Academic journal article Canadian Journal of Experimental Psychology

When Benefits Outweigh Costs: Reconsidering "Automatic" Phonological Recoding When Reading Aloud

Academic journal article Canadian Journal of Experimental Psychology

When Benefits Outweigh Costs: Reconsidering "Automatic" Phonological Recoding When Reading Aloud

Article excerpt

Skilled readers are slower to read aloud exception words (e.g., PINT) than regular words (e.g., MINT). In the case of exception words, sublexical knowledge competes with the correct pronunciation driven by lexical knowledge, whereas no such competition occurs for regular words. The dominant view is that the cost of this "regularity" effect is evidence that sublexical spelling-sound conversion is impossible to prevent (i.e., is "automatic"). This view has become so reified that the field rarely questions it. However, the results of simulations from the most successful computational models on the table suggest that the claim of "automatic" sublexical phonological recoding is premature given that there is also a benefit conferred by sublexical processing. Taken together with evidence from skilled readers that sublexical phonological recoding can be stopped, we suggest that the field is too narrowly focused when it asserts that sublexical phonological recoding is "automatic" and that a broader, more nuanced and contextually driven approach provides a more useful framework.

Keywords: automaticity, phonological recoding, computational models, reading aloud

Words with irregular or exceptional spelling-sound correspondences are read aloud more slowly than words with regular spelling-sound correspondences (e.g., PINT vs. MINT), provided that the words are lower frequency and that the irregularity occurs early in the word (e.g., Roberts, Rastle, Coltheart, & Besner, 2003). The best account of this "regularity" effect is provided by dual route theory (see Roberts et al., 2003, for evidence that parallel distributed processing (PDP) models do not simulate this effect).1 Computational models of the dual route theory are able to simulate the regularity effect (among many others). Currently, the two dominant computational models are the Dual Route Cascaded model (Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001), hereafter, DRC 1 .2 or simply DRC, and the Connectionist Dual Process model (Perry, Ziegler, & Zorzi, 2007), hereafter CDP+2.

The Lexical Route

When presented with a letter string, dual route models identify features that are present in each letter position, and these features activate letters at a letter level. Beyond the letter level, processing diverges along the two routes. Along the lexical route, the letter level activates the orthographic input lexicon (OIL), which contains a single lexical entry for the spelling of each word known to the model. Activation in the OIL then feeds forward to the phonological output lexicon (POL), which contains the stored pronunciation for each word known to the model. Activation from the POL then feeds into the phonemic buffer, where pronunciations are held in preparation for speech. The letter level, OIL, POL, and phonemic buffer are all engaged in interactive activation.

The Sublexical Route

Each model also implements a second route based on sublexical spelling-sound correspondences. In the DRC model, letters are converted sequentially into phonology according to a set of rules explicitly specified by the modeler. In the CDP+ model, sublexical translation is accomplished by a two-layer assembly (TLA) network that has learned these correspondences through exposure to print. The two models differ in other ways, but we dispense with a more thorough discussion, as it has no bearing here. In both models, the pronunciation produced by the sublexical route is passed to the phonemic buffer where it mixes with the pronunciation produced by the lexical route.

A Competition Account of the Regularity Effect

The regularity effect in these models arises because the sublexical route assigns the "regular" pronunciation to items like PINT (rhyming it with MINT), whereas the correct pronunciation is generated by the lexical route. These pronunciations compete in the phoneme buffer. For regular words like MINT, there is no conflict between routes because they activate the same set of phonemes. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.