Computational modeling and elementary process analysis in visual word recognition.

An attempt is made to isolate the assumptions that make a connectionist approach to visual word recognition distinctive. These include the commitment to distributed representations, the claim that there is no distinction between lexical and nonlexical systems in the naming task, and the claim that it is possible to map from orthography to meaning without using localized representations. It is argued that merely demonstrating that a network model can perform these tasks is not sufficient and that a detailed theory of how the network performs its tasks must accompany the simulation, because a simulation is not equivalent to an explanation. It is argued that further progress requires detailed modeling and experimental study of the elementary processes assumed to be involved in networks and that it is premature to dismiss alternative models of lexical access such as serial search models.