Content uploaded by David L. Share
Author content
All content in this area was uploaded by David L. Share on May 30, 2023
Content may be subject to copyright.
Self-teaching in normal and disabled readers
DAVID L. SHARE and CARMIT SHALEV
Faculty of Education, University of Haifa, Israel
Abstract. This study set out to investigate the self-teaching of good and poor readers in
pointed Hebrew – a highly regular orthography. Four groups of children (three groups
in Grades 4 to 6, and one group in Grade 2) were included in this study; poor readers
with large discrepancies between IQ and reading (‘dyslexics’), IQ-nondiscrepant poor
readers (non-dyslexic or ‘garden-variety’ poor readers), chronological-age matched nor-
mal readers, and a group of younger normal readers matched to the older garden-vari-
ety group on both reading and mental age. It was hypothesized that primary deficits in
phonological recoding (decoding) would impair the identification of novel target words
(fictitious names of fruits/towns/stars/coins, etc.) appearing in text, which, in turn,
would lead to deficient orthographic memory for target spellings. Alternative predic-
tions were derived with regard to the degree of orthographic deficiency. According to
the ‘compensatory processing’ hypothesis, orthographic learning was expected to be rel-
atively less impaired among disabled readers compared to normal readers. The alterna-
tive ‘dissociation’ hypothesis, on the other hand, predicts that disabled readers’
orthographic learning would be significantly more impaired than that of normal readers.
Neither hypothesis was supported. Impaired orthographic learning, commensurate with
levels of target decoding success, was evident in the post-test spelling and orthographic
choices of both groups of poor readers. Indeed, a close link was observed between levels
of target word decoding and the acquisition of orthographic information among all
three older groups of children. No qualitative differences between dyslexics and garden-
variety poor readers emerged in patterns of self-teaching. While the data from the three
older groups supported a model of developmental delay rather than deviance, findings
from the younger reading-age/mental-age controls revealed startling qualitative diver-
gence in orthographic learning. No statistically reliable evidence was obtained for
orthographic learning in these younger beginning readers who displayed an essentially
‘surface’ pattern of non-lexical reading. A hybrid ‘orthographic sensitivity’ hypothesis
was proposed to account for these data, according to which an initially surface-style of
word reading engendered by a highly regular orthography gives way to a highly special-
ized print-specific (orthographic) processing advantage that develops in the course of
the second school year as an outgrowth of a critical volume of print experience.
Key words: IQ-discrepancy, Orthographic learning, Phonological recoding, Reading
development, Reading disabilities, Self-teaching
Introduction
The present study examined the printed word learning of good and poor
Hebrew readers within the framework of the self-teaching model
Reading and Writing: An Interdisciplinary Journal 17: 769–800, 2004.
Ó2004 Kluwer Academic Publishers. Printed in the Netherlands.
769
proposed by Share and Jorm (Jorm & Share, 1983; Share, 1995, 1999;
Share & Jorm, 1987). According to this model, phonological recoding
(print-to-sound translation) functions as a self-teaching mechanism
enabling the learner to acquire the detailed orthographic representations
necessary for fast, efficient, visual word recognition and for proficient
spelling. Each successful identification (decoding) of a new word in the
course of a child’s independent reading of text is assumed to provide an
opportunity to acquire the word-specific orthographic information that is
the foundation of skilled visual word recognition. Relatively few (success-
ful) exposures appear to be sufficient for acquiring orthographic represen-
tations, both for adult skilled readers (Brooks, 1977) and young children
(Hogaboam & Perfetti, 1978; Manis, 1985; Reitsma, 1983a, b). In this
way, phonological recoding acts as a self-teaching mechanism or built-in
teacher enabling a child to independently develop both the word-specific
and general orthographic knowledge essential to skilled reading.
The self-teaching hypothesis was first tested among normal second
grade Hebrew readers using an experimental paradigm adapted from
Reitsma (1983a) consisting of multiple presentations of target words
embedded in natural text (Share, 1999). The targets were simply novel
letter strings representing fictitious names for places, animals, fruits etc.
Orthographic targets appeared either 4 or 6 times per text. According
to the self-teaching hypothesis, children who succeed in decoding the
targets will begin to acquire knowledge of their orthographic forms,
such that the correct form will be recognized and recalled beyond
chance when encountered again.
Three days after text reading, orthographic learning was assessed in
three ways. The first, orthographic choice, required children to select
the correct spelling of the target item from among four alternatives (the
original spelling, a homophonic foil, and two non-homophonic spellings
containing either substituted or transposed letters). The second measure
of orthographic learning simply required the children to read aloud a
list of words appearing on a computer screen. Embedded in this list
were both the original spelling and the homophonic spellings of the tar-
gets. Finally, each child was asked to reproduce from memory (write)
the target spelling.
Forty normal second grade readers participated in this first direct
test of the self-teaching hypothesis. Targets were correctly decoded (at
the consonantal level) on 84% of occasions. In the orthographic choice
task, the original target spellings were correctly selected well beyond
chance (74%). Naming times for the original spellings were also signifi-
cant faster for the original spellings than for the homophonic spellings,
with no differences in error rates. Finally, on the spelling measure, the
770 DAVID L. SHARE AND CARMIT SHALEV
correct spelling was reproduced in its entirety on 41% of occasions
compared to only 10% for the homophonic spelling. When spelling was
scored on a per-letter basis as opposed to a whole-word basis, correct
target letters were reproduced on 67% of occasions vs. only 29% for
homophonic letters. The overall pattern of results held for both the
4-exposure and 6-exposure conditions with only small non-significant
advantages for the 6-exposure condition.
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, Experi-
ments 2, 3, & 4) indicating that successful orthographic learning was
determined by what the child said when identifying/decoding the target
item and not what the child merely saw. These findings have now been
replicated and extended to the English language by Cunningham, Perry,
Stanovich and Share (2002). This study also examined sources of vari-
ance in orthographic learning. A substantial correlation (r¼0.52) was
found between orthographic learning and the number of target words
correctly decoded during story reading: verbal and non-verbal intelli-
gence, on the other hand, did not correlate significantly with ortho-
graphic learning.
The primary goal of the present investigation was to examine the
nature of self-teaching among disabled readers. The self-teaching
hypothesis generates a number of predictions with regard to the nature
of the difficulties likely to be experienced by poor readers in learning to
recognize words. First, self-teaching is expected to be less efficient
among poor readers owing to primary word recognition difficulties
(Adams & Bruck, 1993; Perfetti, 1985; Share, 1995; Stanovich, 1988,
1998; Vellutino & Scanlon, 1987). The earlier self-teaching study (Share,
1999, Experiment 3) showed that when the level of decoding accuracy
of the target words is reduced by experimental manipulation, ortho-
graphic learning declines commensurately. And, as already noted,
Cunningham et al. (2002) reported a positive correlation between target
word decoding and orthographic learning. Accordingly, we predicted
that well-known problems in phonological recoding among disabled
readers would impair their ability to decode novel words in text which,
in turn, would lead to deficient assimilation of orthographic informa-
tion.
This study also tested two alternative predictions with regard to the
degree of orthographic impairment expected in disabled readers. Our
first specific prediction derives from reading-level (RL) studies address-
771SELF-TEACHING
ing the issue of whether differences between good and poor readers are
quantitative or qualitative in nature (e.g., Rack, Snowling, & Olson,
1992; Siegel, Share & Geva, 1995; Stanovich & Siegel, 1994). These
studies have demonstrated that, when compared to younger reading-
level matched controls, older poor readers tend to have inferior phono-
logical skills yet equivalent or even superior orthographic knowledge. A
key issue in the current literature concerns the source of the ortho-
graphic ‘advantage’ among disabled readers. One possible explanation,
that we will refer to as the ‘‘compensatory processing’’ hypothesis,
assumes that a basic processing advantage in the visual domain, or,
alternatively, an acquired processing style, permits disabled readers to
assimilate word-specific orthographic information more efficiently than
younger non-disabled readers. However, attempts to establish a specific
link between visual processing and orthographic knowledge as yet have
not borne fruit (see Jackson & Coltheart, 2001; Stanovich, 1998). Sev-
eral researchers (e.g., Castles, Datta, Gayan & Olson, 1999; Stanovich
& Siegel, 1994) have proposed that the relative advantage in ortho-
graphic knowledge may primarily be a product of the greater volume of
print experience among older disabled readers relative to younger nor-
mal readers. To date, this alternative ‘print exposure’ hypothesis has
not been directly tested: It tends to be the default option when specific
visual abilities fail to explain significant variance.
The present study permits a strong test of the compensatory process-
ing hypothesis by examining the on-line acquisition of orthographic
information when print exposure is held constant. Presenting the identi-
cal number of target exposures for both normal and poor readers,
implies that any relative advantage in orthographic processing should
find expression in better than expected levels of orthographic learning,
that is, disabled readers’ orthographic learning of target words should
be significantly less impaired than their ability to decode these same
items. Note that the self-teaching hypothesis argues that because ortho-
graphic learning is so heavily constrained by decoding proficiency, any
decoding deficits must be detrimental to orthographic learning, thus
poor readers will be unable to attain levels of orthographic learning
commensurate with normal decoders. If, of course, it were possible to
entirely bypass decoding difficulties and build an orthographic lexicon
purely on the basis of visual learning and/or contextual guessing, dis-
abled readers should demonstrate the same absolute level of ortho-
graphic learning as normal readers. Such a finding would be an
unequivocal refutation of the self-teaching claim that decoding ability is
asine qua non of reading acquisition. Thus, the processing hypothesis
predicts that when print exposure is controlled, disabled readers relative
772 DAVID L. SHARE AND CARMIT SHALEV
to normal readers will show an interaction between decoding success
and orthographic learning such that orthography will be significantly
less impaired.
Quite a different set of predictions regarding disabled readers’ self-
teaching emerge from empirical investigations that have assessed ortho-
graphic learning among disabled readers either directly or indirectly.
Orthographic learning among disabled readers has been directly exam-
ined in Dutch (Reitsma, 1983b) and in English (Ehri & Saltmarsh,
1995; Hogaboam & Perfetti, 1978).
Hogaboam and Perfetti (1978, Experiment 2) taught skilled and less
skilled fourth graders a set of pseudowords presented either aurally or
in print. The skilled readers were those scoring above the 60th percen-
tile on a test of reading comprehension; the less skilled readers scored
below the 40th percentile. Items were practiced an average of 15 times
over a period of 3 days. A day later, both skilled and less-skilled read-
ers named the items that had been seen and pronounced in printed
form faster than the items that were only heard and spoken, and this
advantage was maintained at 10-week retest. This orthographic learning
effect was replicated in a follow-up study (Experiment 3) with skilled
and less-skilled third graders who received 0, 3, 6, 9, 12, or 18 expo-
sures to pseudowords presented either aurally or visually. Evidence of
orthographic learning was obtained in both reader groups at each of
the five exposure conditions although the differences in naming latencies
did not reach significance possibly owing to the small sample sizes
(n ¼5). There were no group-by-condition interactions in either study.
In Reitsma’s study (Reitsma, 1983b, Experiment 3), 18 first graders
(also split into more and less skilled subgroups) and 13 learning-dis-
abled 3rd grade readers read aloud meaningful sentences containing
words judged to be familiar in spoken but not written form. Test sen-
tences were read either 0 (control), 2, 4 or 6 times across 2 days. Three
days later, target words and their homophonic foils were presented for
naming. Target spellings were named more quickly than foils by both
first grade subgroups for 4 and 6 exposures but not for 2 exposures or
unseen (control) items. The learning disabled group, however, showed
no target/homophone effect. Three weeks later, the reading-disabled
group was given a set of 20 highly familiar items and homophonic foils.
A significant latency advantage for the original spellings revealed sensi-
tivity to orthographic detail in this group. Reitsma’s Dutch data sug-
gest that orthographic learning does occur in disabled readers but
seems to require many more exposures than normal readers.
Ehri and Saltmarsh (1995) taught 30 English-speaking first graders
(also split into more and less skilled readers) and 15 older disabled
773SELF-TEACHING
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, homophonic spellings (e.g., KRADL) and non-
homophonic (e.g., KRATL) spellings. Spelling alterations appeared in
either initial, medial or word-final positions. 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 disabled and
less-skilled beginners both required 9 trials. On the homophonic foils,
both high-ability and low-ability beginners, but not older disabled read-
ers, read the target spellings faster than the homophonic foils. On the
phonetically altered spellings, all three groups showed an advantage for
the original spellings. Beginning readers were sensitive to letter altera-
tions in medial as well as initial and final positions, but disabled readers
were sensitive only to initial and final position changes.
Together these experimental studies suggest that, contrary to the
compensatory processing hypothesis, learning-disabled readers, but not
merely less-skilled readers, of English and Dutch have greater rather
than less difficulties assimilating orthographic detail, requiring much
more exposure and practice than normal readers.
Another body of evidence pertinent to the issue of self-teaching
among disabled readers comes from experimental training studies that
have examined the long-term effects of remedial instruction in phono-
logical skills (phonemic awareness and phonological recoding) not only
on generalized decoding (word attack) skills but also on real-word (and
hence more orthographically-based) word identification (Foorman,
Francis, Winikates, Mehta, Schatschneider, & Fletcher, 1997; Lovett,
Lacerenza, Borden, Frijters, Steinbach, & De Palma, 2000; Olson, Wise,
Ring, & Johnson, 1997; Torgesen, Alexander, Wagner, Rashotte,
Voeller, & Conway, 2001; Torgesen, Wagner & Rashotte, 1997).
The first generation of these carefully controlled evaluations of reme-
dial instruction in phonological skills for disabled readers indicated that
the decoding skills of even severely disabled readers can be significantly
improved, at least according to an accuracy criterion, but failed to find
reliable or enduring transfer to real-word identification (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 demon-
strating sustainable and generalized gains in real-word reading. These
latter studies support the notion of self-teaching among disabled read-
ers but suggest that treatment intensity may be a critical factor in devel-
774 DAVID L. SHARE AND CARMIT SHALEV
oping the automatized decoding skills that seem to be pre-requisite to
reliable growth in orthographically-based reading in severely disabled
readers (see Foorman & Torgesen, 2001; Olson et al., 1997).
These North-American training studies with disabled readers,
together with the two experimental studies discussed earlier (Ehri &
Saltmarsh, 1995; Reitsma, 1983b) suggest that, for disabled readers, the
relationship between decoding ability and orthographic learning may,
in fact, be weaker than among normal readers. We will refer to this
account as the ‘dissociation’ hypothesis. If correct, it predicts that a
given level of decoding among disabled readers should be associated
with even lower levels of orthographic learning compared to normal
readers at the same levels of decoding ability. This prediction is pre-
cisely the opposite of the compensatory processing hypothesis discussed
above.
The present study examined the self-teaching of disabled readers in a
highly regular orthography. In contrast to Germanic-language orthog-
raphies such as English and Dutch, pointed Hebrew adheres strictly to
the principle of one-to-one correspondence; each letter and each vowel
diacritic has one, and only one, phonemic value. For a brief overview
of the Hebrew language and orthography, see Share and Levin (1999).
Our study also addressed one of the most controversial issues in the
definition and classification of reading difficulties, namely, the relevance
or otherwise of IQ. A growing number of theorists have begun to ques-
tion the once sacrosanct role of IQ in formal definitions of reading dis-
ability/dyslexia (see, for example, Aaron, 1997; Fletcher et al., 1994;
Share, McGee, McKenzie, Williams, & Silva, 1987; Share, 1996; Shay-
witz, Escobar, Shaywitz, Fletcher, & Makuch, 1992; Siegel, 1989;
Stanovich, 1991, 1994; Stuebing, Fletcher, LeDoux, Lyon, Shaywitz &
Shaywitz, 2002; Vellutino, Fletcher, Snowling, & Scanlon, 2004; Velluti-
no, Scanlon, & Lyon, 2000). Evidence is accumulating that few if any
differences may exist between normal-IQ poor readers (‘specific reading
disability’ or ‘dyslexia’) and low-IQ poor readers (non-dyslexic or so-
called ‘‘garden-variety’’ poor readers) in the deficient word recognition
processes widely regarded as the major cause of reading failure (see,
e.g., Felton & Wood, 1992; Fredman & Stevenson, 1988; Stanovich &
Siegel, 1994). To date, however, these studies have relied almost exclu-
sively on standard ‘static’ measures of isolated word reading such as
pseudoword reading or orthographic choice that assess previously
acquired knowledge and skills. The self-teaching paradigm (with novel
target words embedded in natural text) offers a tool for comparing the
on-line dynamics of printed word learning by both types of poor read-
ers (dyslexic and non-dyslexic). The comparison of both normal-IQ
775SELF-TEACHING
(dyslexic) and low-IQ (non-dyslexic or garden-variety) poor readers in
the present study provides an additional avenue of information regard-
ing possible differences in the reading processes of these two groups –
differences that are critical to the taxonomic validity of the concept of
dyslexia. On the basis of the extant literature, we predicted that no
qualitative differences would be observed in the word learning of these
two groups.
By including both dyslexic and garden-variety poor readers, this
study permitted a methodological refinement of the now standard 3-
group design in which poor readers are compared with both younger
normally-progressing control readers matched on Reading Level (the
so-called ‘RL’ match) and age-matched normal readers (the chronologi-
cal-age or ‘CA’ match). Although this 3-group design affords both the
twin examination of both RL and CA differences between good and
poor readers, it has one major disadvantage that is often overlooked in
the literature. The key reading level comparison between the older poor
readers and the younger good readers (the one maximally diagnostic on
the issue of developmental deviance versus delay) is confounded by
mental age differences in the case of normal-IQ poor readers. This
means that any reading-level performance advantage among disabled
readers may be partly or wholly an artefact of superior cognitive and/
or metacognitive skills such as planning and monitoring task perfor-
mance, strategy knowledge and, in certain cases, speed-of-processing.
This problem can be remedied by including a low-IQ poor reading
group who are matched to the younger normal readers on both mental
age as well as reading level. A qualitative difference between poor read-
ers and good readers matched for both reading level and mental age
would constitute more definitive evidence on the deviance/delay issue
than is currently obtained from the standard RL comparison flawed by
the mental age confound.
To reiterate, the self-teaching of four groups of readers was com-
pared in this study; poor readers (both dyslexic and garden-variety),
chronological age-matched good readers, and younger reading-level
matched normal readers who were also matched on mental age to the
non-dyslexic (‘garden-variety’ poor readers). More generally, lower lev-
els of orthographic learning among poor readers were expected in light
of their decoding difficulties. More specifically, the ‘dissociation’
hypothesis predicts that depressed target decoding in disabled readers
would be associated with significantly lower levels of orthographic
learning compared to normal readers at the same level of decoding abil-
ity. The opposite prediction is made by the compensatory processing
hypothesis. Under conditions examining the acquisition of orthographic
776 DAVID L. SHARE AND CARMIT SHALEV
information when print exposure is held constant, any compensatory
processing advantage for poor readers should find expression in rela-
tively higher levels of orthographic learning, that is, orthographic learn-
ing should be significantly less impaired than target decoding of the
same items.
Method
Sampling
Because no reliable, standardized, norm-referenced tests existed at the
time this investigation began, it was necessary to develop the relevant
tests and norms in order to establish the diagnosis of specific reading
disability on an objective basis. To this end, we adopted a two-stage
process that included initial identification of potentially disabled read-
ers on the basis of group-wise screening of large numbers of children
in regular classrooms followed up by individual testing of selected
candidate children. In Haifa, with the exception of a single special
school for severely learning disabled children, almost all reading-
disabled children attend regular school classes with occasional with-
drawal by a remedial teacher and/or after-hours remediation in a
remedial Learning Centre.
Screening measures at initial screening
Silent reading comprehension (Ortar, 1987). This test of reading compre-
hension is designed for children from Grades 2 to 8 and includes 68
multiple-choice items graded in difficulty from simple one word items
(to be matched to a picture), short phrases, simple pointed sentences,
short unpointed passages to progressively longer and more complex
passages.
Orthographic parsing (word recognition). In this task, a child is pre-
sented with a sequence of four words strung together without word
spacing. The child’s task is to identify and mark the individual words.
The child was presented with a page of 30 such strings, and instructed
to mark as many words as possible in a 3-minute time period. One
example was given prior to test administration.
Phonological choice (Adapted from Olson, Kleigl, Davidson, &
Foltz, 1985). In this test of phonological decoding, a child is presented
with a pair of pseudowords, one of which is homophonic to a real
777SELF-TEACHING
word. The child is asked to circle the word that sounds the same as a
real word. There were 10 pairs of items on this test. Testing was pre-
ceded by 3 demonstration items.
Peabody Picture Vocabulary (Israeli adoption, Solberg, & Nevo,
1979). This test was adapted for whole-class testing by presenting the
plates, each containing four pictures, on an overhead projector. The test
word was named aloud together with the corresponding number on a
test blank placed in front of each child. A child indicated his/her choice
by checking one of four choices. As it was not possible to establish
individual basal and ceiling levels as is customary in individual testing,
it was decided to administer a subset of 55 items (items 35 through 90)
to all the sample. To discourage copying, two parallel record blanks
(one arrayed horizontally, the other vertically) were given to adjacent
children.
Initial selection of a candidate pool of poor readers
Initial screening took place in 37 classrooms drawn from five schools
and included at total of 1139 children. Valid data were available on
comprehension for 996 children and for 1015 on composite word recog-
nition (phonological choice and orthographic parsing). Complete data
were obtained for 883 children (316 in Grade 4, 303 in Grade 5, and
264 in Grade 6).
Both comprehension and word recognition measures were used to
define an initial pool of potential poor readers. Both sources of infor-
mation regarding reading ability were employed in order to reduce
regression effects and also to broaden the basis of the definition to
include not only the core word recognition deficit of disabled readers
but also the comprehension component of reading regarded as central
in the contemporary Israeli elementary school curriculum.
Our word recognition screening measure was a composite derived by
averaging Grade-level standard scores for orthographic parsing and
phonological choice. Children were then selected who fell at or below
the grade-appropriate 30th percentile on both reading comprehension
(n¼996) and composite word recognition (n¼1015).
1
There were a total of 124 children (39 girls and 85 boys) (or 12% of
the sample) with reading performance below the 30th percentile on both
comprehension and word recognition. The initial screening ‘net’ was
cast liberally with the expectation that parental permission required for
individual testing would be forthcoming for only some of this group,
and that further sample attrition would be expected at follow-up test-
ing.
778 DAVID L. SHARE AND CARMIT SHALEV
The next step was to seek parental permission for this pool of can-
didates and then to select chronological-age (CA) matched control chil-
dren with reading and IQ in the normal range. Poor readers were then
split into (preliminary) high-IQ (dyslexic) and low-IQ (non-dyslexic or
garden-variety) subgroups based on a median split of Peabody scores.
Since our intention was to locate a group of chronological-age
matched children reading at normal levels for their age, it was deemed
necessary to find children matching only the dyslexic readers whose
IQs were in the same normal range. Candidate CA control readers
were selected by choosing a child of the same gender and homeroom
class and with a similar Peabody score to each candidate dyslexic
reader, with the additional proviso that reading comprehension and
word recognition were close to the sample mean. As regards the gar-
den-variety (low-IQ) poor readers, no attempt was made to locate CA
controls with matching IQs since such a group (if they exist in suffi-
cient numbers) would be required to have an inverse discrepancy
between (low) IQ yet normal reading, i.e., a highly abnormal ‘hyper-
lexic’ profile. Younger reading-level controls were then selected who
matched the garden-variety (low-IQ) poor readers on both reading and
mental age and who exhibited a ‘normal’ (non-discrepant) reading/IQ
profile for their age and grade.
Finally, reading and IQ were individually assessed in order to con-
firm the presence of a reading difficulty.
Individual assessment of reading and IQ
Intelligence (Raven’s Progressive Matrices, Raven, Raven, & Court,
1976). This standard measure of non-verbal intelligence was adminis-
tered according to the test manual.
Word recognition. Two measures of word and pseudoword reading
were administered. For each measure both speed and accuracy were
assessed.
Word reading (Shatil, 1997). Children were asked to read aloud, as
quickly and accurately as possible, a list of 112 pointed (fully voweled)
words graded in length from 1 to 5 syllables.
Pseudoword reading (Bentin, Deutsch, & Liberman, 1990). In this
test, the child is required to read aloud a list of 24 pointed pseudo-
words (3–4 letters in length).
Using ‘local’ norms specially developed for these two measures, a
single composite was created by averaging the four scores (speed and
779SELF-TEACHING
accuracy of word and pseudoword reading) after first converting all
scores to Z-scores. As before, children with scores at or below the 30th
percentile on this composite were designated as poor readers. Raven
scores (which correlated 0.42 with the reading composite) were then
used to classify this group as IQ-discrepant (dyslexic) or non-discrepant
(non-dyslexic or garden-variety) poor readers. IQ-reading discrepancies
(in standard errors) ranged from +0.29 (i.e., slight ‘overachievement’)
to –3.33 (severe underachievement). It can be seen from Table 1 that
the dyslexic group are severe underachievers averaging over two stan-
dard errors below prediction. Table 1 also shows that dyslexics had
similar IQs to CA controls, while garden-variety poor readers had sig-
nificantly lower IQs. Dyslexics, however, were reliably poorer readers.
This would appear to be the product of the decision to include only
poor readers in an absolute (and not merely relative) sense. Under the
present selection procedures, larger IQ-reading discrepancies will tend
to be children with higher IQs and lower reading levels, whereas the
garden-variety (non-discrepant) group will tend to be those whose IQ
and reading tend to cluster in the middle of the range. It should be
remarked that the division of these 40 poor readers into two groups of
dyslexic and non-dyslexic groups is an entirely arbitrary subdivision
along a continuum of discrepancy since there was no evidence of depar-
ture from normality.
Although the two groups of poor readers were now ‘unmatched’ on
reading (word recognition) level, it was possible to ‘equate’ the two
groups statistically by covarying reading level (following Stanovich &
Siegel, 1994).
2
Selection of Grade 2 reading-level/mental-age controls
Mental age matching. Early in Grade 2 (November), the Raven test
was administered to 59 children (for whom parental consent was
obtained) from the same five schools as the main sample. Only sets A
and B (not A
B
) of the Standard (black-and-white) Raven were adminis-
tered since these were the only two subtests common to both the Col-
ored Raven (appropriate for this younger age group) and the Standard
Raven for the older children in Grades 4–6. The distribution of Raven
raw scores in this younger group ranged from 8 to 24 compared to 12
to 24 in the older garden-variety group. Of 59 children for whom
parental permission was granted, 12 children with raw scores below 12
(the lowest raw score in the garden-variety group) were dropped. A fur-
ther 12 children with evidence of reading difficulties (particularly slow-
ness) were excluded. Of the remaining children, a group of 20 were
780 DAVID L. SHARE AND CARMIT SHALEV
selected to match the older garden-variety poor readers as closely as
possible on gender, composite word recognition and Raven raw score.
As can be seen in Table 1, the two groups were well matched on Raven
raw scores.
Reading level matching. Test scores for the word and pseudoword
reading measures showed that, as a group, the Grade 2 children had
Table 1. Background data for four groups of readers.
CA controls Dyslexics Garden-variety RL/MA
controls
Grades (4, 5 & 6) n=6,8&6 n=8,6&6 n= 6, 5 & 9 Grade 2
Age 10.9
a
(0.80) 10.9
a
(1.06) 11.1
a
(0.90) 7.6
b
(0.55)
Gender
(girl:boy ratio)
5:15 7:13 6:14 6:14
Comprehension 73% (8.05)
a
56% (12.09)
b
59% (11.91)
b
IQ 103.8 (11.39)
a
101.9 (8.87)
a
92.7 (12.19)
b
Raven raw score
(MA)
18.6 (3.16)
a
18.5 (2.50)
a
IQ/reading
discrepancy
(standardized
residual)
0.04 (.52)
a
)2.07 (.55)
c
).74 (.52)
b
Word recognition
(Z-score composite)
0.23 (.32)
a
)1.70 (.58)
c
).90 (.50)
b
)1.17 (.78)
b
Word reading
accuracy
95% (2.88)
a
81% (7.44)
c
86% (6.91)
b
87% (9.53)
b
Word reading
speed (in s)
92.4 (17.5)
a
143.6 (33.8)
c
112.1 (24.9)
b
163.8 (29.4)
d
Pseudoword
reading accuracy
86% (14.5)
a
57% (13.9)
c
67% (15.1)
b
76% (17.5)
ab
Pseudoword
reading speed
37.1 (7.9)
a
58.5 (15.2)
c
45.9 (14.0)
b
47.1 (9.6)
b
Note: Superscripts indicate significant differences ranging from
a
(highest) to
d
(lowest).
Groups sharing the same superscripts are not significantly different from each other.
Groups with 2 superscripts were not significantly different from the two adjacent (higher
and lower) groups. For example, the pseudoword reading accuracy in the RL/MA
group (marked
ab
) fell between the CA controls and the garden-variety poor readers but
was not significantly different from these two groups (p= 0.06, and 0.075 respectively).
The latter two groups, however, differed significantly as indicated by the two distinct
superscripts.
781SELF-TEACHING
fewer errors but were much, much slower (approximately 2 standard
deviations) than the older poor readers. Nonetheless, averaging all four
components (word and pseudoword speed and accuracy) produced sim-
ilar composite scores (see Table 1). It was decided to use this age group
because younger children would be expected to commit more errors,
but be even slower, while older children would be more comparable in
terms of speed but make even fewer errors. Since the three older groups
were all derived on the basis of the composite measure this seemed the
best choice for controls.
As can be seen in Table 1, although the two groups diverged in
speed of reading real words, they were comparable (and not signifi-
cantly different) on either speed or accuracy of pseudoword reading.
Since the critical independent variable, as far as the self-teaching
hypothesis is concerned, is phonological recoding, this group of average
to above-average second grade readers was considered to be the appro-
priate control group.
Design. The basic paradigm, as in Share (1999), consisted of target
words (fictitious pseudoword names for fruits/towns/coins/stars etc.)
embedded in short passages (see Appendix for a complete list of the
target items). The experimental design was a 2 X 2 design; the two
factors being number of exposures of the target word (either 2 or 6)
and target word spelling (two homophonic spellings of the same
word).
Each child read aloud a total of 10 short texts (5 texts per test
session) averaging 61 words in length (range: 31–86 words). For each
child, five texts contained 2 target exposures and five contained 6
target exposures (exposure order was alternated from text to text).
Exposure was manipulated simply by replacing target items with
either a generic term, anaphoric term, or with a synonym. To con-
trol for order effects, text order was rotated according to a Latin
Square design. Ten children saw one set of target spellings, while the
other ten saw the alternate homophonic spelling. This procedure was
adopted to control for any spurious orthographic learning that might
be attributable to pre-existing spelling preferences. Targets ranged in
length from two to four syllables, and from 4 letters to 6 letters
(‘mothers of reading’ included). Each target contained two homopho-
nic letters. The same target pseudowords were used as in the earlier
study (Share, 1999) with certain minor modifications (two final con-
sonant clusters were simplified by eliminating the penultimate conso-
nant, and some additional vowel alterations were made in the light
of children’s reading errors in the earlier study).
782 DAVID L. SHARE AND CARMIT SHALEV
Materials and measures
As in the earlier self-teaching study, each story was followed by three
open-ended questions testing factual comprehension. These were
designed to ensure that children read with understanding.
Post-test measures assessing orthographic learning.
Three days after reading the test texts, orthographic learning was
assessed with three separate measures:
1. Spelling. The first measure of orthographic learning required the
spelling (free written reproduction) of the target item.
2. Word naming. Each child was presented with two lists each con-
taining 10 words (each of the 5 targets appeared twice) in fixed
random order. One list contained target spellings as seen three
days earlier; the second list contained only alternate homophonic
spellings. List order was counterbalanced to control for practice
effects. As an additional precaution, the two lists were separated
by a time interval of approximately 5– 10 minutes in which sub-
jects read and answered comprehension questions on a simple text.
Naming was also recorded on cassette tape for later error analysis.
3. Orthographic choice. The child was shown two words; one was the
original spelling of the target word read three days earlier, the
other was the alternate homophonic spelling. Location of these
alternatives was rotated across trials.
Procedure
Text reading
Each child was asked to read some short ‘stories’ aloud, and to
decide which one s/he liked most. Children were requested to read
by themselves and to make sure they understood the story because
questions would be asked immediately afterwards. No help was given
except for the title of the passage. Following text reading, the chil-
dren were asked which story they liked most and why. Text reading
was tape-recorded.
Post-test measures
Post-tests for orthographic learning took place 3 days later. The second
set of texts were administered after an interval of at least several days.
783SELF-TEACHING
Post-test measures were administered in a fixed order (spelling, naming,
orthographic choice).
All testing was carried out individually, in school.
Results
Text reading
The texts were found to be well within the reading capabilities of all
groups (see Table 2). Even the weakest readers (dyslexics) read with rel-
atively few errors (90% accuracy) and good comprehension (76%).
Consistent with their performance profile on the initial word reading
measures used for group selection (see Table 1), younger reading-level
(RL) controls committed significantly fewer errors than both groups of
poor readers but had significantly slower reading times. The two groups
Table 2. Accuracy, fluency and comprehension for four groups of readers on the 10
test texts.
Measures Group
CA control Dyslexic Garden-
variety
RL/MA
control
Overall
(n=80)
Comprehension 2.4
a
(0.38)
81%
2.3
a
(0.41)
76%
2.3
a
(0.35)
77%
2.3
a
(0.28)
77%
2.3 (0.36)
78%
Errors 3.1
a
(1.25)
5%
5.9
b
(2.32)
10%
4.4
b
(1.72)
7%
2.5
a
(2.21)
4%
3.97 (2.29)
7%
Time 35.9
a
(4.39) 52.9
b
(13.1) 41.5
a
(6.50) 64.2
c
(15.7) 48.6 (15.34)
Note: Error rates here apply only to non-target words. See Table 3 for target word
decoding.
Table 3. Target decoding success (percentage consonantal accuracy and vowel-plus-
consonant accuracy) for four groups of readers.
Target decoding % CA control Dyslexic Garden-variety RL control
Consonants 96.4
a
(4.69) 80.4
c
(15.53) 87.8
b
(14.42) 92.6
ab
(10.87)
Consonants &
vowels
68.3
a
(19.30) 38.3
b
(17.30) 48.3
b
(22.83) 66.3
a
(20.59)
784 DAVID L. SHARE AND CARMIT SHALEV
of poor readers had a similar error rate and reading comprehension,
but the dyslexic group were significantly slower readers (see Table 2).
Target decoding
Table 3 presents the success rates (in percentages) for decoding the
pseudoword targets. These percentages are presented for both decoding
in which the letter string was correctly decoded (ignoring vowel errors
3
)
as well as for cases in which both vowel diacritics and consonants were
accurately decoded. Older control readers were near ceiling on conso-
nantal accuracy, indeed exactly half of this group achieved perfect
scores, with the younger RL controls slightly but not significantly infe-
rior (8 children attained perfect scores). Both poor reader groups were
significantly below the controls with the dyslexic group the weakest.
Younger RL (garden-variety) controls were not significantly different
from the corresponding garden-variety poor readers.
As expected, accurate decoding of vowel diacritics in addition to con-
sonant decoding was at a far lower level in each of the four groups. This
confirms the consistent finding in the Hebrew-language literature (see
Share & Levin, 1999 for a review) that children commit far more errors
on vowels than on consonants, and extends this finding from isolated
word reading to text reading. Again, the two control groups were at a
similar level with both controls at similarly lower levels of accuracy.
Post-test orthographic learning
Post-test orthographic choices for the four groups are presented in
Table 4.
Table 4. Post-test orthographic choices for four groups of readers.
Group Choices
Overall 2-exposures 6-exposures
CA control 75%
a
73%
a
77%
a
Dyslexic 61%
bc
59%
bc
63%
bc
Garden-variety 67%
ab
65%
ab
70%
ab
Garden-variety controls 57%
c
54%
c
58%
c
Note: All scores above 60% (marked in bold) are statistically significant according to
the binomial test ( p< 0.05).
785SELF-TEACHING
The most striking and unexpected finding was the lack of evidence
for any reliable orthographic learning whatsoever among the younger
garden-variety controls even in the 6-exposure condition. We return to
this issue below but focus, for the moment, on the data for the three
older groups, for whom there was unequivocal evidence of reliable
orthographic learning.
With one exception (the 2-exposure condition among dyslexics),
orthographic choices for the three older groups were all significantly
above the 50% chance level as determined by the binomial test. Fur-
thermore, differences between the 2-exposure and 6-exposure conditions
were consistently small (between 3% and 4% ) and non-significant. Dif-
ferences in the earlier study (Share, 1999) between 4 and 6 exposures,
were of the same magnitude and also non-significant. It seems that the
bulk of orthographic learning occurs on initial encounters with new
words. Performance in the 6-exposure condition among the older con-
trols was at a very similar level to the Grade 2 data reported in Share
(1999, Table 2). Thus, it would also appear that this phenomenon of
Percentage %
100
90
80
70
60
50
Group
Control
GV
Dyslexic
GV control
Decoding Choice
Figure 1. Target decoding and orthographic choice (in percentages) in four groups of
readers (CA-controls, garden-variety (GV) poor readers, dyslexics, and garden-variety
(GV) controls.
786 DAVID L. SHARE AND CARMIT SHALEV
extremely rapid orthographic learning appears to be in place develop-
mentally by the end of the second year of learning to read.
Comparing orthographic choices across the three older groups, con-
trols attained the highest level of orthographic choices and were signifi-
cantly superior to the dyslexics – the poorer of the two poor reader
groups. Data for the garden-variety group bisected the means of the
latter two groups, but fell just short of statistical significance in both
cases. The relative levels of performance on orthographic learning
among these three groups closely tracked the level of target decoding
(see the near-parallel lines for the three groups in Figure 1) – a finding
clearly diverging from the findings reported in English (Ehri &
Saltmarsh, 1995) and Dutch (Reitsma, 1983b).
Consistent with the self-teaching hypothesis, it seems that there is a
fairly close match between levels of target decoding and orthographic
learning among each of these three groups of native Hebrew readers –
a finding also observed among Grade 2 readers in the earlier study in
which differences in decoding levels were created artificially by means
of experimental manipulation. Repeated measures ANOVA examining
the differences between the two measures (target decoding and ortho-
graphic choice) formally confirmed this lack of an interaction between
the three groups (F(2, 57) ¼0.053, ns). As seen in the different line
‘‘heights’’ in Figure 1, there were clear differences in overall levels of
performance as witnessed by a significant main effect for group
(F(2,57) ¼8.09, p¼0.001). However, this main effect vanished once
reading level (pre-existing composite word recognition) was controlled
(F(2,56) ¼0.16, ns). Thus, the differences in target decoding and ortho-
graphic choices among these three groups can be fully accounted for by
pre-existing levels of reading ability, that is, the differences are purely
quantitative. These within-age orthographic choice data, therefore, sug-
gest that the differences in word identification and word learning among
good and poor Hebrew readers (both IQ-discrepant and non-discrep-
ant) in the upper elementary grades appear to fit a picture of develop-
mental delay not deviance. Moreover, the lack of a group-by-measures
interaction among these three groups shows that both IQ-discrepant
and non-discrepant poor readers are qualitatively similar.
The picture alters radically, however, with the inclusion of younger
garden-variety RL/MA controls. Figure 1 shows how the line for this
younger group traverses both the poor reader groups on its descent to
a chance level of post-test orthographic choice. The (cross-over) group-
by-measures interaction (all four groups included) was significant
(F(3,76) ¼3.65, p¼0.016) and remained so even after controlling over-
all reading level (F(3,75) ¼3.54, p¼0.019). Whereas the younger con-
787SELF-TEACHING
trols were more accurate than the older poor readers in decoding tar-
gets, orthographic learning under the present conditions was at chance.
This finding points to a complete dissociation between target decoding
and orthographic learning in this group, a finding clearly contradicting
the claim that self-teaching begins as soon as children are able to
decode new words independently (see Share, 1995). This dissociation,
furthermore, points to fundamental qualitative differences between the
word learning of both groups of older disabled readers (dyslexic and
garden-variety) and younger normal readers.
As already noted, the 1999 study obtained clear evidence of ortho-
graphic learning on all three post-test measures among Grade 2 Hebrew
readers. In the present group of Grade 2 readers, however, there was
no evidence of orthographic learning on any of the three measures (see
below). To what extent might the discrepancy be attributable to sample
differences? Thirteen of the 20 garden-variety controls came from the
same relatively advantaged school as the sample in the 1999 study,
whereas the remaining seven were drawn from two other schools, both
of which were less advantaged. To check whether these sample differ-
ences may have accounted for the difference in results, the scores of the
subset of 13 were re-examined. Orthographic choices and also post-test
spellings for this subgroup (see below) were virtually identical to the
group as a whole (in both cases, in fact, the difference was 1 percentage
point below the overall group mean). It appears then that this ‘surface’
pattern of normal reading may be a genuine characteristic of young
Grade 2 readers in Hebrew’s near perfectly regular orthography. It is
important to note that the present sample were tested during the
months of December and January (approximately half a school year
earlier than the 1999 sample who were tested at the end of the school
year (May/June). Additional data now available from two new samples
of normal end-of-Grade 1 readers confirm the absence of orthographic
learning among beginning Hebrew readers (Share, 2002, 2004). This
finding points to fundamental qualitative changes in reading behavior
occurring in the course of the second year of learning to read a trans-
parent orthography such as pointed Hebrew. We return to this point at
greater length below.
Post-test naming times and naming accuracy
Although target lists were read significantly faster than homophone lists
by dyslexics, no other difference in either accuracy or times approached
significance. The sole difference for dyslexics may therefore be a Type I
788 DAVID L. SHARE AND CARMIT SHALEV
error. In any case, naming times and errors did not prove to be a par-
ticularly discriminating measure, probably because times were not based
on discrete trial latencies but list naming times. No doubt the massive
intra-list (and to a lesser extent inter-list) priming – a product of
repeated presentations of a small set of target items contributed to this
insensitivity. Discrete trial timing was not possible owing to the fact
that testing was carried out concurrently at five different schools and
funds for multiple sets of equipment were unavailable. In the 1999
study, Grade 2 children, tested at the end of the school year, were fas-
ter reading targets than foils when assessed on individual item naming
latencies, so the present results cannot be taken to imply an absence of
orthographic learning among normal readers. Nonetheless, it should be
added that naming performance in the earlier study also proved to be a
less sensitive and less consistent measure of post-test orthographic
learning.
Spelling
Of a total of 1600 target letters (10 words each containing two
homophonic letters times 80 subjects), there were only 15 cases of non-
homophonic letters (all HEY), one case of elision, and five cases of
missing data due to experimenter error. Spelling performance was
tested, as with orthographic choice, with the binomial test (chance level
is 50%). Results appear in Table 5.
Reinforcing the data for orthographic choice, only the spelling
results for the three older groups revealed reliable evidence of ortho-
graphic learning; garden-variety controls were once again at chance.
Target spelling success at 6 exposures among controls was very similar
to the earlier Grade 2 6-exposure data (66%) (Share, 1999). Overall,
Table 5. Post-test spelling of target letters (in percentages) for four groups of readers.
Spelling Control Dyslexic Garden-variety Garden- variety
control
All groups
combined
2-exposures 62%
a
57%
a
55%
a
52%
b
57%
6-exposures 64%
a
54%
b
62%
a
52%
b
58%
Overall 63%
a
56%
b
58%
ab
52%
b
57%
Note: Groups with different superscripts are significantly different from one another.
Figures in boldface type are significantly greater than chance (50%).
789SELF-TEACHING
there was evidence of orthographic learning among the older normal
readers even at 2-exposures confirming the rapid acquisition of ortho-
graphic information observed in the developmental literature (Hogabo-
am & Perfetti, 1987; Reitsma, 1983a, b; Manis, 1985). As with the data
on orthographic choice, the same rank ordering of performance was
reproduced with older controls best, followed by garden-variety poor
readers, then dyslexics. Older controls were significantly superior to
dyslexics but not to garden-variety poor readers who were, as before,
sandwiched in between, and not significantly different from either the
controls or the dyslexics. There were no significant differences between
2- and 6-exposure conditions either overall or for any of the four
groups, replicating the pattern observed among end-of-Grade-2 normal
readers in Share (1999) in which no significant differences were found
between 4- and 6-exposure conditions. As noted above, this outcome
suggests a pattern of orthographic learning in which the bulk of learn-
ing occurs within the first one or two trials, with diminishing gains for
each subsequent exposure (see Share, 2004).
Spelling Decoding
Percentage %
100
90
80
70
60
50
40
Group
control
GV
dyslexic
GV-control
Figure 2. Target decoding and spelling (in percentages) in four groups of readers
(CA-controls, garden-variety (GV) poor readers, dyslexics, and garden-variety (GV)
controls.
790 DAVID L. SHARE AND CARMIT SHALEV
The spelling data also reproduced the picture emerging for the rela-
tionship between target decoding and orthographic choice. Among the
three older groups there was a significant main effect of group (F(2,
57) ¼9.61, p< 0.000) that disappeared entirely once reading level
was controlled (F(2,56) ¼0.094, ns) but no group-by-measures interac-
tion (F(2,57) ¼1.24, ns). This can be seen in the fact that, once again,
the three relevant lines in Figure 2 are nearly parallel. With the youn-
ger controls included, the group-by-measures interaction became sig-
nificant (F(3,76) ¼3.14, p¼0.03), and remained significant even after
controlling for pre-existing reading levels (F(3,75) ¼2.87, p¼0.04).
This is evident in the fact that the line for the younger controls tra-
verses both the lines for the two groups of poor readers as it descends
toward chance. In sum, the spelling data mirror the orthographic
choice data.
Discussion
In this comparison of the self-teaching of good and poor readers, it
was expected that primary deficits in phonological recoding would
directly impair poor readers’ ability to decode novel words encoun-
tered in text. According to the self-teaching hypothesis, orthographic
learning is severely constrained by decoding proficiency, hence any
decoding deficits will be detrimental to orthographic learning. Poor
decoders, therefore, should be unable to attain levels of orthographic
learning comparable to normal decoders. If, on the other hand, it
were possible to entirely bypass decoding difficulties and build an
orthographic lexicon purely on the basis of visual exposure to print
and/or contextual guessing, disabled readers should demonstrate the
same level of orthographic learning as normal readers in spite of their
weak decoding skills. Relative to chronological-age controls, the data
clearly revealed impaired orthographic learning in both groups of
poor readers as expected, and, more generally, a close link between
overall levels of target decoding and orthographic learning. These data
therefore complement the results from the earlier self-teaching study
with normal second grade Hebrew readers (Share, 1999) for whom
orthographic learning was also found to be closely tied to overall lev-
els of target decoding resulting not from naturally occurring differ-
ences in decoding ability but from experimental manipulation (i.e.,
brief presentation of target items coupled with phonological suppres-
sion).
791SELF-TEACHING
Thus far, the data from these three older groups support a develop-
mental delay model not a deviance model. The presence of significant
orthographic learning, commensurate with target decoding success, sug-
gested that poor readers of a highly regular orthography were indeed
able to profit from the opportunities for self-teaching provided by their
attained level of target identification. There was no indication that
orthographic learning was less efficient in comparison with control
readers of the same chronological age. Thus the ‘dissociation’ hypothe-
sis must be rejected. On the other hand, there was no evidence of any
special ‘powers’ with regard to the assimilation of new orthographic
information: The compensatory hypothesis, therefore, must also be
rejected.
Data from the younger reading-level controls, however, painted an
entirely different picture, one that revealed startling qualitative diver-
gence between older poor readers and younger normal readers. These
results are all the more dramatic given that the younger controls and
the older garden-variety poor readers were well matched at pretest on
both accuracy and speed of pseudoword decoding, as well as mental
age according to the Raven test. Among these young Grade 2 readers,
there was no statistically reliable evidence of orthographic learning on
any dependent measure, a finding that stands in stark contrast to the
earlier study of Grade 2 children tested at the end of the school year.
Potential sampling differences did not appear to be responsible for the
discrepant outcomes. Additional self-teaching data now available from
two separate samples of end-of-grade 1children from comparable
schools confirm the lack of orthographic learning among novice readers
(Share, 2002, 2004). In view of the fact that the first 1999 self-teaching
study obtained reliable evidence of orthographic learning in no less
than three samples of children tested at the end of Grade 2, the data
from the present investigation all converge on the hypothesis that there
are fundamental print processing developments occurring between early
and late second grade among native Hebrew speakers/readers. These
data present precisely the inverse pattern of the findings compared to
the Dutch (Reitsma, 1983b, Experiment 3) and English data (Ehri &
Saltmarsh, 1995). In both these studies, first graders demonstrated sig-
nificant and substantial orthographic learning (the pattern observed
here for older disabled Hebrew readers), whereas their older disabled
readers revealed severely impaired orthographic learning (the pattern
obtained in Hebrew for our novice readers).
It is hypothesized that in a highly transparent orthography, such as
pointed Hebrew, the invariant one-to-one relationship between letters
and sounds permits such high levels of decoding (see also Italian, Cossu,
792 DAVID L. SHARE AND CARMIT SHALEV
1999 and German, Wimmer, Landerl & Frith, 1999) so very early in
a child’s orthographic experience, that normal novice readers typically
pass through a phase resembling surface dyslexia. In contrast to both
Dutch and English, each Hebrew letter and vowel diacritic has only
one phonemic value. Moreover, Hebrew orthography has neither vowel
nor consonant digraphs (ea, oa, th, sh), no geminates (ss, ll), no conso-
nant clusters (sp,cl, str) or silent letters. Young novice Hebrew readers
appear able to decode almost any word in a bottom-up, letter-by-letter
or ‘surface’ fashion but are relatively insensitive to surface information,
that is, to the word-specific orthographic form of the printed word
being decoded. The qualifier ‘relative’ acknowledges the fact that novice
readers are not impervious to newly available orthographic informa-
tion. Such an extreme position is clearly refuted by the data showing
significant orthographic knowledge evident in both orthographic and
homophone choice tasks at the end of Grade 1 (Shatil, 1997). On the
other hand, this relative orthographic insensitivity (within the context
of a small number of exposures to novel orthographic targets) revealed
these young novices behaving much like surface dyslexics. It seems
likely that many more exposures would eventually lead to significant
orthographic knowledge as demonstrated by Shatil (1997). Relative to
the older poor readers, however, these novices are certainly processing
the print in a very different manner.
The source of this processing difference, at present, can only be
grounds for speculation. However, a plausible hypothesis is that the
accumulated 4–6 years of print experience (even impoverished experi-
ence relative to chronological-age controls) enable 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, first grade is the sole learning-to-read year, devoted primarily
to the acquisition of the alphabetic code (known as ‘technical’ reading).
Most schools follow a structured sequence of basal-type workbooks
(often with controlled reading vocabulary) in which letters and vowels
are introduced in a fixed sequence. Since the fundamentals are typically
acquired by the end of first grade (Share & Levin, 1999), the reading
curriculum in second grade switches from a learning-to-read to reading-
to-learn mode in which readers, now able to decode virtually any
pointed text, plunge into a wealth of readers, literature, and biblical
texts whose only concession to the novice is the presence of vowel dia-
critics. In view of these 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, it is only until the latter half of the school year that
793SELF-TEACHING
this print exposure brings about a transformation in the orthographic
processing of printed words. It will be worthwhile investigating whether
this transformation is marked by changes in overall reading fluency,
prosody or even comprehension. This calls for a longitudinal study over
the course of second grade examining concurrent changes in ortho-
graphic 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.
Having established that both the dyslexics and garden-variety poor
readers, demonstrate qualitatively distinct patterns of orthographic
learning relative to reading-level controls, it is well to return to the ques-
tion posed in the introduction as to whether the orthographic advan-
tage of disabled readers is simply the product of print exposure or
represents an underlying or learned processing advantage.
In view of the fact that there were a fixed number of exposures to
the targets in this study, and that the two groups (older garden-variety
poor readers and younger controls) were matched for both speed and
accuracy of pseudoword decoding as well as general level of non-verbal
(visuo-spatial) mental ability, it is clear that the superior orthographic
learning among poor readers (obtained in a non-ordinal interaction),
points to a print-specific processing advantage among the older group
in the acquisition of new orthographic information. However, this
print-specificity is very different from the idea that older poor readers
have superior visual memory skills, namely a basic non-print processing
advantage. On the other hand, the traditional print exposure hypothesis
assumes that differences on conventional measures of orthographic
knowledge (such as spelling choice or wordlikeness) imply that the
orthographic superiority of poor readers is attributable to a greater vol-
ume of exposure per se, specifically, that older children have simply
seen the specific test items more often (in the case of tests of word-spe-
cific knowledge) or have seen more exemplars of a more general ortho-
graphic convention in the case of wordlikeness tasks. By this account,
younger normal readers receiving the same amount of exposure to new
words should perform at comparable levels when assessed on word-spe-
cific orthographic information as in the present study. The present
results showed clearly that such was not the case. The conventional
print exposure explanation, therefore, also falls short of an adequate
account of the present data.
Since neither of the two above accounts provide satisfactory expla-
nations of the data, a hybrid hypothesis is proposed by which a highly
specialized print-specific processing advantage (not a basic skill differ-
ence) is acquired as the outgrowth of extensive print experience. This
794 DAVID L. SHARE AND CARMIT SHALEV
hybrid ‘orthographic sensitivity’ hypothesis assumes that a critical vol-
ume of print experience brings about a fundamental change in ortho-
graphic sensitivity. Just as an art or food connoisseur acquires a
sensitivity to subtle nuances that escape the layman, so the accumula-
tion of years of print exposure enable even the older disabled reader
to develop a sensitivity to orthographic patterns that is beyond the
grasp of the novice. At present, this hypothesis is entirely speculative,
and will require either explicit in vivo testing by means of direct con-
current assessment of print experience and changing orthographic abil-
ities, or experimental simulation possibly by means of an artificial
orthography.
On the issue of IQ-discrepant as opposed to non-discrepant reading
difficulties, the data were unambiguous regarding a lack of qualitative
differences on the word recognition processes examined here. Despite
the fact that both groups of poor readers, IQ-discrepant and non-dis-
crepant, were not matched for reading level, the data indicated only
quantitative differences reinforcing the current consensus regarding the
similarity in reading processes of these two groups (e.g., Aaron, 1997;
Share, 1996; Stanovich, 1991; Stuebing et al., 2002). Although dyslexics
performed at a consistently lower level than the garden-variety group,
covarying pre-existing word recognition completely washed out all dif-
ferences on individual measures. The simple correlations, too, between
degree of IQ-reading discrepancy and target decoding (r¼0.31,
p¼0.06), orthographic choice (r¼0.06, ns) and spelling (r¼0.39,
p¼0.01) disappeared entirely once reading level was controlled (all
partial correlations fell to below r¼0.10). Most important, there were
no significant differences (i.e., higher-order interactions) between poor
reader groups on any of the critical repeated measures interactions
examining the relationship between target decoding success and the two
discriminating dependent variables (orthographic choice and spelling).
The significant group main effect in the repeated measures analyses was
also entirely accounted for by pre-existing levels of reading ability. It
seems, therefore, that with regard to their patterns of on-line word
learning, the dyslexic and garden-variety groups do not represent quali-
tatively distinct entities.
Acknowledgements
This research was supported by a grant from the Ministry of Educa-
tion, Office of the Chief Scientist.
795SELF-TEACHING
Appendix
Notes
1. Although it is widely agreed that specific reading disability is best defined on the
basis of regression, this procedure, being purely relativistic, often includes children
with high IQs who are reading at grade-appropriate levels (or even above). While
such children are clearly ‘underachievers’ in the classic IQ-discrepant sense, it is
debatable whether they can be regarded as having a functional reading difficulty.
2. The overall lower absolute level of word recognition among dyslexics does not
appear to be attributable to the slighter greater proportion of younger subjects rel-
ative to the garden-variety poor readers. Although there were significant grade
effects (senior grades scoring better) on word and pseudoword reading accuracy
Target words (in alternate homophonic spellings) appearing in test texts
1
The symbol x represents the laryngeal fricative heard at the end of the word Bach.
796 DAVID L. SHARE AND CARMIT SHALEV
(but not time), garden-variety poor readers were consistently superior to dyslexics
at each of the three grades; moreover, group did not interact with grade for any
measure, including composite word recognition.
3. This study examined orthographic learning of purely consonantal graphemic infor-
mation 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 con-
tent 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 pat-
terns of native Hebrew speakers who often exchange stem-internal vowels in spo-
ken 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 recognition of consonantal roots (Bentin & Frost, 1995; Frost &
Bentin, 1992).
References
Aaron, P. G. (1997). The impending demise of the discrepancy formula. Review of
Educational Research, 67, 461–502.
Adams, M. J., & Bruck, M. (1993). Word recognition: The interface of educational
policies and scientific research. Reading and Writing: An Interdisciplinary Journal,
5, 113–139.
Bentin, S., Deutsch, A., & Liberman, I. Y. (1990). Syntactic competence and read-
ing ability in children. Journal of Experimental Child Psychology, 48, 147–172.
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.
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). Hills-
dale, NJ: Erlbaum Associates.
Cunningham, A. E., Perry, K. E., Stanovich, K. E., & Share, D. L. (2002). Ortho-
graphic learning during reading: Examining the role of self-teaching. Journal of
Experimental Child Psychology, 82, 185–199.
Castles, A., Datta, H., Gayan, J., & Olson, R. K. (1999). Varieties of developmental
reading disorder: Genetic and environmental influences. Journal of Experimental
Child Psychology, 72, 73–94.
Cossu, G. (1999). The acquisition of Italian orthography. In M. Harris & G. Hatano
(Eds.), Learning to read and write: A cross-linguistic perspective (pp. 10–33). Cam-
bridge: Cambridge University Press.
Ehri, L. C., & Saltmarsh, J. (1995). Beginning readers outperform older disabled
readers in learning to read words by sight. Reading and Writing: An Interdisciplin-
ary Journal, 7, 295–326.
Felton, R. H., & Wood, F. B. (1992). A reading level match study of nonword read-
ing skills in poor readers with varying IQ. Journal of Learning Disabilities,25,
318–326.
Fletcher, J. M., Shaywitz, S. E., Shankweiler, D. P., Katz, L., Liberman, I. Y., Stue-
bing, K. K., Francis, D. J., Fowler, A. E., & Shaywitz, B.A. (1994). Cognitive
797SELF-TEACHING
profiles of reading disability: Comparisons of discrepancy and low achievement
definitions. Journal of Educational Psychology,86, 6–23.
Fredman, G., & Stevenson, J. (1988). Reading processes in specific reading retarded
and reading backward 13 year-olds. British Journal of Developmental Psychology,
6, 97–108.
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.
Foorman, B. R., & Torgesen, J. (2001). Critical elements of classroom and small-
group instruction promote reading success in all children. Learning Disabilities
Research & Practice, 16, 203–212.
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.
Hogaboam, T. W., & Perfetti, C. A. (1978). Reading skill and the role of verbal expe-
rience in decoding. Journal of Educational Psychology, 70, 717–729.
Jackson, N. E., & Coltheart, M. (2001). Routes to reading success and failure: Toward
an integrated cognitive psychology of atypical reading. Philadelphia, PA: Psychol-
ogy Press.
Jorm, A. F., & Share, D. L. (1983). Phonological recoding and reading acquisition.
Applied Psycholinguistics,4, 103–147.
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.
Manis, F. R. (1985). Acquisition of word identification skills in normal and disabled
readers. Journal of Educational Psychology,77, 78–90.
Olson, R. K., Kliegl, R., Davidson, B. J., & Foltz, G. (1985). Individual and develop-
mental differences in reading disability. In D. Besner, T. G. Waller, & G. E.
(Eds.), Reading research: Advances in theory and practice (pp. 1–64). New York:
Academic Press.
Olson, R. K., Wise, B., Ring, J., & Johnson, M. (1997). Computer-based remedial train-
ing in phoneme awareness and phonological decoding: Effects on the posttraining
development of word recognition. Scientific Studies of Reading, 1, 235–253.
Ortar, G. (1987). Hava Nikra: Mivchan lehavanat hanikra leveit-hasefer Nosach Yod
[Let’s read: Test of reading comprehension for school, Version 10]. Achi’ever
Press.
Perfetti, C. A. (1985). Reading ability. New York: Oxford University Press.
Rack, J. P., Snowling, M. J., & Olson, R. K. (1992). The nonword reading deficit in
developmental dyslexia: A review. Reading Research Quarterly, 27, 29–53.
Raven, J., Raven, J. C., & Court J. H. (1976). Raven’s progressive matrices. Oxford:
J. C. Raven Ltd., Oxford Psychologists Press.
Ravid, D. (1995). Language change in child and adult Hebrew: A psycholinguistic per-
spective. New York: Oxford University Press.
Reitsma, P. (1983a). Printed word learning in beginning readers. Journal of Experi-
mental Child Psychology,36, 321–339.
Reitsma, P. (1983b). Word-specific knowledge in beginning reading. Journal of
Research in Reading,6, 41–56.
798 DAVID L. SHARE AND CARMIT SHALEV
Share, D. L. (1995) Phonological recoding and self-teaching: Sine qua non of reading
acquisition. Cognition, 55, 151–218.
Share, D. L. (1996). Word recognition and spelling processes in specific reading dis-
abled and garden-variety poor readers. Dyslexia, 2, 167–174.
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. (2002). Lexical and contextual factors in printed word learning.
Research report submitted to the Ministry of Education, Office of the Chief Scien-
tist. April, 2002.
Share, D. L., (2004). Orthographic learning at a glance: On the time course and
developmental onset of self-teaching. Journal of Experimental Child Psychology,
87, 267–298.
Share, D. L., & Jorm, A. F. (1987). Segmental analysis: Co-requisite to reading, vital
for self-teaching, requiring phonological memory. European Bulletin of Cognitive
Psychology,7, 509–513.
Share, D. L. & Levin, I. (1999). Learning to read and write in Hebrew. In M. Harris
& G. Hatano (Eds.), Learning to read and write: A cross-linguistic perspective
(pp. 89–111). Cambridge: Cambridge University Press.
Share D. L., McGee, R., McKenzie, D., Williams, S. M., & Silva P. A. (1987). Further
evidence relating to the distinction between specific reading retardation and general
reading backwardness. British Journal of Developmental Psychology, 5, 35–44.
Shatil, E. (1997). Predicting reading ability: Evidence for cognitive modularity. Unpub-
lished doctoral dissertation, University of Haifa.
Shaywitz, S. E., Escobar, M. D., Shaywitz, B. A., Fletcher, J. M., & Makuch, R.
(1992). Distribution and temporal stability of dyslexia in an epidemiological sam-
ple of 414 children followed longitudinally. New England Journal of Medicine,
326, 145–150.
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.
Siegel, L. S. (1989). IQ is irrelevant to the definition of reading disability. Journal of
Learning Disability, 22, 469–479.
Siegel, L. S., Share, D. L., & Geva, E. (1995). Dyslexics have orthographic skills that
are superior to normal readers. Psychological Science, 6, 250–254.
Solberg, S., & Nevo, B. (1979). Preliminary steps towards an Israeli standardization
of the Peabody Test (in Hebrew). Megamot,3, 407–413.
Stanovich, K. E. (1988). Explaining the differences between the dyslexic and the gar-
den-variety poor reader: The phonological-core variable-difference model. Journal
of Learning Disabilities,21, 590–604.
Stanovich, K. E. (1991). Discrepancy definitions of reading disability: Has intelligence
led us astray? Reading Research Quarterly, 26, 7–29.
Stanovich, K. E. (1994). Does dyslexia exist? Journal of Child Psychology and Psy-
chiatry, 35, 579–595.
Stanovich, K. E. (1998). Refining the phonological core deficit model. Child Psychol-
ogy and Psychiatry Review, 3, 17–21.
Stanovich, K. E., & Siegel, L. S. (1994). Phenotypic performance profile of children
with reading disabilities: A regression-based test of the phonological-core vari-
able-difference model. Journal of Educational Psychology, 86, 24–53.
799SELF-TEACHING
Stuebing, K. K., Fletcher, J. M., LeDoux, J. M., Lyon, G. R., Shaywitz, S. E., &
Shaywitz, B. A. (2002). Validity of IQ-discrepancy classifications of reading dis-
abilities: A meta-analysis. American Educational Research Journal, 39, 469–518.
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 long-term outcomes from two instructional
approaches. Journal of Learning Disabilities, 34, 33–58, 78.
Torgesen, J. K., Wagner, R. K., & Rashotte, C. A. (1997). Prevention and remedia-
tion of severe reading disability: Keeping the end in mind. Scientific Studies in
Reading, 1, 217–234.
Vellutino, F. R., & Scanlon, D. M. (1987). Phonological coding, phonological aware-
ness, and reading ability: Evidence from a longitudinal and experimental study.
Merrill-Palmer Quarterly,33, 321–363.
Vellutino, F. R., Fletcher, J. M., Snowling, M. J., & Scanlon, D. M. (2004). Specific
reading disability (dyslexia): what have we learned in the past four decades? Jour-
nal of Child Psychology and Psychiatry, 45, 2–40.
Vellutino, F. R., Scanlon, D. M., & Lyon, G. R. (2000). Differentiating between diffi-
cult-to-remediate and readily remediated poor readers: More evidence against the
IQ-discrepancy definition of reading disability. Journal of Learning Disabilities,
33, 223–238.
Wimmer, H., Landerl, K., & Frith, U. (1999). Learning to read German: Normal
and impaired acquisition. In M. Harris & G. Hatano (Eds.), Learning to read and
write: A cross-linguistic perspective. (pp. 34-50). Cambridge: Cambridge University
Press.
Address for correspondence: David L. Share, Department of Learning Disabilities Fac-
ulty of Education, University of Haifa, Mt. Carmel, Haifa 31905, Israel E-mail:
dshare@construct.haifa.ac.il
800 DAVID L. SHARE AND CARMIT SHALEV