Factors underlying masked priming effects in competitive network models of visual word recognition

Many models of reading assume that visual word recognition is driven by a competitive activation process. In these models, the effect of a masked prime is to manipulate the competitive process by shifting the balance between the target stimulus and its competitors. Computational modeling makes it possible to explore these effects in detail. To this end, the present chapter develops a precise description of the factors underlying masked priming effects in a specific competitive network model - the interactive activation (IA) model (McClelland & Rumelhart, 1981). Because the resulting expression is formulated in terms of standard psycholinguistic variables, the analysis presented here helps to bridge the divide between purely computational accounts and verbal theories of visual word recognition and priming. This approach is assisted by a framework for partitioning the set of competitors of a target stimulus, and a graphical technique for depicting the course of the competitive process in competitive network models of visual word recognition. The development of simple regression models that are able to fully capture the effects of priming within a complex (interactive, nonlinear, and dynamic) network model is a valuable outcome that has broader implications for the computational modeling of cognition. The analysis of priming effects in the model also leads to a number of predictions that can be tested empirically.