There goes the neighbourhood: Contextual control over the breadth of lexical activation when reading aloud

Abstract

There are currently two computational accounts of how the time to read pseudohomophones (like BRANE) and their nonword controls (like FRANE) varies with changes in context. In Reynolds and Besner's (2005) account, readers vary the breadth of lexical activation in response to changes in context. A competing account proposed by Kwantes and Marmurek (2007) and independently by Perry, Ziegler, and Zorzi (2007) has readers varying their response criterion in response to changes in context. The present work adjudicates between these two accounts by examining how the effect of neighbourhood density changes as a function of list context when reading pseudohomophones aloud. The results of an experiment and simulations from a leading computational model support the lexical breadth account, but are inconsistent with the response criterion account.

Keywords:

• Pseudohomophones
• Dual-route model
• Contextual control
• Reading aloud
• Phonological recoding

View correction statement:

Erratum

Acknowledgments

This work was supported by Grants 341586 to M.G.R. and AO998 to D.B. from the Natural Sciences and Engineering Research Council of Canada. We thank several anonymous reviewers and Peter Kwantes for constructive comments.

Notes

¹ The literature is certainly more complicated than has been detailed here (see Reynolds & Besner's, 2005, review). Here we have focused only on how the base-word frequency effect changes across context.

² The main focus of the present paper is Coltheart and colleagues' (2001) Dual-Route Cascaded (DRC) model. We have focused on the DRC model in order to maintain continuity with previous simulation work reported by Reynolds and Besner (2005) and Kwantes and Marmurek (2007). One disadvantage to using the DRC model is that testing the predictions of these two accounts is affected by failures of the DRC model. Initial simulations of the stimuli used in the experiment revealed that the DRC model did not yield a base-word frequency effect (BWFE) for these items. Although the failure to see a BWFE for these items is probably due to components of the DRC model that are independent of the two accounts of pseudohomophone reading aloud considered here, we felt that it was important to modify the DRC model so that it produced a BWFE for these items. We did this by speeding up the nonlexical route, which reduced the amount of noise feeding into the phonological lexicon. This parameter change did not affect any of the conclusions based on the simulation data.

³ Fodor (1983), the doyen of “modularity”, did not claim that such modules are not “penetrable” per se. Instead, the issue is the nature of the systems that can/cannot penetrate modules, and if they can, when (temporally speaking) they can do so.

⁴ That said, it should be noted that reducing the minimum naming parameter does not eliminate the base-word frequency effect, inconsistent with the null base-word frequency effect observed in mixed list conditions for skilled readers.

⁵ It is important to note that Kwantes and Marmurek (2007) do not make explicit predictions about how context will influence the N effect. Here we examine the predictions of their account in the DRC model on their behalf.

⁶ There is presently considerable controversy surrounding how to interpret item-level effects. Since Clark (1973), it has been widely held that psycholinguistic effects needed to generalize across subjects and items. More recently, this assumption has been questioned (e.g., Forster, 2008; Forster & Masson, 2008). For instance, recent work suggests that when items are matched on a number of variables (as in the present study), subject analyses are sufficient (e.g., Raaijmakers, Schrijnemakers, & Gremmen, 1999). This, of course, creates a problem for computational models of reading aloud that only simulate item-level differences.

⁷One reviewer pointed out that the relative size of different effects in the simulations does not match the relative sizes seen with skilled readers. We have previously referred to this as the “scaling problem” (see Reynolds & Besner, 2004). Although the focus of the current paper is about how an effect changes in magnitude across conditions and not the relative sizes of different effects, we do believe that the scaling problem provides an important diagnostic tool when evaluating computational models.