Journal of Experimental Child Psychology J. Experimental Child Psychology 87 (2004) 267–298 www.elsevier.com/locate/jecp Orthographic learning at a glance: On the time course and developmental onset of self-teaching David L. Share* Department of Learning Disabilities, Faculty of Education, University of Haifa, Mt. Carmel, 31905, Haifa, Israel Received 2 October 2003; revised 20 January 2004 Abstract Experiment 1 examined the time course of orthographic learning among Grade 3 children. A single encounter with a novel orthographic string was suﬃcient to produce reliable recall of orthographic detail. Moreover, newly acquired orthographic information was retained 1 month later. These data support the logistic learning functions featured in contemporary connectionist models of reading rather than a ‘‘threshold’’ model of orthographic learning. Experiments 2 and 3 examined self-teaching among novice readers. In contrast to the ﬁndings from less regular orthographies such as English and Dutch, beginning readers of a highly regular orthography (Hebrew) appear to be relatively insensitive to word-speciﬁc orthographic detail, reading in a nonlexical ‘‘surface’’ fashion. These results suggest fundamental diﬀerences between shallow and deep orthographies in the development of orthographic sensitivity. Ó 2004 Elsevier Inc. All rights reserved. Keywords: Reading; Word recognition; Orthographic learning; Decoding; Self-teaching; Reading acquisition Introduction The hallmark of skilled reading is the rapid and virtually eﬀortless recognition of printed letter strings. This ﬂuency depends, ﬁrst and foremost, on the acquisition of word-speciﬁc orthographic representations linked to phonological, semantic, * Fax: +972-4-824-0911. E-mail address: [email protected] 0022-0965/$ - see front matter Ó 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.jecp.2004.01.001 268 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 morphological, and syntactic information. The development of orthographic representations is a central issue in literacy research and practice. According to the ‘‘self-teaching’’ hypothesis (Jorm & Share, 1983; Share, 1995), the ability to translate unfamiliar printed words into their spoken equivalents (‘‘phonological recoding’’ or simply ‘‘decoding’’) is the central means by which orthographic representations are acquired. The self-teaching model proposes that each successful identiﬁcation of a new word is assumed to provide an opportunity to acquire the word-speciﬁc orthographic information that is the foundation of skilled visual word recognition. Relatively few (successful) exposures appear to be suﬃcient for acquiring orthographic representations for both adult skilled readers (Brooks, 1977) and young children (Hogaboam & Perfetti, 1978; Manis, 1985; Reitsma, 1983a, 1983b). From a self-teaching perspective, exhaustive letter-by-letter decoding (en route to a correct pronunciation) is assumed to be critical for the formation of well-speciﬁed orthographic representations because it draws a childÕs attention to the order and identity of letters. The self-teaching hypothesis assumes that nearly all printed word learning takes place when children independently identify unfamiliar letter strings encountered in the course of everyday reading. Not only are novice readers encountering large numbers of unfamiliar printed words on a daily basis (Nagy & Herman, 1987), but every letter string is at some point unfamiliar. The available evidence (reviewed in Share, 1995) indicates that neither direct instruction nor contextual guessing constitutes a viable means for identifying speciﬁc lexical identities and, hence, for orthographic learning in general. The self-teaching hypothesis proposes that only the ability to translate a printed letter string into its spoken form (i.e., phonological recoding) offers a reliable means of independently identifying new letter strings. By this account, phonological recoding acts as a self-teaching mechanism or built-in teacher, enabling a child to independently develop the word-speciﬁc orthographic knowledge necessary for skilled reading. Extending an experimental paradigm developed by Reitsma (1983a), the selfteaching hypothesis was ﬁrst directly tested by embedding novel target words in short stories (Share, 1999). These targets were simply novel letter strings (i.e., pseudowords) representing ﬁctitious names for places, animals, fruits, and the like. Grade 2 readers were asked to read aloud the test texts and then answer comprehension questions. Three days later, target spellings were identiﬁed more often, named more quickly, and spelled more accurately than were alternate homophonic spellings. A second experiment (Share, 1999, Experiment 2) found that viewing the target letter strings under conditions designed to minimize phonological processing significantly attenuated orthographic learning. A third experiment showed that this reduced orthographic learning was not attributable to alternative nonphonological factors (e.g., brief exposure durations, decontextualized presentation). These results supported the self-teaching hypothesis and indicated that four or even fewer exposures to a target spelling are suﬃcient for orthographic learning to occur. An alternative account of these data in terms of mere visual exposure to target spellings was rejected on the basis of both quantitative and qualitative ﬁndings (Share, 1999), indicating that successful orthographic learning was determined by D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 269 what the child said when identifying/decoding the target item and not what the child merely saw. The self-teaching phenomenon (i.e., rapid orthographic learning contingent on successful phonological recoding) has now been replicated among older normal and disabled Hebrew readers (Share & Shalev, in press) and also among North American English-speaking children (Cunningham, Perry, Stanovich, & Share, 2002). Testing a key prediction of the self-teaching hypothesis, Cunningham et al. (2002) reported a large correlation (r ¼ :65) between orthographic learning and the number of target stimuli correctly decoded. The current series of experiments addressed two issues related to the self-teaching model. The ﬁrst study (Experiment 1) explored the time course of self-teaching, that is, the rapidity (i.e., number of exposures) with which orthographic representations become established and the durability (i.e., retention over time) of these representations. Experiments 2 and 3 sought evidence of self-teaching among beginning (Grade 1) readers. Experiment 1: Parameters of self-teaching—exposure and durability This ﬁrst experiment was designed to reveal how quickly orthographic representations are established and also how durable these representations are. Several studies (Ehri & Saltmarsh, 1995; Reitsma, 1983a, 1983b, 1989; Share, 1999) have shown that among normal young readers, four or more exposures produce reliable orthographic learning. Hogaboam and Perfetti (1978) found reliable orthographic learning after only three exposures. Results for two exposures have been mixed (Reitsma, 1983a, 1983b, 1989). These data seem to suggest a ‘‘threshold’’ model of orthographic learning, with signiﬁcant learning occurring only after some threshold level of experience. In contrast to a threshold model, current connectionist learning algorithms predict signiﬁcant orthographic learning from the very ﬁrst learning trial (Harm & Seidenberg, 1999; Plaut, McClelland, Seidenberg, & Patterson, 1996). This is because changes in connection weights are directly proportional to the discrepancy between the current values and the target values. As weights are adjusted over the course of training, the magnitude of changes gradually diminishes toward asymptotic values. Thus, the most ‘‘powerful’’ single learning trial, according to the connectionist approach, is necessarily the ﬁrst trial, with progressively diminishing returns thereafter. Consistent with these connectionist predictions of single-trial learning are the ﬁndings from studies of repetition priming with pseudowords (Logan, 1988; Scarborough, Cortese, & Scarborough, 1977). For example, Logan (1988) included from 1 to 16 repetitions of pseudowords and words in a lexical decision task and found a strong initial decrease in reaction time, gradually decelerating over subsequent repetitions in the form of a classic power function typical of skill acquisition across a range of domains (Logan, 1988, 2002; Newell & Rosenbloom, 1981). Moreover, the beneﬁts of repeated presentation did not generalize to novel pseudowords and, 270 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 hence, were item speciﬁc (cf. Share, 1995).1 Nevertheless, the applicability of these ﬁndings to developing readers encountering new words in more natural reading contexts remains to be established given that all of this work has been carried out exclusively with highly skilled adult readers (typically university students). The current investigation examined the time course of orthographic learning by including single- and double-exposure conditions in addition to the standard fourexposure condition. Contemporary connectionist learning algorithms predict significant learning after only a single exposure to a new word. A threshold model, in contrast, predicts reliable orthographic learning only after a limited number of learning trials. With regard to durability, previous self-teaching studies obtained strong and clear-cut results when retention of orthographic information was tested 3 days after exposure to targets (Cunningham et al., 2002; Ehri & Saltmarsh, 1995; Reitsma, 1983a, 1983b; Share, 1999; Share & Shalev, in press). Hogaboam and Perfetti (1978, Experiment 2) reported signiﬁcant retention of newly acquired spellings 10 weeks later. To date, no investigation appears to have attempted to replicate such impressive long-term retention of orthographic information. The current study addressed the question of long-term durability by including a range of posttest retention intervals (3 days, 1 week, and 1 month). Method Sample and design A total of 36 Grade 3 children (15 boys and 21 girls, mean age 8.6 years), selected from two schools in a relatively advantaged area in Haifa, Israel, participated in a 3 3 2 design that consisted of three levels of target exposure (one, two, and four exposures), three posttest intervals (3-day, 7-day, and 30-day delays), and two alternate spellings (between-subjects). The order of the two repeated-measures factors (exposure and posttest interval) was fully counterbalanced. Thus, at each of three story-reading sessions (3-day, 7-day, and 30-day sessions), children read three stories (one-exposure, two-exposure, and four-exposure stories) for a total of nine stories. Half of the sample saw one set of spellings, whereas the other half saw the alternate spellings. Targets and texts The targets consisted of 9 of the 10 original pairs of pseudowords from the earlier Grade 2 study (Share, 1999) representing names in the following nine categories (one pseudoword name per category): animals, ﬂowers, fruit, cars, stars, coins, musical instruments, peoples (nations), and personal names. Target length ranged from two to four syllables and from three to ﬁve consonant letters (mean: 4.1). Each target letter string included two letters that represented a consonantal 1 Although the evidence from studies of repetition priming is convincing regarding single-trial learning when participants are performing the same task, transfer of learning across tasks (e.g., from pronunciation to lexical decision) is quite another story (Logan, 1990). D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 271 phoneme that could be transcribed by two alternate graphemes.2 This set of target items included four of the six homophonic grapheme pairs that exist in Hebrew orthography, with alternate letters located at all positions (from initial to ﬁnal) across target strings.3 Each target appeared either four times, twice, or once in short texts ranging in length from 94 to 170 words (mean: 126). All texts were fully pointed (i.e., contained vowel diacritics) and, hence, had near perfect one-to-one grapheme–phoneme correspondence (Feitelson, 1988; Navon & Shimron, 1984; Share & Levin, 1999). Posttest measures of orthographic learning Three measures of orthographic learning were administered either 3, 7, or 30 days after text reading. Spelling. This ﬁrst test of orthographic learning required children to reproduce the target spelling in writing. Naming. Children were next asked to name a series of words presented on a computer screen one at a time. The target spellings, both original and homophonic, were embedded in a longer list of 60 items designed to reﬂect the natural range and distribution of word frequency in childrenÕs reading material. Thus, several high-frequency function words appeared in the list several times, as in natural text. Each list contained all targets seen 3, 7, or 30 days earlier, together with their homophone foils. Each spelling (both target and homophone) was presented twice. To control for diﬀerential priming, the order of presentation in each list was target–homophone–homophone–target for half of the sample and homophone–target–target–homophone for the other half. This arrangement equates the total number of times each of the two alternate spellings is phonologically and orthographically primed. Each word was presented in fully pointed form in the center of a computer screen and remained visible until removed by activation of the voice key. The intertrial interval was 1000 ms. Orthographic choice. Each child was presented with both homophonic spellings of the target word. For half of the sample, one of the spellings was the original or 2 This study examined orthographic learning of purely consonantal graphemic information because Hebrew orthography is a consonantal alphabet in which (optional) vowel diacritics (or ‘‘points’’) have only a subsidiary status (appearing mostly below letters). This reﬂects the fact that in Semitic morphology, the semantic core of most content words is represented by a purely consonantal root, with vowel information conveying mostly grammatical inﬂections such as person, number, and gender. This diﬀerence between consonants and vowels is reﬂected not only in the speech patterns of native Hebrew speakers who often exchange stem-internal vowels in spoken word production (Ravid, 1995) but also in the fact that skilled readers have been shown to be largely insensitive to vowel identity (Shimron & Navon, 1981–1982), relying on direct recognition of consonantal roots (Bentin & Frost, 1995; Frost & Bentin, 1992). 3 There were several other cases of letters that could be transcribed by alternate graphemes (e.g., BET/ VAV), but in most of these cases, the alternate grapheme was very infrequent in young childrenÕs spelling. Thus, the probability of correctly reproducing the complete array of target letters was just under 50%. 272 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 ‘‘correct’’ target spelling seen in text; for the other half, the correct target was the alternate spelling. Items were presented in a slightly diﬀerent font from that used in the original texts. Procedure Testing was conducted on an individual basis in a quiet resource room adjacent to the Grade 3 classrooms. The basic testing procedure involved the reading of three texts followed 3, 7, or 30 days later by posttesting. There was an interval of at least several days between a posttest and subsequent text reading. In the ﬁrst text-reading session, the task was explained as follows: ‘‘I want you to read aloud some stories and tell me which one you liked best. Try to read them all by yourself. Be sure you understand the stories because IÕm going to ask you some questions afterward. Okay?’’ The only assistance given was with reading the title of the passage. No further help of any kind—neither praise nor corrective feedback— was given during story reading. Children who explicitly sought help in identifying a word were asked to try their best to read it by themselves. All story reading was tape-recorded. Following text reading, children were asked three factual questions that could be answered only on the basis of text content and not from general knowledge. Texts were read ﬂuently (mean time: 46 s), accurately (mean accuracy: 98%), and with good comprehension (mean: 76%). Posttests The three measures of orthographic learning were administered at delays of either 3, 7, or 30 days to determine the longevity of newly acquired orthographic information. These three tasks were administered in a ﬁxed order: spelling, naming, and orthographic choice. Spelling. Each child was ﬁrst asked whether he or she remembered the story about the fruit/town/ﬂower and then was requested to write this name. Every attempt was made to elicit the childÕs own representation of the target words. First, after reminding the child about the topic of a particular story (e.g., ‘‘Do you remember the story you read to me about the hottest town in the world?’’), the examiner asked the child to write the name of the fruit/town/ﬂower. If the child was unable to recall the name, the ﬁrst syllable was supplied. If this also failed to elicit the cued word, the target was then supplied in full. No attempt was made to praise, modify, or correct any written response. Naming. Immediately following spelling, the child was seated in front of a desktop computer and told that he or she was going to read aloud some words that would appear in the middle of the screen. The child was asked to read each word as quickly and accurately as possible. Twelve practice trials preceded the test list. Naming responses were recorded manually by the experimenter, who sat next to the child. Naming responses were also tape-recorded for later cross-checking. No feedback was given during word naming. D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 273 Orthographic choice. Following naming, the two alternate spellings of the target word were presented: ‘‘Here are two words that both look very much alike, but if you look carefully youÕll see that theyÕre diﬀerent. One of these words, and only one, is the same as the name of the town/fruit/ﬂower you read in the story 3/7/30 days ago. Make sure to look very carefully at each word and then tell me which one is the right one.’’ No corrective feedback was given in response to the childÕs choice. Locations of the two alternatives were switched from child to child. The order of presenting target sets was similarly varied across children. Statistical tests As in previous work (Cunningham et al., 2002; Share, 1999; Share & Shalev, in press), the dichotomous categorical (success/failure) data from the orthographic choice task and the spelling production measures were tested using binomial tests for the diﬀerences between proportions (including divergence from predesignated chance-level proportions of 50%) and were reported as z scores. Naming latencies (and accuracy) were tested using the conventional parametric tests (i.e., repeatedmeasures analysis of variance [ANOVA]). Results Identiﬁcation of targets during text reading Overall decoding accuracy for target word consonants (vowel errors were ignored) was very high (93%). This result aﬃrms the assumption that in a regular orthography such as pointed Hebrew, at least, readers normally apply their knowledge of symbol–sound relationships when confronted with an unfamiliar letter string rather than simply guessing or skipping these items. Spelling Results for spelling (homophonic) target letters appear in Table 1. Overall, the spelling data produced clear-cut evidence of orthographic learning. Levels of success in reproducing the target letters were well beyond chance (61%), z ¼ 5:40. Turning Table 1 Mean percentages and raw scores of target letters correctly spelled at posttest by Grade 3 children: Experiment 1 Number of exposures 3 days 7 days 30 days Overall 1 2 4 61 (44/72) 61 (44/72) 58 (42/72) 61 (44/72) 64 (46/72) 63 (44/70) 62 (42/68) 63 (43/68) 53 (36/68) 61 (130/212) 63 (133/212) 58 (122/210) Overall 61 (130/216) 63 (134/214) 59 (121/204) 61 (385/634) Note. Raw scores (in parentheses) in each cell represent the number of individual target (homophonic) letters correctly reproduced out of a possible 72. Each of the 36 children in Experiment 1 read one story in each of the nine conditions, with each story containing a single pseudoword target with two homophonic letters, hence a total of 72 (36 2) observations in each cell. (There are fewer than 72 observations per cell in cases where target letters were omitted in spelling productions.) 274 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 ﬁrst to the results for exposure (see marginal [column] means in Table 1), each of the three conditions was signiﬁcantly beyond chance, with z values of 3.30, 3.71, and 2.35 for one, two, and four exposures, respectively. It appears that even a single encounter with an unfamiliar word is suﬃcient to produce reliable orthographic learning. There appears to be little gain with additional exposures, z values < 1:0. Each of the three posttest intervals also yielded signiﬁcant orthographic learning, with z values of 2.99, 3.69, and 2.66 for 3-day, 7-day, and 30-day intervals, respectively. Remarkably, newly acquired orthographic information was retained up to 1 month later and was not signiﬁcantly inferior to spelling success at 3-day and 7-day intervals, z values for diﬀerences < 1:0. Although the number of data points from any one cell in this 3 3 design have questionable reliability,4 it is noteworthy that even the single-exposure result (62%) was marginally signiﬁcant at 30 days, z ¼ 1:94, p ¼ :052. Looking at the body of Table 1, it is evident that orthographic learning in the various conditions was surprisingly uniform, with the largest absolute diﬀerence in either exposure conditions or posttest delays being a mere 4%. The ﬁndings also generalized across all but one of the pairs of homophonic letters. With the exception of the ALEF/AYIN (glottal stop) pair with an overall success rate (summing over the nine conditions) of 57% (p ¼ :065, one-tailed), each of the remaining three homophone pairs was spelled at a rate signiﬁcantly beyond chance (CHET/CHAF, 73%; KAF/KUF, 61%; and TET/TAF, 58%). The spelling data demonstrate that orthographic learning is both rapid and robust. Naming accuracy Overall naming accuracy was 76% for both targets and homophones. A 3 (exposure) by 3 (posttest interval) by 2 (target/foil) repeated-measures ANOVA failed to reveal any signiﬁcant main eﬀects or interactions, all F values < 1:0. As can be seen from Table 2, naming accuracy was fairly similar across number of exposures and across posttest intervals. Naming latencies Vocalization latencies less than 300 and more than 3000 ms were ﬁrst discarded, and then mean reaction times were calculated for correct pronunciations of targets and foils with missing values replaced by individual subject means. These data are summarized in Table 3. Overall, there was a small but nonsigniﬁcant 16-ms advantage for targets, F ð1; 35Þ ¼ 1:73, p > :05. As with the accuracy data, there were small and inconsis4 Some indication of the power of these binomial tests can be gauged by asking what degree of deviation from a 50% chance level is required to attain signiﬁcance with an alpha of .05 (two-tailed). For the spelling data in Experiment 1, the ﬁgures are 3% (i.e., more than 53% or less than 47%) for the grand mean (N ¼ 634), 7% for the marginal means (n ¼ 212), and 13% for the individual cell means (n ¼ 72). For the orthographic choice data, the ﬁgures are 6% for the grand mean (N ¼ 318), 10% for the marginal means (n ¼ 106), and 19% for the individual cell means (n ¼ 36). D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 275 Table 2 Mean percentages and standard deviations of posttest naming accuracy of targets and foils by Grade 3 children: Experiment 1 Number of exposures 3-day interval 7-day interval 30-day interval Overall 1 Target Homophone Target Homophone Target Homophone 80 79 77 81 74 77 81 75 79 79 62 77 71 66 74 68 82 77 77 73 77 76 74 77 Target Homophone 77 (38.0) 79 (36.2) 2 4 Overall (36.8) (35.4) (37.4) (34.9) (40.9) (39.4) (32.3) (37.4) (37.2) (32.8) (43.4) (37.4) 75 (38.3) 77 (35.5) (40.3) (39.8) (37.4) (41.1) (34.6) (35.0) 76 (37.3) 71 (38.5) (36.6) (37.6) (36.9) (36.4) (39.9) (36.6) 76 (37.7) 76 (36.8) Note. Standard deviations are in parentheses. Table 3 Means and standard deviations of posttest naming latencies (in milliseconds) of targets and homophone foils by Grade 3 children: Experiment 1 Number of exposures 3-day interval 7-day interval 30-day interval Overall 1 Target Homophone Target Homophone Target Homophone 772 800 778 784 797 808 814 843 789 753 796 783 766 800 771 807 767 810 784 814 779 781 787 800 Target Homophone 782 (215) 797 (237) 2 4 Overall (201) (184) (183) (275) (258) (242) (234) (257) (231) (201) (193) (156) 800 (218) 793 (210) (191) (203) (192) (225) (205) (249) 768 (194) 806 (224) (219) (216) (202) (235) (219) (218) 783 (213) 799 (223) Note. Standard deviations are in parentheses. tent diﬀerences favoring the target in most cases, but none reached signiﬁcance. A 3 3 2 repeated-measures ANOVA revealed no reliable main eﬀects or interactions. It would seem that in the context of the current paradigm with relatively few trials (two for each target and foil) and considerable intralist priming (both phonological and orthographic), naming latency might not be the most sensitive measure of orthographic learning. In previous studies with this same paradigm (Share, 1999; Share & Shalev, in press), naming latency was consistently found to be the least reliable of all three measures of posttest orthographic learning and also the least sensitive.5 5 In Share (1999, Experiment 1), latency diﬀerences between targets and foils were signiﬁcant in the four-exposure condition but not in the six-exposure condition (there were no accuracy eﬀects). In Experiment 2, there were no signiﬁcant eﬀects in either latency or accuracy, yet the other two measures (spelling and orthographic choice) revealed signiﬁcant orthographic learning. Finally, in Experiment 3, signiﬁcant diﬀerences were found in accuracy but not in latency—precisely the opposite pattern to that observed in Experiment 1. 276 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 Table 4 Mean percentages and raw scores for posttest orthographic choices by Grade 3 children: Experiment 1 Number of exposures 3-day interval 7-day interval 30-day interval Overall 1 2 4 56 (20/36) 75 (27/36) 72 (26/36) 75 (27/36) 75 (27/36) 72 (26/36) 65 (22/34) 71 (24/34) 74 (25/34) 65 (69/106) 74 (78/106) 73 (77/106) Overall 67 (73/108) 74 (80/108) 70 (71/102) 70 (224/318) Note. Raw scores are in parentheses. Orthographic choice Results for the orthographic choice task appear in Table 4. The overall level of correct orthographic choices (70%) was well above the 50% chance level, z ¼ 7:28, p < :001. At each of the three exposure conditions (summing over posttest intervals), choices were also signiﬁcantly greater than chance, with z values of 2.33, 4.08, and 3.89 for one, two, and four exposures, respectively. Replicating the spelling data, a single encounter with the target string produced reliable orthographic learning.6 With an additional exposure, there was a small but nonsigniﬁcant gain, z ¼ 1:33, whereas all other comparisons were nonsigniﬁcant, z values < 1:0. This picture of rapid initial orthographic learning with small but often nonsigniﬁcant increments accruing at additional exposures accords well with the earlier Grade 2 ﬁnding (Share, 1999) of small nonsigniﬁcant gains in orthographic learning from four to six exposures. Together, the spelling production and spelling recognition data suggest that the most decisive point in the process of assimilating orthographic informa6 It might be queried whether the orthographic choice data can be used to support the claim that a single exposure to a novel letter string leads to signiﬁcant orthographic learning. Because the orthographic task was administered immediately after the naming task in which targets and their homophonic spellings were each presented twice, it might be argued that the orthographic choice posttest actually took place after a minimum of three exposures to the target rather than just one exposure. However, it is important to note that in the naming task, not only were both homophonic spellings of the target presented, but order was counterbalanced across participants. In these circumstances, it is assumed that any advantage gained by seeing the original target spelling in the naming task is oﬀset by equal exposure to the homophonic foil. That is, the two exposures ‘‘cancel each other out,’’ as it were, leaving the initial exposure as the deciding factor in the orthographic choice. Looked at another way, one can think of the ﬁnal orthographic choice task as evaluating the net orthographic ‘‘gain’’ of three exposures (in the case of the original spelling) compared with only two exposures to the alternate homophonic spelling, with the diﬀerence again being a single exposure. Most important, the data from several studies employing the self-teaching paradigm support this working assumption. In the 1999 study with late Grade 2 readers (Share, 1999, Experiment 1), the orthographic choice task was the ﬁrst of the same three posttests with 74% correct choices. The Cunningham et al. (2002) replication used the same posttesting procedure (also with late Grade 2 readers) and produced a ﬁgure of 75%. The current study (Grade 3) with orthographic choice last (rather than ﬁrst) yielded the same success rate (75%). Orthographic choice was also the last of the three posttests in Share and ShalevÕs (in press) study that included a group of normal readers in Grades 4, 5, and 6, once again with the same ﬁgure of 75% obtained when the choice task followed the spelling and naming tasks. All of these studies support the working assumption that the order of these posttests makes little diﬀerence with regard to the outcomes for orthographic learning. Nevertheless, it must be conceded that, in the strictest sense, only the spelling data can be taken as unambiguous support for the claim regarding one-shot (‘‘logistic’’) orthographic learning. D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 277 tion is the very ﬁrst encounter with a novel word, in contrast to earlier developmental work that suggested that orthographic learning occurs only after a certain threshold number of exposures. Pooling across the three exposure conditions, the data from the three posttest intervals produced essentially the same pattern of results. Orthographic memory was signiﬁcantly greater than chance and equally robust at each interval, with z values of 3.66, 5.00, and 3.89 for 3-day, 7-day, and 30-day delays, respectively. Consistent with the spelling data, orthographic information not only was retained 1 month later but also was not inferior to performance at either the 3-day or 7-day posttests, with both z values less than 1.0. The small diﬀerence between the 3-day and 7-day data was not signiﬁcant, z ¼ 1:48. As was the case with the spelling data, the result for a single exposure at 30 days (65%) approached signiﬁcance, z ¼ 1:71, p ¼ :083. Repeated-measures ANOVA revealed no interaction between exposure and delay, F < 1:0, indicating that the rates of successful orthographic choices were consistent across exposures and posttest intervals. In sum, the orthographic choice data, like the spelling data, suggest that orthographic learning is both rapid and robust. Discussion The rapid single-trial learning revealed in the spelling and choice tasks is consistent with the learning algorithms featured in contemporary connectionist models, whereby the greatest changes in connection weights occur earlier on in training because weight changes are directional proportional to the discrepancy between current weights and target weights (Harm & Seidenberg, 1999; Plaut et al., 1996). As weights are adjusted in the course of training, weight changes gradually diminish toward asymptotic values. Thus, the strongest single learning trial is necessarily the ﬁrst trial, with progressively diminishing returns thereafter, as in the current case of orthographic learning. Thus, current instantiations of connectionist models with logistic learning functions ﬁt the data better than does a threshold model positing signiﬁcant learning only after some threshold level of experience. It would clearly be of interest to map the learning function more completely by including a wider range of exposures. The current study included only three levels of exposure and found no signiﬁcant diﬀerences among the three. Although performance at two exposures was numerically superior to that in the one-exposure condition, simple pairwise comparisons found no signiﬁcant gains after a single exposure on either choice or spelling, and the data for four exposures did not show any reliable trend toward further improvement. In the 1999 Grade 2 study, comparison between four and six exposures, for both choice and whole-word spelling (but not target letter spelling or naming latencies), produced nonsigniﬁcant gains from four to six exposures. When the results for a single exposure in the current study (65%, 69/106) are compared directly with the six-exposure result in the 1999 study (78.5%, 157/200), there was a signiﬁcant diﬀerence for orthographic choice, z ¼ 2:56, p < :01, but not for target letter spelling (66 vs 61%, z ¼ 1:18). 278 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 The lack of reliable diﬀerences for additional presentations of target items suggests the possibility that orthographic learning may be more context speciﬁc than item speciﬁc. In all of the studies of orthographic learning discussed previously, target stimuli were presented within one and the same context such as a sentence or passage. This suggests the possibility that frequency of exposure within a single coherent context may be less critical than presentation across multiple contexts or occasions. Alternatively, if learning is totally item speciﬁc, each individual encounter with a novel letter string, whether in a single context or in multiple contexts, would be treated as an equivalent learning trial. If the developmental data on orthographic learning are found to ﬁt the item-speciﬁc logistic learning function posited by the connectionist models, such a ﬁnding would subsume the printed word learning data within the broader context of skill learning. Growing proﬁciency in practiced tasks across a range of skill domains is well described by a power function, with initially large gains diminishing with further practice (Logan, 1988; Newell & Rosenbloom, 1981). As an example of skill acquisition, the current orthographic learning data are well accommodated within the framework of LoganÕs (1988, 2002) instance theory, according to which single-trial item-speciﬁc learning follows from the claim that ‘‘attending to a stimulus is suﬃcient to commit it to memory. ... Subjects store and retrieve representations of each individual encounter with a stimulus’’ (Logan, 1988, p. 501). The accumulation of separate episodic traces with experience is taken to explain the transition from eﬀortful algorithmic processing to memory-based processing proﬁciency—in the current context, the transition from word identiﬁcation through ‘‘algorithmic’’ grapheme– phoneme translation of an unfamiliar letter string to word recognition of the now no longer unfamiliar string. Behaviorally as well, more ‘‘eﬀort’’ appears to be invested in decoding at the very ﬁrst encounter with a novel string. When children read aloud the text passages in this study, the letter-by-letter sounding out and blending observed on the initial encounter with a new target would typically be replaced by a smooth uninterrupted pronunciation by the second or third exposure. The online dynamics of the decoding process would seem to be a valuable topic for future investigation. Despite a huge research literature demonstrating the central role of decoding in reading development, there has been surprisingly little systematic documentation of precisely how developing readers arrive at, or fail to arrive at, the correct spoken form of a printed word and what the implications are, if any, for the construction of new orthographic and lexical knowledge. Nearly all researchers have been content to rely on measures of simple accuracy or vocalization onset latency—measures that are easily gleaned but surely rather crude superﬁcial measures of what is likely to be a complex process (cf. Berent & Perfetti, 1995). Clearly, there are also practical implications of single-trial learning. If ‘‘ﬁrst impressions’’ are indeed the most potent, a decoding (or spelling) error on the very ﬁrst attempt at a new word should be more detrimental to long-term orthographic learning than should an error committed at a later point. The common classroom practice of ignoring spelling errors in the early written products of beginning readers (when the primary focus of learning and instruction is the acquisition of the alphabetic D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 279 principle and/or communicative intent) suggests that greater eﬀort might need to be expended later to alter faulty orthographic representations created at the initial encounters with novel words. An important point about the self-teaching hypothesis is that pronouncing an unfamiliar letter string necessitates a type of processing that is uniquely conducive to the development of well-speciﬁed orthographic representations. The exhaustive letter-by-letter processing, often involving more than a single ‘‘pass’’ along the letter string, may be one of the key beneﬁts of phonological recoding. Furthermore, because processing is generally more exhaustive and eﬀortful at the ﬁrst exposure, this may result in a more powerful learning experience. It would be of interest to systematically manipulate the nature of decoding depth and exhaustiveness. In the third experiment (Share, 1999), decoding depth (and accuracy) was manipulated by reducing exposure time to only 300 ms compared with self-paced (i.e., unlimited) exposure in Experiment 1. Not surprisingly, time-limited exposure produced a lower rate of decoding success and also signiﬁcantly attenuated orthographic learning. It seems safe to assume that processing was also less thorough in this particular manipulation. The retention data also proved to be impressive, with clear-cut retention even at 30 days. Even 1 month later, the strength of orthographic memory was not signiﬁcantly poorer than after 3 days. These data corroborate the earlier ﬁndings of Hogaboam and Perfetti (1978), who reported signiﬁcant retention at 10 weeks. However, it might not be entirely pertinent to ask how long orthographic information is retained given that reading, as an everyday activity for most literate individuals, involves recurrent exposure to printed words of varying frequencies. If Nagy and Herman (1987) are correct in estimating that children are exposed to millions of printed words each year, this implies that even rare words are appearing often enough to refresh decaying representations. Hence, the question of how long newly acquired orthographic information is retained may be somewhat misconstrued. It is interesting to note that for both choice and spelling, performance at 7 days was actually superior, although not signiﬁcantly, to performance at 3 days, with the hint of a decline at 30 days. This seemingly inverted-U function was certainly not anticipated and suggests that orthographic memory is consolidating over the ﬁrst week before declining. Of course, these observations remain entirely speculative, based purely on nonsigniﬁcant trends. Nevertheless, the possibility of an invertedU function in long-term retention may be worth examining in future research. Experiment 2: Early onset—self-teaching among beginning readers? The second experiment reported here tested the hypothesis that beginning reading is beginning self-teaching (Share, 1995). A growing number of studies have suggested that some rudimentary self-teaching skills, perhaps suﬃcient to establish primitive orthographic representations of the kind discussed by Perfetti (1992), may exist at the very earliest stages of learning to read, even before a child possesses any decoding skill in the conventional sense of being able to sound out and blend simple pseudowords (Ehri & Wilce, 1985, 1987; Morris, 1992; Stuart & Coltheart, 1988). This view 280 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 runs counter to many prevailing accounts of printed word learning that propose an initial logographic or visual stage prior to later developing phonological recoding (see, e.g., Frith, 1985; Gough & Juel, 1991). This early self-teaching is assumed to depend on three factors: letter–sound knowledge, some minimal phonological sensitivity, and the ability to use contextual information to determine exact word pronunciations on the basis of partial decodings. Ehri and others (Ehri & Sweet, 1991; Ehri & Wilce, 1985, 1987; Rack, Hulme, Snowling, & Wightman, 1994; Scott & Ehri, 1990; Treiman & Rodriguez, 1999; Treiman, Tincoﬀ, & Richmond-Welty, 1996) have demonstrated that even kindergarten children are capable of learning words on a phonetic basis rather than a visual basis provided they have some knowledge of letter–sound or letter–name relations. For example, knowledge of the names of the letters J and L may enable a child to read the word JAIL even in the absence of blending skill. However, a partial decoding strategy based solely on letter–sound knowledge is in itself insuﬃcient. It necessarily depends on the ability to recognize identity between learned letter names or sounds and phonological segments in spoken words. A child who is able to generate words beginning with a given sound, and who has also acquired a basic knowledge of letter– sound correspondences, will be in a position to generate a plausible candidate for a novel item. A child oblivious to the phonemic structure of speech (i.e., for whom spoken words are indivisible wholes), will have no way of generating a candidate pronunciation for an unfamiliar letter string. The joint role of letter–sound knowledge and phonological sensitivity is consistent with the wealth of evidence indicating that these two factors are critical co-requisites in reading acquisition (e.g., Bradley & Bryant, 1983; Ehri & Sweet, 1991; Hatcher, Hulme, & Ellis, 1994; Share, 1995; Tunmer, Herriman, & Nesdale, 1988). Consistent with these observations regarding early decoding skill and the possibility of self-teaching among beginning readers, two studies—one in Dutch (Reitsma, 1983b) and one in English (Ehri & Saltmarsh, 1995)—have reported evidence of rapid orthographic learning among Grade 1 children. Reitsma (1983b, Experiment 3) examined printed word learning in Dutch in a sample of 18 Grade 1 readers (split into more and less skilled subgroups) and 13 learning-disabled Grade 3 readers. These children read aloud meaningful sentences containing words judged to be familiar in spoken but not written form. Test sentences were read either zero (control), two, four, or six times across 2 days. Three days later, target spellings were named more quickly than homophonic foils by both Grade 1 subgroups for four and six exposures but, as noted previously, not for two exposures or unseen (control) items. Ehri and Saltmarsh (1995) taught 30 English-speaking Grade 1 readers (also split into more and less skilled readers) and 15 older disabled Grade 3 readers to read target words with simpliﬁed spellings (e.g., CRADL, MESNGER) that were practiced between 10 and 12 times (over 2 days). Three days later, children named three types of items: the original spellings (CRADL), homophonic spellings (e.g., KRADL) and nonhomophonic spellings (e.g., KRATL). The more skilled beginning readers took an average of four practice trials before their ﬁrst error-free reading of the complete list of targets, whereas the less skilled beginners required an average of nine trials. D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 281 Each of the two subgroups of beginners read the target spellings faster than it did either the homophonic or the phonetically altered foils. It is important to note that English and Dutch are linguistic ‘‘cousins,’’ with both belonging to the Germanic family of languages characterized by numerous closed syllables with complex consonant clusters. Like other Indo-European language families such as Romance (French, Spanish, and Italian), Germanic orthographies diﬀer in orthographic ‘‘depth,’’ that is, the 1:1 consistency of letter–phoneme mapping. English is widely regarded as exceptional in its degree of irregularity (Seymour, Aro, & Erskine, 2003), a claim reinforced by direct comparisons of decoding accuracy for matched word and pseudoword lists (see, e.g., Landerl, 2000; Seymour et al., 2003; Wimmer & Goswami, 1994), but Dutch is not highly regular either compared with many other Indo-European scripts such as German, Italian, Finnish, and Greek. A consortium of European researchers (COST A8) recently classiﬁed Dutch at an intermediate level of regularity (Seymour et al., 2003). Comparisons of word and pseudoword decoding among Grade 1 children in 13 European orthographies supported this intermediate ranking of Dutch orthography. The current study examined whether the Dutch and English ﬁndings of early orthographic learning generalize to a highly regular orthography, pointed Hebrew, a script that transcribes a non-European language with relatively simple syllable structures and few consonant clusters. The early-onset hypothesis formulated by Share (1995) predicts that the high levels of decoding aﬀorded by HebrewÕs near perfectly regular orthography (Share & Levin, 1999; Shatil & Share, 2003) should lead to signiﬁcant orthographic learning even among beginning readers. Method Sample and design A total of 32 Grade 1 children (18 girls and 14 boys, mean age 7.0 years) participated in this study. This group was drawn from the same two schools as were the Grade 3 children in Experiment 1. The design was a 2 2 fully repeated-measures design, with exposure and posttest interval each consisting of two levels. The two exposure conditions were set conservatively at two and four exposures. Similarly, retention intervals were 3 and 7 days. Materials and procedure Eight of the nine target items used in the previous experiment were employed again in the current study. Stories were simpliﬁed and shortened somewhat for this younger group, averaging 64 words in length (range: 51–77). Median reading time was 92 s, accuracy averaged 97% correct, and comprehension was 61%. Each child participated in two text-reading sessions that were followed either 3 or 7 days later by posttesting. In each text-reading session, children read aloud four stories that were followed by orally administered comprehension questions. Both story order and session order were counterbalanced. The same posttesting format used in Experiment 1 was adopted in this study: spelling followed by naming followed by orthographic choice. 282 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 Owing to limited resources, this study employed only a single spelling for each orthographic target. Selected spellings were those found in Experiment 1 to be the less preferred of the two alternate spellings (nonpreferred [64.9%] vs preferred [72.3%] in the orthographic choice task, 50.0% vs 70.8% in spelling production). These included four instances of the Hebrew letter TAF (/t/), four TET (/t/), one KUF (/k/), one KAF (/k/), three AYIN (/§/—glottal stop), one ALEPH (/§/), and two word ﬁnal CHAF (/x/ [ch as in Bach]). Estimates for base rate orthographic preferences among Grade 1 children for these selected spellings were then obtained in a comparison sample of Grade 1 children. An independent sample, rather than the experimental group, was used to evaluate base rate preferences in view of the ﬁnding (see Experiment 1) that only a single exposure is suﬃcient to induce orthographic learning. Spelling and orthographic choice tasks were administered to a group of just over 100 Grade 1 children from a school in a comparable area with regard to socioeconomic status. These 100 children were asked to spell the eight target words. Spelling dictation was carried out as a whole-class exercise. All phonologically plausible spellings of each target phoneme (there were cases of up to three alternate graphemes per target phoneme) were then tallied to give an estimate of the ‘‘orthographic preference’’ for the target spelling. For spelling production, the overall frequency of individual target letters was 46.6%. A week later, these same 100 children were given an orthographic choice task in which both alternate spellings of each of the target words were presented side by side. In this Grade 1 sample, target spellings adopted for use in the current study with Grade 1 children were selected at a rate of 51.8% (426/823). Because neither of these ﬁgures diﬀered signiﬁcantly from the 50% chance level, orthographic learning in Experiment 2 was judged to be statistically signiﬁcant only if it exceeded the 50% chance level. Results Target decoding accuracy averaged 93% for consonantal letters (matching the rate achieved by the Grade 3 children) and 77% for consonants and diacritics (compared with only 57% for the Grade 3 children). Decoding rates were identical for the twoexposure and four-exposure conditions. Decoding accuracy for both consonants and diacritics in Grade 1 was signiﬁcantly higher than that in Grade 3 children, a result consistent with other studies (e.g., Ravid, 1996) showing that knowledge and/or reliance on diacritics is greater in Grade 1 when children are taught to read fully pointed text, often with direct instruction in the phonemic values of the diacritical marks. By Grade 3, however, children have become accustomed to unpointed text and are less certain about the phonemic value of these markings or are less reliant on them owing to more eﬀective use of orthographic information. Spelling production Table 5 presents the rates for successfully reproducing the critical target letters. The overall result (46%) was very close to chance and nearly identical to the rate D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 283 Table 5 Mean percentages and raw scores of target letters correctly spelled at posttest by Grade 1 children: Experiment 2 Number of Exposures 3-day interval 7-day interval Overall 2 4 45 (57/126) 48 (61/128) 48 (62/128) 45 (57/128) 47 (119/254) 46 (118/256) Overall 46 (118/254) 46 (119/256) 46 (237/510) Note. Raw scores are in parentheses. observed in the Grade 1 comparison sample (47%). As can be seen in Table 5, the diﬀerences between 3-day and 7-day outcomes, as well as between the two-exposure and four-exposure outcomes, were negligible. This pattern of surprisingly uniform spelling outcomes was also observed in the Grade 3 data. In stark contrast to both the Grade 3 results presented previously and the Grade 1 data reported for Dutch and English (Ehri & Saltmarsh, 1995; Reitsma, 1983b), there was no evidence whatsoever of orthographic learning. Naming accuracy and latency As in Experiment 1, a 2 (number of exposures) by 2 (posttest interval) by 2 (alternate spelling) repeated-measures ANOVA indicated no signiﬁcant overall diﬀerence between naming accuracy for targets (88%) compared with homophonic foils (87%), F < 1:0, nor did it indicate any signiﬁcant main eﬀects or interactions. Vocalization latencies less than 300 and more than 5000 ms were ﬁrst discarded, and then mean reaction times calculated for correct pronunciations of targets and foils with missing values replaced by individual subject means. As with the accuracy data, vocalization onset times failed to indicate any reliable diﬀerences (targets: 1586 ms; foils: 1580 ms; F < 1:0), nor were there any signiﬁcant main eﬀects or interactions. In view of the questionable reliability and sensitivity of reaction time data in this particular paradigm, it would not be prudent to conclude that orthographic learning did not take place in these young children solely on the basis of the latency data. Orthographic choice These data appear in Table 6. The overall success rate of 52% was nearly identical to the ﬁgure obtained in the Grade 1 pilot sample (51.8%) and was not signiﬁcantly beyond the 50% chance level, z < 1:0. Neither the two-exposure nor the Table 6 Mean percentages and raw scores for posttest orthographic choices by Grade 1 children: Experiment 2 Number of exposures 3-day interval 7-day interval Overall 2 4 38 (24/63) 56 (36/64) 59 (38/64) 55 (35/64) 49 (62/127) 55 (71/128) Overall 47 (60/127) 57 (73/128) 52 (133/255) Note. Raw scores are in parentheses. 284 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 four-exposure results exceeded chance, z < 1:00 and z ¼ 1:24, respectively. The 3-day result was clearly nonsigniﬁcant, whereas the 7-day data (57%) approached signiﬁcance, z ¼ 1:59, p ¼ :133. In sum, the results for the orthographic choice task also provide little support for the early-onset hypothesis. Discussion In contrast to the Grade 3 data, there was no reliable evidence of orthographic learning on any of the posttest measures. Performance was consistently at chance. Only the orthographic choice results for four exposures (55%) and for the 7-day condition (57%) produced results that deviated slightly (but nonsigniﬁcantly) from chance. It would appear that among normal beginning readers of pointed Hebrew, high levels of decoding are not associated with the storage of word-speciﬁc orthographic detail. Additional ﬁndings from a recent study of self-teaching among older disabled readers corroborate this result (Share & Shalev, in press). These data were collected from a group of 20 children tested early in their second school year (Grade 2) who were selected as reading level controls for older disabled readers in a study of self-teaching among dyslexic and garden variety poor readers. Grade 2 control readers were matched to the older readers on speed and accuracy of reading both real words and pseudowords and were exposed to the same set of targets from Share (1999) that appeared in the original unabbreviated version of the stories presented to the current Grade 1 sample. The only other diﬀerences between the current Grade 1 data and those from the Grade 2 group were the use of two-exposure and six-exposure texts (rather than two-exposure and four-exposure texts), a single posttest interval of 3 days (rather than 3 and 7 days), and both alternate spellings of the targets. The young Grade 2 group in the Share and Shalev (in press) study decoded the targets at the same level of accuracy as did the current Grade 1 children (93%), with decoding accuracy for both vowels and consonants (66%) falling approximately midway between the level obtained by the Grade 3 sample (57%, Experiment 1) and the current Grade 1 sample (77%, Experiment 2), a ﬁnding that is also consistent with the declining accuracy of the reading of diacritics across Grades 1 to 3 (Ravid, 1996). Orthographic choices failed to exceed chance levels (57%), no eﬀects were found on either naming accuracy or naming times, and target letter spelling was also very close to chance (52%). However, all other groups in the Share and Shalev (in press) study demonstrated reliable and signiﬁcant orthographic learning despite the fact that decoding accuracy was actually a little lower (80% for dyslexics and 88% for garden variety poor readers). The presence of statistically signiﬁcant crossover interactions between decoding success, on the one hand, and orthographic choice and spelling, on the other, conﬁrmed that these normal Grade 2 readers were unable to recognize or recall orthographic detail despite high levels of decoding success. Taken together, both sets of data indicate that the early-onset hypothesis must be rejected at least in the case of pseudowords presented in HebrewÕs highly regular pointed script. The evidence of early orthographic learning reported in the deeper orthographies of Dutch and English (Ehri & Saltmarsh, 1995; Reitsma, 1983b) does not appear to generalize to the shallow pointed Hebrew script. D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 285 The absence of rapid orthographic learning in Grade 1 Hebrew readers stands in stark contrast to the data obtained from the group of Grade 2 children tested during the ﬁnal months of the school year (Share, 1999). Potential sampling diﬀerences do not appear to be responsible for the discrepant results. In view of the fact that the earlier 1999 self-teaching study obtained reliable evidence of orthographic learning in three separate samples of children tested at the end of Grade 2, the data converge on the hypothesis that there are fundamental print processing developments taking place among young Hebrew readers between early Grade 2 and late Grade 2. Before embarking on speculation regarding the source of this intriguing ﬁnding, it must be acknowledged that the current Grade 1 data were derived entirely from pseudoword stimuli, whereas both the English and Dutch studies employed real words or pseudo-homophones of real words judged to be familiar to children in spoken form but not in printed form. Thus, the possibility that beginning readers of Hebrew may exhibit self-teaching with real words cannot be ruled out a priori. Perhaps items such as pseudowords with relatively few associative links in long-term memory are less likely to be retained by novices than are real words. It might be speculated that at the onset of reading development, much printed word learning involves a considerable degree of meaning seeking with relatively little attention paid to the nonsemantic dimension of print such as orthographic detail. Alternatively, children at this age may simply require a greater number of target exposures; there were only a maximum of four presentations per item in Experiment 2. Experiment 3 addressed both of these issues empirically in a study of Grade 1 children reading both real words and pseudowords, each in alternate spellings and with a greater number of exposures (eight as well as four). All posttests were administered after an interval of 7 days. Experiment 3: Self-teaching in Grade 1 readers with real words and pseudowords Method Sample A total of 64 children participated in this study (35 girls and 29 boys, mean age 7.1 years). This sample was drawn from ﬁve schools spanning a wide range of socioeconomic backgrounds and nine separate Grade 1 classes. Design This study included words as well as pseudowords, with each item in alternate homophonic spellings, and also eight exposures in addition to the standard four. A total of 64 children participated in this 2 (word type) by 2 (spellings) by 2 (exposures) by 2 (replications) design. The fourth factor, replications, was adopted because a total of 16 passages (eight words and eight pseudowords) administered over four storyreading sessions, each followed by a posttest session, was considered excessive in view of the exigencies of testing young children in school settings. Consequently, it was decided to limit the testing demands to the more manageable proportions used 286 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 in previous studies (i.e., two text-reading sessions and two posttests for a total of four testing sessions). Therefore, each child read a total of only eight texts, four stories containing real-word targets and four stories containing pseudowords, with the order of these word/pseudoword sessions counterbalanced. Within each session, a child read two stories with four target exposures and the other two stories with eight target exposures. Because story order (either within or across sessions) had not been found to inﬂuence the results in previous work, story order within session was ﬁxed. Splitting the 16 test items into two groups of eight and assigning them to two independent (‘‘item replication’’) subsamples, each with identical conditions, made it possible to check the generalizability of results across items. (Owing to an experimenter error in which one child received the wrong subset of items, 33 children saw one set of words and pseudowords and 31 children saw the second set.) Target words and texts The real words and texts in this study were developed in a three-stage procedure. An initial pool of candidate words deemed to exist in a Grade 1 childÕs oral vocabulary but unfamiliar in written form was ﬁrst generated. Each of these words contained from one to four homophonic letters. A word-by-word survey of the reading scheme used at the school where the pilot work was carried out reduced this initial set to 67 words; any word that appeared in print, even once, was discarded. Next, four Grade 1 teachers and a literacy specialist (the regional pedagogical adviser) were asked to rate the degree to which Grade 1 children are likely to be familiar with the printed form of these words on a 5-point scale (‘‘almost certainly seen in print,’’ ‘‘probably seen,’’ ‘‘possibly seen,’’ ‘‘probably not seen,’’ and ‘‘almost certainly not seen’’). Only words rated in the lowest two categories (‘‘probably not seen’’ and ‘‘almost certainly not seen’’) by at least three of the four judges were included in the next step. A total of 28 words remained at this point. All 28 items were then given in a spelling dictation task (two separate sessions of 14 items) to a sample of 69 children at this pilot school. Testing was carried out with small groups of up to ﬁve children at a time. Children were also asked whether they knew the meaning of each word as it was dictated. From the subset of items in which the homophonic letters were spelled with the incorrect target letters close to 50% or more of the time, 8 items were selected for inclusion in the ﬁnal test set: KETEM (stain), AVAZ (goose), KIVUN (direction), TA’UT (mistake), TA’ARUXA (exhibition), KASKASIM (dandruﬀ), ATALEF (bat, winged), and KONXIYA (conch, shell). These items each included between two and four homophonic letters, with the critical letters appearing in all letter positions from the initial to ﬁnal positions.7 This set also included at least one exemplar of each of the six homophonic letter pairs that exist in Hebrew script. Of the 8 target items, 5 included three homophonic letters, but in all 5 cases, only two of the three letters met the criterion of 50% homophone 7 In Hebrew, homophonic letters are very common (more so than in English) because approximately one-third of all phonemes can be transcribed by alternate graphemes. Thus, the vast majority of Hebrew words contain at least one homophonic letter. D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 287 substitutions in the pilot sample. One item (KASKASIM) included four homophonic letters. Item length ranged from two to four syllables and from three to six letters. Texts Eight short stories were then composed with the target words as the central theme (see examples in the Appendix). Story length ranged from 65 to 84 words. The stories were then split into two sets of four, with speciﬁc letter identities equated as much as possible. For the pseudoword stimuli, eight of the nine items used in Experiment 2, together with their accompanying passages, were used again in the current study. Procedure The procedure was the same as in Experiment 2 with two exceptions. The posttest naming task was dropped, leaving only spelling and orthographic choice, and only a single posttest interval (7 days) was included. Results Story-reading performance Children took an average of just over 2 min to read these passages (median: 125 s [SD ¼65 s], with an average of 5.7 errors [SD ¼ 6:1]) and comprehension at 80% [SD ¼ 29%]. Target decoding Overall target decoding accuracy was 93% (SD ¼ 19%) for consonantal letters only and 75% (SD ¼ 35%) for consonants and vowels. Not unexpectedly, there were substantial diﬀerences between words and pseudowords. For real words, target consonantal decoding accuracy averaged 95%, which was signiﬁcantly greater than the decoding accuracy (90%) for pseudoword targets, t(510) ¼ 2.96, p ¼ :003. For the combined consonant and vowel accuracy, there was also a signiﬁcant advantage for real words (89% vs 61%), t(510) ¼ 9.92, p < :001. These data suggest that decoding, and particularly the vowel decoding of novel words in text, might not be purely a matter of nonlexical, bottom-up grapheme–phoneme translation in the classic dualroute sense but is partly assisted by item-speciﬁc knowledge of known phonological forms. Orthographic choice Comparing, ﬁrst, the outcomes for the same pseudoword targets used in both this study (Table 7) and Experiment 2, it can be seen that the overall nonsigniﬁcant result for the pseudowords in this study (55%), z < 1:0, replicated the Experiment 2 data at the same 7-day posttest interval (57%) (collapsing over exposure), thereby conﬁrming the surprising lack of sensitivity to orthographic information among these readers. Doubling the number of exposures from four to eight did not alter this picture, z < 1:0. Nor did the inclusion of real words deemed familiar to these young readers succeed in uncovering any convincing evidence of orthographic learning. 288 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 Table 7 Mean percentages and raw scores for posttest orthographic choices by Grade 1 children: Experiment 3 4 exposures 8 exposures Overall Words Pseudowords 56 (72/128) 59 (76/128) 57 (73/128) 50 (64/128) 57 (145/256) 55 (140/256) Overall 58 (148/256) 54 (137/256) 56 (285/512) Note. Raw scores are in parentheses. * p < :05. Whereas the overall result for real words (57%) was signiﬁcant, p ¼ :04, individually, neither the four-exposure condition nor the eight-exposure condition reached significance. Above all, the inclusion of real words with familiar meanings appeared to make little diﬀerence to the overall pattern of results. There was no overall diﬀerence between real words and pseudowords, z < 1:0, rebutting the supposition that the use of target items devoid of meaning in Experiment 2 somehow masked true orthographic abilities. It must be acknowledged, nonetheless, that the overall result for words (57%) was, formally, statistically signiﬁcant. Indeed, nearly all the ﬁgures in Table 7 exceeded the 50% chance level, consistently clustering around the borderline of statistical signiﬁcance. The grand mean (56%) also was signiﬁcant, so there are clearly ‘‘glimmerings’’ of orthographic learning here. But comparison of these marginal data with the unambiguously robust ﬁndings seen in Experiment 1 and in all previous self-teaching studies undertaken with children from late Grade 2 upward reveals fundamental differences in the degree of orthographic learning. The 7-day data for the Grade 3 group from Experiment 1 averaged 74%, the end-of-Grade 2 sample averaged 74% (Share, 1999, Experiment 1), Grades 4, 5, and 6 averaged 75% (Share & Shalev, in press), and end-of-Grade 2 English speakers averaged 75% (Cunningham et al., 2002). These data demonstrate, in a very consistent manner, eﬀects of a diﬀerent order of magnitude for orthographic choice. Looking at the generalizability of results across the two subsamples of children presented with diﬀerent subsets of items, no signiﬁcant diﬀerences were found either on the grand mean, on marginal means (i.e., four or eight exposures, words or pseudowords) or on individual cell means (i.e., no z value exceeded unity). It is important to emphasize that these poor results for orthographic learning were obtained in spite of high levels of target decoding—levels that were actually higher than those recorded for the older samples for whom orthographic learning was unmistakably present. The spelling data presented next conﬁrmed this apparent dissociation between good (i.e., accurate) decoding and poor orthographic learning. Spelling The spelling data produced an unambiguous result (Table 8). All cells, marginal means, and even the grand mean were consistently and unequivocally nonsigniﬁcant. In contrast to the orthographic choice data, there was no suggestion at all of orthographic learning. As in the case of orthographic choice, the pattern for the complete D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 289 Table 8 Mean percentages and raw scores of target letters spelled correctly at posttest by Grade 1 children: Experiment 3 Words Pseudowords Overall 4 exposures 8 exposures Overall 46 (117/256) 53 (135/256) 49 (252/512) 51 (130/256) 51 (131/256) 51 (261/512) 48 (247/512) 52 (266/512) 50 (513/1024) Note. Raw scores are in parentheses. sample (N ¼ 64) was reproduced in both subsamples with very small and nonsignificant diﬀerences between the two subgroups at all levels of analysis. Discussion In spite of the inclusion of familiar known words in addition to pseudowords and doubling the number of target presentations, the outcomes of this third study conﬁrmed the pattern emerging from Experiment 2 in that memory for word-speciﬁc orthographic detail was negligible. This ﬁnding, replicated across independent groups of participants and items, raises both cross-linguistic and developmental (intralinguistic) issues. Turning ﬁrst to the startling developmental diﬀerences between the outcomes for Grade 1 (Experiments 2 and 3) and those for Grade 3 (Experiment 1), it is hypothesized that in an orthography as transparent as pointed Hebrew, which permits such high levels of decoding so very early in a childÕs orthographic experience (cf. Cossu, 1999; Seymour et al., 2003; Wimmer & Goswami, 1994), normal novice readers typically pass through a phase resembling surface dyslexia. Competent novices of regular orthographies are capable of decoding nearly any word in a ‘‘bottom-up’’ or ‘‘surface’’ fashion but appear to be relatively insensitive to surface information, that is, to word-speciﬁc orthographic detail. The qualiﬁer ‘‘relative’’ acknowledges that novice readers are not impervious to orthographic information. The overall result in Experiment 3 for orthographic choice (but not spelling) suggested some embryonic orthographic learning. More generally, the Hebrew data showing signiﬁcant orthographic knowledge evident in both successful orthographic choice and homophone choice tasks with high-frequency real words at the end of Grade 1 (Shatil, 1997) clearly refutes any categorical conclusions regarding the presence or absence of orthographic learning. Studies of early spelling development in Hebrew further conﬁrm that even Grade 1 children are able to spell most of the homophonic target letters in familiar words correctly (Ravid, 2001; Shatil, Share, & Levin, 2000). Therefore, it seems reasonable to assume that many more exposures than the current four or eight would eventually lead to signiﬁcant orthographic knowledge. In contrast, the dramatic diﬀerences between Grade 1 novices and the older, more proﬁcient readers studied in Experiment 1 and in prior work with this paradigm revealed these beginners to be relatively insensitive to word-speciﬁc orthographic detail, reading in a highly surface fashion. Relative to the older readers (late Grade 2 onward) these beginners are clearly processing print in a very diﬀerent manner. 290 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 The source of this processing diﬀerence does not appear to be the availability or nonavailability of word meaning. The inclusion of meaningful words in Experiment 3 was speciﬁcally aimed to address the possibility that the lack of orthographic learning in Experiment 2 was somehow due to the use of meaningless pseudowords. However, although it is true that the overall level of orthographic choices for words, but not pseudowords, was signiﬁcant in this third study, these two rates were nearly identical (55% vs 57%) and, most important, were not signiﬁcantly diﬀerent. Moreover, in the four-exposure condition, orthographic learning of pseudowords was found to be statistically signiﬁcant (59%), and in the spelling data, the pseudoword results were marginally superior, although again the diﬀerences never attained significance. In sum, these data are consistent with earlier ﬁndings reported by Hogaboam and Perfetti (1978) and Share (1999) indicating that word meaning per se does not appear to play a signiﬁcant role in word-speciﬁc orthographic learning. The role of familiar phonological/oral forms as a base for securing spellings may be a more promising avenue with regard to orthographic learning. Although the current study did not take up this issue, a recent study with this same experimental paradigm (Somech & Share, in preparation) found much poorer orthographic memory for words that Grade 3 readers were unable to recall orally at posttest compared with successfully recalled items. Thus, the availability of a familiar phonological form may be a signiﬁcant factor in orthographic learning, at least among older, more skilled readers. However, among novice readers, the phonological availability hypothesis seems unlikely to shed much light on orthographic learning for the same reason that the meaning hypothesis was discounted: Real and highly familiar spoken words failed to confer any advantage for Grade 1 readers in terms of orthographic learning relative to newly learned, and probably only partially recalled, pseudowords. Currently, the source of this processing diﬀerence can only be grounds for speculation. One possibility is that some ‘‘critical mass’’ of print exposure by more skilled readers enables the uptake of word-speciﬁc orthographic detail in a way that is qualitatively diﬀerent from that of novices. In this regard, it is important to note that in Israeli schools, Grade 1 is the sole ‘‘learning-to-read’’ year, devoted primarily to the acquisition of the code (known as ‘‘technical’’ reading). Most schools follow a structured sequence of basal-type workbooks (often with controlled reading vocabulary) in which letters and diacritics are introduced in a more or less ﬁxed sequence. Because the fundamentals are typically acquired by the end of Grade 1 (Feitelson, 1989; Share & Levin, 1999), the reading curriculum in Grade 2 switches from a learning-to-read mode to a reading-to-learn mode in which readers, now able to decode virtually any pointed text, plunge into a variety of readers, literature, and biblical texts whose only concession to the novice is the presence of vowel diacritics. In view of these basic curricular changes, it seems reasonable to assume that Grade 2 witnesses a major transition in both the volume and variety of print experience. It is hypothesized that for the majority of normal readers of a highly regular script, the latter half of Grade 2 witnesses the emergence of a highly specialized print-speciﬁc processing advantage (not a basic skill diﬀerence) acquired as an outgrowth of extensive print experience. This ‘‘orthographic sensitivity’’ hypothesis D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 291 assumes that a critical volume of print experience brings about a fundamental change in orthographic sensitivity. Just as an art or food connoisseur becomes sensitive to subtle nuances that escape the layperson, the accumulation of print exposure enables the more experienced reader to develop a sensitivity to orthographic detail that is beyond the grasp of the novice. The concept of orthographic sensitivity is very diﬀerent from contemporary thinking regarding the role of print exposure in the development of orthographic knowledge. It is generally assumed that diﬀerences between more skilled readers and less skilled readers on conventional measures of orthographic knowledge, such as orthographic choice, are attributable to a greater volume of exposure per se; that is, older children simply have seen the speciﬁc test items more often. By this account, younger readers receiving the same number of exposures to new words should perform at comparable levels when assessed on word-speciﬁc orthographic information as in the current paradigm. The current Grade 3 data, together with converging data from additional samples (Share, 1999; Share & Shalev, in press) show clearly that a conventional print exposure account does not capture the current ﬁndings. The notion of orthographic sensitivity is also distinct from (but in no way is at odds with) traditional accounts of orthographic development that typically stress recurrent multiletter sequences (e.g., -ight) and morphological patterns (e.g., -ed). It should be emphasized that the current study examined orthographic learning solely in the context of word-speciﬁc letter-level information and not general orthographic conventions that, in Hebrew, are tied to morphological constraints governing noun and verb formation. The reason that only item-speciﬁc letter-level orthographic learning was assessed is simply that homophonic letters marking general morpho– phonological structures are extremely common, consistent, and well known even among Grade 1 children. For example, the grammatical feminine gender is marked solely by the aﬃxed letter TAF; the homophonic letter TET is never part of any grammatical aﬃx or any morpho–phonological pattern. In contrast, individual root letters in a speciﬁc lexical morpheme can be either TET or TAF, with no general rule governing these alternations. The focus on item-speciﬁc orthographic detail implies that a putative lack of morphological sensitivity/awareness among Grade 1 readers cannot account for their lack of orthographic learning in the present investigation. The orthographic sensitivity hypothesis could be tested in a longitudinal study examining concurrent changes in orthographic learning (in both reading and writing), reading ﬂuency, and prosody as well as ﬁne-grained documentation of both the quantity and quality of print exposure. Alternatively, experimental simulation may be possible by means of an artiﬁcial orthography. Turning to the cross-linguistic aspects of these ﬁndings, it is clear that an orthographic sensitivity account resting on quality and quantity of print exposure may explain the Hebrew-language data but not the cross-linguistic diﬀerences in the Grade 1 ﬁndings. As noted previously, neither Dutch nor English can be considered highly regular orthographies (Seymour et al., 2003), and the comparative data support this assertion (Landerl, 2000; Seymour et al., 2003; Wimmer & Goswami, 1994). In terms of irregularity, English might best be classiﬁed as an ‘‘outlier’’ orthography. Even words that most English speakers would consider to be ‘‘regular’’ (e.g., ship, tell, 292 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 speed, make) could be considered irregular in comparison with the strictly one-onone letter–sound correspondences that characterize highly regular scripts (and in which digraphs and geminates typically are not found). In contrast to both Dutch and English, fully voweled (pointed) Hebrew orthography has neither vowel nor consonant digraphs (e.g., sh, ee), no geminates (e.g., ss, ll), no consonant clusters (e.g., sp, cl, str), and no silent letters (with one exception). The adherence of pointed Hebrew to the one-to-one mapping principle is nearly perfect (Navon & Shimron, 1984; Share & Levin, 1999; Shimron, 1993).8 Possibly because simple one-to-one letter– sound relationships have such far-reaching utility, simple nonlexical correspondences suﬃce to decode nearly any pointed letter string, whether high frequency or low frequency and whether word or pseudoword. In contrast, deeper orthographies oblige the developing reader to look beyond low-level phonology and consider higher order regularities that are often word speciﬁc or ‘‘lexicalized’’ (Share, 1995). Indeed, many of EnglishÕs oldest and most common words (e.g., have, was, to, of, come, would) are so idiosyncratic that word-speciﬁc orthographically based memorization seems to be the only reasonable option. The notion that readers of deeper orthographies must pay greater attention to word-speciﬁc visual–orthographic information has been formalized in Katz and FrostÕs (1992) orthographic depth hypothesis. This theory maintains that ‘‘shallow orthographies are more easily able to support a word recognition process that involves the languageÕs phonology. ... Deep orthographies encourage a reader to process printed words by referring to their morphology via the printed wordÕs visual–orthographic structure’’ (p. 71). Extending this notion to early reading acquisition, Seymour et al. (2003) argued that learning to read in a shallow orthography may be based primarily on a single alphabetic/phonological process, whereas a deep orthography requires both phonological and visual–orthographic (‘‘logographic’’ in SeymourÕs terminology) processes. Consistent with this view, Seymour and colleagues found signiﬁcantly greater lexicality eﬀects (word reading superior to pseudoword reading) among novice readers of deeper orthographies (e.g., Dutch, Danish, English) compared with shallower orthographies (e.g., German, Italian, Greek). A growing number of cross-linguistic studies now demonstrate that children learning to read English rely to a greater extent on so-called ‘‘large’’ orthographic units, such as multiletter word bodies and whole-word strategies, than do readers of more regular orthographies (Goswami, Ziegler, Dalton, & Schneider, 2001, 2003; Landerl, 2003; Landerl, Wimmer, & Frith, 1997; Wang & Geva, 2003; Ziegler, Perry, Ma-Wyatt, Ladner, & Schulte-Korne, 2003). If beginning readers of deeper orthographies, such as English and Dutch, are indeed more reliant on word-speciﬁc orthographic information in printed word learning, the relative insensitivity to orthographic information observed among Grade 1 8 The only exceptions are two diacritical marks. The patax (ganuv) appearing under a word-ﬁnal CHET (as in tapuax [apple]) is pronounced before, rather than after (as is conventional), the CHET. The second exception, a vowel diacritic called kamatz katan, occurs when the kamatz (normally pronounced /a/) is combined with the schwa to produce the sound /o/ in a limited number of high-frequency words. Neither of these exceptions accounts for more than a few dozen words in Hebrew. D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 293 readers in pointed Hebrew could be interpreted as exclusive reliance on simple letter– sound translation for reading any and all classes of printed words.9 Comparative studies are needed to evaluate this hypothesis. Returning to the role of phonological recoding in the acquisition of word-speciﬁc orthographic representations, it is clear that in HebrewÕs highly regular script at least, decoding accuracy per se is not suﬃcient to ensure orthographic learning. Absolute levels of decoding accuracy were actually greater among the very group (Grade 1) showing the poorest orthographic learning. Furthermore, it seems safe to assume that the decoding proﬁciency of the Hebrew-speaking Grade 1 children surpassed the levels for the English and Dutch children in the Reitsma (1983b) and Ehri and Saltmarsh (1995) studies. But rather than asking whether children can successfully decode novel orthographic forms (not nearly as challenging a task in Hebrew as in English), it may be more important to ask how they succeed in decoding them and, in particular, how rapidly and eﬃciently the decoding process is executed. Slow, laborious, letter-by-letter decoding may leave little capacity for attending to multiletter and word-level graphemic detail (Bowers, 1989). A child might not even be looking at the printed letters when attempting to blend the string of phonemes. Even in English, the evidence from a number of recent training studies with severely disabled readers suggests that decoding accuracy alone is not suﬃcient to ensure reliable long-term orthographic learning. These studies are relevant to the issue of orthographic learning and self-teaching because they examined the long-term eﬀects of remedial instruction in phonological skills (phonemic awareness and phonological recoding) not only on generalized decoding (word attack) skills but also on real-word identiﬁcation and, hence, more orthographically based word identiﬁcation (Foorman et al., 1997; Lovett et al., 2000; Olson, Wise, Ring, & Johnson, 1997; Torgesen et al., 2001; Torgesen, Wagner, & Rashotte, 1997). The ﬁrst generation of these carefully controlled evaluations of phonologically based remedial instruction indicated that the decoding skills of even severely disabled readers can be signiﬁcantly improved, at least according to an accuracy criterion, but reliable and enduring transfer to real-word identiﬁcation did not occur (Foorman et al., 1997; Olson et al., 1997; Torgesen et al., 1997). More recently, however, several second-generation training studies (e.g., Lovett et al., 2000; Torgesen et al., 2001) that have included more intensive training in phonological awareness and decoding skills have succeeded in demonstrating sustainable and generalized gains in realword reading. This later work supports the notion of self-teaching among disabled readers but suggests that treatment intensity may be a critical factor in developing the automatized decoding skills that seem to be a prerequisite to reliable growth in orthographically based reading in severely disabled readers (Olson et al., 1997). These English-language training studies, together with the apparent dissociation between accurate decoding and orthographic learning among beginning 9 This line of reasoning predicts that in the case of regular orthographies with one-to-one letter–sound relations but one-to-many phoneme-to-grapheme relations (e.g., Greek, Hebrew), spelling should also be less strongly linked to reading than in the case of deeper orthographies because this depends heavily on word-speciﬁc orthographic knowledge. 294 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 readers of Hebrew, underscore the point raised earlier regarding the need to go beyond readily available measures of decoding, such as accuracy and even vocalization onset latencies, and employ more sophisticated measures of the temporal dynamics of decoding. Decoding a novel word is not merely a sequential letterby-letter process of retrieving phonemes one-by-one and then stringing these together. Going beyond decoding accuracy and speed, EhriÕs (1978, 1992) amalgamation hypothesis, Bowers and WolfÕs (1993) timing hypothesis, and PerfettiÕs (1992) discussion of representation redundancy all emphasize the fact that printed words, and particularly multisyllabic words, are not merely chains of letters; rather, they form coherent hierarchically organized subgroupings of letters that must be unitized into syllabic and morphemic units. Consider the implications for orthographic learning of a child who decodes the word bending as ‘‘ben-ding.’’ It is these groupings that are perceived as orthographically integrated whole-word units or whole morpheme units by skilled readers (Adams, 1990; Carlisle, 1995, 2000; Mann, 2000; Stanovich, 1992). The accurate but slow and nonautomatic decoding of adult dyslexics (Bruck, 1990; Felton, Naylor, & Wood, 1990) may be an expression of an inability to amalgamate decoded letters into cohesive units. A variety of factors may inﬂuence the childÕs ability to integrate and amalgamate the sounded letters of a novel orthographic string into an integrated unit in a way that promotes orthographic learning. Inadequate sensitivity to higher order phonological and morphological structures in oneÕs spoken language may impair the developing readerÕs ability to merge the individual phonemes of a novel letter string into the larger units required for well-speciﬁed and well-integrated orthographic representations. As discussed previously, these considerations call for research to go beyond simple accuracy and vocalization onset times and to probe the nature of online decoding. Little is known about how exactly sequences of letters coalesce in the minds of young readers as they decode. Systematic documentation of the online dynamics of decoding may help to illuminate some of the puzzles of orthographic learning. Acknowledgments This research was supported by a grant from the Israel Science Foundation. The author thanks Carmit Shalev and Limor Golan for their invaluable assistance in collecting the data in these experiments. Appendix A. Sample Grade 1 text with embedded pseudoword target (‘‘Takunia’’) The hottest town In Australia is the hottest town in the world. This town is called Takunia and itÕs right in the middle of the desert. In Takunia, the temperature can reach 60 °C. ItÕs so hot that even the ﬂies drop dead. You can even fry an egg on the roof of your car. D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 295 The people in Takunia drink lots of beer. ItÕs very strong beer. If youÕre not used to drinking beer youÕd better watch out! Would you like to live in Takunia? (51 Hebrew words) Appendix B. Sample Grade 1 text with embedded real-word target (‘‘bat’’) A strange animal One day a bat was sitting in a tree, beside a birdÕs nest. The bird got a fright and chirped, ‘‘Go away you big bird!’’ ‘‘IÕm not a bird, IÕm a mouse,’’ said the bat. Then the bat jumped down and sat on the grass in the garden below. The cat saw him, came closer and said, ‘‘IÕll catch you, mouse.’’ ‘‘IÕm not a mouse, IÕm a bird’’ said the bat, ‘‘Look at my wings.’’ Then he spread his wings and ﬂew up and away. What a strange animal, some of the time a bird, other times a mouse. (63 Hebrew words) References Adams, M. J. (1990). Beginning to read. Cambridge, MA: Bradford. Bentin, S., & Frost, R. (1995). Morphological factors in visual word identiﬁcation in Hebrew. In L. B. Feldman (Ed.), Morphological aspects of language processing (pp. 271–292). Hillsdale, NJ: Lawrence Erlbaum. Berent, I., & Perfetti, C. A. (1995). A rose is a REEZ: The two-cycle model of phonology assembly in reading English. Psychological Review, 102, 146–184. Bowers, P. G. (1989). Naming speed and phonological awareness: Independent contributors to reading disabilities. In S. McCormich & J. Zutell (Eds.), Cognitive and social perspectives for literacy research and instruction: 38th yearbook of the National Reading Conference (pp. 165–172). Chicago: National Reading Conference. Bowers, P. G., & Wolf, M. (1993). Theoretical links among naming speed, precise timing mechanisms, and orthographic skill in dyslexia. Reading and Writing, 5, 69–85. Bradley, L., & Bryant, P. E. (1983). Categorizing sounds and learning to read: A causal connection. Nature, 301, 419–421. Brooks, L. (1977). Visual pattern in ﬂuent word identiﬁcation. In A. S. Reber & D. L. Scarborough (Eds.), Toward a psychology of reading (pp. 143–181). Hillsdale, NJ: Lawrence Erlbaum. Bruck, M. (1990). Word recognition skills of adults with childhood diagnoses of dyslexia. Developmental Psychology, 26, 439–454. Carlisle, J. F. (1995). Morphological awareness and early reading achievement. In L. B. Feldman (Ed.), Morphological aspects of language processing (pp. 189–209). Hillsdale, NJ: Lawrence Erlbaum. Carlisle, J. F. (2000). Awareness of the structure and meaning of morphologically complex words: Impact on reading. Reading and Writing, 12, 169–190. Cossu, G. (1999). The acquisition of Italian orthography. In M. Harris & G. Hatano (Eds.), Learning to read and write (pp. 10–33). Cambridge, UK: Cambridge University Press. Cunningham, A. E., Perry, K. E., Stanovich, K. E., & Share, D. L. (2002). Orthographic learning during reading: Examining the role of self-teaching. Journal of Experimental Child Psychology, 82, 185–199. Ehri, L. C. (1978). The development of orthographic images. In U. Frith (Ed.), Cognitive processes in spelling (pp. 311–338). London: Academic Press. Ehri, L. C. (1992). Reconceptualizing the development of sight word reading and its relationship to recoding. In P. B. Gough, L. C. Ehri, & R. Treiman (Eds.), Reading acquisition (pp. 107–144). Hillsdale NJ: Lawrence Erlbaum. 296 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 Ehri, L. C., & Saltmarsh, J. (1995). Beginning readers outperform older disabled readers in learning to read words by sight. Reading and Writing: An Interdisciplinary Journal, 7, 295–326. Ehri, L. C., & Sweet, J. (1991). Fingerpoint-reading of memorized text: What enables beginners to process the print. Reading Research Quarterly, 24, 442–462. Ehri, L. C., & Wilce, L. S. (1985). Movement into reading: Is the ﬁrst stage of printed word learning visual or phonetic? Reading Research Quarterly, 20, 163–179. Ehri, L. C., & Wilce, L. S. (1987). Cipher versus cue reading: An experiment in decoding acquisition. Journal of Educational Psychology, 79, 3–13. Feitelson, D. (1988). Facts and fads in beginning reading. Norwood, NJ: Ablex. Feitelson, D. (1989). Reading education in Israel. In W. Ellis & J. Hladez (Eds.), International handbook on reading education (pp. 163–181). Westport, CT: Greenwood Praeger. Felton, R. H., Naylor, C. E., & Wood, F. B. (1990). Neuropsychological proﬁle of adult dyslexics. Brain and Language, 39, 485–497. Foorman, B. R., Francis, D. J., Winikates, D., Mehta, P., Schatschneider, C., & Fletcher, J. M. (1997). Early interventions for children with reading disabilities. Scientiﬁc Studies of Reading, 1, 255–276. Frith, U. (1985). Beneath the surface of developmental dyslexia. In K. E. Patterson, J. C. Marshall, & M. Coltheart (Eds.), Surface dyslexia (pp. 301–330). Hillsdale, NJ: Lawrence Erlbaum. Frost, R., & Bentin, S. (1992). Reading consonants and guessing vowels: Visual word recognition in Hebrew. In R. Frost & L. Katz (Eds.), Orthography, phonology, morphology, and meaning (pp. 27–44). Amsterdam: North-Holland. Goswami, U., Ziegler, J. C., Dalton, L., & Schneider, W. (2001). Pseudohomophone eﬀects and phonological recoding procedures in reading development in English and German. Journal of Memory and Language, 45, 648–664. Goswami, U., Ziegler, J. C., Dalton, L., & Schneider, W. (2003). Nonword reading across orthographies: How ﬂexible is the choice of reading units? Applied Psycholinguistics, 24, 235–247. Gough, P., & Juel, C. (1991). The ﬁrst stages of learning to read. In L. Rieben & C. A. Perfetti (Eds.), Learning to read: Basic research and its implications (pp. 47–56). Hillsdale, NJ: Lawrence Erlbaum. Harm, M. W., & Seidenberg, M. S. (1999). Phonology, reading acquisition, and dyslexia: Insights from connectionist models. Psychological Review, 106, 491–528. Hatcher, P., Hulme, C., & Ellis, A. W. (1994). Ameliorating early reading failure by integrating the teaching of reading and phonological skills: The phonological linkage hypothesis. Child Development, 65, 41–57. Hogaboam, T. W., & Perfetti, C. A. (1978). Reading skill and the role of verbal experience in decoding. Journal of Educational Psychology, 70, 717–729. Jorm, A. F., & Share, D. L. (1983). Phonological recoding and reading acquisition. Applied Psycholinguistics, 4, 103–147. Katz, L., & Frost, R. (1992). Reading in diﬀerent orthographies: The orthographic depth hypothesis. In R. Frost & L. Katz (Eds.), Orthography, phonology, morphology, and meaning (pp. 67–84). Amsterdam: North-Holland. Landerl, K. (2000). Inﬂuences of orthographic consistency and reading instruction on the development of nonword reading skills. European Journal of Psychology of Education, 15, 239–257. Landerl, K. (2003). Dyslexia in German-speaking children. In N. Goulandris (Ed.), Dyslexia in diﬀerent languages: Cross-linguistic comparisons (pp. 15–32). London: Whurr. Landerl, K., Wimmer, H., & Frith, U. (1997). The impact of orthographic consistency on dyslexia: A German–English comparison. Cognition, 63, 315–334. Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review, 95, 492–527. Logan, G. D. (1990). Repetition priming and automaticity: Common underlying mechanisms? Cognitive Psychology, 22, 1–35. Logan, G. D. (2002). An instance theory of attention and memory. Psychological Review, 109, 376–400. Lovett, M. W., Lacerenza, L., Borden, S. L., Frijters, J. C., Steinbach, K. A., & De Palma, M. (2000). Components of eﬀective remediation for developmental reading disabilities: Combining phonological and strategy-based instruction to improve outcomes. Journal of Educational Psychology, 92, 263–283. D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 297 Manis, F. R. (1985). Acquisition of word identiﬁcation skills in normal and disabled readers. Journal of Educational Psychology, 77, 78–90. Mann, V. A. (Ed.) (2000). Introduction to special issue on morphology and the acquisition of alphabetic writing systems [special issue]. Reading and Writing, 12, 143–147. Morris, D. (1992). Concept of word: A pivotal understanding in the learning-to-read process. In S. Templeton & D. R. Bear (Eds.), Development of orthographic knowledge and the foundation of literacy: A memorial Festschrift for Edmund H. Henderson (pp. 53–77). Hillsdale, NJ: Lawrence Erlbaum. Nagy, W. E., & Herman, P. A. (1987). Breadth and depth of vocabulary knowledge: Implications for acquisition and instruction. In M. McKeown & M. Curtis (Eds.), The nature of vocabulary acquisition (pp. 19–35). Hillsdale, NJ: Lawrence Erlbaum. Navon, D., & Shimron, Y. (1984). Reading Hebrew: How necessary is the graphemic representation of vowels?. In L. Henderson (Ed.), Orthographies and reading: Perspectives from cognitive psychology, neuropsychology, and linguistics (pp. 91–102). London: Erlbaum. Newell, A., & Rosenbloom, P. S. (1981). Mechanisms of skill acquisition and the law of practice. In J. R. Anderson (Ed.), Cognitive skills and their acquisition (pp. 1–55). Hillsdale, NJ: Lawrence Erlbaum. Olson, R. K., Wise, B., Ring, J., & Johnson, M. (1997). Computer-based remedial training in phoneme awareness and phonological decoding: Eﬀects on the posttraining development of word recognition. Scientiﬁc Studies of Reading, 1, 235–253. Perfetti, C. A. (1992). The representation problem in reading acquisition. In P. B. Gough, L. C. Ehri, & R. Treiman (Eds.), Reading acquisition (pp. 145–174). Hillsdale, NJ: Lawrence Erlbaum. Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56–115. Rack, J., Hulme, C., Snowling, M., & Wightman, J. (1994). The role of phonology in young childrenÕs learning to read words: The direct mapping hypothesis. Journal of Experimental Child Psychology, 57, 42–71. Ravid, D. (1995). Language change in child and adult Hebrew: A psycholinguistic perspective. New York: Oxford University Press. Ravid, D. (1996). Accessing the mental lexicon: Evidence from incompatibility between representation of spoken and written morphology. Linguistics, 34, 1219–1246. Ravid, D. (2001). Learning to spell in Hebrew: Phonological and morphological factors. Reading and Writing, 14, 459–485. Reitsma, P. (1983a). Printed word learning in beginning readers. Journal of Experimental Child Psychology, 36, 321–339. Reitsma, P. (1983b). Word-speciﬁc knowledge in beginning reading. Journal of Research in Reading, 6, 41–56. Reitsma, P. (1989). Orthographic memory and learning to read. In P. G. Aaron & R. M. Joshi (Eds.), Reading and writing disorders in diﬀerent orthographic systems (pp. 51–73). Dordrecht/Norwell, MA: Kluwer Academic. Scarborough, D. L., Cortese, C., & Scarborough, H. S. (1977). Frequency and repetition eﬀects in lexical memory. Journal of Experimental Psychology: Human Perception and Performance, 3, 1–17. Scott, J. A., & Ehri, L. C. (1990). Sight word reading in prereaders: Use of logographic vs. alphabetic access routes. Journal of Reading Behavior, 22, 149–166. Seymour, P. H. K., Aro, M., & Erskine, J. M. (2003). Foundation literacy acquisition in European orthographies. British Journal of Psychology, 94, 143–174. Share, D. L. (1995). Phonological recoding and self-teaching: Sine qua non of reading acquisition. Cognition, 55, 151–218. Share, D. L. (1999). Phonological recoding and orthographic learning: A direct test of the self-teaching hypothesis. Journal of Experimental Child Psychology, 72, 95–129. Share, D. L., & Levin, I. (1999). Learning to read and write in Hebrew. In M. Harris & G. Hatano (Eds.), Learning to read and write (pp. 89–111). Cambridge, UK: Cambridge University Press. Share, D. L., & Shalev, C. (in press). Self-teaching in dyslexic and normal readers. Reading and Writing. Shatil, E. (1997). Predicting reading ability: Evidence for cognitive modularity. Unpublished doctoral dissertation, University of Haifa. 298 D.L. Share / Journal of Experimental Child Psychology 87 (2004) 267–298 Shatil, E., & Share, D. L. (2003). Cognitive antecedents of early reading ability: A test of the cognitive modularity hypothesis. Journal of Experimental Child Psychology, 86, 1–31. Shatil, E., Share, D. L., & Levin, I. (2000). On the contribution of kindergarten writing to Grade 1 literacy: A longitudinal study in Hebrew. Applied Psycholinguistics, 21, 1–21. Shimron, J. (1993). The role of vowels in reading: A review of studies of English and Hebrew. Psychological Bulletin, 114, 52–67. Shimron, J., & Navon, D. (1981–1982). The dependence on graphemes and on their translation to phonemes in reading: A developmental perspective. Reading Research Quarterly, 17, 210–228. Somech, C., & Share, D. L. (in preparation). The contribution of morphology to decoding and orthographic learning. Unpublished manuscript, University of Haifa. Stanovich, K. E. (1992). Speculations on the causes and consequences of individual diﬀerences in early reading acquisition. In P. Gough, L. Ehri, & R. Treiman (Eds.), Reading acquisition (pp. 307–342). Hillsdale, NJ: Lawrence Erlbaum. Stuart, M., & Coltheart, M. (1988). Does reading develop in a sequence of stages? Cognition, 30, 139–181. Torgesen, J. K., Alexander, A. W., Wagner, R. K., Rashotte, C. A., Voeller, K. K. S., & Conway, T. (2001). Intensive remedial instruction for children with severe reading disabilities: Immediate and longterm outcomes from two instructional approaches. Journal of Learning Disabilities, 34, 33–58. Torgesen, J. K., Wagner, R. K., & Rashotte, C. A. (1997). Prevention and remediation of severe reading disability: Keeping the end in mind. Scientiﬁc Studies in Reading, 1, 217–234. Treiman, R., & Rodriguez, K. (1999). Young children use letter names in learning to read words. Psychological Science, 10, 334–338. Treiman, R., Tincoﬀ, R., & Richmond-Welty, E. D. (1996). Letter names help children to connect print and speech. Developmental Psychology, 32, 505–514. Tunmer, W. E., Herriman, M. L., & Nesdale, A. R. (1988). Metalinguistic abilities and beginning reading. Reading Research Quarterly, 23, 134–158. Wang, M., & Geva, E. (2003). Spelling performance of Chinese children using English as a second language: Lexical and visual–orthographic processes. Applied Psycholinguistics, 24, 1–25. Wimmer, H., & Goswami, U. (1994). The inﬂuence of orthographic consistency on reading development: Word recognition in English and German children. Cognition, 51, 91–103. Ziegler, J. C., Perry, C., Ma-Wyatt, A., Ladner, D., & Schulte-Korne, G. (2003). Developmental dyslexia in diﬀerent languages: Language-speciﬁc or universal? Journal of Experimental Child Psychology, 86, 169–193.
© Copyright 2020