BRAIN AND LANGUAGE 54, 136–167 (1996)
ARTICLE NO. 0063

Deep Dyslexic Phenomena in a Letter-by-Letter Reader
LAUREL J. BUXBAUM
Moss Rehabilitation Research Institute, Philadelphia, Pennsylvania
AND

H. BRANCH COSLETT
Department of Neurology, Temple University Hospital, Philadelphia, Pennsylvania;
and Moss Rehabilitation Research Institute, Philadelphia, Pennsylvania
Numerous accounts of pure alexia have suggested that prelexical impairment precludes rapid access to orthographic information in patients with the disorder. We
report a patient with features of both pure and partially recovered deep dyslexia in
whom we demonstrate prelexical deficits in maintaining a reliable abstract representation of the right side of letter arrays, as well as in modulating a ‘‘spotlight’’ of
visual attention. These deficits, we suggest, encourage the patients’s use of a letterby-letter reading strategy; despite them, however, he demonstrates rapid, accurate
reading of some, but not all classes of words. Furthermore, the patient’s reading is
influenced by both prelexical and lexical–semantic factors such that speed and accuracy are optimal for high imageability nouns of few letters. Finally, the patient
accurately names orally spelled words of all classes. Taken together, these data are
consistent with the hypothesis that rapid reading may be enabled by lexical–semantic support from a right hemisphere-mediated processing system which recognizes
words as wholes, thereby mitigating the effect of the prelexical deficits.  1996
Academic Press, Inc.

Patients with pure alexia exhibit deficits in rapid, automatic word recognition, frequently relying upon a compensatory letter-by-letter reading strategy.
This serial approach to reading results, typically, in a monotonic relationship
between word length and reading latency. While accounts of the disorder
have posited processing deficits at a variety of cognitive levels, many recent
Supported by NIH Grants 1T32HD07425-01 and RO1-AG08870 and by the Moss Rehabilitation Research Institute. We are grateful to JH and his wife for their patience and cooperation,
as well as to Eleanor Saffran, Myrna Schwartz, and Mike Montgomery for helpful comments
and suggestions. Presented in part at the International Neuropsychological Society Meeting,
Cincinnati, OH, 1994. Address reprint requests to Laurel J. Buxbaum at the Moss Rehabilitation Research Institute, 1200 West Tabor Road, Philadelphia, PA 19141. E-mail: Buxbaum@
shrsys.hslc.org.
136
0093-934X/96 $18.00

Copyright  1996 by Academic Press, Inc.
All rights of reproduction in any form reserved.

LETTER-BY-LETTER DEEP DYSLEXIA

137

investigations have confirmed the presence of relatively peripheral prelexical
deficits which may impede the parallel processing of letter information. For
example, based upon observations that reading may be influenced by the
number (Kinsbourne & Warrington, 1962), visual quality (Farah & Wallace,
1991) and spatial position (Rapp & Caramazza, 1991) of stimuli, it has been
argued that pure alexics exhibit an impairment in the allocation of visual
attention, akin to simultanagnosia (Kinsbourne & Warrington, 1962; Levine & Calvanio, 1977) or hemispatial neglect (Rapp & Caramazza, 1991).
In contrast, other investigators have suggested that pure alexia is attributable
to deficits in higher level processes mediating letter and word recognition
(Patterson & Kay, 1982; Kay & Hanley, 1991; Arguin & Bub, 1993; Howard, 1990). Thus, for example, it has been suggested that slowing (ReuterLorenz & Brunn, 1990) or imprecision (Arguin & Bub, 1993) of activation
of abstract single letter identity results in the characteristic letter-by-letter
reading.
Their differences notwithstanding, these accounts of pure alexia share the
assumption that the impairment ultimately impedes automatic and parallel
access to a stored visual word representation through visual input. From the
perspective of most information processing models of reading (see, for example, Coltheart, 1980; Glosser & Friedman, 1990; Saffran, 1985), the consequence of such a deficit is an impairment in subsequent rapid semantic activation. In fact, previous failures to demonstrate the effect of such variables
as word class or lexical status on reading accuracy have been cited as support
for the assumption of a locus of processing deficit that is relatively peripheral
(Coltheart, 1980; Shallice & Warrington, 1980; but see Friedman, Beeman,
Lott, Link, Grafman, & Robinson, 1994). In contrast, such variables have
been shown to influence the reading of such patients as deep dyslexics, in
whom the processing deficit is thought to be more central (Shallice & Warrington, 1980).
There have been several recent cases of pure alexia, however, which have
provided evidence of effects arising at the level of the visual word form or
semantics.1 It is in accommodating such data that many prelexical accounts of
the disorder, in their purest forms, may be challenged. Coslett and Saffran
(1989) for example, described two pure alexic patients who, after partial
recovery, demonstrated superior reading of nouns as compared to functors
and high as compared to low imageability words with brief (249 msec)
presentations. In addition to these effects on explicit reading abilities, there
is evidence that at least some pure alexics may ‘‘implicitly’’ process at least
categorical semantic information from rapidly presented words which they
cannot explicitly identify (e.g., Landis, Regard, & Serrat, 1980; Howard,
1990; Shallice & Saffran, 1986; Coslett and Saffran, 1989; Coslett, Saffran,
1
For convenience, the term lexical-semantic will refer to levels of representation including
the visual word form and semantic systems.

138

BUXBAUM AND COSLETT

Greenbaum, & Schwartz, 1993). Taken together, these cases suggest that
while prelexical deficits in visual, attentional, or orthographic processing
may exist, such peripheral impairment does not preclude rapid lexical access
in at least some patients with pure alexia.
At least two accounts of data demonstrating implicit reading in pure alexia
have been proposed. On one hypothesis, degradation or impaired access to
the visual word form system of the left hemisphere permits activation of an
entry in the visual word form system, but with insufficient activation strength
to support explicit word identification (Friedman et al., 1994; Shallice &
Saffran, 1986). An alternative hypothesis is that rapid access to semantics,
when it occurs, involves an alternate reading strategy which does not depend
on the inaccessible or degraded left hemisphere visual word form system.
Proponents of the latter account have suggested that the observed part of
speech effects and relative impairments in explicit as compared to implicit
responses are consistent with the linguistic capabilities of the right hemisphere (e.g., Schweiger, Zaidel, Field, & Dobkins, 1989). Based in part on
such findings, Coslett, Saffran, and colleagues (1989, 1993) have argued that
the reading of pure alexics is right hemisphere mediated.
The similarity of these putative characteristics of the right hemisphere
lexicon to the reading of patients with deep dyslexia, another acquired disorder of reading, has also been noted (Saffran, Boygo, Schwartz, & Marin,
1980; Coltheart, 1980; Schweiger et al., 1989). Deep dyslexia is characterized by impaired reading of abstract words, functors, bound morphemes, and
pseudowords as well as errors of semantic substitution. Thus, Coltheart
(1980) and Saffran et al. (1980), among others, have suggested that deep
dyslexics read by accessing a lexical–semantic entry from the right hemisphere, which is then transmitted to left hemisphere processors, enabling
phonologic output.
If deep dyslexia and the rapid and ‘‘recovered’’ reading of pure alexia are
both mediated by the same right hemisphere visual word form and semantic
system, as has been suggested, then the two disorders might be expected to
co-occur subsequent to left hemisphere damage. There are infrequent reports,
in fact, of pure alexics who make semantic errors in comprehension and in
reading aloud (Albert, Yamadori, Gardner, & Howes, 1973; Caplan &
Hedley-White 1974), show effects of imageability on reading speed (Kay &
Hanley, 1991), or demonstrate part of speech effects on reading accuracy
(Coslett & Saffran, 1989, 1994; Doctor, Sartori, & Saling, 1990). Recently,
Friedman et al. (1994) reported a transcortical sensory aphasic whose reading
showed length effects, part of speech effects and impaired reading of nonwords in the context of preserved recognition of these types of stimuli when
they were spelled aloud, thereby demonstrating a modality-specific phonologic alexia.
We report a patient who, although similar to the patient reported by Friedman et al. (1994), nevertheless differs in a number of important respects.

LETTER-BY-LETTER DEEP DYSLEXIA

139

Early in the post-infarct course, the present patient demonstrated a severe
and disabling alexia, characterized by letter-by-letter reading, part of speech
and imageability effects, inability to read nonwords or affixed words, and
semantic substitution errors in the context of normal oral language. These
findings, then, are consistent with the concurrent diagnoses of pure and deep
dyslexia. In the series of investigations reported here, we focus in turn upon
two aspects of reading performance. In the first set of experiments, we
demonstrate that the patient exhibits a number of prelexical visual deficits
which, we suggest, impair rapid parallel access to the visual word form
system and encourage the use of a letter-by-letter reading strategy. In the
second set of experiments, we show that in spite of these deficits, some
classes of words (e.g., concrete nouns) may be read rapidly and accurately,
suggesting that the prelexical impairment does not preclude rapid access to
lexical–semantic information for some classes of words. In addition, we
show that an interaction of prelexical and lexical–semantic factors determines
the pattern of performance such that reading is optimized with short, highly
imageability nouns. Finally, we demonstrate that the patient can name orally
spelled words normally, indicating that an intact lexical–semantic system is
accessed from auditory input. Taken together, these data are consistent with
the hypothesis that the patient’s rapid reading, when it occurs, is facilitated
by a lexical–semantic system which partially mitigates the effect of the
prelexical deficits.
PATIENT DESCRIPTION

JH is 59-year-old right-handed businessman with a high-school education
who noted the sudden onset of mild right-sided weakness and loss of vision
on the right. Neurological examination performed 4 weeks after the onset
of symptoms revealed a mild right hemiparesis and a right homonymous
hemianopia. The clinical diagnosis of stroke was confirmed by MRI scan
(see Figs. 1 and 2) performed 5 days after the onset of symptoms which
demonstrated infarction of the left inferior temporo-occipital cortex, with
involvement of the parahippocampal gyrus. The infarct also appeared to involve the left lateral geniculate.
PRELIMINARY INVESTIGATIONS

Excessive and receptive language functions, as assessed by the Boston
Diagnostic Aphasia Examination (Goodglass & Kaplan, 1972) 1 month after
the stroke were normal except for a mild impairment in pointing to named
letters and numbers and a profound dyslexia. Spelling to dictation, recognition of orally spelled words, and spontaneous writing were normal. With
unlimited exposure JH read a list of 20 concrete, high-frequency nouns with
40% accuracy, often using a slow, laborious, explicit letter-by-letter strategy.
Letter misidentifications were frequent, and he occasionally required more

140

BUXBAUM AND COSLETT

FIG. 1. A coronal MRI scan demonstrating infarction in the vicinity of the left lateral geniculate and involving the parahippocampal gyrus.

than 30 sec to provide a response. He provided relatively rapid responses
without letter naming to approximately 20% of words.
JH’s reading performance was assessed at both rapid (100 msec) and unlimited exposures 2 months after the stroke. As can be seen in Table 1, his
reading was approximately as accurate with brief as with prolonged exposures. Additionally, at both exposures he was unable to read nonwords, and
demonstrated effects of imageability and part of speech. Thus, for example,
he could rapidly read many high-imageability nouns such as ‘‘village,’’ but
was unable to read such functor words as ‘‘else,’’ ‘‘shall,’’ ‘‘why’’, or ‘‘un-

LETTER-BY-LETTER DEEP DYSLEXIA

141

FIG. 2. A coronal MRI scan demonstrating infarction of the inferior portion of the left occipital lobe.

til.’’ Finally, he exhibited explicit letter-by-letter reading with unlimited exposure durations.
Analysis of error type revealed that JH produced not only visual errors
like most pure alexics, but semantic paralexic errors as well. Adopting a
very stringent criterion for defining semantic errors (that is, no clear visual
similarity), 7% of errors were semantic paralexias at 100 msec (e.g., navy
→ sailor; free → sample; door → roomful; insect → small; check → tab;
leisure → relax). Using a less stringent criterion in which semantically related but more visually similar words were scored as semantic errors, an

142

BUXBAUM AND COSLETT

TABLE 1
Reading Accuracy for Words and Nonwords at Two Exposure Durations:
Percentage Correct
Stimulus type
High imageability nouns
(n 5 40)
Low imageability nouns
(n 5 40)
Regular nouns
(n 5 40)
Exception nouns
(n 5 40)
Functors
(n 5 40)
Nonwords
(n 5 25)

100 msec

Unlimited

68

75

30

30

45

43

53

43

15

25

4

8

additional 5% of errors at 100 msec could be characterized as semantic paralexias (e.g., principle → king . . . prince; slope → slide; theory → thought;
peril → plunder). Thus, using the most lenient criterion, 12% of errors at
100 msec were semantic. Note that the observed rate of semantic errors at
brief exposures is well within the range of that reported by McCarthy and
Warrington (1990, p. 225) in deep dyslexics.
Note also the significance of the fact that the rate of semantic errors was
dramatically different at the two stimulus durations. JH produced 13 semantic errors with 100 msec stimulus duration, but only a single (,1%) semantic
error (fact → thought) with prolonged stimulus presentation. This pattern of
performance, that is, semantic paralexias at rapid but not unlimited exposures, has been previously reported in a group of five pure alexic patients
(Coslett & Saffran, 1994, p. 311). Coslett and Monsul (1994) have argued
elsewhere that a lack of semantic errors at unlimited exposures reflects constraints imposed on phonologic production by residual letter recognition abilities.
Unlike semantic errors, other error types were not affected by the exposure
duration. Thus, visual paralexias (e.g., although → athlete, skit → shirt;
sword → world; outside → longsided) were the most frequent errors, both
with brief (46%) and unlimited (51%) exposure; mixed errors (e.g., mother
→ mom . . . monetary; bouquet → boutonniere) comprised 5% of the total
with brief and 6% with unlimited exposure; whereas errors which may be
morphologically based (e.g., plates → plate; ticking → ticket; flower →
flowery; poetry → poem) constituted 33% with brief exposure and 38% with
unlimited exposure. Approximately 80% of errors involved word endings,
but there were occasional errors on the beginnings of words as well.
To formally investigate the effect of word length on reading latency, JH

LETTER-BY-LETTER DEEP DYSLEXIA

143

FIG. 3. Latency to read high imageability nouns: Effect of length.

was asked to read a list of 60 words presented on a microcomputer; stimuli
were 10 words each with lengths of three to eight letters, matched for frequency at each length. All were high imageability nouns (.5, Paivio, Yuille, & Madigan, 1968), a word class for which he had demonstrated the
greatest proficiency in previous evaluations. To measure reaction time, JH
was asked to depress a key when he identified the word. As reported by a
number of previous investigators (e.g., Warrington & Shallice, 1980; Coslett & Saffran, 1989), there was an effect of word length on reading latency
(see Fig. 3)
There was no clinical evidence of deficits in multiple-object processing
as assessed by description of pictures of complex scenes (e.g., ‘‘Cookie
Theft’’ from the Boston Diagnostic Aphasia Examination; Goodglass &
Kaplan, 1972), nor was there evidence of neglect on line bisection or letter
cancellation tasks (Wilson, Cockburn, & Halligan, 1987).
Thus, preliminary investigations suggested that JH was a letter-by-letter
reader whose performance also demonstrated the typical features of deep
dyslexia. These observations prompted our further exploration of JH’s letter
and word processing abilities.
EXPERIMENTAL INVESTIGATIONS
Part 1: Investigation of Prelexical Visual Processing

Experiment 1: Report of Letters in Arrays
Numerous investigators (see, e.g., Rapp & Caramazza, 1991; Levine &
Calvanio, 1977; Kay & Hanley, 1991) have demonstrated position effects

144

BUXBAUM AND COSLETT

FIG. 4. Accuracy of letter report by patient and control: Effect of letter position.

on the accuracy with which their pure alexic subjects reported letters in
strings. In some cases, reduced accuracy has been shown for report of the
central letters in arrays (Reuter-Lorenz & Brunn, 1990; Bub, Black, & Howell, 1989), a finding which has generated the hypothesis that some pure alexics process words in ‘‘ends-in’’ fashion. Other investigators (e.g., Kay &
Hanley, 1991; Rapp & Caramazza, 1991) have reported patients with letter
processing impairment that is most marked for the right side of letter strings.
We investigated JH’s identification of letters as a function of position in the
following experiment.
Procedure. For this and all subsequent experiments, stimuli were presented by means of
PsychLab software (Bub & Gum, 1988) on a Macintosh computer. JH was seated at a distance
of approximately 18 inches from the monitor screen. Stimuli were presented in black lowercase
letters in 24-point Monaco font on a white background. A central fixation point was presented
for 500 msec, followed by a 1000-msec delay. All stimuli were presented unmasked at midline;
exposure durations were varied according to the task. Where indicated, control data were
collected with age- and education-matched male subjects.
In the present experiment, 40 three-letter and 40 five-letter orthographically illegal and unpronounceable letter strings were presented in a randomized sequence. Each array was composed of lowercase letters separated by a single character space and centered at midline; thus,
for example, the third letter from the left in the five-letter arrays occupied the same midline
location as the second letter from the left in the three-letter arrays. Three-letter arrays spanned
approximately 1° of visual angle whereas five letter arrays spanned approximately 2° of visual
angle. Exposure time for patient JH was 249 msec and for control subject RS was 50 msec.
Subjects were asked to report as many letters as possible without regard to serial order.

Results and discussion. As demonstrated in Fig. 4, JH demonstrated a
steep left to right gradient in reporting accuracy as a function of letter position for both three- and five-letter arrays. The overall accuracy of JH’s performance was superior for the three-letter (65% correct) as compared to fiveletter arrays (34% correct) (χ2 5 29.2, p , .0001).

LETTER-BY-LETTER DEEP DYSLEXIA

145

The overall accuracy of the control subject, RS, was comparable in the
three-letter (96% correct) and five-letter conditions (93% correct) (χ2 5 1.1,
p . .05). On both five-letter and three-letter arrays, accuracy was relatively
uniform across array positions.
Several aspects of JH’s performance are noteworthy. First, while most
deficient in reporting right-sided letters, he demonstrates impairment in reporting letters even in initial left-sided positions. Second, he is more impaired
in reporting letters from any given spatial position in the five as compared
to three-letter arrays. Finally, he is likewise more impaired in reporting letters
from any given serial position in the five as compared to three letter arrays.
Thus, for example, his report of the third letter from the left was superior
in the three-letter arrays (40% correct) as compared to the five-letter arrays
(13% correct). These effects may reflect the contribution of a number of
factors, including impairments in the speed of letter identification, deficient
registration of right-sided visual information, and visual attentional biases
or limitations. These possible explanations for the observed pattern of performance were explored in subsequent experiments.
Experiment 2: Report of Single Letters in Varying Locations
One possibility is that JH’s impairment on the letter reporting task of Experiment 1 is in part the result of a deficit in rapid letter identification (see,
e.g., Reuter-Lorenz & Brunn, 1990; Arguin & Bub, 1993; Patterson & Kay,
1982). Thus, after initiating processing of the left-most letter in the array,
JH may have had insufficient time to identify the right-sided letters. Alternatively, the position effects observed in Experiment 1 may be, at least in part,
a consequence of JH’s right visual field deficit. Although the impairments
in reporting accuracy for even left-sided positions, as well as the relative
impairments for the same spatial positions in five-letter arrays as compared
to three-letter arrays suggests that neither of these accounts completely explains the pattern of JH’s performance, an additional experiment was performed to assess the possible contributions of both slowed letter processing
and right hemianopia.
Procedure. JH was required to report 40 single letters, each presented for 100 msec, at one
of five different spatial locations. Eye position was monitored by the examiner to ensure midline fixation before initiation of each trial. Stimuli were presented in one of the five locations
probed in Experiment 1 (midline; 1° and 2° to the left and right of midline).

Results and discussion. JH’s performance was errorless, suggesting that
slowing in the speed of individual letter identification was unlikely to have
been a source of his deficient performance in Experiment 1. At 100 msec
exposure and with unpredictable stimulus locations, saccades to target positions are precluded; thus, positions 4 and 5 were located in the ‘‘hemianopic’’
region, yet JH reported even these letters with 100% accuracy. The failure
to demonstrate positional effects on report of single letters at rapid exposures

146

BUXBAUM AND COSLETT

suggests that the observed gradients in the three- and five-letter arrays of
Experiment 1 may not be solely attributable to a deficit in the registration
or rapid processing of visual information from right-sided retinotopic spatial
locations. Rather, the data suggest that JH’s impairment is attributable, at
least in part, to higher level visual deficits in visual attention or spatial representation. In the following experiments, we sought to determine the nature
of the impairment(s).
Experiment 3: Two-Letter Arrays
An attentional bias to attend to the most eccentric ipsilesional stimulus
has frequently been reported in patients with neglect (DeRenzi et al., 1989;
Ladavas, Petronio, & Umilta, 1990). Such an asymmetry in visual attention
across arrays might be caused by engagement of visual attention by the ipsilesional side of the stimulus array such that the appearance of left-sided
letters results in obligatory shifts of attention to the sites containing those
letters, or, alternatively, deficits in shifting attention from those sites. With
respect to the present case, an impairment of this nature might result in relative inattention to sites furthest from the left side of the array (Kinsbourne,
1987; Behrmann, Moscovitch, Black, & Mozer, 1990). Consistent with the
pattern of results demonstrated in Experiment 1, this account predicts that
letters in the final position of the five-letter arrays would be reported more
poorly than those in the final position of the three-letter arrays.
If visual attention is indeed biased to the leftmost stimulus, then the mere
presence of a nontarget letter on the left side of the array should adversely
affect responses to a target letter appearing to the relative right. Conversely,
a nontarget letter on the right of the array should not disrupt processing of
a target letter on the left. Thus, if an attentional bias is present, the position
of the nontarget letter relative to the target letter should influence the pattern
of performance.
Procedure. Eight arrays composed of two lowercase letters were presented in a single block.
The two letters appeared simultaneously with equal probability in two of the five positions
utilized previously in the five-letter array condition of Experiment 1. Thus, for example, one
such array comprised letters in spatial positions 2 and 5. Prior to presentation of each twoletter array, an underline probe appeared for 249 msec which signaled the forthcoming location
of a single target letter to be reported from the two-letter array. Each of the five positions
was probed in this manner a total of 16 times. For positions 2, 3, and 4, the probed target letter
appeared with equal probability to the left or right of the flanking letter. Thus, for example, a
probe in position 2 was followed on eight trials by an array of letters in positions 1 and 2
(probed target letter on relative right, referred to hereafter as ‘‘right’’) and on eight trials by
an array of letters in positions 2 and either 3, 4, or 5 (probed target letter on relative left,
referred to hereafter as ‘‘left’’). A stimulus exposure duration of 249 msec was selected for
patient JH for the purpose of comparison with previous experiments; exposure duration was
33 msec for subject RK. Subjects were informed in advance of the block of trials that they
would view an underline probe followed by two letters and were asked to report only the
letter appearing in the probed location.

147

LETTER-BY-LETTER DEEP DYSLEXIA

TABLE 2
Percentage Accuracy of Report of Pre-probed Single Letters in Two-Letter Arrays:
Effect of Spatial Position and Relative Position of Probe
Relative
position
of probed
letter

Spatial position of probed letter

1

2

3

4

5

Left
Right

94
—

88
88

88
100

100
75

—
25

Results and discussion. The control subject responded with perfect accuracy to all probed stimuli. As indicated in Table 2, the side on which the
probed target letter appeared relative to the flanking letter had no overall
effect on JH’s reporting accuracy. This is evident in the comparison of the
‘‘left’’ and ‘‘right’’ results for JH for positions 2, 3, and 4; that is, the comparative accuracy of report when the probed target letter was located to the
relative left or right of the flanker. If visual attention is indeed biased toward
left-sided stimuli, then accuracy of report should be superior when the probed
target letter appears to the relative left of the flanker. The data demonstrate
no such effect. At least within the range of the two-letter arrays employed
here, then, attentional biases toward ipsilesional stimuli are not observed.
Experiment 4: Letter Spacing
An additional possibility is that JH’s deficits in processing letter arrays
are at least in part the result of interference with the processing of a given
letter by the letters flanking it. Such lateral interference effects have been
demonstrated in normal subjects at brief exposure durations (Bouma, 1973;
Campbell & Mewhort, 1980), as well as in ‘‘attentional dyslexics’’ (Shallice & Warrington, 1977; Saffran & Coslett, 1996). In some reported cases,
disrupted processing, most severe for central letters in an array, has been
reduced when the distance between the letters is increased (Price & Humphreys, 1993). To investigate the possibility that flanker interference is a
factor in JH’s letter-by-letter reading, we performed an additional experiment
in which the distance between the letters in arrays was manipulated.
Methods. Forty three-letter and 40 five-letter arrays were presented in two randomized
blocks. The positions of the letters differed from those of Experiment 1 by virtue of the introduction of an additional character space between each letter. Three letter arrays thus comprised
letters located at midline and approximately 2° to the left and right of midline such that letters
in positions 1 and 3 occupied the locations of letters in positions 1 and 5 in Experiment 1.
Five-letter arrays utilized positions approximately 4° to the left and right of midline in addition
to the above. Each array was presented for 249 msec, and JH was asked to report as many
letters as possible without regard to serial order.

148

BUXBAUM AND COSLETT

FIG. 5. Accuracy of letter report by patient JH in three- and five-letter arrays: unspaced and
spaced letter conditions.

Results and discussion. As indicated in Fig. 5, increasing the distance between letters does not benefit reporting accuracy, but instead results in performance that is remarkably similar, both for any given position and with
respect to overall accuracy, to that for the unspaced letter arrays of Experiment 1.
These data suggest that in the case of JH, deficits in the processing of
arrays are unlikely to result from the interference of letters in nontarget locations. In addition, the arrays in the present experiment subtended greater
degrees of visual angle than those of Experiment 1, yet the results with respect to any given position are comparable, suggesting once again that JH’s
impairment in processing right-sided stimuli is not attributable to a failure
to represent visual feature information from the right visual field.
There are at least two possible interpretations of these data. The first is
that the number of elements in the array and their relative (as opposed to
retinotopic) spatial position may be the most relevant determinants of reporting accuracy. Alternatively, the results may indicate that reporting accuracy is largely independent of spatial position and instead reflects the influence of sequential reporting factors.
The latter hypothesis is consistent with a number of investigators’ suggestions that pure alexics rely on a strategy of sequential letter identification
as a consequence of impairment in the parallel processing of orthographic
information (e.g., Reuter-Lorenz & Brunn, 1990; Friedman & Alexander,
1994; Howard, 1990). On such accounts, processing begins on the left side
of a letter string and proceeds rightward, resulting in monotonic increases
in reading latency as a function of word length. Thus, with brief exposure
durations, there may be insufficient time for visual attention to reach the end

LETTER-BY-LETTER DEEP DYSLEXIA

149

of multiple-letter strings. While preliminary evidence from JH suggests that
he does, in fact, rely upon an overtly sequential strategy on reading tasks,
neither the data from Experiment 1 nor the present experiment can be readily
accommodated by invoking such a mechanism alone. Such an explanation,
for example, cannot account for the observed impairments in report even for
initial positions. Additionally, on a serial account, accuracy of report of a
given letter in the series should be independent of the size of the array in
which it is encountered; contrary to this supposition, JH’s report of each
letter in the series was superior in the three-letter arrays as compared to the
five-letter arrays.
In the following experiment, we directly pitted the serial account against
a spatial interpretation of the letter-reporting gradients by dissociating the
serial and spatial positions of the letters to be reported.
Experiment 5a: Report of Sequential Letters—Right to Left Presentation
If poor report of right-sided letter positions is the result of demands imposed by serial report, and is independent of spatial location, then report of
terminal positions in letter sequences should be relatively impaired even
when these terminal positions are located in the ‘‘good’’ ipsilesional field.
Thus, on the sequential hypothesis, reading right to left should result in relatively poor report of left-sided letters. In contrast, if the deficit in report of
arrays is spatially based, then right to left reading should not ameliorate the
deficient report of right-sided letters. In an attempt to ensure that JH did, in
fact, read letters right to left, we used a rapid serial visual presentation
(RSVP) paradigm in which letters appear sequentially beginning on the right.
Procedure. Forty five-letter random strings were presented individually. Each letter was
presented for 100 msec, and the SOA between the letters was 250 msec. Each letter appeared
in right to left sequence in one of the five different spatial positions used in Experiment 1;
thus, for example, the letters in serial position 1 in the present experiment appeared on the
right of the array in the same spatial location as did the letters in position 5 in Experiment
1. JH was informed that letter strings would appear on the screen beginning on the right and
moving leftward, and that he was to report as many letters as possible without regard to serial
order.

Results. JH accurately reported 25% of the letters from serial position 1
(spatial position 5), 65% from serial position 2, 72% from serial position 3,
78% from serial position 4, and 85% from serial position 5 (spatial position
1). In all cases, he attempted to report letters in the sequence in which they
appeared.
Experiment 5b: Report of Sequential Letters—Left to Right Presentation
For the purpose of comparison, JH was asked to report sequentially appearing letters which proceeded from left to right. With the exception of this

150

BUXBAUM AND COSLETT

difference in the direction of letter presentation, the procedure was identical
to that used in Experiment 5a.
Results. JH accurately reported 75% of the letters from serial position 1
(spatial position 1), 60% from serial position 2, 43% from serial position 3,
53% from serial position 4, and 35% from serial position 5 (spatial position 5).
Discussion. These results thus further disconfirm the possibility that the
steep gradients in reporting accuracy observed in Experiment 1 solely reflect
the sequential order in which attention is deployed across the visual array.
Specifically, when attention is induced to proceed from right to left, as in the
first portion of the present experiment, there is failure to report the initially
processed stimuli; in contrast, stimuli processed from left to right in the second portion of the experiment demonstrate reporting that is most impaired
for final sequential positions. These data thus support the hypothesis that
spatial and not sequential factors are of importance in JH’s impaired report
of right sided letters.
Perhaps the most important aspect of these findings, however, lies in their
comparison with the results of Experiment 2, in which it was demonstrated
that JH’s report of single letters presented in the same spatial locations and
at the same exposure durations as those in the current experiment was flawless. In the present experiment, then, impaired report of right-sided letters
from the same locations must reflect the influence of previously or subsequently appearing left-sided letters and cannot be attributed solely to a failure
to register visual feature information from the right visual field. These results
corroborate the hypothesis, proposed in response to the results of Experiment
4 with spaced arrays, that JH exhibits deficits in the processing of right-sided
information as right is defined with respect to other visual information in
the array. Thus, a form of ‘‘neglect’’ or ‘‘extinction’’ of right-sided locations
occurs after a representation of the entire array has been encoded as a unitary
entity in a viewer-centered ‘‘map’’ of spatial locations (Stark, Coslett, and
Saffran, 1996) in which information from different sites is maintained.
It is possible, however, that this deficit in the spatial ‘‘map’’ is not the
sole source of JH’s impairment in processing letter arrays. As we have noted,
a number of investigators (e.g., Levine & Calvanio, 1977; Rapp & Caramazza, 1991) have described deficits in the allocation of visual attention in
patients with pure alexia. On one such account, for example, limitations in
attentional capacity impede the processing of feature information from multiple spatial locations in parallel (Levine & Calvanio, 1977); on this account,
focusing attention more narrowly on a single letter location may increase
the discriminability of visual information in that location.
Consider, then, the possibility that JH may exhibit similar limitations in
the resolving power of a ‘‘spotlight’’ of visual attention (Erickson & Yeh,
1985; Erickson & St. James, 1986), such that processing is optimal only
within a relatively restricted region. Such a deficiency in attentional alloca-

LETTER-BY-LETTER DEEP DYSLEXIA

151

tion might coexist with impairment in the processing of relative right sided
locations to produce letter-by-letter reading.
Experiment 6: Detection of Letters in Hierarchical Stimuli
One consequence of a restriction in the region to which attention may be
deployed is relatively better detection of small, local elements as compared to
larger, global forms constructed from these features. By using hierarchically
constructed stimuli (Navon, 1977) in which target letters appeared with equal
probability as either global or local forms, we sought in this experiment to
confirm the hypothesis that a deficit in the spotlight of visual attention is
one source of JH’s impairment in the parallel processing of letter arrays.
The performance of the control subject RK was also evaluated.
Methods. One hundred fifty-six hierarchical stimuli comprising letters of approximately 4
degrees of visual angle composed of smaller letters of approximately .5° of visual angle were
presented individually at midline. For example, one such stimulus was a letter ‘‘H’’ composed
of small ‘‘J’s’’. The small constituent letters were spaced apart a distance of approximately
.5°. The subject was instructed to respond to the presence of a named letter by depressing a
reaction time key. In advance of each trial, the examiner notified the subject of the target
letter. On one-third (52) of the trials, the target was the global, large letter; on 52 trials it was
the local, small letter; and on 52 trials, the target named by the examiner was absent. Stimuli
were presented for 4000 msec or until the subject responded. Stimuli to which subjects failed
to respond within 4000 msec were counted as misses.

Results and discussion. JH responded correctly to 25/26 local target trials
and 24/26 global target trials; accuracy was thus equivalent in response to
each target type. He made two false positive errors on target-absent trials.
Mean response times to detect local and global stimuli, respectively, were
1138 msec (6488 msec) and 1631 msec (6585 msec). The control subject
RK responded accurately to all 52 target-present trials and made no false
positive errors on target absent trials. Mean response times to detect local
and global stimuli, respectively, were 570 msec (6188 msec) and 679 msec
(6193 msec).
The data from the two subjects were entered into an ANOVA with target
size (global, local) and subjects as factors. The analysis revealed main effects
of both target size (F(1, 96) 5 14.6, p , .001) and subject (F(1, 96) 5
93.6, p , .0001). Of greatest relevance, however, was a significant interaction between target size and subject (F(1, 96) 5 5.9, p , .01), as can be
seen in Fig. 6. Post-hoc analyses of simple main effects revealed an effect
of target size on the reaction times of JH (F(1, 96) 5 19.2, p , .0001) but
not control RK (F(1, 96) 5 .98, p . .1). Thus, only JH is impaired in the
processing of global as compared to local forms.
These data thus support the hypothesis that JH exhibits an impairment in
the spotlight of attention with the consequence that he is unable to achieve an
integrated perceptual representation over large and/or complex spatial arrays.
Based upon evidence that the resolution of the attentional spotlight is in-

152

BUXBAUM AND COSLETT

FIG. 6. Speed of detection by patient and control of letters at local and global levels.

versely related to its diameter (Erickson & St. James, 1986; Erickson & Yeh,
1985), strategic spatial restriction of the spotlight may serve to improve letter
recognition from sites within the spotlight’s beam. Thus, such a capacity
limitation may have relevance for JH’s tendency to produce letter-by-letter
responses to whole word stimuli.
Summary of Experiments Assessing Prelexical Processing

The results of Experiments 1 through 6 provide evidence for at least two
prelexical deficits which might disrupt reading either singly or—as we believe more likely—in combination. First, we have shown that JH is unable
to generate a reliable abstract representation of the right-sided portion of
letter strings. Critically, data from Experiment 2 demonstrate that this deficit
is not attributable to a failure to construct a low-level retinotopic representation of visual feature information, as he reports single letters from right-sided
locations at the same exposure duration with 100% accuracy. Additionally,
results from Experiment 3 suggest that the pattern of letter report cannot be
attributed to a bias to attend to left-sided stimuli. Finally, as demonstrated
in Experiment 5, the deficit in reporting right-sided letters is apparent irrespective of the direction of presentation, suggesting that the tendency to
omit right-sided letters is not attributable to memory or other task-specific
demands. Taken together, these data demonstrate that JH is unable to reliably
generate or maintain an accurate egocentric ‘‘spatial map’’ (Stark, Coslett, &
Saffran, 1996); thus, the representation of right-sided stimuli is not maintained sufficiently to activate letter identification systems to the point required
for conscious awareness.
Second, the data from Experiment 6 suggest an impairment in a spotlight
of visual attention. If, as suggested by a number of investigators, attention

LETTER-BY-LETTER DEEP DYSLEXIA

153

may be likened to a spotlight or zoom lens in which area attended and resolving power are inversely related, a reduction in the capacity of the spotlight
would be expected to decrease the domain over which attention may be distributed with sufficient resolution to permit accurate letter identification. On
this account, explicit letter-by-letter responses in oral reading may reflect a
reduction in capacity which causes JH to limit the aperture of the spotlight
to encompass sublexical units or even single letters, a strategy which would
also serve to minimize errors in letter identification.
To this point, we have primarily discussed the implications of these prelexical visual deficits for the reading of nonword letter strings. Abundant data
from normals and brain-damaged subjects (e.g., Coslett & Saffran, 1991)
suggest, however, that the processing of familiar words may differ in important respects from that of nonword letter strings. Thus, while there is evidence that the constituent letters of a word must be recognized for its identification (Johnston & McClelland, 1980), recognition of the letters themselves
is frequently superior when they are presented in the context of words
as compared to random letter arrays (McClelland and Rumelhart, 1981;
Reicher, 1969; Wheeler, 1970). Further, the identity of some words may be
derived from identification of sublexical units, such as syllables or morphemes, in combination with such orthographic and/or semantic factors as
word frequency or familiarity. Therefore, deficits in letter identification of
the sort demonstrated by JH do not a priori predict obligatory scrutiny of
each letter of a word in laborious sequential fashion unless there is complete
failure to access support from ‘‘higher’’ level orthographic, morphologic,
and semantic information.
In the case of JH, preliminary clinical evidence suggested that lexical–
semantic factors indeed influenced word recognition. In the second portion
of this paper we explore the nature of these and other lexical–semantic effects, and examine their interaction with prelexical factors.
Part 2: Reading

As we have noted, the initial assessment of JH’s reading demonstrated a
pattern of performance consistent with deep dyslexia. The following experiments were performed to more formally examine his reading performance
at brief exposures.
Experiment 7: Oral Reading of Words and Nonwords
a. Effect of imageability and frequency. The effects of imageability and
frequency were assessed by asking JH to read 80 nouns varying in these two
dimensions. Twenty words were of high imageability (.5, Paivio et al.,
1968) and high frequency (.60, Francis & Kucera, 1982), 20 of high imageability and low (,20) frequency, 20 of low (,3) imageability and high frequency, and 20 of low imageability and low frequency. Words in the four

154

BUXBAUM AND COSLETT

groups were matched for length and varied from three to seven letters. Stimuli were presented in random order for 249 msec.
JH correctly read 80% of words of high imageability and frequency, 70%
of words of high imageability and low frequency, 45% of words of low
imageability and high frequency, and 20% of words of low imageability and
low frequency. Thus, he correctly read 75% of high imageability words and
only 32.5% of low imageability words matched for frequency and length
(χ2 5 14.5, p , .0001). He read high frequency words (63%) more reliably
than low frequency words (45%) but this difference was not significant (χ2 5
2.4, p . .1).
Errors on high imageability words included 90% visual errors (e.g., cart
→ carton; gravy → grape; hand → hard) and 1 mixed error (slope → square).
Errors on low imageability words included 63% visual errors (e.g., satire →
saturday; concept → corn); 33% ‘‘don’t know’’ or ‘‘I know but cannot say
it’’ responses, and 4% (1) apparently unrelated response (attitude → testing).
Thus, imageability (but not word frequency) strongly influences JH’s reading
accuracy.
b. Effect of regularity. The effect of regularity was assessed by asking JH
to read 30 regular (e.g., plank) and 30 irregular (e.g., sword) nouns matched
for imageability, frequency, grammatical class, and length (Kay, Lesser, &
Coltheart, 1992). Stimuli were presented for 249 msec in a random sequence.
JH read 60% of regular and 67% of irregular words. Thus, regularity does
not affect reading accuracy.
c. Part of speech effects. To assess JH’s reading of nouns and functors,
30 high imageability nouns (.5, Paivio et al. 1968) and 30 functors matched
for length and frequency (Kay, Lesser, and Coltheart, 1992) were presented
for 249 msec in mixed and randomized order. JH correctly read 80% of
nouns and only 20% of functors, a difference which is highly significant
(χ2 5 21.6, p , .0001). Errors on functors included 4 (13%) functor substitutions, 12 (40%) substitutions of a noun, of which 92% were visually similar
(e.g., whether → wheat; okay → kayak), and 8 (27%) ‘‘don’t know’’ or
‘‘can’t pronounce’’ responses. Subsequent analysis of JH’s errors demonstrated that the mean frequency of the functors to which JH produced paralexic responses was 352 counts per million, whereas the mean frequency of
his paralexic responses to these same words was lower (121 counts per million). This suggests that JH’s deficit in functor reading cannot be attributed
to a tendency to produce responses that are higher in frequency than the
target stimuli.
d. Reading of suffixed words. To assess JH’s processing of affixes, he was
asked to read 25 suffixed and 25 pseudosuffixed words matched for length
and frequency. Words were generated in pairs from the same potential root
morpheme (e.g., show); one of the words of the pair was suffixed (e.g.,
showed), whereas the other word ended in a letter sequence that could be a
suffix in a different context (e.g., the ‘‘er’’ in shower). Words were presented

LETTER-BY-LETTER DEEP DYSLEXIA

155

for 249 msec in random order. JH correctly read 13/25 (52%) of pseudosuffixed but 0/25 (0%) of the suffixed words, a difference that is highly significant (χ2 5 17.6, p , .0001 by Fisher’s Exact Test). The 25 errors with
suffixed words included 12 (48%) affix omissions, 3 (12%) affix substitutions
(e.g., ticked → ticking), 6 (24%) substitutions of a pseudosuffixed word with
the same root morpheme (e.g., towed → towel), and 4 (16%) visual errors.
The 13 errors with pseudosuffixed words included 5 (38%) pseudoaffix
omissions, 4 (31%) pseudoaffix substitutions (e.g., pennant → penny), and
3 (23%) visual errors.
e. Mirror-reading of suffixed words. The relative impairment in JH’s reading of suffixed as compared to pseudosuffixed words containing the same
potential root morpheme (e.g., ‘‘show’’) is suggestive of deficits in the processing of suffixes rather than a neglect of the right side of visual stimuli.
To further assess the nature of JH’s impairment, we asked him to read suffixed words printed in a mirror-reversed format. Fifteen suffixed and 15 pseudosuffixed words generated from the same potential root morphemes (as
above) were hand-printed in 1/2-inch high letters and centered on individual
3 3 5-inch cards in mirror-reversed print such that word endings were on
the cards’ left. JH correctly read 73% of pseudosuffixed and only 20% of
suffixed words (χ2 5 8.6, p , .01 by Fisher’s Exact Test). Errors on suffixed
words included 4 (33%) affix omissions (1 of which was produced after JH
correctly spelled the word aloud: tables → T.A.B.L.E.S. . . . table), 5 (42%)
substitutions of a pseudosuffixed word with the same root morpheme (e.g.,
flowed → flower), 2% (17%) visual errors (e.g., morals → morsel), and 1
mixed visual/pseudosuffix substitution error (tricked → ticket). Errors on
pseudosuffixed words included 3 (75%) pseudoaffix deletions and 1 pseudoaffix substitution (liver → lives). Though the data from this study are not
extensive, the relative impairment with suffixed as compared to pseudosuffixed mirror-reversed words confirms the results of the previous study: the
deficit in reading suffixed words is not attributable to failure to perceive the
right side of visual stimuli.
f. Reading of nonwords. JH’s processing of nonwords was assessed by
asking him to read 24 pronounceable pseudowords (6 words each of lengths
three to six letters) created by changing one or two letters of high frequency
nouns. JH correctly pronounced only 4 (17%) of the pseudowords; he produced lexicalization errors (e.g., smope → smoke . . . smock; shid → ship,
but with a ‘‘d’’ on the end . . . ship; ked → key . . . K.E.D. . . . kid) to 20
(84%) of the stimuli.
Discussion. Thus, JH’s performance is largely consistent with the initial
clinical diagnosis of deep dyslexia. He demonstrates effects of imageability
and part of speech but not frequency in oral reading; furthermore, he is impaired in reading nonwords and suffixed words, and he makes morphologic
errors which cannot be attributed to failure to register the right side of stimuli.
Semantic substitution errors were present in reading evaluations conducted

156

BUXBAUM AND COSLETT

1 month prior to these formal investigations. As noted previously, the absence of semantic substitutions in the present evaluation may reflect partial
recovery in JH’s letter recognition ability, with consequent constraints on
phonologic production (see Coslett & Saffran, 1991; Coslett & Monsul,
1994). For example, a response such as ‘‘small’’ is unlikely to occur to the
stimulus ‘‘insect’’ if the initial letter ‘‘i’’ is recognized. This possibility is
consistent with initial observations that semantic substitutions occurred almost solely in conditions of rapid presentation, when letter-naming procedures are least accurate and therefore unlikely to effectively constrain responses.
The data demonstrating that JH is able to process at least some types of
letter strings (e.g., concrete nouns) to the point of explicit word identification
even with brief exposures stands in striking contrast both to his poor performance in the processing of certain other classes of words (e.g., low imageability nouns and functors), as well as his impaired performance in processing nonword letter strings. The disparity in JH’s success in reading these
different types of letter strings, we suggest, is attributable to the differential
availability of orthographic and/or semantic support for these stimulus types.
Thus, high imageability nouns may benefit from the presence of orthographic
and/or semantic information that mitigates the effects of the prelexical deficits. At least two possible accounts of these data may be proposed. One
possibility is that the previously demonstrated deficits in the parallel processing of visual arrays are not sufficient to preclude rapid access to at least
some visual word forms. This rapid contact with visual word forms serves
to partially mitigate the prelexical impairment by enabling interpretation of
the letter string as a unitary perceptual entity. Alternatively, as suggested by
Friedman et al. (1994), the mitigation of the prelexical deficit may reflect
support from the semantic system rather than visual word forms. On this
account, feedback from semantics to weakly activated visual word forms
boosts the activation strength of the latter, thus enabling whole-word recognition. On either account, when confronted with a word or nonword letter string
for which visual word forms and semantic representations are inaccessible
(or nonexistent), the prelexical impairment induces JH to revert to a default
letter-by-letter strategy.
Both accounts of the mechanism by which processing is facilitated at the
level of the visual word forms lead to the same set of predictions with respect
to the interaction of semantic and prelexical factors. First, reading of words
for which orthographic and semantic information can be accessed in parallel
should be more rapid and accurate than reading of words for which there is
no such parallel activation. In addition, as such words should be identified
as wholes, performance should not be influenced by length. Conversely,
words for which stored representations are not available may be read by
means of a serial letter-by-letter strategy; the reading of these words could
be expected to be influenced by word length. In the following experiment

LETTER-BY-LETTER DEEP DYSLEXIA

157

FIG. 7. Reading accuracy for high and low imageability nouns: Effect of length.

we investigated this prediction by asking JH to read words varying in length
and imageability.
Experiment 8: Reading Latency and Accuracy with High and Low
Imageability Words of Varying Lengths
Materials and Procedure. JH was asked to read a series of 236 high imageability (.5,
Paivio et al., 1968) and 240 low imageability (,3.5, Paivio et al. 1968) nouns presented in
a random sequence. For each of the two imageability levels, there were 40 words each of
lengths of three to seven letters; there were only 36 high imageability words of eight letters.
Words were presented individually with a microcomputer, and response latencies were measured by a voice activated reaction time key. JH was instructed to read each word aloud as
rapidly and accurately as possible.

Results. Figure 7 presents the percentage of correct responses for both
high and low imageability words as a function of length. The percentage
correct data, transformed by an arcsin square root function, were entered
into an analysis of variance with length (3, 4, 5, 6, 7, 8) and imageability
(high, low) as factors. There was a main effect of imageability (F(1, 5) 5
31.3, p , .001) such that overall accuracy of responding was better for high
as compared to low imageability words. In addition, there was an interaction
of length and imageability by Tukey’s procedure for one observation per
cell (Winer, 1971) (F(1, 4) 5 3.2 . critical value p(.25)). Simple main
effects could not be computed due to the single observation per cell. However, as can be clearly seen in Fig. 7, the interaction of length and imageability arises as a result of a greater effect of increasing number of letters on
the accuracy of reading low as compared to high imageability words.
An ANOVA performed on the reaction times of the 339 correct responses
with length (3, 4, 5, 6, 7, 8) and imageability (high, low) as factors revealed

158

BUXBAUM AND COSLETT

FIG. 8. Latency to read high and low imageability nouns: Effect of length.

a significant interaction (F(5, 327) 5 4.63, p , .001), as can be seen in Fig.
8. There were simple main effects of number of letters on both high (F(5,
327) 5 15.7, p , .001) and low (F(5, 327) 5 2.9, p , .05) imageability
words. Post-hoc testing (Sheffé with p , .05) of the effect of number of
letters at each level of imageability showed that for the low imageability
condition, words of three and four letters were read significantly faster than
words of eight letters and words of four letters were read significantly faster
than words of six letters. For the high imageability condition, none of the
comparisons reached significance.
Finally, post hoc analysis of the mean and median word frequency of the
238 high imageability and 240 low imageability words was conducted. The
low imageability words, on which JH’s performance was more impaired,
were slightly higher in frequency (mean 5 98 counts per million; median
5 27 counts per million) than were the high imageability words (mean 5
78 counts per million; median 5 26 counts per million).
Discussion. These data suggest that word length interacts with imageability in a similar manner with respect to accuracy as well as reaction time.
Both measures reveal a greater effect of length upon low imageability as
compared to high imageability words. Thus, shorter low imageability words
are read significantly more rapidly and accurately than longer low imageability words. High imageability words, in contrast, are less subject than the low
imageability words to the effect of length on reading latency or accuracy.
Results of the analysis of word frequency suggests that this factor does not
account for this significant interaction.
The disparity between the two imageability conditions, we suggest, results
from differences in the amount of orthographic and/or semantic support
available for each word type. Results of experiments presented in the first

LETTER-BY-LETTER DEEP DYSLEXIA

159

portion of this paper suggest that JH is impaired in allocating visual attention
across arrays, and is unable to construct a stable internal representation of
letter strings. Consistent with a number of information-processing models
of reading, we assume that following visual analysis of the letter string,
words are identified by matching the visual input to stored visual word forms
(cf. Warrington & Shallice, 1980). Semantic and phonologic information is
subsequently activated. From the perspective of such models, the consequence of JH’s prelexical deficits is an impairment in the quality of visual
input to the visual word form and semantic systems. The available visual
information is nonetheless sufficient to permit the activation of information
represented in the visual word form and/or semantic systems, which may
then feed back to support the most plausible interpretation of the degraded
visual information. Most high imageability words, then, are read in parallel
by matching to a stored visual word form. Thus, a form of ‘‘top-down’’
support, emanating either from the word-form and/or the semantic representation, enables correct and rapid reading of (most) high imageability words
regardless of length.
The relative impairment with low imageability words, we suggest, reflects
the fact that when the mode of input is visual, orthographic entries for most
low-imageability words are unavailable or receive insufficient activation to
achieve word recognition. Thus failing to receive lexical–semantic support,
the majority of low imageability words must be read sublexically, often
letter-by-letter, with the result that reading low imageability words is slower,
as well as increasingly inaccurate as the probability of making errors in letter
identification increases on longer words.
Experiment 9: Naming of Orally Spelled Words
One issue of relevance in the present context is the relationship between
JH’s oral reading and naming of orally spelled words. Although Shallice and
Warrington (1980) proposed that naming of orally spelled words is mediated
by the spelling system operating in reverse, a number of recent investigations
in normal subjects support the claim that recognition of orally spelled words
is accomplished not through the spelling system, but by means of the same
processes that mediate reading of printed words (Monsell, 1987). On the
latter account, identification of an orally spelled word occurs subsequent to
transformation of the auditory input into a ‘‘visual’’ image which is identified
by accessing the visual word form system.
Investigations with alexic patients have also supported the latter hypothesis. Several investigators have reported letter-by-letter readers who demonstrated discrepant performance on reading and spelling tasks, lending support
to the notion that different processing systems mediate each type of task.
Katz (1989), for example, described an acquired dyslexic patient who demonstrated parallel performance on tests of reading and naming of orally

160

BUXBAUM AND COSLETT

spelled words, in the context of a distinctive and more severe impairment
in spelling. Rapcsak, Rubens, and Laguna (1990) reported a letter-by-letter
reader with ‘‘surface dysgraphia’’ for whom oral presentation resulted in a
significant improvement in word recognition. A third patient, described by
Hanley and Kay (1992), demonstrated phonological regularizations in reading but not spelling.
In the following experiment we investigated JH’s ability to name orally
spelled words which he had previously attempted to read.
a. High and low imageability words. The 20 high imageability, high frequency nouns and 20 low imageability, high frequency nouns presented in
Experiment 8 were utilized in a separate testing session to evaluate JH’s
ability to recognize orally spelled words. The words were spelled aloud by
the examiner at the rate of one letter per second. JH correctly named 100%
of high imageability words and 95% of low imageability words.
b. Suffixed and pseudosuffixed words. The 25 suffixed and 25 pseudosuffixed words presented in Experiment 8 were spelled aloud by the examiner
at the rate of one letter per second. In contrast to his grossly impaired reading
of these stimuli (e.g., 0/25 suffixed words), JH’s naming of these words was
errorless.
c. Pseudowords. The 24 pseudowords presented in Experiment 8 were
spelled aloud by the examiner at the rate of one letter per second. JH correctly
pronounced 58% of the orally presented pseudowords, as compared to 12%
with visual presentation (χ2 5 11.0, p , .001).
As indicated in Fig 9, then, JH’s ability to name orally spelled words and
nonwords is superior to his reading of these same words.
At least two accounts of these data may be identified. First, the data are
consistent with the hypothesis that JH exhibits a visual modality-specific
deficit in access to a single visual word form system which supports both
reading and naming of orally spelled words. Such an account was offered
by Friedman et al. (1994) to explain similar discrepancies in their patient
with pure alexia. On this notion, degraded visual input results in weakened
activation of an intact orthographic entry. Part of speech effects arise because
words with rich semantic representations, such as nouns, may be activated
on the semantic level even with weakened input from the visual word form.
Words with limited semantic representations, such as functors, may be recognized only with relatively strong input from the word form system. With
auditory input, the visual word form is strongly activated, resulting in recognition of both word types.
The present data are also consistent with the hypothesis that by virtue of
JH’s lesion in the left tempero-occipital cortex, which serves to impede normal access by visual input to the visual word and semantic systems of the
left hemisphere, oral reading must be accomplished by a ‘‘default’’ system.
To account for the dramatic effects of imageability and part of speech, one
must postulate that this ‘‘default’’ system possesses an impoverished repre-

LETTER-BY-LETTER DEEP DYSLEXIA

161

FIG. 9. Performance accuracy as a function of input modality: Naming of orally spelled
words and reading.

sentation for low imageability words, functors, and bound morphemes. Note
that on this hypothesis, access to the visual word form and/or semantic systems of the left hemisphere remains intact for auditory input; thus, this system can be used in a normal fashion to decode orally spelled words.
Unfortunately, our data do not definitively adjudicate between these alternative accounts. There is at least one initially puzzling aspect of the data,
however, which we believe supports the second account. Given that JH can
accurately name orally spelled words, why does he fail, in reading tasks, to
use the letter names derived from letter-by-letter visual inspection of the
words to name the word? We suggest that this seemingly paradoxical finding
is attributable to task demands and the information processing procedure that
each task induces. Thus, we suggest that the instruction to read a visually
presented word automatically invokes a cognitive operation which attempts
to match the visual stimulus to an entry in the visual word form system. In
the event that the word cannot be matched to an existing visual word form,
JH attempts to identify the word by means of a serial letter identification
strategy. Critically, however, even after naming the constituent letters, he
fails to switch from a visually mediated ‘‘reading’’ procedure to one based
on phonologic access to the visual word form system. Thus, JH—unlike
other studied pure alexics—fails to name the word, we assume, because he
does not utilize the phonologically encoded letter names which he correctly
produces.

162

BUXBAUM AND COSLETT

Note that on this account the discrepancy betwen reading letters aloud and
naming of orally spelled words would only be anticipated if JH accessed
different lexical–semantic systems for the two tasks. If JH consulted the
same lexical–semantic system when naming orally spelled words and ‘‘reading’’ words by identifying the constituent letters, dissociation in performance
on these tasks would not be expected.
Summary of Experiments Assessing Reading

We have demonstrated that, at least early in the post-infarct course, JH
demonstrated the typical features of deep dyslexia—semantic substitutions,
part of speech effects, morphologic errors, and inability to read nonwords.
We have shown, furthermore, that JH is able to read some classes of words
at exposures too rapid for explicit letter-by-letter reading. We have also demonstrated an interaction between word length and imageability such that
length has a greater effect on reading speed and accuracy for low imageability as compared to high imageability words. Finally, we have shown that
JH’s performance is superior for naming of orally spelled words as compared
to reading of the same words and nonwords, suggesting a modality-specific
deficit in visual access to morphologic and phonologic information. In the
following section, we provide an account of JH’s reading which accommodates these findings, as well as those described in the letter processing experiments in the first portion of the paper.
GENERAL DISCUSSION

The data from experiments assessing both the visual attentional and word
processing abilities of JH, a patient who demonstrates features of both pure
and partially recovered deep dyslexia, enable us offer an account of his pattern of reading which incorporates both prelexical and lexical–semantic
deficits. In general agreement with several other investigators (Kinsbourne & Warrington, 1962; Levine & Calvanio, 1977; Farah & Wallace,
1991; Rapp & Caramazza, 1991) we have localized the deficits causing
letter-by-letter reading to ‘‘early’’ visual and attentional processing levels.
More specifically, we have suggested that JH exhibits two principal prelexical impairments: a restriction in the capacity of a spotlight of visual attention,
which prohibits the development of an integrated representation across letter
arrays, and a deficit in establishing or maintaining an abstract spatial representation of the right side of arrays. Despite these deficits, however, JH exhibits partially recovered deep dyslexia, suggesting that the attenuated visual
information is sufficient to permit activation of existing lexical–semantic
entries. Finally, we have demonstrated an interaction of prelexical and lexical–semantic factors in reading speed and accuracy, and have suggested that
information represented in the lexical–semantic system may support the interpretation of information degraded by virtue of impaired prelexical pro-

LETTER-BY-LETTER DEEP DYSLEXIA

163

cessing, thus permitting rapid reading of repesented words independent of
length.
An interaction of prelexical and lexical–semantic factors in reading such
as that invoked here has been noted both in normal subjects (Behrmann et
al., 1991) and in patients with neglect dyslexia. For example, Sieroff et al.
(1988) and Brunn and Farah (1990) have reported neglect patients for whom
deletion of initial letters is greater for nonwords as compared to words. Additionally, Behrmann et al. (1990, 1991) reported two neglect dyslexics who
demonstrated a double dissociation of effects of peripheral, visual stimulus
attributes and lexical factors. Thus, for example, the reading accuracy of one
patient was strongly determined by the number of letters and spatial extent
of words, but was relatively unaffected by lexical status, while the reverse
pattern was evident for the other patient. In response to these data and related
findings from normal subjects, Mozer and Behrmann (1990) and Behrmann
et al. (1990, 1991) suggested that prelexical inputs that are only mildly degraded by visual attentional deficit may be sufficiently intact to permit the
activation of lexical information, while more severe peripheral impairment
blocks such access. A form of ‘‘lexical superiority’’ similar to that demonstrated by neglect dyslexics, we suggest, may account for the disparity between JH’s profound deficits in reporting the terminal letters of nonword
letter strings and his relatively better incorporation of terminal letters in word
reading.
The account offered above, however, does not explain the specific pattern
of JH’s reading. Thus, for example, it is not clear on this account why prelexical deficits would selectively compromise the reading of low as compared
to high imageability or suffixed as compared to pseudosuffixed words. If,
as was previously suggested, JH is able to read some familiar words because
of the support provided by an intact lexical–semantic entry, his poor performance with functors, low imageability words, and bound morphemes suggests that these elements are not represented in the lexicon supporting his
reading. Based on a number of lines of evidence (see below) suggesting that
the right hemisphere supports a lexicon with precisely these characteristics,
we suggest that our data are consistent with the hypothesis that JH is reading
with his right hemisphere.
There are several sources of evidence that the right hemisphere may, in
fact, have the capacity to read in a limited manner. Data from commissurotomy and hemispherectomy patients, for example, suggest that the right hemisphere can match printed words to objects and make a number of types of
semantic judgments (Zaidel & Peters, 1981; Patterson, Vargha-Khadem, &
Polkey, 1989), but that it is limited in its ability to process low imageability
words, grammatical morphemes, and functors, and is phonologically impoverished (see Baynes, 1990, for a review). Based on findings such as these,
a number of investigators have suggested that deep dyslexics read by accessing right hemisphere mechanisms. Support for this hypothesis was provided

164

BUXBAUM AND COSLETT

by Rapcsak et al. (1992), who reported a patient in whom deep dyslexia
was observed subsequent to a severe stroke which lesioned most of the left
hemisphere.
An additional line of evidence in support of right hemisphere reading abilities is derived from several investigations of the ‘‘implicit’’ and recovered
reading of pure alexic patients. For example, Coslett and Saffran (1994) reported five patients with pure alexia who were able to perform lexical decision
tasks and semantic category judgments at exposures too rapid to support explicit reading; however, their performance demonstrated insensitivity to the
appropriateness of affixes and sensitivity to imageability. Furthermore, the
recovered oral reading of two of these patients showed part of speech and
imageability effects, as well as impaired processing of affixes, consistent with
the putative lexical/semantic capabilities of the right hemisphere. Finally, and
perhaps most convincingly, a recent study by Coslett and Monsul (1994) demonstrated that the reading capability of one of these patients, JG, was disrupted
by transcranial magnetic stimulation of the right, but not left hemisphere.
Thus, we suggest that JH, like other reported pure alexics, may derive
lexical–semantic information through two different procedures—a rapid,
whole word route which is reliant upon a right hemisphere lexicon, and a
default letter-by-letter strategy. Pseudowords, low imagery words, bound
morphemes, and words that primarily serve grammatical purposes, such as
functors, cannot be correctly read through the whole word procedure because
of the absence of lexical–semantic entries for these words in the right hemisphere lexicon. Alternatively, reading of words unsuccessfully recognized
through the parallel procedure may be attempted through a serial processing
strategy. Without the support of lexical–semantic representations, JH’s prelexical deficits in processing arrays result in high error rates when this procedure is used.
The notion that patterns of reading performance in pure alexia, as in neglect dyslexia, reflect an interaction between prelexical and lexical–semantic
factors may in part account for the fact that although a number of investigators have sought evidence of rapid ‘‘implicit’’ processing in their pure alexic
patients, it has not consistently been found. Thus, we suggest, patients such
as JH for whom peripheral degradation of stimuli nevertheless permits letter
identification are able to access the right hemisphere visual word form system
rapidly and in parallel, while other patients exhibiting more severe prelexical
deficits may be unable to engage orthographic and/or semantic support. The
variable severity of patients’ prelexical deficit, then, may in part account for
the presence of ‘‘implicit’’ processing abilities in some, but not all patients
with pure alexia.
REFERENCES
Albert, M. L., Yamadori, A., Gardner, H., & Howes, D. 1973. Comprehension in alexia. Brain,
96, 317–328.

LETTER-BY-LETTER DEEP DYSLEXIA

165

Arguin, M., & Bub, D. N. 1993. Single-character processing in a case of pure alexia. Neuropsychologia, 31, 435–458.
Baynes, K. 1990. Language and reading in the right hemisphere: Highways or byways of the
brain? Journal of Cognitive Neuroscience, 2, 159–179.
Behrmann, M., Moscovitch, M., Black, S. E., & Mozer, M. 1990. Perceptual and conceptual
mechanisms in neglect dyslexia. Brain, 113, 1163–1183.
Behrmann, M., Moscovitch, M., & Mozer, M. C. 1991. Directing attention to words and nonwords in normal subjects and in a computational model: Implications for neglect dyslexia.
Cognitive Neuropsychology, 8, 213–248.
Bouma, H. 1973. Visual interference in the parafoveal recognition of initial and final letters
of words. Vision Research, 13, 767–782.
Bub, D. Black, S. E., & Howell, J. 1989. Word recognition and orthographic context effects
in a letter-by-letter reader. Brain and Language, 39, 405–427.
Bub, D., & Gum, T. 1988. PsychLab software.
Campbell, A. J., & Mewhort, D. J. K. 1980. On familiarity effects in visual information processing. Canadian Journal of Psychology, 34, 134–154.
Caplan, L. R., & Hedley-Whyte, T. 1974. Cueing and memory dysfunction in alexia without
agraphia: A case report. Brain, 97, 251–262.
Coltheart, M. 1980. Deep dyslexia: A right hemisphere hypothesis. In M. Coltheart, K. Patterson, & J. C. Marshall, (Eds.), Deep dyslexia. London: Routledge & Kegan Paul.
Coslett, H. B., & Monsul, N. 1994. Reading with the right hemisphere: Evidence from transcranial magnetic stimulation. Brain and Language, 46, 198–211.
Coslett, H. B., & Saffran, E. M. 1989. Evidence for preserved reading in ‘pure alexia.’ Brain,
112, 327–359.
Coslett, H. B., & Saffran, E. M. 1991. Simultanagnosia: To see but not two see. Brain, 114,
1523–1545.
Coslett, H. B. & Saffran, E. M. 1994. Mechanisms of implicit reading in alexia. in M. Farah &
G. Ratcliffe (Eds.), The neuropsychology of high-level vision. Pittsburgh: Carnegie Mellon
Univ. Press.
Coslett, H. B., Saffran, E. M., Greenbaum, S., & Schwartz, H. 1993. Reading in pure alexia:
The effect of strategy. Brain, 116, 21–37.
DeRenzi, E., Gentilini, M., Faglioni, P., & Barbieri, C. 1989. Attentional shift towards the
rightmost stimuli in patients with left visual neglect. Cortex, 25, 231–237.
Doctor, E. A., Sartori, G., & Saling, M. M. 1990. A letter by letter reader who could not read
nonwords. Cortex, 16, 247–262.
Erickson, C. W., & St. James, J. D. 1986. Visual attention within and around the field of focal
attention: A zoom lens model. Perception and Psychophysics, 40, 225–240.
Erickson, C. W., & Yeh, Y. 1985. Allocation of attention in the visual field. Journal of Experimental Psychology: Human Perception and Performance, 11, 583–597.
Farah, M. J., & Wallace, M. A. 1991. Pure alexia as a visual impairment: A reconsideration.
Cognitive Neuropsychology, 8, 313–334.
Francis, W. N., & Kucera, H. 1982. Frequency analysis of English usage. Boston: HoughtonMifflin.
Friedman, R. B., & Alexander, M. P. 1984. Pictures, images, and pure alexia: A case study.
Cognitive Neuropsychology, 1, 9–23.
Friedman, R. B., Beeman, M., Lott, S. N., Link, K., Grafman, J., & Robinson, S. 1994. Modality-specific phonological alexia. Cognitive Neuropsychology, 10, 549–568.
Geschwind, N. 1965). Disconnexion syndromes in animals and man. Brain, 88, 585–644.
Glosser, G., & Friedman, R. B. 1990. The continuum of deep/phonological alexia. Cortex,
26, 343–359.
Goodglass, H., & Kaplan, E. 1972. The assessment of aphasia and related disorders. Philadelphia: Lea & Febiger.
Hanley, J. R., and Kay, J. 1992. Does letter by letter reading involve the spelling system?
Neuropsychologia, 30, 237–256.

166

BUXBAUM AND COSLETT

Howard, D. 1990. Letter by letter readers: Evidence for parallel processing. In D. Besner and
G. Humphreys (Eds.), Basic processes in reading. Hillsdale, NJ: Erlbaum.
Johnston, J. C., & McLelland, J. L. 1980. Experimental tests of a hierarchical model of word
identification. Journal of Verbal Learning and Verbal Behavior, 19, 503–524.
Kay, J., Lesser, R., & Coltheart, M. 1992. Psycholinguistic assessments of language processing
in aphasia (PALPA). Hove, England: Erlbaum.
Kay, J., & Hanley, R. 1991. Simultaneous form perception and serial letter recognition in a
case of letter by letter reading. Cognitive Neuropsychology, 9, 249–273.
Kinsbourne, M. 1987. Mechanisms of unilateral neglect. In M. Jeannerod (Ed.), Neurophysiological and neuropsychological aspects of spatial neglect. Amsterdam: NorthHolland.
Kinsbourne, M., & Warrington, E. K. 1962. A disorder of simultaneous form perception.
Brain, 85, 461–486.
Ladavas, E., Petronio, A., & Umilta, C. 1990. The deployment of visual attention in the intact
field of hemineglect patients. Cortex, 26, 307–317.
Landis, T., Regard, M., & Serrat, A. 1980. Iconic reading in a case of alexia without agraphia
caused by a brain tumor: A tachistoscopic study. Brain and Language, 11, 45–53.
Levine, D. N., & Calvanio, R. 1977. A study of the visual defect in verbal alexia-simultanagnosia. Brain, 101, 65–81.
McCarthy, R. A., & Warrington, E. K. 1990. Cognitive neuropsychology: A clinical introduction. San Diego: Academic Press.
Mclelland, J., & Rumelhart, D. 1981. An interaction model of content effects in letter perception. 1. An account of the basic findings. Psychological Review, 88, 375–807.
Monsell, S. 1987. On the relation between lexical input and output pathways for speech. In A.
Allport, D. Mackay, W. Prinz, & E. Scheerer (Eds.), Language perception and production:
Relationships between listening, speaking, reading, and writing. London: Academic Press.
Mozer, M., & Behrmann, M. 1989. On the interaction of selective attention and lexical knowledge: A connectionist account of neglect dyslexia. Journal of Cognitive Neuroscience,
1, 96–123.
Navon, D. 1977. Forest before trees: The precedence of global features in visual perception.
Cognitive Psychology, 9, 353–383.
Paivio, A., Yuille, J. C., & Madigan, S. A. 1968. Concreteness, imagery, and meaningfulness
values for 925 nouns. Journal of Experimental Psychology, 76, (Monograph Supplement
1), 1–25.
Patterson, K., & Kay, J. 1982. Letter by letter reading: Psychological descriptions of a neurological syndrome. Quarterly Journal of Experimental Psychology, 34A, 411–441.
Patterson, K., Vargha-Khadem, F., & Polkey, C. 1988. Reading with one hemisphere. Brain,
112, 39–63.
Price, C. J., & Humphreys, G. W. 1993. Attentional dyslexia: The effect of co-occuring deficits. Cognitive Neuropsychology, 10, 569–592.
Rapcsak, S. Z., Beeson, P. M., & Rubens, A. B. 1992. Reading with the right hemisphere.
Journal of Clinical and Experimental Neuropsychology, 14, 39.
Rapcsak, S. Z., Rubens, A. B., & Laguna, J. F. 1990. From letters to words: Procedures for
word recognition in letter by letter reading. Brain and Language, 38, 504–514.
Rapp, B. C., & Caramazza, A. 1991. Spatially determined deficits in letter and word processing. Cognitive Neuropsychology, 8, 275–311.
Reicher, G. M. 1969. Perceptual recognition as a function of meaningfulness of stimulus material. Journal of Experimental Psychology, 81, 275–328.
Reuter-Lorenz, P. A., & Brunn, J. L. 1990. A prelexical basis for letter-by-letter reading:
A case study. Cognitive Neuropsychology, 7, 1–20.
Saffran, E. 1985. Acquired dyslexia: Implications for models of reading. In T. G. Waller &
G. E. Mackinnon (Eds.), Reading research: Advances in theory and practice. New York:
Academic Press. Vol. 4.
Saffran, E. M., Boygo, L. C., Schwartz, M. F., & Marin, O. S. M. 1980. Does deep dyslexia

LETTER-BY-LETTER DEEP DYSLEXIA

167

reflect right-hemisphere reading? In M. Coltheart, K. Patterson, & J. C. Marshall (Eds.),
Deep dyslexia. London: Routledge & Kegan Paul.
Saffran, E. M., & Coslett, H. B. 1996. ‘‘Attentional dyslexia’’ in Alzheimer’s Disease: A case
study. Cognitive Neuropsychology, 13, 205–228.
Saffran, E. M., Fitzpatrick-DeSalme, E. J., & Coslett, H. B. 1990. Visual disturbances in
dementia. In M. F. Schwartz (Ed.), Modular deficits in Alzheimer-type dementia. Cambridge: MIT Press.
Schweiger, A., Zaidel, E., Field, T., & Dobkins, B. 1989. Right hemisphere contribution to
lexical access in an aphasic with deep dyslexia. Brain and Language, 37, 73–89.
Shallice, T., & Saffran, E. M. 1986. Lexical processing in the absence of explicit word identification: Evidence from a letter-by-letter reader. Cognitive Neuropsychology, 3, 429–458.
Shallice, T., & Warrington, E. K. 1980. Single and multiple component central dyslexic syndromes. In M. Coltheart, K. E. Patterson, & J. C. Marshall (Eds.), Deep dyslexia. London:
Routledge & Kegan Paul.
Shallice, T., & Warrington, E. K. 1977. The possible role of selective attention in acquired
dyslexia. Neuropsychologia, 15, 31–41.
Sieroff, E., Pollatsek, A., & Posner, M. I. 1988. Recognition of visual letter strings following
injury to the posterior visual spatial attention system. Cognitive Neuropsychology, 5, 427–
449.
Stark, M., Coslett, H. B., & Saffran, E. M. 1996. Impairment of an egocentric map of space:
Implications for perception and action. Cognitive Neuropsychology, 13.
Warrington, E. K., Cipolotti, L., & McNeal, J. 1993. Attentional dyslexia: A single case study.
Neuropsychologia, 31, 871–885.
Warrington, E. K., & Shallice, T. 1980. Word form dyslexia. Brain, 103, 99–112.
Wheeler, D. D. 1970. Processes in word recognition. Cognitive Psychology, 1, 59–85.
Wilson, B., Cockburn, J., & Halligan, P. 1987. Rivermead Behavioral Inattention Test. Titchfield: Thames Valley Test Company.
Winer, B. J. 1971. Statistical principles in experimental design. New York: McGraw-Hill.
Zaidel, E., & Peters, A. M. 1981. Phonological encoding and ideographic reading by the
disconnected right hemisphere: Two case studies. Brain and Language, 14, 205–234.