Regularity and Length Effects in Word Naming: A Test of the Dual Route Cascaded Model

According to some dual-route theories of word naming, an inhibitory effect of length is caused by assembled phonology. Exception words are read via addressed phonology, and so should be immune from length effects. Analyses reported here show, however, that there is an inhibitory length effect for both regular and exception words, and this effect is no smaller for exception than regular words. The DRC, a computational implementation of dual-route theory, does not produce this pattern, instead showing a facilitatory length effect for exception words. This is partially due to the correlation between length and position of irregularity, but length effects are predicted to be smaller for exception words than regular words at all positions of irregularity by the DRC, contrary to the data. This suggests that length effects occur at the letter identification stage and a smaller length effect for words than nonwords may be due to lengthsensitivity of word superiority effects. The dual-route theory of reading (e.g., Monsell, Patterson, Graham, Hughes, & Milroy, 1992) proposes that there are two mechanisms for converting print to sound: a lexical route that retrieves known pronunciations (addressed phonology), and a non-lexical route that assembles pronunciations from grapheme-phoneme correspondences (assembled phonology). These mechanisms compete to produce the pronunciation of a written word. Regular words are read equally well by both routes; exception words are only read correctly by the lexical route; and nonwords are read by the non-lexical route. The dual-route cascaded (DRC) model of reading aloud (Coltheart, Curtis, Atkins, & Haller, 1993; Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001) is a computational instantiation of the dual-route theory. Its lexical route is an extension of the interactive-activation model of McClelland and Rumelhart (1981; Rumelhart & McClelland, 1982), which processes letters in letter strings in parallel, and its non-lexical route processes letter strings sequentially left-to-right with a set of grapheme-phoneme rules. According to Coltheart et al. (2001), the slowing of exception words relative to regular words is due to the conflict between the pronunciations generated by the rule-based and lexical routes. That is, exception words are read primarily by the lexical route, but regular words are read with a contribution from the non-lexical route. Moreover, in the DRC, length effects are the signature of the non-lexical route, as later letters influence the production of pronunciation by the nonlexical route slower due its seriality: the phonemes that they generate lag behind the other phonemes in activation, and are hence slower to reach threshold, delaying the pronunciation of the entire word. Since pseudowords are read entirely by the non-lexical route, length effects are predicted to be greater for pseudowords than words, as is indeed the case (Weekes, 1997). Given the balance between the two routes in this dual route model, one could (possibly simplistically) predict that whilst the influence of the sequential, non-lexical route will cause an effect of orthographic length for regular words, these effects will be absent or reduced for exception words, as their pronunciation is largely driven by the lexical route. Here, regression analyses were used to examine the predictions of the DRC for such an effect, and a comparison was made with the same effect in the Spieler and Balota (1997; Balota & Spieler, 1998) word naming databases.