Two routes or one in reading aloud? A connectionist dual-process model.

Abstract

A connectionist study of word reading is described that emphasizes the computational demands of the spelling–sound mapping in determining the properties of the reading system. It is shown that the phonological assembly process can be implemented by a two-layer network, which easily extracts the regularities in the spelling–sound mapping for English from training data containing many exception words. It is argued that productive knowledge about spelling–sound relationships is more easily acquired and used if it is separated from case-specific knowledge of the pronunciation of known words. It is then shown how the interaction of assembled and retrieved phonologies can account for the combined effects of frequency and regularity–consistency and for the reading performance of dyslexic patients. It is concluded that the organization of the reading system reflects the demands of the task and that the pronunciations of nonwords and exception words are computed by different processes.

On the dominance of whole-word knowledge in reading aloud

“…It remains to be seen whether these results can be simulated by a different class of connectionist models, such as those advanced by Plaut et al (1996) and Zorzi et al (1998).…”

“…The difficulty with both the Plaut et al (1996) and the Zorzi et al (1998) models in the present context 3 is that, currently, there is nothing implemented that produces a repetition effect. As it stands, the training period of the models ends when errors reach a criterion level.…”

“…There are, of course, other implemented frameworks that have been advanced to account for naming performance and, more specifically, the regularity 3 frequency interaction (e.g., Plaut et al, 1996;Zorzi, Houghton, & Butterworth, 1998). Both of these accounts are dualroute.…”

Spelling–sound consistency effects in disyllabic word naming

“…The most influentid theories and computational models of phonologicd processing and word recognition prirnarily, or exclusively, address the processing of monosyilabic words (Coltheart, C u d s , Atkins, & Haller, 1993;Coltheart & Rastle, 1994;Plaut, McCleUand, Seidenberg, & Patterson, 1996;Grainger & Jacobs, 1996;Zorzi, Houghton, & Butterworth). Dud-route theonsts have developed a phonologicd processing mechanism that includes a set of grapheme-phoneme conversion rules which variables that influence the processing of these words.…”

How Do PDP Models Learn Quasiregularity?

“…In this state, the network's input-output mappings can no longer be viewed as nonlinear since the activation function mimics a linear function as its inputs approach zero, as shown by the thick solid part of the function. That is, for most inputs, the mapping to outputs becomes essentially a linear transformation.A model that is reduced completely to a linear map can still read most regular words, and while making errors on many exception words, it retains the ability to read nonwords, thus generalizing well (Zorzi et al, 1998). To further understand why a linear map behaves in this way, let us consider its general property.…”