ABSTRACT
The necessary role of phonological computation in visual word perception is acknowledged today by most researchers investigating word recognition. Thus, in contrast to the dominant view of the early 1980s (e.g., McCusker, Hillinger, & Bias, 1981), it is by now well established that the computation of a phonological code from print occurs as a rule, regardless of subjects’ fluency, the difficulty of the reading material, or the experimental task (see Frost, 1998, for a review). Current models of visual word recognition diverge, however, with respect to the relative contribution they assign to prelexical phonological computation in modeling lexical access or reading aloud. Two opposing theories should be considered in this debate: the strong phonological theory and the dual-route theory. The strong phonological theory (e.g., Frost, 1998; Lukatela & Turvey, 1994a, b; Van Orden, Pennington, & Stone, 1990) posits that phonological representations are the principal building blocks of the mental lexicon. According to the non-neutrality assumption of the theory, the core lexical representations of the mental lexicon are phonological by definition (Frost, 1998, p. 75). This architecture necessarily dictates that fast phonological computation is the primary process launched mandatorily by the cognitive system for processing printed information. In contrast to this view, the dual-route theory suggests a lexical architecture in which an orthographic input lexicon plays a major role. Although the dual-route theory has generated various models in the last three decades (e.g., Coltheart, 1980; McCusker et al., 1981; Paap, Noel, & Johnasen, 1992; Zorzi, Houghton, & Butterworth, 1998), in this chapter I focus on the most modern version of the theory, the dual-route cascaded (DRC) model proposed by Coltheart and his colleagues (e.g., Coltheart, Curtis, Atkins, & Haller, 1993; Coltheart, et al. 2001). At first blush, it seems inappropriate to compare a verbal model, such as the strong phonological theory, to a computational model such as the DRC model. This is because a computational model makes many more specific assumptions than a verbal model, and thereby presents an additional step taken to test the validity of the verbal model. Our discussion therefore focuses on the general assumptions underlying the strong phonological view and dual-route view regarding the structural and dynamic properties of the mental lexicon.
