Consonants and vowels in orthographic representations

Abstract

It has been argued that orthographic representations—the abstract mental representations of the spellings of words—include orthography-specific information regarding the consonant/vowel (CV) identity of the individual letters that make up a word's spelling. This hypothesis has been used to explain the finding that the substitution errors in the spelling of certain dysgraphic individuals exhibit a striking tendency to preserve the CV identity of the target letters. In this paper, we evaluate the adequacy of two alternative hypotheses that do not posit orthography-specific CV representations. One hypothesis proposes that constraints on the nature of letter substitution come from the phonological representation of a word, and a second hypothesis contends that CV-preserving substitutions are driven by orthotactic knowledge—knowledge of the well-formed letter sequences in the orthography of a language. We present novel tests of these hypotheses using data from four case studies of dysgraphic individuals. The results clearly adjudicate in favour of the claim that orthographic representations contain orthography-specific CV information. In this way, the results support the more general claim that abstract categories are represented within the language system.

Acknowledgments

This research was supported by IGERT Research and Training fellowship for Adam Buchwald and NIMH grant R29MH55758 to Brenda Rapp.

Portions of this work were presented at annual meetings of the Academy of Aphasia (Vienna, 2003) and the Psychonomics Society (Vancouver, 2003). The authors would like to thank Dr Michael McCloskey for access to data of HE and CM, as well as for thoughtful comments on this research and an earlier draft of this paper. We would also like to thank two anonymous reviewers and Alfonso Caramazza for their comments on earlier versions of this manuscript. Finally, we thank BWN and RSB for their participation in this research.

Notes

Grapheme is used here to refer to the abstract representation of a single “orthographic” letter, such that the word SHE contains three graphemes (S, H, and E), but only two phonemes (/∫/ and /i/).

As noted, this representation synthesises components of previous proposals. In both Caramazza and Miceli's (1989, 1990) and Badecker's (1996) formulation, the orthographic CV tier is dominated by an orthographic syllable tier, which encodes the syllable boundaries in each word. The orthographic syllable representational level requires that orthographic CV information be represented. Because the existence of an orthographic CV representation is the question posed here, it seems premature to include a syllabic level in the present discussion. Additionally, Caramazza and Miceli represent geminates with a separate marker on a letter quantity tier such that a geminate letter has only one representation on the letter identity tier. In contrast, McCloskey et al. represent letter quantity on a tier that encodes position, such that a doubled letter has one entry on the letter identity tier and two corresponding entries on the letter position tier (see McCloskey et al., for a detailed discussion of the different predictions made by these representations). Because substitution errors maintain the appropriate number of letters, the issue of how letter quantity is represented is outside the scope of the present discussion (see also Tainturier & Caramazza, 1996). Finally, McCloskey et al. represent orthographic CV status on the same representational level as letter identity information. However, it is clear in their discussion that information regarding orthographic CV status is dissociable from information regarding letter identity. Thus, the present formulation is compatible with their representation.

There are two cases of mixed polygraphs in Italian, LB's (Caramazza & Miceli, 1990) native language: the cluster gi is a mixed polygraph representing the /d/ when it is followed by a front vowel, and the cluster ci is a mixed polygraph representing the sound/t∫/ in the same contexts. In principle, one could examine the substitutions on the ‘i’ of the Italian mixed polygraphs.

The term orthotactics is used in analogy to phonotactics, which encode the legal sequences and positions of sounds in a spoken language. Orthotactics, by contrast, encode the legal sequences and positions of letters in a written language.

Although a length effect is the hallmark of a graphemic buffer deficit, damage to other components of the spelling architecture could lead to decreased word accuracy (but not letter accuracy) for longer words. For example, in the case of a lexical deficit, spelling may rely on proper functioning of the phonology–orthography conversion system, which may have more opportunities for error in a longer word. However, there is no principled account for a length effect on letter accuracy in a lexical deficit. In contrast, damage to the graphemic buffer should lead to a length effect on letter accuracy, as longer words place greater demands on the buffer, leading to an increase in the proportion of errors.

All chi-square tests with 1 degree of freedom are continuity adjusted.

Note that word type frequency can also be thought of as a certain sort of letter token frequency, which is not weighted by word frequency.

It is worth noting that the CV preservation rates for the excluded items do not significantly differ from those for the included items, for any of the analyses discussed below.

The reason for this limitation was entirely pragmatic, but it should not bias the results against the orthotactic hypothesis for the following reasons. It is clear that there are different orthotactic constraints for different positions in words. For example, many consonant clusters are not available in word-initial and/or word-final positions, but are permitted word-medially (e.g., sd never occurs at the beginning or ends of words, but can occur word-medially, as in Thursday). In fact, others are available word-finally (e.g., tack) but not word-initially, and vice versa (e.g., wrench). Restricting the analysis to these word-initial and word-final positions should benefit the orthotactic hypothesis; because these positions are more constrained orthotactically, an orthotactic account should often favour preservation in these contexts.

Window size and placement are related the grain problem (see Townsend & Bever, 2001, for a description), which is the problem of determining the appropriate level of granularity over which variations are significant.

This argument does not directly bear on the OCV-PCV debate as framed in this paper, as each of these hypotheses represent the number of letters in a word on the letter identity tier. Therefore, both accounts represent the < sc > cluster as two letters in length.

It is worth noting that Monaghan and Shillcock (2003) do not simulate the selective deficits found in the patients' performance.

The proposal that orthographic CV information forms a part of the spelling process is not inconsistent with all connectionist implementations. For example, Glasspool and Houghton (unpublished) assume explicit CV representations in their connectionist account of CV identity preservation in spelling. Note, however, that this requires an explicit rather than an emergent representation of CV identity.