BRAIN

AND

LANGUAGE

39, 405-427 (1990)

The Evolution of Pure Alexia: A Longitudinal
Study of Recovery
M. BEHRMANN
Department

of Psychology,

University of Toronto, and Neurology,
Centre, University of Toronto

Sunnybrook

Medical

S. E. BLACK
Neurology,

Sunnybrook

Medical

Centre, University

of Toronto

AND

D. BUB
Department

of Neurolinguistics,

Montreal Neurological
Cote-de-Neiges

Institute

and Centre Hospitalier

This case report documents the partial recovery, over a 1Zmonth period, of
pure alexia in an adult female following a left occipital infarction. Measures of
speed and accuracy were obtained on an oral reading and a lexical decision task
immediately postonset and then on 10 subsequent occasions. Explicit letter-byletter reading was observed only during the first week poststroke but a significant
effect of word length was seen in all testing sessions. Reading accuracy was
relatively good at all stages and reading latency showed a remarkable decrease
over time but did not reach normal reading rates. The inability to use higherorder orthographic knowledge, as manifest in the absence of a word superiority
effect, was still noted at one year postonset. We therefore concluded that the
change in behavior was attributable to increased proficiency in the use of the
adaptive letter-by-letter procedure rather than to the resolution of the underlying
The first author is a fellow of the Ontario Ministry of Health Personnel Development
Award Program. This research was supported, in part, by a grant from the Medical Research
Council of Canada and from the Sunnybrook Medical Research Trust Fund. The authors
thank Dr. Gary Snow for the standardized neuropsychologicaf assessments and our subject,
DS, for her cooperation. We also thank the reviewers for insightful comments and suggestions about this work. Address all correspondence and reprint requests to M. Behrmann,
Rotman Research Institute, Baycrest Centre for Geriatric Care, 3560 Bathurst Street, North
York, Ontario, Canada M6A 2El.
405
0093-934x&o $3.00
Copyright
Q 1990 by Academic Press, Inc.
All rights of reproduction
in any form reserved.

406

BEHRMANN,

BLACK, AND BUB

deficit. It is suggested that longitudinal neurobehavioral studies add to our understanding of the alexic deficit and provide insight into the recovery process.
8 1990 Academic

Press, Inc.

Systematic longitudinal studies of recovery from alexia are sparse,
unlike the many detailed reports of recovery from aphasia. The early
anecdotal accounts of alexia recovery are brief and provide contradictory
accounts, with some patients demonstrating minimal or no recovery and
others improving with specialized training (Benton, 1964). More recently,
Marshall and Newcombe and their colleagues (Marshall, Newcombe, &
Hioms, 1975; Newcombe, Hiorns, & Marshall, 1976; Newcombe & Marshall, 1973; Newcombe, Marshall, Carrivick, & Hioms, 1976) presented
a series of investigations in which they plotted recovery curves in several
patients with acquired dyslexia. Both qualitative and quantitative differences emerged over time, with changes in the shape of the recovery
curve and in the nature of the error patterns. A fundamental point made
by these authors is that in-depth longitudinal studies provide a crucial
source of evidence which increases our understanding of physiological
recovery mechanisms. Knowledge conerning the potential for and constraints on recovery is critical for the purpose of prognosis and remedial
intervention. In this paper, we track the recovery of a single subject
with acquired letter-by-letter dyslexia in order to document subtle
changes in reading performance over time. In addition to providing longitudinal empirical data, we also attempted to examine whether the
changes in performance were attributable to the resolution of the underlying functional lesion which gave rise to the reading deficit or whether
the observed changes resulted from increased efficiency in the use of
the adaptive letter-by-letter reading procedure.
Recent research on spontaneous remission in acquired neurobehavioral
disorders has generally included reading alongside other language functions. These studies have documented the reading deficit in relation to
Broca’s’and Wemicke’s aphasia and have tended to focus on the overall
pattern of performance in relation to etiology, lesion size, and other
possible prognostic factors (Kertesz, 1979, 1984). Since reading shares
some, but not all, cognitive processes with other language functions, the
recovery of reading, in many instances, is poorly aligned with the recovery of these other functions (Gardner, Denes, & %-if, 1975). The
study of the recovery of reading therefore merits independent consideration. Pure alexia, in which the reading disorder is largely unaccompanied by other cognitive deficits, provides an important opportunity for
such study. For example, Newcombe et al. (1974, 1976) systematically
documented the behavior of MB, a patient with residual dyslexia over
a period of 160 weeks, following evacuation of a large left occipital
intracranial abscess. At 3 months postsurgery, MB was able to name

RECOVERY

OF PURE ALEXIA

407

letters but not to read words, a pattern typical of alexia without agraphia.
Over time, the number of MB’s error responses decreased, reaching an
asymptote around 60 weeks postsurgery. Interestingly, the type of error
observed changed across testing periods, somewhat resembling the pattern of initial reading acquisition in children. Initially, MB attempted to
spell the words out loud and failures of visual analysis were noted. As
time passed, more phoneme-grapheme conversion errors were observed,
reflecting MB’s attempt to deal with the complex sound-spelling patterns
of English.
Our subject, DS, differed from MB in a fundamental way as her reading
performance was relatively accurate even immediately after the stroke.
Measures of accuracy were thus redundant and, instead, we tracked the
evolution of her pure alexia using computer presentation of stimuli and
on-line reaction time paradigms. Like MB, our subject, DS, exhibited
signs of explicit serial letter encoding in the initial stages poststroke.
This form of acquired dyslexia, in which individual letters are slowly
and sequentially sounded out, has been termed spelling dyslexia (Kinsbourne & Warrington, 1962), word-form dyslexia (Warrington & Shallice,
1980), and more recently, letter-by-letter reading (Patterson & Kay,
1982). The plethora of labels reflects the numerous theories surrounding
this deficit.
Before the various explanations put forward to account for this reading
impairment can be examined, the cardinal symptoms of the disorder
must be specified. The hallmark of this reading disorder is a word length
effect whereby words are identified through their constituent parts and
response latency is a monotonic function of the length of the stimulus
array. Although time is the critical feature in letter-by-letter reading, no
firm criteria or time boundaries have been set down. Reading times
ranging from 1.8 to 16.9 set for a 3- to 4-letter word and from 3.0 to
47.5 set for a 7- to S-letter word have been reported for different letterby-letter readers (Patterson & Kay, 1982; Warrington & Shallice, 1980).
Although there is much variability in the range of times observed in these
patients, even the fastest times reported are considerably slower than
the 400 msec required by a normal adult reader to read a common word
out loud (Rayner & Pollatsek, 1989).
Despite the fact that the symptoms of letter-by-letter reading are relatively well accepted, there is little consensus on the mechanism responsible for the impairment and lesions originating in peripheral as well
as more central reading processes have been implicated. One explanation
for the disorder suggests that letter-by-letter readers have a fundamental
problem in perceptual analysis; consequently, the encoding of single
letters is abnormally difficult and gives rise to elevated perceptual thresholds (Friedman & Alexander 1984). An alternative account suggests that
the problem is not in the recognition of single items but rather in the

408

BEHRMANN,

BLACK,

AND

BUB

synthesis or integration of multiple visual forms. This more general visual
disturbance, termed simultanagnosia, manifests as the reduced ability to
apprehend separate visual forms, leading to individual item analysis (Coslett & Saffran, 1989a; Kinsbourne & Warrington, 1962). According to
Warrington & Shallice (1980), the deficit in the automatic synthesis of
individual units is not general, but is produced by a primary impairment
in the word-form system in which the orthographic representations of
previously encountered words are stored. Because the representations
of words are affected, their perceptual status is relegated to that of
nonwords and they are no longer rapidly identified. Variables such as
concreteness or word frequency do not influence performance and words
cannot be comprehended unless the individual constituents are processed
in this sequential manner. Since the permanent description of words
cannot be activated, reading is mediated by an auditory spelling mechanism which accepts letter names as input. Thus, the overt strategy of
letter-by-letter reading “is an attempt to compensate for damage to a
stage in the reading process prior to phonological or semantic analysis”
(Warrington & Shallice, 1980, p. 108). Patterson and Kay (1982) refute
the claim that the deficit is in the word-form system itself and provide
empirical evidence from their letter-by-letter reader that the individual
graphemic analyzers and the word-form system are both intact. Their
results show that the deficit lies somewhere between these two systems,
disconnecting them in the way suggested by Dejerine (1892) in his classical interpretation of alexia without agraphia. The disconnection manifests as an inability to map abstract letter identities in parallel onto
existing word-level representations.
Support for the view that the word-form is not destroyed is provided
by the observation that some letter-by-letter readers are capable of making correct semantic judgments even when the target word has been
presented at intervals too brief for explicit identification (Coslett & Saffran, 1989b; Shallice & Saffran, 1986). In other words, whole-word codes
are preserved and can assist in determining the identity of the stimulus
item. Bub, Black, and Howell (1989) have shown that, in addition to
semantic knowledge, higher-level orthographic influences may be preserved in these subjects. Adopting a paradigm which has been used in
normal subjects to demonstrate the superiority of words relative to nonwords (Reicher, 1969; Wheeler, 1970), they showed that their letter-byletter, JV, reveals better performance for words relative to nonwords in
a forced choice task in which he has to decide which of two probed
letters appeared in the original briefly exposed target. This influence of
orthographic context, presumably the result of activating higher-level
codes however, does not operate very strongly in a free report task in
which the subject names the letters in the target, suggesting that the

RECOVERY

OF PURE ALEXIA

409

existing orthographic information may not be sufficiently detailed or
activated for an overt response.
The diverse explanations cited above suggest that letter-by-letter reading is an overt behavior which may be produced by a deficit at any one
of a number of stages in the reading process. Like the regularization
errors in surface dyslexia which may be the result of a deficit in any
one of seven locations (Coltheart & Funnell, 1987), letter-by-letter reading could also arise because of deficits at several possible lesion sites.
The laborious encoding strategy could be an adaptive strategy arising
subsequent to damage at very early stages of reading (Reuter-Lorenz
& Brunn, 1990; Farah & Wallace, 1990) or at later stages (Warrington &
Shallice, 1980) or at any one of a number of places between the two
(Kay & Hanley, 1990; Patterson & Kay, 1982). The goals of our study,
therefore, were to document the site of the underlying deficit in our
subject and to obtain fine-grained details of her letter-by-letter reading.
We then aimed to take measurements over time and to compare them
to the baseline data to determine whether any recovery took place. We
hypothesized that any recovery might be the product of a change in the
underlying deficit itself, or alternatively, that the underlying impairment
may not resolve but that the use of the adaptive strategy may become
more efficient over time. This distinction is important. In the first case,
reading would be more similar to the normal process whereas, in the
latter, the disorder would persist but would be obscured by better use
of the alternative, abnormal behavior. To differentiate between these two
possible sources, we aimed to reexamine the competence of the reading
system at one year poststroke to determine whether the underlying deficit
had resolved.
Although letter-by-letter reading has been described in some detail in
the literature to date, the interest has been on the nature of the disorder
rather than its change over time (Bub et al., 1989; Coslett & Saffran,
1989b; Friedman & Alexander, 1984). Although Friedman and Alexander
(1984) tested their letter-by-letter reader over a 2-year period, their focus
was on the residual deficit. They provide few details of its evolution,
except for remarking that at 22 months postsurgery “both reading accuracy and reading speed have improved at a slow but steady rate” (p.
13), and that “reading has not yet achieved automaticity.” Since recovery
has important implications for remediation which attempts to exploit the
natural pattern of recovery and since little is known about the progression
of acquired dyslexia, we evaluated the performance of our pure alexic
subject over several months using chronometric measures and employing
reaction time as an index of recovery. Longitudinal data of this type
have not been reported in the literature and they may contribute to our
understanding of the underlying disorder.

410

BEHRMANN,

BLACK, AND BUB

CASE REPORT
DS, a 32-year-old right-handed nurse who was born in Bombay was
admitted to hospital in November 1986 with sudden onset of left occipital
headache and dizziness. DS’s mother tongue was English and although
she had spoken the vernacular during her childhood in India, she was
schooled in English and speaks it exclusively at present. On initial examination, she had a right homonyous hemianopsia and mild right hemiparesis. On four-vessel angiography, she had an occlusion of the posterior
cerebral artery and a corresponding left occipital infarction was seen on
CT scanning. A migrainous etiology was inferred in view of a previous
and subsequent history of migraine headache and by exclusion of other
possible contributing factors by extensive investigation.
On detailed assessment of her language function shortly after her
stroke, her primary difficulty was with reading. She used a sequential
letter-by-letter procedure to decipher words. Her auditory comprehension and spoken language production were good. Occasional word-finding
difficulty detected on the Boston Naming Test was attributed to cultural
bias; for example, she could not name hammock or pretzel. DS’s hemiparesis resolved quickly and she was discharged home a few weeks after
admission. She continued on an out-patient occupational therapy program
which focused on activities of daily living and homemaking.
Visual field testing by Goldman perimetry shortly after her stroke
revealed a congruous right homonymous field defect, denser in the superior quadrant. This had improved to a right upper quadrantanopsia at
8 months follow-up. Neuropsychological evaluation 8 months after her
stroke revealed deficits in attention and concentration, in verbal learning
and recall, and in some visuospatial functions. Her verbal IQ (88) fell
into the lower 25th percentile and her performance IQ (74) in the lower
5th percentile on the Wechsler Adult Intelligence Scale Revised. Performance on the Wechsler Memory Scale (83) was commensurate with
her full-scale IQ (80). Again, there were no aphasic deficits noted and
reading comprehension was normal although reading was noted to be
slow. A reevaluation 18 months after onset showed improvement in her
performance IQ (89), which now fell in the lower 25th percentile. Her
verbal IQ (91) and her full scale score (89) showed a minor improvement.
Although DS continued to show mild impairments in attention and concentration, verbal learning, and recall, she had resumed her normal lifestyle and her occupation as a homemaker. She could read children’s
books to her two children but found everyday reading laborious and
unenjoyable.
Lesion Localization

A CT scan repeated 3 months postonset (see Figs. la, b, and c) was
unchanged from the admission scan except that the hypodensity in the

RECOVERY

OF PURE ALEXIA

411

occipital region was more clearly defined. It involved the medial occipital
temporal region including the fusiform, lingual, and cuneate gyrus, the
calcarine cortex, and the hippocampus. Thus, Brodmann’s areas 17, 18,
28, 19, and 37 were partially affected by the lesion. The hypodense area
also extended into the occipital white matter including the optic radiations
and patchy involvement of the forceps major of the splenium. It is noteworthy that a lesion involving the inferior paraventricular white matter
was also found because this area is considered critical for the development of pure alexia (Damasio & Damasio, 1983). The posterior lateral
thalamus and posterior internal capsule also appeared to be partially
affected.
TESTING
The testing was divided into three stages: documenting the letter-byletter pattern shortly after the onset of the stroke, locating the underlying
site of the lesion, and, finally, evaluating the changes in performance
over time.
A. Documenting DS’s Reading Behavior
These tests, conducted within the first week poststroke, were designed
to characterize the severity of the initial deficit and to set the baseline
for later comparison. Single words were presented to DS’s intact left
visual field using an Apple IIE computer and monitor linked to a millisecond reaction time clock. DS sat at a comfortable distance of 50 cm
from the monitor. The computer presentation consisted of a central fixation point which remained on the screen for 1 set and was followed,
after a delay of 500 msec, by the target stimulus which subtended a
vertical angle of 0.5”. Reaction times as well as accuracy scores were
collected. This general procedure was also used for subsequent testing.
For the single word testing, a list of 120 uppercase words, 40 three
letters, 40 five letters, and 40 seven letters in length, were presented
individually following the central fixation point. The mean frequency of
the words was controlled across lengths (mean 52, SD 70). Half the list
contained low-frequency words appearing less often than 20 times per
million and the other half contained high-frequency words, appearing
more than 20 times per million (Kucera & Francis, 1967). Half the list
consisted of abstract words and the other half, concrete words. All the
words had regular sound-spelling correspondences and there was one
homophone (sum).
The words appeared for unlimited exposure duration until DS responded by activating a voice relay key or by depressing a response key
depending on the task (see below). The stimuli were right-aligned so that
the final character of all words fell in the space immediately adjacent to
fixation. The visual angles subtended were 1.5”, 2.4”, and 3.6” for 3-,

412

BEHRMANN,

BLACK, AND BUB

FIG. 1. Axial C.T. scan at 15” to orbitomeatal line obtained on a GE9800, reveals a
left occipital infarction. The details of localization are discussed in the text.

5, and 7-letter words, respectively. A 3-set delay, measured from response time, preceded the following trial.
1. Oral reading. DS was instructed to read the words aloud as quickly
and as accurately as possible. A microphone-activated voice relay mechanism was used to measure reaction time and accuracy was recorded
manually. A set of practice trials was administered. DS read 100/120
words correctly. We noted that she named each individual letter out
loud before pronouncing the whole word. The four errors all arose from
letter misidentifications, e.g., joy + “jay.” There was no significant
effect of frequency (Fisher exact test high vs. low frequency, p = .14)
nor word type (Fisher exact test concrete versus abstract p = .36) on
her oral reading.
2. Lexical decision. The 120 words described above were combined
with 120 orthographically legal and pronounceable nonwords (for example, foy, trisk) derived by changing 1 or 2 letters of the companion
real word. The stimuli were presented individually following the fixation
point and DS was instructed to use her dominant hand to depress the
“yes” key for words and the “no” key for nonwords. Practice trials

BEHRMANN,

0

;

BLACK, ANT BUB

I

I

1

3

5

7

WORD LENGTH

2. DS’s mean reaction time per word length on oral reading and lexical decision
one week postonset.

were provided. Accuracy of judgment was high, with 110/120 of the
words and 102/120 of the nonwords correctly classified, slightly worse
than the performance of a group of 10 normal subjects who read half of
the items from the same task and scored 59.5/60 on the words and 59/60
on the nonwords. As on oral reading, there were no effects of frequency
nor word type on DS’s performance (Fisher exact test p = .08, p =
.18, respectively).
The average response times per word length for both oral reading and
naming latency are illustrated in Fig. 2.
Speed of reading is obviously very slow, as reflected in the long latencies on both tasks. A significant linear relationship between time and
word length was demonstrated (r = .99, p < .05 oral reading; r = .98,
p < .05 for lexical decision real words), reflecting a strong length effect.
The longer times for nonwords relative to words on lexical decision are
consistent with the findings that nonwords have different status in reading
(Rayner & Pollatsek, 1989). Nonwords, therefore, served only as foils
and were not analyzed further.

RECOVERY

OF PURE ALEXIA

415

DS required, on average, 1.2 set per letter in oral reading and 1.8 set
per letter in lexical decision. These times place her approximately in the
middle of the severity range set by Shallice (1988) in which the least
severe letter-by-letter reader required 1.8 set for a 3- to 4-letter word
and the most severe required 16.9 sec.
B. Locating the Underlying Impairment
Several stages of reading were investigated in an attempt to locate the
functional lesion giving rise to the letter-by-letter reading.
1. Single letter analysis. A list of 60 single upper case letters was
presented for letter naming. Each letter appeared individually for the
briefest possible exposure duration (subject to screen refresh time and
phosphor decay on the Apple IIE monitor) following a central fixation
point. DS was able to identify 58 of the 60 letters even at this brief
duration, suggesting that her letter-by-letter reading could not be attributed to a perceptual failure in recognizing individual items.
2. Integrating multiple items. Two- and three-item letter arrays (a single
character space between letters) were also presented to DS in the same
way as the single items described above. She recognized 45/60 items of
the two-item arrays at 32 msec and 47/60 of the three-item arrays at 48
msec. Thus, although she was able to identify the multiple item arrays
correctly at unlimited exposure durations, she did experience difficulty
integrating separate units at very brief exposures.
3. Lexical access. In order to determine whether DS, like other subjects
(Coslett & Saffran, 1989b; Shallice & Saffran, 1986), retained implicit or
unconscious knowledge about words which were presented too briefly
for explicit report, we constructed a semantic classification task which
we presented at limited exposure duration. The fact that, in some subjects, implicit knowledge is preserved although not necessarily utilized
in explicit report tasks is theoretically significant. Dejerine’s (1892) original theory proposed a total visual-verbal disconnection, thus preventing
any information from reaching the left hemisphere language area. The
fact that visual information may be gaining access to verbal mediation
argues against this view and, instead, is more consistent with the recent
notion that the right hemisphere may be responsible for lexical access
in these patients (Coslett & Saffran, 1989b).
Our task involved the presentation of three lists of 60 words (5-7
letters in length) using an exposure duration (3 set) at which DS would
be unable to read the entire word explicitly. She was instructed to report
the word if possible or if she could not read it, to judge (or guess if
necessary) whether the word was related to work or leisure (list 1, e.g.,
tennis, salary), whether it was associated with positive or negative emotion (list 2, e.g., malice, honesty), or whether it referred to an animal
or object (list 3, e.g., beaver, pencil). DS was not significantly different

416

BEHRMANN,

BLACK,

AND

BUB

from chance (50%) in her judgments of those words (n = 118) she could
not identify explicitly (x2 = .34, p > .lO), suggesting an absence of
preserved implicit semantic knowledge.
4. Orthographic
context effects. In order to assess whether DS was
able to synthesize any higher-level orthographic units from perception,
a finding which would favor adequate access and use of the word-form,
we adopted a paradigm which has been used extensively with normal
subjects (McClelland, 1976; McClelland & Johnston, 1977) and more
recently with subjects with acquired dyslexia (Bub et al., 1989; Kay &
Hanley, 1990; Reuter-Lorenz & Brunn, 1990). The word superiority effect
(WSE), in which a letter in a real word is recognized more quickly than
a letter in a nonword, which, in turn is recognized more quickly than a
letter in a random string, is taken as evidence that higher-level orthographic representations play a role in perception. While whole-word
activation may account for the former, the activation of subword units
is thought to account for the superiority of pseudowords over random
strings. Bub et al. (1989) have shown that their letter-by-letter reader,
JV, was able to recognize more letters in words and pseudowords than
in nonwords in all positions of a I-letter display when a forced choice
procedure was employed. In a free report task, in which the subject was
instructed to report all the letters, however, his performance was significantly poorer in the last two positions compared to the first two
positions of the array. The fact that a WSE was observed, albeit only
on the forced choice task, led the authors to conclude that whole-word
codes are preserved but that the activation from print, especially in the
final positions, is weak and may not be available for explicit naming in
the free report task.
We used the same forced choice procedure as Bub et al. (1989) to
assess whether DS could activate higher-level codes from print. We
predicted that if she was able to use the intact word representations
and/or orthographic knowledge, she would be able to report more letters
in words than in pseudowords and would show poorest performance on
random strings irrespective of letter position in the array.
Three groups of 40 four-letter strings were obtained, one of highfrequency words (more than 55 per million), one of pseudowords, and
one of random strings. The pseudowords were matched to the real words
in bigram frequency and the letter sequences all had a zero bigram
frequency. Each string was presented in DS’s intact visual field for a
limited exposure duration following a central fixation point. A random
pattern mask followed for 500 msec. The target duration was set at 450
msec, the exposure at which DS scored 70% accuracy at letter detection
in pseudowords. The relevant measure, then, was the accuracy of letter
detection as a function of the different target types. The forced choice
probe was conducted in two ways:

RECOVERY

OF PURE ALEXIA

417

a. Whole-word condition. Two 4-letter strings appeared after the mask,
one of which was the target item and the other a distractor, differing
from the target by a single letter. The different letter appeared equally
often in each of the four letter positions. In the case of words, the
distractor was also an English word; for example, when the target was
BOOK, the distractor was BOOT. The position of the target (either left
or right) was counterbalanced across trials. The task was blocked with
letter strings being presented first, followed by pseudowords and then
words.
b. Single-letter condition. Two single letters appeared after the string,
with their relative word position marked, for example, for the target
BOOK, the probes were ---T or---K.
Each letter position was probed
an equal number of times and in the case of words, the distractor letter
could also complete an English word as in the example provided above.
The numbers of correct letter detections on both whole-word and single
letter probes are shown in Fig. 3.
There was no significant difference between DS’s ability to detect
letters in words relative to pseudowords or random strings on either the
whole-word trials (x2 = 1.39, p = 5) or on the single-letter trials (x2 =
.8, p > .5). There was, however, a significant difference in her ability
to detect the target letter in the first two positions of the array compared
to the last two positions (positions 1 and 2 vs. positions 3 and 4, collapsed
across target types, x2 = 10.67, p < .Ol). These findings suggest that
DS was not able to use higher-order knowledge to assist in the perceptual
synthesis of the 4-letter displays on either procedure. The results are
consistent with the findings of Kay and Hanley (1990) whose subject,
PD, showed no word superiority and who was unable to identify more
than one letter presented simultaneously. While DS had sufficient time
to encode the first two letters serially, proceeding from left to right,
leading to better performance in these positions, the exposures were too
brief to allow sequential encoding of all four positions.
In sum, DS demonstrated, in the acute stage of her illness, the classical
symptoms of a severe letter-by-letter reader with relatively intact peripheral and visual analysis. While she was able to access higher units
of lexical semantic information when the target appeared for unlimited
duration (leading to high accuracy scores in the lexical decision and oral
reading tasks), she was unable to access this information rapidly when
the target appeared too briefly for explicit encoding of the component
letters.
C. Longitudinal Testing
Following the preliminary testing, DS was assessed on 10 subsequent
occasions over a 1Zmonth period. The oral reading and lexical decision
tasks described above were administered on each occasion. An additional

418

BEHRMANN,

BLACK, AND BUB

40

0

q

WHOlETSINGLE ITEMS

0
lfctms

fJsEu-

STRINGS

TARGET TYPE

FIG. 3. Number letters correctly detected in words, pseudowords, and random strings
on a forced-choice task.

five lists of words were compiled (and intermixed with nonwords for
lexical decision) using the same criteria for length, frequency, and word
type as the original list. The lists were presented in random order over
time. The administration procedure mirrored that of the preliminary testing and accuracy and latency measures were obtained in each session
for oral reading and for lexical decision.
I. Oral reading. DS’s accuracy in oral reading (expressed as percentage
errors) was consistently high across all time periods and all word lengths,
ranging between 92 and 100% correct. Reaction times in oral reading,
expressed as a function of time since onset, are illustrated in Fig. 4.
A repeated measures analysis of variance with time since onset as a
between factor and word length as a within factor was conducted. There
was a significant main effect of time (F(9, 18) = 19.1, p < .OOl), with
reaction time decreasing with an increase in number of weeks since onset

419

RECOVERY OF PURE ALEXIA

v

I
0

1’0

2b

3’0

4’0

Q

3LElTER

+
9

SLEl-lER
7LEllER

sb

TIME POST-ONSET (WEEKS)

FIG. 4. DS’s mean reaction time in oral reading as a function of time period.

of the stroke. A significant effect of word length was also noted (F(2,
18) = 4.89, p < .05), with time for 3-letter words (1635.1. msec) being
significantly faster than those for Sletter words (1983.3 msec) which, in
turn, were faster than those for 7-letter words (2340 msec). The recovery
curves for each word length, plotted as a function of time since onset,
showed the best fit to curvilinear regression lines (r = 0.8 for 3 letters,
0.9 for 5 letters, and 0.85 for 7 letters), with negative deceleration from
30 weeks postonset. In addition, a significant linear relationship between
word length and response latency, with an increase in processing time
with every additional letter, was observed in each individual testing
session with the exception of 21 weeks postonset (r = .09) at which
point, the 7-letter words converged on the 3-letter words. The regression
slopes and linear trends plotted for word length on reaction time for the
10 testing sessions are presented in Appendix I.
In the final testing session, at 51 weeks postonset, DS’s mean reading
times of 896.3, 1138, and 1442 msec for 3-, 5-, and 7-letter words, respectively, showed a significant linear trend (r = 1.00) with an intercept
of 612.3 and a slope of 273 msec for each additional letter. These times

420

BEHRMANN,

BLACK, AND BUB

Q
+

3LEl-rER.s
5LErrERS

-o- 7LElTERS

6000

TIME POST-ONSET (WEEKS)
FIG. 5.

DS’s mean reaction time on lexical decision as a function of time period.

far exceed the slopes of 30 msec per letter found in skilled readers when
a stimulus word is presented in the left visual field (Bub & Lewine, 1988)
but are more consistent with the l-l.3 set per letter time of JV, another
letter-by-letter reader (Bub et al., 1989). These findings suggest that
although there has been a dramatic improvement in response latency as
a function of time since the stroke, DS still shows elevated reaction times
and the length effect which is characteristic of letter-by-letter reading.
2. Lexical decision. As on oral reading, the accuracy scores on lexical
decision were consistently high, ranging between 90 and 98% correct for
“yes” responses. The reaction times, as a function of testing period,
are illustrated in Fig. 5.
An analysis of variance with time as a between factor and word length
as a within factor was conducted. The results reveal a significant effect
of testing session on reaction time (F(9, 18) = 18.8, p < Ml) and a
significant effect of word length (F(2, 18) = 5.3, p < .05). The mean

RECOVERY

OF PURE ALEXIA

421

reaction times per word length (collapsed across testing sessions) were
2486.5, 3317.6, and 3815.2 for 3-, 5, and 7-letter words, respectively.
The data were best fit to a curvilinear regression line demonstrating
rapid decrease in response latency up to 20 weeks postonset with more
gradual decrease thereafter. The monotonic relationship between word
length and reaction time was observed in all but one testing session (4
weeks postonset) as seen on the regression slopes and linear trends (see
Appendix I). The negative deceleration in latency is confirmed by the
relatively small diffference in the number of seconds taken to read a
single letter at 39 weeks postonset compared to the number of seconds
required at 51 weeks postonset. Observation of DS over an additional
4 months (up to 75 weeks) revealed no major changes in her performance,
providing further support for the asymptote we documented.
The results of the lexical decision task, then, like the oral reading
task, demonstrate rapid decrement in response latency over time, with
performance approaching an asymptote in the final periods. In order to
obtain an estimate of normal reading speeds on oral reading and lexical
decision, we administered our tasks using the identical administration
procedure to a normal control subject, matched to DS in age, sex, and
level of education. Mean reaction times on oral reading were 589, 551,
and 579 msec for 3-, 5-, and 7-letter words, revealing a nonsignificant
regression slope of 4 msec (r = .63, p > .05). Mean reaction times for
lexical decision were 497, 505, and 553 msec for 3-, 5-, and 7-letter
words, respectively, producing a nonsignificant regression slope of 11
msec for every additional letter (r = .77, p > .05). It is clear that DS’s
mean reaction times were considerably slower than those of a matched
control subject and that, unlike the normal control, she continued to
show a linear trend in her data even in the final testing session.
3. Orthographic context effects. At one year postonset, we reexamined
the nature of the underlying reading deficit in DS to determine whether
the changes observed in her reading performance might be attributable
to her being able to gain rapid access to the visual word-form system.
Instead of repeating the forced-choice probe procedure employed initially, we utilized a free-report paradigm which, according to Bub, Black,
and Behrmann (1990), is a more sensitive measure of whole-word activation. The use of a binary response in the forced-choice procedure
renders the task relatively insensitive because the level of accuracy, by
chance alone, is 50%. Guesswork in the free report measure, in which
the subject names the perceived letters, is minimized by using target
items whose identity cannot be guessed on the basis of three letters.
In the following task, we used this free-report paradigm to assess DS’s
ability to access the word-form system. The strongest prediction was
that if the change in reading performance were attributable to the resolution of the primary deficit, then we might expect to see the emergence

422

BEHRMANN,

BLACK,

AND

BUB

of a word-superiority effect (WSE). If however, there was no recovery
of this fundamental inability to rapidly access word forms, then no WSE
would be evident and the changes in DS’s reading performance could
not be attributed to the resolution of the underlying deficit.
A set of 4-letter words (n = 120) was used (taken from Bub, Black,
& Behrmann, 1990). All words yielded at least two plausible responses
if completed on the basis of their first three letters; for example, the
target SEEN could be completed as SEEM, SEEK, SEED. Nonsense
words whose summed bigram frequencies were either equal to or greater
than the counterpart word were formed by changing one letter of the
target. In this task, DS was instructed to report the letters without
guessing the identity of the unseen letters. The targets were presented
to her intact left field and followed by a random pattern mask for 500
msec. The targets were all presented at 150 msec, the exposure duration
at which DS obtained 70% correct letter detection in pseudowords. The
presentations of word and pseudoword targets were blocked. Two measures were taken: the total number of letters detected correctly from the
target and the total number of full words (all four letters) correctly
reported. Using this measure, Bub, Black, and Behrmann (1990) found
that normal readers showed a clear difference (35%) between words and
pseudowords, when the number of whole-words perceived correctly is
measured. The difference between the words and pseudowords was confirmed on an item analysis which showed a significant difference across
all items (f(119) = 14.01, p < .OOl). According to these authors, this
technique provides a sensitive measure of a word-form deficit, indicating
whether or not the lexical representations are accessed and utilized. The
results of these measures for words and nonwords are shown in Fig. 6.
As is evident from this figure, there is no overall significant difference
in DS’s ability to detect letters or entire targets on words compared to
pseudowords on either of the two measures; i.e., there is no overall
WSE (Fisher exact test collapsed across all frequencies; letters x2 =
.07, p > 5; entire targets x2 = 5.13, p > .5). The mean single letters
correctly reported were 76.1% for words and 75% for nonwords while
the mean whole words correctly reported were 43 and 39.5% for words
and nonwords, respectively. The absence of a WSE suggests that even
at one year postonset, DS was still unable to use higher-level orthographic
context to assist in perception.
DISCUSSION

We have described the evolution of a reading deficit over a 1Zmonth
period of a subject with pure alexia, largely unaccompanied by any other
significant neurobehavioral deficit. In the first week following the stroke,
DS read laboriously and sequentially and was classifiable as a letter-byletter reader of moderate severity according to the continuum of reading

423

RECOVERY OF PURE ALEXIA
80

LEllERs
WHCIEWXDS
60I-

64
5

4cI-

%

2tI-

I3-r

TARGET TYPE

FIG. 6. Percentage letters correctly detected in words and pseudowords on a full-report
task at one year poststroke.

times set by Shallice (1988). DS was able to recognize letters accurately,
provided that enough time was given for explicit encoding. Her poorer
performance on multiple-item arrays at brief exposure durations suggests
an impairment in rapidly identifying several items presented simultaneously (see Kay & Hanley 1990, for a similar case). DS was able to
encode the first two letters of a 4-letter array reliably but was not able
to process the final two letters under brief exposure durations. The
absence of a word superiority effect reflects the inability to access higherorder orthographic structure for detecting letters in familiar words, pseudowords, and random strings.
Over time, DS showed remarkable improvement in her reading performance: her reaction times in oral reading and lexical decision tasks
decreased dramatically although, even in the final testing session, her
reading times were still considerably longer than those reported for nor-

424

BEHRMANN,

BLACK,

AND

BUB

ma1subjects. DS’s reading times were also slower than the times reported
for a group of nonalexic left-occipital damaged patients tested on the
same tasks with the same materials (Black, Bub, & Behrmann, 1987).
Mean reaction times for a group of five patients with single left inferior
occipital lesions for 3-, 5-, and 7-letter words, respectively, were 860,
895, and 950 msec (DS at 51 weeks: 896.3, 1138, 1442) on oral reading
and 884, 887, and 1003 msec (DS at 51 weeks 1303, 1555, 1712 msec)
on lexical decision. The mean rate of processing in this group was 30
msec per letter, which is far less than the 168.5 msec (lexical decision)
or 273 msec (oral reading) per letter required by DS in the final testing
session. DS’s performance is clearly abnormal even as it approaches an
asymptotic level. In addition, she still showed an effect of word length
on response latency. This persisting monotonic relationship between
length and time suggests that she was still a letter-by-letter reader.
In view of the fact that the alteration in reading performance appears
to be more quantitative than qualitative, to what underlying mechanism
can we attribute the difference? A reanalysis of the word-form deficit
still failed to reveal a word superiority effect even when a more sensitive
free-report procedure was used in place of the forced-choice task. The
absence of higher-order contextual influences suggests that access to the
word-form had not improved and the observed changes in DS’s performance were therefore unlikely to be the result of the recovery of the
primary deficit. Rather, we infer that her improvement reflected increased
efficiency in the use of sequential letter-by-letter spelling as a form of
compensation. Thus, although DS still seemed to be processing letters
serially rather than in parallel, increased speed in naming individual
letters or in exploiting kncwledge of orthotactic patterns may well have
led to the improvement in her reaction times In other words, she seemed
to have refined her compensatory procedure but had still not achieved
automatic access to the word-form system. Support for this interpretation
is obtained from another letter-by-letter reader (Friedman & Alexander,
1984) who, despite changes in behavior over a 2-year period, had still
not achieved automaticity and was still defined as a letter-by-letter reader.
An important question which arises from these findings concerns the
way in which automaticity may be achieved. One possible avenue for
reacquisition is through the mediation of this sequential letter-by-letter
strategy. Whether such an identification procedure is an adequate means
for ensuring access to the visual word-form, however, is unclear. Some
data from the childhood developmental literature suggest that it is. Seymour and Porpodas (1980), for example, have shown that a word length
effect is a critical dimension in reading acquisition. They described word
lengths in the order of 200 msec per letter in a group of normal 7-yearolds. This length effect seemed to be critically involved in the process
of constructing word representations. As reading skills were refined with

RECOVERY

OF PURE ALEXIA

425

age and word forms were stabilized, so this length effect appeared to
decrease. It is questionable, however, whether the laborious sequential
reading process utilized in letter-by-letter reading facilitates the reacquisition of word forms or not. The fact that DS’s performance was
approaching an asymptote at levels well beyond the normal times (and
beyond the usual period of spontaneous recovery) suggests that this might
not be the case.
The recovery curve mapped out in our subject has important implications for remediation. An alternative, wholistic approach might be
necessary to ensure rapid access to the word-form system since the
adopted compensatory strategy does not seem to suffice. Intervention
of this sort, of course, is possible only if letter-by-letter readers are
detected clinically and correctly diagnosed. This is often not the case.
In fact, it has been suggested that letter-by-letter reading is not well
recognized in a clinical setting for a number of reasons. It is relatively
rare in its pure form, conspicuous neurobehavioral problems are absent,
and the loss of the reading ability is often not considered limiting (Marks
& De Vito, 1987). Indeed, if DS were to be seen in a clinical setting at
one year postonset, it is unlikely that her dyslexia would have been
discovered on routine clinical testing, as reading accuracy was good.
The overall inefficiency of her reading ability, however, continued to
take its toll, making reading burdensome. Chronometric measures such
as those described in this paper may therefore serve a useful diagnostic
purpose indicating the nature of the reading disturbance and documenting
the changes in performance during the recovery process.

426

BEHRMANN,

BLACK,

AND

BUB

APPENDIX I
REGRESSIONSLOPESAND LINEAR TRENDS FOR ORAL READINGAND
LEXICAL DECISION IN RELATIONTO TIME POSTONSET
Oral reading

Lexical decision

Time postonset
(weeks)

Slope
(msec)

Linear trend
r

Slope
(msec)

Linear trend
r

2
3
4
8
11
14
21
26
39
51

1128
575
192.5
471
162
405.5
10
316
263
273

0.99
1.00
0.94
0.95
1.00
0.99
0.09
1.00
1.00
1.00

2795
1607
72
861.5
445
136.5
218.5
160.5
179
168.5

0.98
0.98
0.23
0.87
0.81
O.%
1.00
0.81
0.93
0.94

REFERENCES
Benton, A. 1964. Contributions to aphasia before Broca. Cortex, 4, 314-327.
Black, S. E., Bub, D., and Behrmann, M. 1987. Asymmetries of lexical access-A study
of patients with occipital damage. Phoenix: Academy of Aphasia.
Bub, D., Black, S. E., and Behrmann, M. 1990. Visual word-form deficits in acquired
dyslexia. Manuscript submitted for publication.
Bub, D., Black, S. E., and Howell, J. 1989. Word recognition and orthographic context
effects in a letter-by-letter reader. Brain and Language, 36, 357-376.
Bub, D., and Lewine, J. 1988. Different modes of word recognition in the left and right
visual fields. Brain and Language, 33, 161-188.
Coltheart, M., and Funnell, E. 1987. Reading and writing: One lexicon or two? In D. A.
Allport, D. G. MacKay, W. Prinz, and E. Scheerer (Eds.), Language perception and
production: Shared mechanisms in listening, reading and writing. London: Academic
Press.
Coslett, H. B., and Saffran, E. M. 1989a. Simultanagnosia: To see but not two see.
Manuscript under review.
Coslett, H. B., and %&ran, E. M. 1989b. Evidence for preserved reading in “pure alexia.”
Brain, 112, April, pt. 2, 327-359.
Damasio, A. R., and Damasio, H. 1983. The anatomic basis of pure alexia. Neurology,
33, 1573-1583.
Farah, M. J., and Wallace, M. A. 1990. Pure alexia as a visual impairment: A reconsideration. Cognitive Neuropsychology, in press.
Friedman, R., and Alexander, M. 1984. Pictures, images and pure alexia: A case study.
Cognitive Neuropsychology, 1, 9-23.
Gardner, H. Denes, G., and Zurif, E. 1975. Critical reading at the sentence level in aphasia.
Cortex, 11, 60-72.
Johnston, J. C. 1978. A test of the sophisticated guessing theory of word perception.
Cognitive Psychology, 10, 123-153.

RECOVERY OF PURE ALEXIA

427

Kay, J., and Hanley, R. 1990. Simultaneous form perception and serial letter recognition
in a case of letter-by-letter reading. Cognitive Neuropsychology, in press.
Kertesz, A. 1979. Aphasia and associated disorders. New York: Grune & Stratton.
Kertesz, A. 1984. Recovery from Aphasia. In F. C. Rose (Ed.), Advances in neurology.
Volume 42. Progress in aphasiology. New York: Raven Press.
Kinsboume, M., and Warrington, E. K. 1962. A disorder of simultaneous form perception.
Brain, 85, 461-486.
K&era, H., and Francis, T. 1967. Computational analysis of present-day American English. Providence, RI: Brown University Press.
McClelland, .I. L. 1976. Preliminary letter identification in the perception of words and
nonwords. Journal of Experimental Psychology: Human Perception and Performance,
2, 80-91.
McClelland, J. L., and Johnston, J. C. 1977. Preliminary letter identification in the perception of words and nonwords. Journal of Experimental Psychology: Human Perception and Performance, 2, 80-91.
Marks, R. L., and De Vito, T. 1987. Alexia without agraphia and associated disorders:
Importance of recognition in the rehabilitation setting. Archives of Physical Medicine
and Rehabilitation,

68, 239-243.

Marshall, J. C., Newcombe, F., and Hioms, R. W. 1975. Dyslexia: Patterns of disability
and recovery. Scandinavian Journal of Rehabilitation Medicine, 7, 37-43.
Newcombe, F., Hioms, R. W., and Marshall, J. C. 1976. Acquired dyslexia: Recovery
and training. In Y. Lebrun and R. Hoops (Eds.), Recovery in aphasics. Neurolinguistics 4. Amsterdam, Swets & Zeitlinger.
Newcombe, F., and Marshall, J. C. 1973. Stages in recovery from dyslexia following a
left cerebral abcess. Cortex, 9, 329-332.
Newcombe, F., Marshall, J. C., Carrivick, P. J., and Hiorns, R. W. 1974. Recovery curves
in acquired dyslexia. Journal of the Neurological Sciences, 24, 127-133.
Patterson, K. E., and Kay, J. 1982. Letter-by-letter reading: Psychological descriptions
of a neurological syndrome. Quarterly Journal of Experimental Psychology, 34, 411442.
Rayner, K., and Pollatsek, A. 1989. The psychology of reading. Englewood Cliffs, NJ:
Prentice-Hall.
Reicher, G. M. 1969. Perceptual recognition as a function of meaningfulness of stimulus
material. Journal of Experimental Psychology, 81, 274-280.
Reuter-Lorenz, P. A., and Brunn, J. L. 1990. A prelexical basis for letter-by-letter reading:
A case study. Cognitive Neuropsychology,
7, l-20.
Seymour, P. H. K., and Porpodas, C. D. 1980. Lexical and nonlexical processing of spelling
in developmental dyslexia. In U. Frith (Ed.), Cognitive Processes in Spelling. London:
Academic Press.
Shallice, T. 1988. Prom neuropsychology to mental structure. Cambridge: Cambridge University Press.
Shallice, T., and Saffran, E. 1986. Lexical processing in the absence of explicit word
identification: Evidence from a letter-by-letter reader. Cognitive Neuropsychology,
3,
429-458.

Warrington, E. K., and Shallice, T. 1980. Word-form dyslexia. Brain, 103, 99-l 12.
Wheeler, D. D. 1970. Processes in word recognition. Cognitive Psychology, 1, 59-85.