token object representations are able to persist after focal attention is withdrawn from an object, and can be reactivated when the same object is reattended to shortly afterwards. Kahneman et al. (1992) used the term “object-files” to refer to these working memory representations. In the years since the publication of these seminal results many subsequent studies have further probed this type of memory. However, the question of the neural mechanisms which implement object-files is still largely an open one. In this thesis I attempt to examine these mechanisms, asking what is required computationally for such memory traces to persist and be reactivated, and how and why they might be implemented in the brain. My study has three distinct parts. Firstly, I present a computational analysis of the problem, the results of which provide number of avenues of further enquiry. Secondly, I present a neural network implementation of a mechanism, hypothesised in the previous analysis, that is able to account for results of the object-reviewing paradigm. Finally, I present the results of a series of human psychophysical experiments, investigating the relationship of object-reviewing performance with the ability to track multiple moving objects, first observed by Pylyshyn and Storm (1988). The results of these experiments would appear to have implications for the theoretical foundations of both object-reviewing and multiple object tracking research.