A parallel activation model with a sequential twist

Recognizing a word involves finding a match between a coded version of the input stimulus and an internalized lexical representation. A central issue in the investigation of this process concerns the mechanism by which a match is found. On the one hand, parallel activation models based on the Interactive Activation model (McClelland & Rumelhart, 1981) involve a system in which each letter increases the activation level of every word unit that contains that letter in the correct position, and decreases the activation level of every word unit that does not contain that letter in that position. The word unit that receives the most activation is the correct word unit. This approach essentially assumes that the input letter string is compared with every lexical representation in the lexicon simultaneously. The dominant examples of this approach are the Dual Route Cascaded Model (DRC) developed by Coltheart, Rastle, Perry, Langdon, and Ziegler (2001) and the Multi-Level Readout Model (MROM) developed by Grainger and Jacobs (1996). At the opposite extreme, search models assume that the comparison with the input is carried out one word at a time in a sequential fashion (Forster, 1976). In what follows, we will consider a third type of model that occupies a position intermediate between these two extremes, one that incorporates both parallel and sequential features.