Orthographic Knowledge 1 an Investigation into the Structure and Acquisition of Orthographic Knowledge: Evidence from Cross-script Kanji-hiragana Priming

Numerous studies have demonstrated that orthographic knowledge is coded in an abstract format in English (e.g., the perceptually dissimilar words READ/read map onto a common abstract orthographic representation). However, it is unclear at present whether this mapping occurs at the letter or word level. Two experiments investigate this issue in a language (i.e., Japanese) where words can be written in perceptually unrelated scripts (Kanji and Hiragana), and crucially, where there are no letter correspondences between scripts. Using the long-term priming paradigm, robust priming was obtained when study/test words were depicted in Hiragana/Kanji, and vice versa. Furthermore, little priming was obtained following a study/test modality shift. The modality specific nature of this priming suggests that corresponding words in the two scripts share common orthographic representations. A model is out-lined that describes how abstract orthographic knowledge is acquired. orthographic knowledge 3 An investigation into the structure and acquisition of orthographic knowledge: Evidence from cross-script Kanji-Hiragana priming. A central issue for theories of reading concerns how letters and words are represented orthographically . That is, how written materials are perceptually coded within the visual system. According to most accounts, orthographic knowledge is coded in an abstract format such that different exemplars of a given stimulus map onto the same representation (e.g., Coltheart, 1981; McClelland, 1976). For example, the visual patterns A/a or READ/read are thought to map onto common orthographic codes, despite the perceptual dissimilarity of the items. This is assumed to be the case even for items that are arbitrarily related in their perceptual form, as in the examples above. Evidence for the existence of abstract orthographic codes comes from a variety of sources. Coltheart (1981), for example, describes a conduction aphasic patient who could not name pseudowords (e.g., nega), but who nevertheless was able to match upper/lower pseudowords that were perceptually dissimilar (e.g., NEGA/nega) without difficulty. Given that these items are (a) meaningless, (b) perceptually dissimilar in upper/lower case, and (c) unpronounceable by the patient, Coltheart concluded that the patient accessed abstract orthographic codes in order to perform the task. Consistent with this conclusion, McClelland (1976) reported that the word superiority effect (WSE) is equally large for words presented in case uniform and mixed conditions; for example the words FADE and fAdE were both better identified than the matched pseudowords GADE and gAdE in a task in which participants attempted to identify briefly presented targets. Given that mixed-case words are unfamiliar visual patterns, these results suggest that the WSE is mediated by word representations coded in an abstract fashion (also see Bowers, Bub, & Arguin, 1996). In addition, Bowers (1996) found long-term priming to be equally robust for words repeated in the same and different case, even though the different-case words were perceptually dissimilar at study and test (e.g., READ/read). This cross-case priming was attributed to orthographic knowledge, since the priming was dramatically reduced following a study/test modality shift in which words were studied auditorily and tested visually (for additional evidence in support the existence of abstract orthographic knowledge, see Besner, Coltheart, & Davelaar, 1984; Bowers, Arguin, & Bub, 1996; Evett & Humphreys, 1981; Rayner McConkie, & Zola, 1980, among others; but see Boles, 1992). Taken together, these findings strongly support the conclusion that orthographic knowledge is coded abstractly. Under this assumption, two related questions need to be addressed, namely, (a) where within the orthographic system are these abstractions learned (i.e., the letter or word level), and (b) how are these codes acquired. Clearly, answering (a) greatly constrains the solutions to (b), and accordingly, it is first necessary to determine the specific loci at which abstractions takes place. One possibility is that learning selectively occurs at the letter level, which would lead to abstract word codes, but only in virtue of letters serving as input to word representations. Alternatively, some learning procedure may operate at multiple levels, and accordingly, the existence of abstract word knowledge would not be contingent on, nor necessarily the product of, abstract letter codes.letter codes. orthographic knowledge 4 Current findings suggest that, at minimum, a learning process acts directly on letter representations to create abstract letter codes. Perhaps the best evidence in support of this conclusion is that the word superiority effect extends to words presented in mixed case letters (McClelland, 1976). Mixed-case words are unfamiliar (and unrecognizable) when considered as complete visual patterns, and the only familiar (and recognizable) features of the words are the component letters themselves. Thus, the WSE must reflect the activation of abstract letter codes, which in turn leads to the activation of abstract word codes (for additional evidence, see Besner & Jolicoeur, submitted; Rynard & Besner, 1987; for further discussion of this issue, see Bowers, Vigliocco, & Haan, in press). However, these and related findings cannot be used to determine whether learning is restricted to the letter level. The reason, of course, is that upperand lower-case words are composed of upperand lower-case letters, and accordingly, any demonstration that words are coded abstractly can be attributed to learning that took place at the letter level. In order to avoid this ambiguity, it is necessary to determine the structure of word representations under a specific condition, namely, when: (a) words can be depicted in dissimilar formats, and (b) the different formats do not share letter-by-letter correspondences. If these types of words map onto common orthographic codes, then it must be concluded that the learning process interacts with word level knowledge (since common abstract letter codes do not exist for these items). In fact, a study by Brown, Sharma and Kirsner (1984) provides some evidence compatible with this view. These authors assessed priming for Indian words displayed in Hindi/Urdu scripts: The spoken forms of these scripts are identical under normal circumstances, but their written forms are unrelated, and they do not share a simple letterby-letter correspondence. Employing the lexical decision task, they observed similar (nonsignificantly different) amounts of priming when items were studied and tested in the same (113ms) and different (93 ms) script conditions, consistent with the conclusion that Hindi/Urdu words map onto common orthographic word representations. It is important to note, however, that the authors did not compare the size of cross-script priming (which can be attributed to orthographic, semantic or phonological codes) to cross-modal priming (which can be attributed to semantic or phonological codes), and accordingly, the relative contribution of modality specific (orthographic) and modality independent (nonorthographic) codes cannot be determined. In fact, Kirsner, Dunn, & Standen (1989) attribute this priming to phonological representations. Two experiments are reported below that attempt to clarify this issue by comparing cross-script to cross-modal priming for words that do not share letter-by-letter correspondences in the two scripts; specifically, Kanji and Hiragana scripts of Japanese. Kanji characters (based on Chinese ideograms) represent morphemes, whereas Hiragana characters represent syllables, and the same set of words are depicted in both scripts. (Children tend to learn Hiragana first, and later Kanji.) In Experiment 1, test words were presented in Hiragana, and in Experiment 2, test words were presented in Kanji, providing an opportunity to assess cross-script priming in both the Hiragana-Kanji and KanjiHiragana study/test directions. See Figure 1 for a partial list of the experimental items used in Experiment 1. If robust cross-script priming is obtained in the face of minimal orthographic knowledge 5 cross-modal priming, then the priming would appear to be mediated by abstract orthographic word codes.1 The two experiments are described together.