Masked phonological priming effects in English - Center for Reading ...

128 K. Rastle, M. Brysbaert / Cognitive Psychology 53 (2006) 97–145cost in terms of total orthographic activation, however, a cost which would have to be takeninto account in the reformulation of any lexical decision rule.Further analyses of the DRC model under this parameterization reveal another veryserious cost: These alterations to the model leave it unable to read aloud exception words.The DRC model used in Simulation 5 was presented with the 88 exception words and the88 matched regular words developed by Rastle and Coltheart (1999) for reading aloud.Though the model read aloud 85/88 regular words correctly, it read aloud 0/88 exceptionwords correctly. These errors comprised regularizations (69/88; e.g., ‘books’fi/buks/), lexicalizations(3/88; e.g., ‘tsar’fi/tai/), and other kinds of error (16/88; e.g., ‘aft’fi/ææt/).Given the massive contribution of assembled phonology required to simulate fast phonologicalpriming effects on lexical decision, it is not surprising that exception words posesuch difficulty for the DRC model under this parameterization.6.7. Simulation 6These simulations leave us in a difficult position. On the one hand, the DRC modeloperating under the standard set of parameters for lexical decision (Simulation 1) doesnot come close to simulating fast phonological priming effects. On the other hand, theDRC parameterization that shows some scope for simulating fast phonological primingeffects on lexical decision (Simulation 5) cannot correctly read aloud exception words. Ifour evaluation of the DRC model requires that it simulate fast phonological primingeffects on lexical decision using a set of parameters that can also read aloud, then the modelseems almost certainly false. An alternative approach would be to justify using theparameters developed in Simulation 5 for lexical decision while adopting the standardset of parameters (Coltheart et al., 2001; Rastle & Coltheart, 1999) for reading aloud.Indeed, Coltheart et al. (2001) used a parameter set for simulating lexical decision that differedvery slightly (on a single parameter) from that used for reading aloud—justifying thissmall parameter change as a strategic response to the specific demands posed by the lexicaldecision task.In Simulation 6, we investigated quantitatively whether we could offer a specific justificationfor adopting the parameter set used in Simulation 5 for lexical decision. Our logicwas simple. We reasoned that the lexical decision task requires readers to discriminatebetween word and nonword stimuli. As such, any strategic variation in the parametersused for lexical decision should maximize—not minimize—the model’s ability to performthis discrimination. If the parameters used in Simulation 5 are to be justified in terms of astrategic variation due to the demands posed by the lexical decision task, then words andnonwords should produce a larger difference in the sources of activation used to make alexical decision when the model is controlled by the parameters used in Simulation 5 thanwhen it is controlled by the standard parameters for reading aloud (Coltheart et al., 2001;Rastle & Coltheart, 1999). We therefore presented two parameterizations of the DRCmodel with the 112 word targets and the 112 nonword targets used in Experiment 1,and monitored the two sources of activation currently used to make a lexical decision(i.e., the total activation of the orthographic lexicon and the maximum activation ofany single unit; see Coltheart et al., 2001) for a 100 cycle period within each parameterization.If the parameters developed in Simulation 5 are to be justified for use in lexicaldecision on strategic grounds, then they should work to maximize the model’s ability todiscriminate between words and nonwords.