Transposed-Letter Confusability Effects in Masked Form Priming
Undoubtedly, at some processing stage, letters must be encoded with respect to their position within a string because otherwise it would be impossible to determine the relative ordering of letters and, hence, to distinguish between words like trial and trail. Thus, all models must have some way of doing this. Computational models of visual word recognition such as the interactiveactivation model (IAM; McClelland & Rumelhart, 1981), the activationverification model (AVM; Paap, Newsome, McDonald, & Schvaneveldt, 1982), and the multiple read-out model (MROM; Grainger & Jacobs, 1996) use a channel-specific scheme: Each letter is immediately assigned a position-specific channel and then is processed completely independently within its own specific channel. As a result, fist is no more similar to fits than it is to fire as all three words have identical letters in only two of the four letter positions. Thus, this coding scheme would have great difficulty explaining the presence of TL confusability effects. Furthermore, in this coding scheme, the lexical selection processes for, say, four-and five-letter words would be essentially independent. That is, the word hose would only activate the lexical representations for four-letter words; therefore, it would not activate the lexical representation for house. Such a prediction is clearly inconsistent with the available evidence (e.g., de Moor & Brysbaert, 2000; Humphreys, Evett, & Quinlan, 1990; Perea & Carreiras, 1998; Peressotti & Grainger, 1999).