Intrusion of orthographic knowledge on phoneme awareness: Strong in normal readers, weak in dyslexic readers

In three typical phonological awareness tasks it was found that children with normal reading development sometimes give responses that are based on orthographic rather than phonological information. In dyslexic children, the number of occurrences of such orthographic intrusions was significantly lower. This effect cannot be explained by positing a lower degree of orthographic knowledge in dyslexic children since a group of younger children who had the same spelling level as the dyslexics also showed more orthographic intrusions. A plausible explanation for this difference between normal and dyslexic readers is that, in normal readers, phonological and orthographic representations of words are so closely connected that they are usually coactivated, even if such a coactivation is misleading. In dyslexics this connection is less strong, so that orthographic representations interfere less with phonemic segmentation. The relevance of this finding with respect to recent assumptions about the importance of phonology in establishing orthographic representations is discussed. Phonological awareness tasks have become a stock-in-trade in reading research, and yet these tasks are still poorly understood. Since Bruce (1964) introduced the phoneme deletion paradigm ("Can you say TENT without the tuh?"), innumerable studies have used this and similar tasks (e.g., "How many sounds are there in the word TEN?"). In general, these studies have confirmed Bruce's original finding, namely, that increased reading competence is reflected in improved performance on phoneme awareness tasks. Thus, 5-year-old English-speaking children do relatively poorly on pho© 1996 Cambridge University Press 0142-7164/96 $9.00 + .10 Applied Psycholinguistics 17:1 2 Landerl et al.: Orthographic intrusions neme deletion tasks, but by age 9 they are at ceiling. Because the stimulus presentation in these tasks is auditory and in theory does not depend on reading ability, these tasks have been used with illiterates (Morais, Cary, Alegria, & Bertelson, 1979), nonalphabetic readers (Read, Zhang, Nie, & Ding, 1986), and prereaders. In these groups performance tends to be poor, and consequently debates have arisen about the mutual interaction of alphabetic knowledge and phoneme awareness (The Onset of Literacy, 1986; Segmental Analysis of Literacy, 1987). The phenomenon that, in skilled adult readers, orthographic information intrudes and interferes with performance on auditory tasks such as rhyme judgment was first shown by Seidenberg and Tanenhaus (1979) and Donnenwerth-Nolan, Tanenhaus, and Seidenberg (1981). These important studies suggested that, in skilled adult readers, phonological and orthographic information in words are closely connected, so that these two types of representation are automatically coactivated, even when this is not advantageous, as in a rhyme judgment task. The question then arose as to whether such a close link between orthographic and phonological representations also exists for beginning readers. Ehri and Wilce (1980) used a phoneme counting task with fourth graders where they compared stimulus words such as R-I-CH and P-I-TCH, which have the same number of phonemes but a different number of graphemes. They found that, just like the adults in Seidenberg and Tanenhaus's (1979) study, young readers' phonological judgment was based on orthographic representations. Thus, they heard more sounds in PITCH than in RICH, even though phonetically they have the same number of sounds. Subsequently, similar findings were reported even for first graders (Barron, 1994; Tunmer & Nesdale, 1982). Thus, we can conclude that there is early orthographic interference in phonemic awareness tasks. A recent study by Bruck (1992) examined the link between phonological and orthographic representations in dyslexics. Surprisingly, her findings suggest that, for dyslexics at all ages, this link is less strong than in normal readers. Bruck confronted normal and dyslexic readers with two conditions of a phoneme counting and a phoneme deletion task. One condition consisted of nonwords that contained the same number of letters as phonemes (e.g., TISK) and the other condition consisted of nonwords that contained one digraph so that these items contained more letters than phonemes (e.g., OTH contains three letters but only two phonemes). In accordance with other studies (Bradley & Bryant, 1978; Bruck, 1992; Manis, Szeszulski, Holt, & Graves, 1988; Olson, Wise, Connors, & Rack, 1990), Bruck found that, on the nondigraph items, dyslexics made significantly more errors than reading level matched children. The interesting finding, with respect to the link between phonological and orthographic representations, was that normal readers were more misled by the way the digraph nonwords would be spelled than were the dyslexics. On the phoneme counting task, the large majority (between 83 and 99%) of normal reader's errors were overshoots, that is, overestimations of the number of sounds in line with the number of letters. For the dyslexics, however, the percentage of overApplied Psycholinguistics 17:1 3 Landerl et al.: Orthographic intrusions shoots was considerably lower (between 51 and 77%). Similarly, on the phoneme deletion task, the normal readers showed a stronger tendency than the dyslexics to delete only one letter of the digraph instead of the complete phoneme (e.g., THOACE becomes HOACE). On the basis of these data, Bruck (1992) concluded that dyslexics did not use orthographic information to the same extent as normal readers, suggesting that there is an independence between orthographic and phonological codes. These findings are extremely interesting with respect to recent conceptions of the orthographic lexicon as a store of phonologically underpinned orthographic representations (Ehri, 1992; Perfetti, 1992; Stuart & Coltheart, 1988). Phonological underpinning means that there are close connections between graphemes and grapheme strings and the according phonological segments, so that phonology is always coactivated if a word spelling is presented. The evidence on orthographic intrusions in phoneme awareness tasks shows that, in the normal reader, coactivation also takes place the other way around, that is, orthography is coactivated when phonological judgments must be made. Interestingly, for dyslexic readers there seems to be less coactivation, presumably in both directions. However, other evidence somewhat contradicts Bruck's findings. Rack (1985), for example, suggests that dyslexics use orthographic information as a kind of compensatory strategy for their poor phonological processing and therefore rely more on orthographic information in their phonological judgment than do normal readers. In his study, dyslexic children needed more time to decide whether two auditorily presented words rhymed for word pairs that were orthographically dissimilar (FOOD-SHOULD) than for word pairs that were orthographically similar (GOOD-STOOD). Unexpectedly, the group of reading age control children in Rack's study did not show this effect. This is surprising in light of the findings of orthographic interference even in very young normal readers and leaves doubts about the validity of Rack's data. Campbell and Butterworth (1985) found evidence for the use of an orthographic compensatory strategy in a single case study of a highly educated, adult dyslexic subject. While performance on a variety of phonemic awareness tasks was still poor, the strategy the subject used appeared to be based on orthographic information. However, the subject that Campbell and Butterworth examined was highly literate and may therefore not be typical of dyslexics in general. Finally, Perin (1983) argued that she found evidence for the influence of orthographic knowledge on phonological awareness tasks for both normal readers and children with reading problems. Three groups of children (i.e., good readers/good spellers, good readers/poor spellers, and poor readers/ poor spellers) were asked to count the number of phonemes in mismatching words (which had more letters than phonemes), matching words (which had a one-to-one phoneme-letter match), and nonwords derived from the latter. Perin analyzed the children's performance on those items for which it was clear that they had an orthographic representation (i.e., they could produce a correct spelling) and found no reliable interaction between word Applied Psycholinguistics 17:1 4 Lander! et al.: Orthographic intrusions type and group. All three groups may have used orthographic knowledge in order to solve the task. However, an inspection of the error percentages shows some difference between normal children's and readingand spellingdisabled children's performances. While, for the good readers/good spellers, the error percentage for the mismatching words was considerably higher than for the other two word types (37% vs. 22%), there was less difference between the mismatching words and the other two word types for the poor readers/poor spellers (41% vs. 37%). In addition, Perin found that, on a spoonerism task, the normal readers snowed a stronger influence of orthographic knowledge than the children with reading and/or spelling problems. Normal readers gave more incorrect responses than the other two groups on those items where the initial phonemes were mapped by letter clusters rather than single letters (e.g., CHUCK BERRY became /bAk keri:/ instead of /bAk t/eri:/). Thus, these data confirm Bruck's assumption that there is less coactivation of the orthographic code in phoneme awareness tasks in dyslexic than in normal readers. In view of its relevance for theories of dyslexia, the present study attempts to explore further the extent to which orthographic information intrudes on dyslexic children's performance on phoneme awareness tasks. We made several methodological modifications to Bruck's (1992) study. First, a well-defined sample of dyslexic children was examined using a main comparison group of spelling level matched children. Spelling level is probably a better indicator of children's knowledge of orthographic patterns than reading level, since to be able to provide a correct spelling of a particular word in English, the exact letter-by-letter knowledge of the word is necessary. In addition, we required that the two groups of dyslexic and spelling level matched children were not only comparable on a standardized spelling test, but also able to produce the same number of correct spellings for the words that we used as stimuli in the phonological awareness tasks. Third, we used words rather than nonwords. Words of high frequency were used so that there was a good chance that the subjects would know the word spellings (i.e., have accurate orthographic representations). Further, we compared sets of rhyming stimuli that were similar phonologically but different orthographically (LORD VS. SWORD). And finally, whereas Bruck (1992) examined normal and dyslexic subjects' performance on items including a digraph, we expected our subjects to be misled by silent letters. A series of studies by Ehri and Wilce (1982) showed that silent letters are salient in memory representations of word spellings. Thus, we reasoned that overshoot errors might be readily provoked for words including a silent letter. Since our hypothesis predicts a lack of overshoot errors in dyslexics, it was important to create a strong temptation for such errors. However, we very carefully explained to the children, through the practice examples, that the tasks were about sounds and not about letters. Three typical, widely used phonological awareness tasks a phoneme counting task and two phoneme deletion tasks were administered. Each task consisted of a control condition, which included words that were phonologically transparent (e.g., HAM, HOT) but could contain regular diApplied Psycholinguistics 17:1 Landerl et al.: Orthographic intrusions Table 1. Mean scores of subject characteristics, with ranges in parentheses