There are several models of visual word recognition, which attempt to provide a theoretical account of the many empirical findings revealed by lexical decision tasks (for a review see Jacobs & Grainger, 1994). They all more or less agree on the following basic assumptions: A visually presented stimulus (i.e., a letter string) initially activates word representations in the mental lexicon that are orthographically and/or phonologically similar to the presented stimuli (Wagenmakers, Steyvers, Raaijmakers, Shiffrin, van Rijn, & Zeelenberg, 2004). If the letter string forms a word — or a pseudoword — activation increases over time. A "WORD" response is given, when the activation exceeds a certain criterion value. Grainger and Jacobs (1996) proposed a computational model of orthographic processing in visual word recognition (MROM), which is based on the interactive activation model, the prototype of all localist connectionist models, and includes a familiarity judgment mechanism as well (IAM; McClelland & Rumelhart, 1981).
As can be seen in the Figure above, Grainger and Jacobs (1996) propose three different criteria that can determine responses in a lexical decision task: The first criterion (M) is fixed and refers to the activation of a single lexical word node. When this criterion is reached by any of the word nodes, the stimulus is identified as a specific word, resulting in a "WORD" response. However, it has been proposed that correct lexical decisions can also be made without such lexical access to a certain word representation. This so-called first-pass judgment or fast-guess mechanism is generally said to be based on stimulus familiarity (Jacobs, Graf, & Kinder, 2003). The MROM therefore implemented a second, global criterion ( S ), which is based on a summed lexical activation over all word nodes and can be flexibly set. Thus, exceeding this criterion does not depend on identification of a particular word, which is why the criterion is specifically important for the lexical decision task. When the summed unit criterion is reached, again the "WORD" response is given. A "NONWORD" response is given when the third temporal deadline criterion (T) is reached before either the local or the global criteria are reached. In the MROM errors to word stimuli occur – i.e., words are falsely responded to as "NONWORD" – when the temporal deadline is set too low or both of the other criteria are set too high.