An operator-based model of human covert visual attention

Over the last thirty years there has been much research into the nature of covert visual attention, which is the portion of visual attention that does not involve movement of the head or eyes. Surprisingly, though the role of covert visual attention has been shown to be diverse, most attentional theories are not able to explain more than one observed behavior. For instance, a theory of how attention is involved in identification may not be able to explain how attention is involved in search. Of the theories that do attempt to explain more than one attentional phenomenon, there are few that are specified to the extent that they might be considered computational. I have developed an operator-based computational model of human covert visual attention in two dimensions that is based on the Model Human Processor. My theory, which is called "NOVA" (Not Overt Visual Attention), is unique because it accounts for behavior in seven of the most significant classes of visual attention experiments from the psychology literature. These include precuing, crowding, decay, illusory conjunctions, search, counting, and detection experiments. NOVA task models are composed of operator traces. Each operator trace is composed of a linear sequence of operators, which are deliberate acts, such as a shift of attention or an act of recognition. Timing calculations are made by summing durations of component operator applications, perceptual dependencies and motor actions. Task performance is explained by operator traces in all seven experiments, and in five of the seven experiments timing performance is explained. In the other two experiments, timing was not presented as part of the experimental results. Since operators apply in approximately 50 msec, that is the limit to the timing resolution of NOVA. I have implemented NOVA and all of my task models in the Soar cognitive architecture. Soar is well suited as a platform for implementing a model based on the Model Human Processor, since it possesses many of the Model Human Processor's most important aspects. Implementation is a check of NOVA's specification and evidence that NOVA could be integrated with more complex models of behavior.