Serial effects 1 Running head: SERIAL EFFECTS IN READING ALOUD Serial Effects in Reading Aloud can be Produced by Parallel Connectionist Models

Coltheart and Rastle (1994) and Rastle and Coltheart (1999) observe that their Dual Route Cascaded (DRC) model of single-word reading has a left -to-right grapheme-tophoneme correspondence mechanism, distinguishing it from connectionist models of word naming which possess no such serial mechanism. In the DRC an irregular grapheme to phoneme correspondence near the beginning of a word (e.g. chef), compared with one near the end of a word (e.g. glow), produces the most disruptive competing analysis to the early output of the lexical route, which takes time to generate its own output. Coltheart and Rastle showed the same behaviour with human subjects, and argued that this Position of Irregularity (PoI) effect is evidence against connectionist models of reading. Zorzi (2000) has shown that one connectionist model of naming, the Connectionist Dual Process model (Zorzi, Houghton & Butterworth, 1998), can simulate the PoI effect and he reanalyses the effect in terms of a confound between consistency and syllable position – a reanalysis that Rastle and Coltheart reject. We argue that the PoI effect can be expected to emerge from any simple connectionist network that takes fixation position into account, and we demonstrate that this is indeed the case with two computational experiments using different versions of such networks. We show that the PoI effect emerges most parsimoniously from a new model of word recognition, the split-fovea model, which is based on the precise vertical splitting of the human fovea.