Neural network simulation of a discrete model of continuous effects of irrelevant stimuli

A general simulation method based on minimal neural network representations of nonmathematical, structural models of information processes is presented in some detail. That is, the time-dependent behavior of each component in a given structural model is represented by a simple, noncommittal equation that does not affect the theoretical structure of the original model. Special attention is given to the distinct ways in which the stochastic nature of response times can be represented in this way. The neural network simulation method then is applied to the linear stage model of Sternberg (1969), McClelland's (1979) cascade model, and to Miller's (1988) discrete model of continuous effects of irrelevant stimuli. In view of the results thus obtained it is argued that there exists a formal equivalence between discrete and continuous models of information processing in that both types of models can be represented as instances of the same neural network with variable thresholds. Furthermore, in the simulation of Miller's discrete model graded effects of irrelevant stimuli are obtained that are usually cited in support of a continuous flow model. The latter simulation study also shows, however, that Miller's model in its present form does not generate fast errors such as have been reported in the pertinent experimental literature. The main reason for this particular findings is that a sufficient number of irrelevant stimuli have to be processed by the decision stage before an erroneous response is given, whereas the processing of only a single target stimulus is already sufficient for the occurrence of the correct response. Both the extent of, as well as a possible remedy for, this problem are discussed.