A werd is not quite a word: On the role of sublexical phonological information in visual lexical decision

Abstract

To establish the relative contribution of phonological and orthographic information to visual word recognition, we varied the instruction how to respond to the pseudohomophones in a Dutch lexical decision task. One participant group was asked to base their word/nonword decisions on spelling and therefore reject pseudohomophones together with the nonhomophonic nonwords; the other group had to base their decisions on phonology and therefore accept pseudohomophones together with real words. Rejecting pseudohomophones (ignoring phonological information) was accompanied by costs in speed and accuracy for the pseudohomophones but not for other items. Accepting pseudohomophones (ignoring orthographic information) led to speed and accuracy costs for pseudohomophones and nonhomophonic nonwords that were approximately ten times larger than those for rejecting pseudohomophones. The simultaneous costs for pseudohomophones and nonhomophonic nonwords contradict an explanation of pseudohomophone acceptance in terms of a postlexical spelling check. The results indicate that phonological information can be ignored much more easily than orthographic information. Therefore, they fail to support a primary role of sublexical phonological assembly in lexical decision. This conclusion was further supported by strong effects of phonological consistency that were found in a naming experiment but were completely absent in lexical decision.

The paper is based on experiments conducted during the first author's employment at the Nijmegen Institute for Cognition and Information. We are grateful to Herbert Schriefers, Anja Ischebeck, Kristin Lemhoefer, Marc Brysbaert, and Natasha Warner and several anonymous reviewers for helpful comments and discussions.

Notes

¹Of course, pseudohomophones are nonwords too. However, for the sake of simplicity we will use the term nonwords to indicate nonhomophonic nonwords.

²All item-statistics were based on type-counts of all monosyllabic Dutch words in the CELEX database(Baayen et al., 1993) with a frequency of at least 1 per million. The word-frequencies were based on form representations.

³After testing 20 participants in the reject condition, it appeared that there was a substantial subgroup that did not show phonological effects (7 out of 20 participants read pseudohomophones equally fast or even faster than nonwords). To test whether there would be systematic differences between participants who showed phonological effects and those who do not, we tested 20 more participants. However, all of the new participants read normal nonwords faster than pseudohomophones and we dropped that question. The statistical consequence of having unequal group-sizes for the reject and the accept condition is a decrease in power when comparing these groups (Maxwell & Delaney, 2000).

⁴The differences between the groups might be related to differences in reading ability, and several studies seem to indicate that phonological effects are larger in less skilled readers (e.g., Jared, Levy & Rayner, 1999; Seidenberg et al., 1996). Therefore, we repeated our previous analyses on a subset of 40 participants, that were matched in terms of the RTs on the first session, word RT:t ₁<1; t ₂(46) = 1.6; p=.108; nonword RT: t ₁<1; t ₂<1; word errors: t ₁(38) = 1.1, p=.299; t ₂(46) = 1.1; p=.283; nonword errors: t ₁(38) < 1; t ₂<1. The pattern of results was the same as the one reported above.

⁵Note, however, that there was no speed-accuracy trade off. The correlation between RTs and errors was. 67 (p=.035), indicating that response times and errors measure the same thing.

⁶We thank Natasha Warner for pointing out this explanation.

⁷Another approach was simply to exclude all feedback-inconsistent words from the analyses. The results with the remaining 47 feedback-consistent words were the same as in Table 4.

⁸Ziegler, Jacobs, and Klueppel (2001) suggested that pseudohomophones are subjected to a spelling check before they can be rejected in a lexical decision task. Note that our data and interpretation do not contradict this idea. We merely wish to point out that whatever spelling check is conducted is not responsible for the dramatic problems in accepting pseudohomophones.