R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

77

COGNITION
Cognition 82 (2001) 77–126
www.elsevier.com/locate/cognit

Deep dyslexia in the two languages of an Arabic/
French bilingual patient
Renée Béland a,b,c,*, Zohra Mimouni a,d
a

b

Centre de Recherche, Institut Universitaire de Gériatrie de Montréal, Montréal, Canada
Groupe de Recherche en Neuropsychologie Expérimentale, Université de Montréal, Montréal, Canada
c
École d’Orthophonie et d’Audiologie, Université de Montréal, Montréal, Canada
d
Cégep Montmorency, Laval, Canada
Received 29 December 2000; accepted 11 June 2001

Abstract
We present a single case study of an Arabic/French bilingual patient, ZT, who, at the age of
32, suffered a cerebral vascular accident that resulted in a massive infarct in the left perisylvian region. ZT’s reading displays the characteristics of the deep dyslexia syndrome in both
languages, that is, production of semantic, visual, and morphological errors, and concreteness
effect in reading aloud and impossibility of reading nonwords. In the first part of this paper,
using a three-route model of reading, we account for the patient’s performance by positing
functional lesions, which affect the non-lexical, the semantic lexical and the non-semantic
lexical routes of reading. Phonological priming observed in a cross-language visual lexical
decision task indicates that implicit assembled phonological recoding is possible. The above
lesions and implicit nonword reading characterize the output form of deep dyslexia. However,
error distribution reveals dissociations across languages (e.g. the semantic error rate is higher
in French whereas translations are more frequent in the Arabic testing) that cannot be
accounted for within a three-route model. In the second part, extensions to Plaut and Shallice’s connectionist model (Cognitive Neuropsychology, 10 (5) (1993) 377) are proposed to
account for the translinguistic errors observed. ZT’s error distribution is compared to that
obtained by Plaut and Shallice after lesions had been applied at different locations through the
40–60 network. The overall syndrome of deep dyslexia found in both languages is explained
as resulting from lesions along the direct (O ! I) and output (S ! Ip, Ip ! P) pathways of
reading. Lesions along the output pathway mostly affecting S ! Ip connections in French and
Ip ! P connections in Arabic account for discrepancies in ZT’s error pattern across tasks and
languages. This case study demonstrates the superiority of a connectionist approach for

* Corresponding author. Centre de Recherche, Institut Universitaire de Gériatrie de Montréal, 4565,
Chemin Queen Mary, Montréal, Québec, Canada H3W 1W5. Tel.: 11-514-340-3540; fax: 11-514-3403530.
E-mail address: renee.beland@umontreal.ca (R. Béland).
0010-0277/01/$ - see front matter q 2001 Elsevier Science B.V. All rights reserved.
PII: S 0010-027 7(01)00148-2

78

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

predicting the error pattern in deep dyslexia. q 2001 Elsevier Science B.V. All rights
reserved.
Keywords: Deep dyslexia; Bilingualism; Arabic; French; Connectionism

1. Introduction
Early accounts of deep dyslexia (Marshall & Newcombe, 1973) describe it as a
symptom-complex characterized by semantic errors in reading, together with the cooccurrence of derivational and visual errors, substitutions of function words, and
difficulty in or even impossibility of reading nonwords. Coltheart (1980a) extends to
the initial list a set of other co-occurring symptoms, i.e. lexical derivation of phonology from print impaired, abstract words harder to read aloud than concrete words,
writing spontaneously or to dictation impaired, auditory-verbal memory impaired,
reading of a word dependent on its context, verbs harder to read than adjectives
which are harder than nouns, function words harder to read than content words. The
list is then reviewed and edited in Coltheart, Patterson, and Marshall (1987) who
take out from the list the three last features, thus limiting to eight the number of
symptoms claimed to co-occur mandatorily with semantic errors in deep dyslexia.
Initial attempts to account for the deep dyslexia syndrome were proposed within
cognitive neuropsychological models. The multiple-impairment hypothesis postulates that several deficits along the lexical route (semantic and non-semantic routes
for reading) and the non-lexical route are responsible for the co-occurrence of the
types of errors observed in the deep dyslexic patient’s reading performance. More
specifically, within the three-route model (Coltheart et al., 1987; Morton & Patterson, 1980; Shallice & Warrington, 1980), semantic and morphological errors in
reading aloud, difficulty in reading abstract words and inability to read nonwords
aloud are claimed to arise through impairments at four independent lesion sites
within the language-processing system: (1) lesion to the assembled pathway that
allows the identification of sub-word orthographic segments and their conversion
into phonological segments, then into the oral production of a letter string; (2)
damage to the addressed pathway which allows reading aloud of words with mediation of the semantic system; (3) impairment of the addressed pathway which allows
reading of words without mediation of the semantic system; and (4) damage to the
semantics of abstract words.
In subsequent research, however, criticisms have been raised leading to a weakening of the “syndrome-complex” theory. More specifically, even though semantic
errors have been found to represent the central symptom of deep dyslexia as claimed
initially, the close association and mandatory co-occurrence of the other symptoms
have been questioned. For example, Allport and Funnell (1981) reported an absence
of part-of-speech effect; Caramazza and Hillis (1990) suggested that some of the
symptoms described do not necessarily appear together with semantic errors.
Another functional interpretation of deep dyslexia has been offered by Coltheart

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

79

(1980a,b, 1983, 2000) and Saffran, Bogyo, Schwartz, and Marin (1980) who claim
that in deep dyslexia, a very large lesion in the left hemisphere destroys the patient’s
reading ability, hence leading to reading by an intact but weak language system
located in the right hemisphere rather than reading via an impaired language system
located in the left hemisphere. This hypothesis has been empirically based on similarities in reading performance of deep dyslexic patients and of patients assumed to be
reading with the right hemisphere. Right hemisphere reading has been observed as
having limitations, which include incapacity in reading abstract words and function
words, in distinguishing among semantically related words, and in converting an
orthographic code into a phonological code. Such limitations should explain the
deep dyslexic’s difficulty in reading abstract and function words, nonwords and
bound morphemes. However, this hypothesis has been frequently questioned on
several grounds. Normal subjects tested in reading tasks involving the left visual
field (LVF) presentation do not behave like deep dyslexic patients (e.g. Patterson &
Besner, 1984). When compared to split-brain patients, deep dyslexic patients have
shown better reading in LVF presentation (Michel, Hénaff, & Intriligator, 1996;
Patterson & Besner, 1984). In addition, Roeltgen (1987) reported the case of a patient
who lost symptoms of deep dyslexia following a second left hemisphere lesion.
Glosser and Friedman (1990) examined deep dyslexia cases with relatively small
size lesions, hence refuting the claim that the large size of the lesion in the left
hemisphere in deep dyslexic patients automatically induces right hemisphere
compensation. More recently, a study by Price et al. (1998) in which neuroimaging
techniques were used has provided no support for an exclusive reliance on right
hemisphere reading by deep dyslexics (but see Coltheart, 2000 for conflicting views).
Connectionist approaches to deep dyslexia have also been attempted. Hinton and
Shallice (1991) and then Plaut and Shallice (1993) have succeeded in reproducing
the co-occurring errors found in deep dyslexia by lesioning a connectionist network
at different sites. More specifically, the model designed by Plaut and Shallice (1993)
is capable of simulating most of the simultaneous characteristics of the syndrome by
lesioning various connections in a network trained to map orthography onto semantics and semantics onto phonology. The authors showed that lesions made at different sites produce the co-occurrence of visual, semantic and phonological errors as
well as most of the lexical effects observed in deep dyslexia. In their model, the cooccurrence of these different error types derives from the nature of the computation
that the whole system carries out. In contrast with the multiple functional impairment approach, in the Plaut and Shallice (1993) connectionist model, lesions in
different parts of the network may result in the same overall syndrome. At the
same time, other aspects of the syndrome may differ between lesion sites. As an
example, Plaut and Shallice (1993, p. 487) showed that a lesion simulated on the
direct pathway produced visual and semantic errors together with a concreteness
effect. Lesions to the clean-up network produced the same error pattern without the
concreteness effect.
Although the deep dyslexia syndrome has incited considerable research and
resulted in numerous publications, very little has been reported on bilingual deep
dyslexic patients. Most of the studies have dealt with errors either in monolingual

80

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

patients or in one of the languages of the patients described. In this paper, we report
the first case of an Arabic/French bilingual patient who, in addition to showing all of
the symptoms of output deep dyslexia in his two languages, produces translations
into French, his second language. Wydell and Butterworth (1999) report the case of
a 16-year-old English/Japanese bilingual boy who showed the symptoms of developmental dyslexia in English only. Byng, Coltheart, Masterson, Prior, and Riddoch
(1984) present the case of a deep dyslexic patient whose first language was Nepalese
but who could also read in English. The patient showed most of the features of the
deep dyslexia syndrome in English but reading was almost impossible in Nepalese.
Most of the patient’s oral productions were translations into English. The existing
theoretical approaches to deep dyslexia such as the multiple-impairment interpretations, the right hemisphere hypothesis or the connectionist models, have not
provided any account for such types of errors either because they have been addressing monolingual patients or because of an implicit understanding that the interpretations offered for monolingual patients would equally apply to patients speaking
two or more languages. Furthermore, such approaches have not made any provision
for languages with different types of alphabets.
In the case we are presenting, the patient’s two languages are highly contrastive.
Arabic and French display two different alphabets (Latin vs. Arabic) and two different writing modes (right-to-left vs. left-to-right). In addition, French is characterized
by a deep or opaque orthography, i.e. orthography with an irregular spelling in that
sound and spelling do not always correspond in the language. Arabic, on the other
hand, exhibits two types of orthography: (1) an orthography “with diacritics”, i.e. in
which consonants and long vowels are represented by letters whereas short vowels
are represented by symbols; and (2) an orthography “without diacritics”, i.e. where
only consonants and long vowels are represented, short vowels being absent. The
two types of orthography have been referred to in the literature on Semitic languages
as shallow and deep, respectively (Bentin & Ibrahim, 1996). However, it is important to point out that Arabic orthography, whether deep or shallow, is highly regular
in that there is a one-to-one correspondence between letters, diacritics, and
phonemes. A detailed description of the Arabic orthographic system is presented
in Section 2.
It has been suggested in the literature on reading languages with deep and shallow
orthographies (Bentin & Ibrahim, 1996) that the relative involvement of the phonological and the lexical routes in reading depends on the degree of regularity or
transparency in the spelling-to-sound correspondence. According to this claim, readers in languages with deep orthography such as English or French would rely more
on the lexical route whereas readers in languages with shallow orthographies such as
Spanish or Portuguese would be using non-lexical mechanisms (Bentin & Frost,
1987; Katz & Feldman, 1983). Readers of languages with both types of orthography
such as Arabic or Hebrew are, therefore, expected to use either route depending on
the type of orthography presented. According to the orthographic depth hypothesis
(Katz & Frost, 1992; Wydell & Butterworth, 1999), the lexical route (addressed
routine) predominates in alphabetical languages with deep orthography whereas the
non-lexical route (assembled routine) predominates in languages with shallow

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

81

orthography. In Hebrew, for instance, experiments by Frost (1994) in naming and
lexical decision tasks (LDTs) conducted with normal subjects showed that subjects
behave differently in naming tasks depending on whether stimuli are written in deep
or shallow print. Larger frequency effects and semantic priming effects are observed
when the subjects are presented with deep orthography. Therefore, one may expect
deep dyslexic patients to produce more semantic errors when reading stimuli
presented in deep orthography.
To our knowledge, only two cases of acquired reading disorders have been
published on languages with two types of orthography. A first study was conducted
by El Alaoui-Faris et al. (1994) who reported the case of an Arabic-speaking patient
suffering from alexia without agraphia. The patient showed no difference whether
tested in shallow or deep orthography. However, given that the patient recovered
from her reading impairment despite the presence of left cerebral lesions, the authors
concluded to a bi-hemispheric participation in reading. Their interpretation was
based on studies of pure alexic patients (Coslett & Saffran, 1989) and on studies
in reading acquisition (Seymour & Elder, 1986), both suggesting a participation of
the right hemisphere in reading. According to El Alaoui-Faris et al., the participation
of the right hemisphere is maintained in adult Arabic readers because of the level of
complexity of the visuo-spatial activity involved in reading deep orthography of
Arabic. The second case was reported by Birnboim and Share (1995) who presented
a Hebrew-speaking surface dyslexic producing more reading errors in the deep than
in the shallow orthography, but this effect did not reach significance.
Studies on deep dyslexic patients speaking languages with shallow orthography
have presented opposing results. Ardila (1991) examined 41 Spanish-speaking aphasic patients who did not produce semantic errors. He concluded that deep dyslexia
never develops in speakers of languages with shallow orthography. This claim was,
however, refuted by the subsequent report of three deep dyslexic Spanish-speaking
patients (JMK in Ferreres & Miravalles, 1995; ON and MG in Ruiz, Ansaldo, &
Lecours, 1994), of one deep dyslexic Portuguese-speaking patient (MAC in
Delgado, 1998) and of one phonological dyslexic Spanish-speaking patient (AD
in Cuetos, Valle-Arroyo, & Suárez, 1996).
With respect to languages with deep orthography, extensive research has been
undertaken on English-speaking deep dyslexic patients, resulting in a distinction
between the input, the central and the output forms, depending on the lesion site, or
the route responsible for the production of semantic errors. In the input and central
forms, semantic errors arise from an impairment affecting respectively access to the
semantic system, and the semantic system itself. In the output form, semantic errors
arise from impairment in accessing phonology from semantics. Four cases of output
deep dyslexic English-speaking patients have been documented (JA in Katz &
Lanzonni, 1992; GR in Hildebrandt & Sokol, 1993; JC in Buchanan, Hildebrandt,
& MacKinnon, 1994; and LW in Newton & Barry, 1997). In three of these cases, the
authors aimed at demonstrating that implicit access to the sub-lexical phonology of
words (JA and GR) and nonwords (JC) is preserved in output deep dyslexia.
Surprisingly, very few cases of acquired reading disorders have been reported in
French-speaking patients. One case of phonological dyslexia was examined by

82

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Derouesné and Beauvois (1985). In a case of deep dyslexia reported by Lecours,
Lupien, and Bub (1990) analysis was limited to the morphological errors produced
by the patient. Finally, the literature contains two cases of deep dysphasia: Cardebat,
Puel, Démonet, and Nespoulous (1991), whose patient presented a mild deficit in
auditory lexical decision in the absence of a central semantic impairment as well as a
severe impairment of verbal short-term memory (STM); and Valdois, Carbonnel,
David, Rousset, and Pellat (1995) who presented an exhaustive case study of a
French-speaking patient with surface dyslexia, deep dysphasia and deep dysgraphia.
In repetition, their patient produced the same types of errors as found among deep
dyslexics in reading aloud.
To summarize, cases of deep dyslexia have been reported in languages with both
deep and shallow orthography as well as in languages with the two types of orthography. However, with respect to languages with shallow orthography and those
with the two types of orthography, the existing evidence is still largely insufficient
and does not provide a clear picture of whether and how the type of orthography may
influence deep dyslexia. Furthermore, the very few studies reporting on acquired or
developmental dyslexic bilingual subjects reveal conflicting results. In the two cases
reported, one notices that only one of the two languages is affected, and that the
language affected may be either the first or the second language.
The present study is divided into two parts. First, we localize the functional
lesions responsible for the reading impairments of the Arabic/French bilingual
patient by resorting to a three-route model of reading. In the second part, we propose
accommodations to the Plaut and Shallice (1993) connectionist model in order to
account for the characteristics of bilingual deep dyslexic patients. Then, we compare
the multi-impairment hypothesis and the connectionist approach with regards to
their ability to account for the deep dyslexia syndrome.
Before proceeding into a detailed investigation of this case, we first propose a
concise description of the most relevant linguistic features of Arabic and French. We
also emphasize the linguistic properties that are common to the two languages
spoken by the patient as well as their contrastive aspects.

2. Linguistic features of Arabic and French
2.1. The alphabets and the writing systems
Arabic and French have an Arabic and a Latin alphabetical system, respectively,
written on a horizontal axis. However, whereas French is written from left to right,
Arabic is written from right to left in a connected fashion. Furthermore, in contrast to
French writing, Arabic does not contrast upper- vs. lower-case or cursive vs. script
written forms. With respect to their phonemic inventories, the two languages share
14 consonants and three vowels.
One major feature of Arabic writing lies in the various context-determined
graphemic representations taken by each of the 28 letters making the Arabic alphabet. More specifically, each letter may have from two to four distinct allographs

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

83

Table 1
Examples of Arabic orthographic representation with and without vowel diacritics
With vowel diacritics

Without vowel diacritics
kataba

‘he wrote’

kutiba

‘it was written’

kutubun

‘books’

depending on its graphemic position in word-initial, -medial, -final, or in isolation
(e.g. the letter in
,
,
, and , respectively).
Arabic writing involves the use of a set of symbols called diacritics that appear on
top or below letters to represent short vowels, gemination, case, and silence. 1
However, diacritic symbols do not normally appear in written Arabic. They are
found in the Holy Koran, in children’s books, and in pedagogical materials used
in the first years of the reading and writing process. They progressively disappear
from written texts as the child starts mastering reading and writing. Thus, the Arabic
word is restricted to its consonants and long vowels. The absence of short vowel
information in Arabic usually results in the presence of homographs in the language
that are semantically and phonologically ambiguous. As shown in the examples
presented in Table 1, Arabic words without diacritic symbols may carry several
meanings and pronunciations, thus requiring the adult reader to rely on the context in
which these words appear. This is why diacritic symbols are provided in cases where
word ambiguity may distort understanding.
2.2. The morphological systems
In French the prototypical process of word formation is linear affixation, i.e.
words are made up of sequences of morpheme segments that are concatenated
together in a linear order as shown in the word trans-form-at-ion.
In contrast to French, a Semitic language like Arabic exhibits nonconcatenative
morphological processes. Simple words are commonly formed on the basis of a
discontinuous root of three or four consonants (e.g. /k-t-b/) between which sets of
vowels are inserted (e.g.
[katab] ‘he wrote’;
[kitaab] ‘a book’;
[kutub] ‘books’, etc.). In addition to this infixation process, suffixation and prefixation are also typical morphological operations (e.g. [maktab] ‘office’; [maktaba]
‘library’; [maktabat] ‘libraries’). Within recent linguistic accounts such as autosegmental approaches (McCarthy, 1981, 1982), the word
[katab] ‘he wrote’,
for instance, consists of three different morphemes: a discontinuous consonantal root
/k-t-b/ which specifies the semantic field of the family of related lexical items (‘the
act of writing’ in the above example), a discontinuous set of vowels /i-aa-/ which
1
Silence or sukuun, which takes a small circle as a diacritic mark, indicates the absence of a vowel
either in a CC sequence or in a word-final position.

84

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Fig. 1. Internal representation of the word [kitaab] ‘book’ in Arabic.

identifies the lexical entry and specifies its functional/grammatical categories (the
perfective in our example) and a prosodic template CVCVC which allows for
distinction between morphological patterns (e.g. for verbs, pattern 1 CVCVC
[katab] ‘wrote’, pattern 3 CVVCVC [kaatab] ‘corresponded’, etc.). As illustrated in Fig. 1, within this autosegmental approach, word formation involves the
mapping of units of the consonantal root and the vocalic morpheme with the
template.
Within morpheme-based linguistic models of the Arabic lexicon, lexical items are
morphologically related when they share a consonantal root carrying the same
meaning (e.g. the consonantal root /b-r-d/ carries the meaning of ‘cold’ in [barada]
‘to cool’ and [buruuda] ‘the cold’). Words sharing the same consonantal root but
whose meanings are different (e.g. /b-r-d/ in [bariid] ‘mail’ vs. [buruuda] ‘the
cold’) are claimed to appear in separate lexical entries and, hence, are not morphologically related (see Prunet, Béland, & Idrissi, 2000 for details of various views on
the Arabic lexicon).

3. Case history
ZT was born in 1962. A right-handed male, ZT is a native speaker of Lebanese
Arabic. He started studying Arabic and French at school at the age of 4 in Lebanon. In the school and university he attended, the scientific subjects were taught in
French whereas all other subjects were taught in Arabic. After he graduated, he
moved to Montreal, Canada in 1984. He then earned a Master’s degree in a
French-speaking university and worked as an engineer from 1985 to 1993 in
two different companies where French was used as a main language. ZT spoke
Arabic at home and with friends. Even though ZT demonstrates fluency in his two
languages, he considers Arabic as his dominant language. Therefore, in the present
study, we will refer to ZT’s two languages as L1 and L2 for Arabic and French,
respectively.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

85

3.1. Neurological assessment
At the age of 32, ZT suffered a cerebral vascular accident, which resulted in a
severe mixed aphasia, a right hemiplegia, a severe bucco-facial apraxia, a transient
right facial palsy and a transient homonymous lateral hemianopsia. Diagnosis
revealed thrombosis of the left middle cerebral artery. CT scan performed a few
days after his admission to the hospital showed an infarction of the left peri-sylvian
region. Another scan performed a year later revealed an evolution toward chronicity
of the massive infarct in the left peri-sylvian territory (see Fig. 2A). The frontotemporo-parietal hypodensity also included the insula cortex, the corresponding
corona radiata and the left temporal pole (see Fig. 2B).
3.2. Neuropsychological assessment
Neuropsychological testing conducted three weeks post-onset revealed no impairment in construction praxis nor memory problems in complex visual recall tasks.
Short- and long-term memory as well as judgment and reasoning capacities were
well preserved. On a psychological level, the patient, even though eager to understand his illness, showed signs of frustration, anger, and sadness when unable to
communicate.
At a subsequent neuropsychological evaluation performed 6 months post-onset,
ZT started using the written mode to communicate. He had become more aware of
his severe language disturbances and seemed to have accepted the fact that he would
not be able to return to his engineering job.
On the Mesulam test, ZT’s score was perfect, thus showing no attention disturbances. The results on the Raven matrices showed an IQ well above normal (110).
On the complex figure of Rey, the patient’s performance reflected difficulties in the
production of details.
With respect to verbal memory, ZT showed well-preserved capacities in learning
and in delayed recall of pairs of words from the Wechsler battery, but a poor
performance on non-semantically related pairs. The patient’s verbal STM was evaluated in French. His performance was then compared to that of three bilingual
(Arabic/French) controls matched for age, sex and education. Verbal STM capacity
was assessed using span tasks. Different types of stimuli were used: digits, monosyllabic words, letters and nonwords. For each stimulus type, lists of two to eight
items were created. Lists of increasing length were presented in both visual and
auditory modalities for immediate oral recall. The patient’s span was defined as the
longest item list recalled in correct order in at least half of the four trials. Results
indicated a reduced verbal span in both modalities. In the auditory modality, ZT
could recall only three digits (range for controls ¼ 5–8), two words (range for
controls ¼ 4–5), two letters (range for controls ¼ 5–6), and two nonwords (range
for controls ¼ 3–4). In the visual modality, the patient recalled three digits (range
for controls ¼ 6–8), three words (all controls recalled four words), and three letters
(range for controls ¼ 4–6). To summarize, in French ZT showed a limited verbal
STM in both visual and auditory modalities.

86

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Fig. 2. (A,B) CT scan performed a year post-onset.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

87

3.3. Language assessment
Early language evaluation reported severe mixed aphasia. One year post-onset,
language assessment in Arabic and French showed mildly impaired oral comprehension. Oral production was found to be severely impaired with the exception of
repetition of simple words in which the patient produced very few errors. Writing
to dictation for single letters, words, and nonwords was severely impaired both in
Arabic and French. Testing in reading aloud showed a large proportion of semantic
errors in both languages, a central feature of the deep dyslexia syndrome. A comprehensive study was then undertaken to investigate whether the patient would produce
the other symptoms characterizing deep dyslexia, and whether these symptoms
would be present in both languages mastered premorbidly by the patient.

4. Methodology
Native speakers of the two languages tested the patient. Evaluation was conducted
in parallel in the two languages, with only one language used in each testing session.
To ensure that ZT was assessed in the two languages at the same period of his
evolution, tasks in the two languages were administered in alternation. In Arabic,
we tried – in as much as the language permits – to control for dialect varieties and
hence, to select stimuli that show uniformity across Standard Arabic and Lebanese,
the patient’s dialect. When relevant, ZT was tested in both Arabic orthographies,
shallow and deep. In such experimental settings, two versions of the same set of
stimuli were designed, one with vowel diacritics (w
w) and one without (wo
wo), which
were administered to the patient in alternation and at different times throughout
testing.

5. Localization of ZT’s functional lesions within a three-route model
We took as a basis for our investigation the functional three-route model for word
recognition proposed by Buchanan and Besner (1993). We then brought to this
model the two following modifications. First, we combined the input and output
phonological lexicons into a single phonological lexicon. A single phonological
lexicon presupposes that phonological representations are neither acoustic nor
articulatory, but rather abstract representations (see Allport, 1984; Béland & Paradis, 1997). Second, we added route J for the repetition of nonwords. The revised
version of this model is illustrated in Fig. 3.
As in Buchanan and Besner (1993), capital letters have been used to identify the
different pathways between the different subcomponents of the model. This model
comprises three routes for reading: (1) a semantic lexical route B-E-F-D; (2) a nonsemantic direct lexical route B-C-D; and (3) a non-lexical route A. It also comprises
three routes for repetition: (1) a lexical semantic route I-H-F-D; (2) a non-semantic
direct lexical route I-D; and (3) a non-lexical route J.

88

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Fig. 3. Functional three-route model for word recognition adapted from Buchanan and Besner (1993).

5.1. Input processing
In deep dyslexia, visual and acoustic analyses (see Fig. 3) are assumed to be
intact. It is crucial, however, to assess their integrity before conducting any further
investigation.
5.1.1. Visual discrimination
The Arabic stimuli set comprised 80 identical and 80 different pairs of mono- and
bi-syllabic words and nonwords. Diacritic symbols were not presented in this task.
The patient was instructed to respond orally yes when the stimuli of a pair were
identical and no when they were not. When non-identical, members of a pair differed
by one consonant located either in the initial, medial or final position. The nonidentical pairs were divided into four subsets: (1) 20 pairs involving graphemically
close but phonologically distant consonants; (2) 20 pairs involving graphemically
distant but phonologically close consonants; (3) 20 pairs involving graphemically
and phonologically close consonants; and (4) 20 pairs involving graphemically and

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

89

phonologically distant consonants. The four subsets were created in order to control
for the potential influence of phonological similarity in the patient’s performance. If
phonological similarity had any influence on the patient’s judgment, a lower score in
subsets (2) and (3) than in subsets (1) and (4) is to be expected. ZT’s performance on
this test was almost flawless (158/160).
The discrimination test in French comprised 160 pairs (80 identical and 80
different) of monosyllabic CV and V words and nonwords presented in lowercase letters. When non-identical, the members of a pair differed by one consonant
located either in the initial or final position. As in the Arabic testing, the 80 different
pairs were divided into four subsets. ZT obtained a score of 159/160. The results
show that ZT’s ability in written discrimination is almost flawless in both Arabic and
French, an indication that the visual analysis component is unimpaired.
5.1.2. Auditory discrimination
The stimuli were presented auditorily to the patient who was instructed to answer
yes when the two members of a pair were identical and no when they were not.
In Arabic, the stimuli set consisted of 90 pairs (45 identical and 45 non-identical)
of monosyllabic words and nonwords. When non-identical, members of a pair
differed by only one phoneme in the word-initial or word-medial position. The
phonological differences between the two phonemes involved one of the following
feature contrasts for consonants: voiced, nasal, and pharyngeal, and one of the
feature contrasts for vowels: long, high, and round. ZT’s score on this test was
almost perfect (88/90).
The French stimuli set comprised 36 pairs (18 identical and 18 non-identical) of
mono- and bi-syllabic words differing by one consonantal phoneme in either the
word-initial or word-final position. ZT’s performance on this test was close to
perfect (35/36).
5.2. Access to the orthographic input lexicon
One of the characteristics of deep dyslexic patients is that their performance is
surprisingly good in a LDT given the degree of their word reading impairment
(Coltheart, 1980a; Patterson, 1979). Therefore, a visual LDT was designed and
administered to the patient to test his abilities to distinguish words from nonwords
via visual input.
Version w presented stimuli written with vowel diacritics, whereas in version wo
the stimuli were without. Stimuli were printed and randomly administered to the
patient without time constraints. He was instructed to respond orally yes if he
recognized the stimulus as a word and no if he did not. In the Arabic version w,
the stimuli set was made up of 80 mono-, bi- and tri-syllabic words (n ¼ 40) and
nonwords (n ¼ 40). Nonwords were of two types: (1) nonwords created by changing
one to two phonemes in real words of the language; (2) nonwords that were not
created from words but which respected both the orthographic and phonological
constraints of the Arabic language without containing a real consonantal root.
ZT scored 69/80 (86%) producing nine errors on nonwords and two on words. The

90

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

performance of three Lebanese control subjects who were Arabic/French bilinguals
and who were matched for age, educational level and level of bilingualism was 79/
80 (99%), 75/80 (94%), and 78/80 (98%) for each of the subjects.
In version wo,
wo the stimuli set was the same as in version w. ZT’s score was 71/80
(89%). The control subjects scored 79/80 (99%), 78/80 (98%), and 76/80 (95%).
In French, the stimuli consisted of 80 mono-, bi- and tri-syllabic words (n ¼ 40)
and nonwords (n ¼ 40). The nonwords were created by replacing one consonant of a
word by one consonant whose sound correspondence is phonologically close. ZT
obtained a score of 63/80 (79%). Three Arabic/French bilingual control subjects
matched for age and educational level scored 72/80 (90%), 73/80 (91%), and 71/80
(89%).
A difference in the performance was observed between Arabic and French in ZT
as well as in the controls, most probably reflecting the predominance of Arabic over
French in all the subjects. However, ZT’s performance was lower than that of the
control subjects in both Arabic and French, an indication that access to the orthographic input lexicon is partially impaired.
5.3. Access to the phonological lexicon
An auditory LDT was designed for each language. In Arabic, the stimuli set
consisted of 48 mono-, bi- and tri-syllabic words and 48 nonwords. Nonwords
were created by replacing one consonant of a word by a phonologically close
consonant which shared all feature values but one. The feature contrasts involved
were voiced, nasal, emphatic, anterior, continuant, and guttural. ZT’s score was 83/
96 (86%). The control subjects scored 86/96 (90%), 82/96 (85%), and 75/96 (78%).
In French, the stimuli set was the same as the one used in the visual modality. ZT
scored 65/80 (81%). The control subjects scored 63/80 (79%), 70/80 (88%), and 70/
80 (88%).
5.4. Reading aloud
Several sets of stimuli (letters, syllables, words and nonwords) were designed in
Arabic and French to test the patient’s performance in reading aloud. Furthermore,
when appropriate, Arabic tests included two lists of stimuli, one with vowel diacritics and one without.
5.4.1. Reading aloud of the alphabet
The patient was requested to read aloud the printed 28 letters of the Arabic
alphabet that were presented to him in a randomized fashion. ZT produced 15 errors;
four consisted of words beginning with the target letter (e.g. !
[qird]
‘monkey’), two contained the target letter (e.g. !
‘Sahara’), six were
substitutions (e.g.
! ), two were attempts to produce the target letter (e.g.
! [em]), and one remained without response.
With respect to the 26 letters of the French alphabet which were presented to him
in a random order and printed in upper case, ZT produced ten errors, eight of which
consisted of words beginning with the target letter (e.g. G ! [gã] gant ‘glove’;

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

91

N ! nous [nu] ‘we’), or containing the target letter (e.g. X ! [taksi] taxi ‘cab’).
The two other errors were attempts to produce the letter sound (C ! [k, s]), and a
semantic association (W ! [lOkOmOtiv] locomotive ‘locomotive’ possibly
because the letter W is close to the logo of a train transportation company, or
because it starts the word wagon ‘wagon’).
5.4.2. Reading aloud of syllables
In Arabic, it was difficult to construct a large set of syllables that would not
correspond either to a real word or to a letter name of the alphabet. Therefore,
syllable reading aloud was tested only in French.
The French stimuli set was made up of 69 monosyllabic CV and V words and
nonwords printed in upper case (e.g. QUE, KA, VU, LO). The patient obtained a
score of 34/69 (49%). Many of these errors were comparable to the ones produced in
reading aloud of single letters, i.e. ZT’s erroneous responses corresponded to French
words beginning with the target initial letter sequence (e.g. CU ! [kŁizin] cuisine
‘kitchen’). Other errors consisted in incorrect grapheme-to-phoneme conversion
(e.g. AU [o] was read [u]).
5.4.3. Reading aloud of concrete and abstract words
In Arabic, the stimuli set was made up of a total of 108 words; 54 were concrete
and 54 were abstract. The stimuli were presented without vowel diacritics. The list is
an adaptation of the one used in French (see below) since there exist no reliable
frequency table and no reference for the degree of concreteness of words in Arabic.
As indicated in Table 2, ZT’s error rate is higher on abstract (42=54 ¼ 78%) than on
concrete (31=54 ¼ 57%) words. The difference was significant (x 2 with continuity
correction ¼ 4:22, P , 0:05). 2
The French stimuli set was taken from Gagnon (1988) and Valdois et al. (1995). It
consisted of 60 abstract and 60 concrete words matched for lexical frequency and
length in the number of letters and phonemes. The 120-word list was presented to 34
undergraduate students who were requested to rate the words for concreteness on a 1
to 5 scale. The mean rating was 2.25 (SD 0.46) for abstract and 4.79 (SD 0.28) for
concrete words; the difference between two word types was significant
(tð118Þ ¼ 236:47, P , 0:01). The patient produced more errors on abstract
(44=60 ¼ 73%) than on concrete words (26=60 ¼ 43%). The difference was significant (x2 ¼ 9:90, P , 0:01).
5.4.4. Reading aloud of function words
The Arabic stimuli set comprised 35 open-class items (nouns) and 35 closed-class
items (conjunctions, prepositions and pronouns) presented without diacritics. The
patient produced 19 errors (54%) on open-class items and 33 errors (94%) on closedclass items (see Table 2). The difference in the proportion of errors was significant
(x2 ¼ 12:63, P , 0:001).
2

All x 2 values are reported with continuity correction.

92

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Table 2
Number and percentage of errors for different word categories in Arabic (w
w and wo)
wo and French in reading
aloud
Arabic

Concreteness effect
Concrete (wo
wo)
Abstract (wo
wo)
Function word effect
Open class items (wo
wo)
Closed class items (wo
wo)
Grammatical class effect
Nouns (wo
wo)
Verbs (wo
wo)
Adjectives (wo
wo)
Nouns (w
w)
Verbs (w
w)
Adjectives (w
w)

French

No.

% errors

No.

% errors

31/54
42/54

57
78

26/60
44/60

43
73

19/35
33/35

54
94

18/30
30/30

60
100

117/166
60/64
36/56
109/158
60/64
35/56

70
94
64
69
94
62

23/40
30/40
23/40

57
75
57

The French stimuli set was taken from Gardye, Béland, and Nespoulous (1990). It
comprised 30 open-class and 30 closed-class items matched for lexical frequency
and length in number of phonemes and letters. ZT produced an error rate of 60% (18/
30) on open-class items and of 100% (30/30) on closed-class items. The difference in
the proportion of errors was significant (x2 ¼ 12:60, P , 0:001).

5.4.5. Reading aloud of nouns, verbs, and adjectives
In Arabic, the stimuli set, made up of nouns, verbs, and adjectives both with and
without vowel diacritics, was presented to the patient. The distribution of ZT’s
reading errors for the three grammatical classes reported in Table 2 shows that in
stimuli wo the error rate varies with the grammatical class of the stimuli in the
following way: error rate on verbs (60=64 ¼ 94%) . error rate on nouns
(117=166 ¼ 70%) . error rate on adjectives (36=56 ¼ 64%); the difference in the
percentage of errors is significant between verbs and nouns (x2 ¼ 12:82,
P , 0:001), between verbs and adjectives (x2 ¼ 14:42, P , 0:001) but not between
adjectives and nouns. A similar pattern of errors is observed in stimuli w: more
errors on verbs (60=64 ¼ 94%) than on nouns (109=158 ¼ 69%) (x2 ¼ 14:03,
P , 0:001), more errors on verbs (94%) than on adjectives (35=56 ¼ 62%)
(x2 ¼ 15:84, P , 0:001), and no significant difference in error rate between
nouns and adjectives.
The French stimuli set was made up of 40 verbs, 40 nouns and 40 adjectives
matched for lexical frequency and length in number of syllables. Analysis revealed
no significant grammatical class effect, but as can be seen in Table 2, the error rate
distribution shows a tendency towards nouns and adjectives being read better than
verbs.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

93

5.4.6. Reading aloud of nonwords
Two lists of 63 and 22 Arabic nonwords, respectively without and with diacritics,
were presented to the patient. Nonwords were created by changing one to two
phonemes in real words and respected the phonotactic constraints of the language.
ZT failed to produce any correct responses in reading aloud nonword stimuli
presented with vowel diacritics. He could read only three nonwords without diacritics.
In French, a set of 84 nonwords controlled for phonological complexity, and length
in number of letters, syllables and phonemes (see Béland, Bois, Seron, & Damien,
1999) was presented to the patient. He produced 74 errors for an error rate of 88%.
5.4.7. Distribution of error types in reading aloud
An analysis of the errors produced on words and nonwords was conducted. All
stimuli used in the reading aloud tests and those used in early language assessment
were included in this analysis. As indicated in Tables 3 and 4, 16 error types were
identified: circumlocution, gestural, morphological, morphological-then-translation,
no response, other, phonological, semantic, semantic-then-translation, translation,
visual/phonological, visual/phonological-and-semantic, visual/phonological (root),
visual/phonological-then-morphological, visual/phonological-then-semantic, and
visual/phonological-then-translation. Examples and definitions of error types are
presented in Appendix A.
Table 3
Correct responses and error distribution in word reading aloud in Arabic wo and w and in French
Response type

Arabic wo (n ¼ 610)

Arabic w (n ¼ 277)

French (n ¼ 553)

No.

%

No.

%

No.

%

0.82
35.90
0
13.28
0.49
10.82
3.28
5.90
10.16
3.93
4.75
9.51
0

2
74
0
56
1
27
11
21
28
14
11
31
0

0.72
26.71
0
20.22
0.36
9.75
3.97
7.58
10.11
5.05
3.97
11.19
0

15
200
5
24
1
29
10
96
86
1
8
55
9

2.71
36.17
0.90
4.34
0.18
5.24
1.81
17.36
15.55
0.18
1.45
9.95
1.63

0.33
0

0
0

0
0

0
1

0
0.18

0

1

0.36

11

1.99

0.82

0

0

2

0.36

Circumlocution
5
Correct
219
Gestural
0
Morphological
81
Morphological-then-translation
3
No response
66
Other
20
Phonological
36
Semantic
62
Semantic-then-translation
24
Translation
29
Visual/phonological
58
Visual/phonological-and0
semantic
Visual/phonological (root)
2
Visual/phonological-then0
morphological
Visual/phonological-then0
semantic
Visual/phonological-then5
translation

94

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Table 4
Correct responses and error distribution in nonword reading aloud in Arabic wo and w and in French
Response type

Correct
No response
Other
Phonological
Visual/phonological
Visual/
phonological-andsemantic
Visual/
phonologicalthen-morphological
Visual/phonologicalthen-semantic
Visual/phonologicalthen-translation

Arabic wo (n ¼ 63)

Arabic w (n ¼ 22)

French (n ¼ 84)

No.

%

No.

%

No.

%

3
32
3
6
16
0

4.74
50.56
4.74
9.48
25.28
0

0
15
0
2
4
0

0
68.18
0
9.09
18.18
0

10
11
3
34
22
1

11.90
13.09
3.57
40.47
26.18
1.19

0

0

0

0

1

1.19

2

3.16

1

4.54

2

2.38

1

1.58

0

0

0

0

It should be pointed out here that most of the error types mentioned above have
been reported as characterizing deep dyslexia except for the translation type which
has been observed in only one deep dyslexic bilingual patient (Byng et al., 1984).
The percentage of correct responses in Arabic, reported in Table 3, is significantly
higher for words presented without than with vowel diacritics (x2 ¼ 6:85, P , 0:01).
5.4.8. Summary
As has been suspected in ZT’s early language testing, his performance in reading
aloud displays the features of the deep dyslexia syndrome in both languages. Results
obtained in the various reading tasks revealed that ZT’s reading aloud is impaired at
the sub-lexical (letters and syllables) and the lexical (words) levels. Some of the
lexical effects that characterize deep dyslexic reading are found in both languages,
whereas others are present in one language only. The concreteness effect (concrete
words read better than abstract words), the function word effect (more errors on
function words) and the lexicality effect (reading of nonwords almost impossible)
are observed in both his languages. The grammatical class effect (nouns and adjectives read better than verbs) is found only in Arabic. 3
3

One of the reviewers drew our attention to the fact that nouns are read better than verbs in Arabic only.
This may represent the first documented case in which the grammatical class effect is restricted to a
specific language in a bilingual aphasic subject. We attribute ZT’s poor performance in Arabic verbs to the
heavy morphological load characterizing Arabic verbs; indeed, Arabic verbs are marked for aspect,
person, gender and number whereas adjectives and nouns carry only gender and number morphological
markers. Aphasic Arabic speakers have already been reported as experiencing difficulties with verbs,
which they tend to simplify through the process of affix omission (Mimouni & Jarema, 1997).

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

95

Moreover, the results show a better performance in reading Arabic wo.
wo The higher
error rate in Arabic w may be due to the low frequency effect – Arabic speakers are
trained to read in deep orthography at an early age (about 9), thus developing the
addressed routine, which might be less vulnerable to language impairment.
As indicated in Table 3, the error types characterizing deep dyslexia are observed
in the two Arabic orthographies. The pattern of error distribution is also similar in
both orthographies. More specifically, semantic errors occur with the same proportion in both orthographies, thus refuting the prediction that more semantic errors
should occur when reading deep orthography.
5.5. Semantic processing
Three different sources have been reported as being responsible for semantic
errors in deep dyslexia, namely, impairment affecting access to the semantic system
from visual print (input form), impairment to the semantic system itself (central
form), or impaired access to phonology from the semantic system (output form). The
input form of deep dyslexia is characterized by a higher proportion of visual over
semantic errors and a better comprehension of spoken than written words; the
central form is characterized by a higher proportion of semantic over visual errors
and a poor comprehension of both spoken and written words; finally, in the output
form, the proportion of semantic errors is higher than that of visual errors whereas
comprehension is good for spoken and written words.
To further investigate the presumption that the patient presented the output form
of deep dyslexia, the patient’s semantic processing was assessed in both languages
by resorting to written and auditory word–picture matching tasks and a task testing
his morpho-semantic judgment from visual input.
5.5.1. Auditory word–picture matching
The patient was tested with an Arabic translation of the auditory word–picture
matching test from the Caplan and Bub (1990) Psycholinguistic Assessment of
Language (PAL) battery comprising 60 stimuli. In this task, the patient had to
choose one of two pictures as a match to the spoken word. His score was almost
perfect (59/60). Furthermore, the patient’s performance in the auditory word–picture
matching of the Paradis (1991) Bilingual Aphasia Test (BAT) battery was flawless
(18/18).
In French, an adaptation of the PAL was used. The patient produced only one
error (score ¼ 59=60).
5.5.2. Written word–picture matching
The list of stimuli used in this task was a written adaptation of the PAL battery.
The patient had to choose one of two pictures that matched the written word
presented to him. He scored 50/54 (93%) and 55/60 (92%) in Arabic and French,
respectively.

96

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

5.5.3. Morpho-semantic judgment
As indicated in Table 3, ZT produced a high proportion of morphological errors.
This proportion is even higher in Arabic. Given that in both Arabic and French,
morphologically related words are also semantically related (refer to linguistic
description above), it is impossible to know whether morphological errors are triggered by the semantic or the morphological similarity of words. There are also, in
both languages, words that are visually and phonologically related without being
morphologically related. To assess ZT’s ability to discriminate between morphologically related words and visually/phonologically related pairs, a morpho-semantic
test was designed and administered. The patient was presented with three words: two
were morphologically and semantically related whereas the third one was visually/
phonologically related only (e.g. in Arabic:
[barada] ‘to cool’;
[buruuda] ‘the cold’;
[bariid] ‘mail’; in French: greffer ‘to transplant’,
greffon ‘transplanted organ’, and greffier ‘legal clerk’). The patient was requested
to identify the word that was not morphologically related. The stimuli consisted of
24 triads in Arabic presented without diacritics and 72 triads in French. The patient’s
score was 15/24 (62.5%) and 59/72 (82%) in Arabic and French, respectively. The
performance of three Arabic/French bilingual control subjects and of six Frenchspeaking control subjects matched for age and educational level was almost flawless
on this test. In Arabic, the three bilingual controls produced two, four and three
errors for a mean performance of 21/24 (87.5%). In French, four of the controls
produced no errors and two produced three errors for a mean score of 71/72 (98.6%).
5.5.4. Summary
The present results show that ZT’s semantic processing is mildly impaired. From
ZT’s performance in word–picture matching, one may conclude that the semantic
errors in reading result from an impairment in the retrieval of the phonological
representation subsequent to correct activation of the concept in the semantic
system, pointing to an output form of deep dyslexia.
5.6. Sub-lexical processing
A characteristic of deep dyslexic patients is their inability to derive phonology
from print sub-lexically. French has a deep writing system whereas Arabic has, to
some extent, a less deep writing system since consonantal graphemes have one and
only one sound correspondence, a feature that rules out the possibility of constructing a rhyme judgment task in the language. Word endings of rhyming words being
necessarily spelled the same way in Arabic, the derivation of phonology is not
needed to perform the task, and thus, a rhyming task in Arabic is no different
from a discrimination task. For this reason, rhyme judgment was tested only in
French.
5.6.1. Rhyme judgment from auditory input
The stimuli set comprised 116 pairs of words subdivided into four categories: (1)
word pairs with homographic non-rhyming endings (e.g. aquarium [akwariOm]

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

97

‘aquarium’ vs. parfum [parfœ̃] ‘perfume’); (2) word pairs with heterographic nonrhyming endings (e.g. gâteau [gato] ‘cake’ vs. tiroir [tirwar] ‘drawer’); (3)
word pairs with homographic rhyming endings (e.g. souper [supe] ‘supper’ vs.
laver [lave] ‘to wash’); and (4) word pairs with heterographic rhyming endings
(e.g. balai [bal1] ‘broom’ vs. sommet [sOm1] ‘summit’). Therefore, in this rhyming
test, half of the yes and half of the no responses corresponded to homographic
stimuli, whereas the other half corresponded to heterographic stimuli. The patient
was requested to answer yes if the word endings in the pairs rhymed and no if they
did not. If the patient’s judgment were to be based solely on word spelling, then the
patient should be 100% correct in categories (2) and (3) and 100% incorrect in
categories (1) and (4).
ZT obtained the following scores in each of the categories: (1) 21/29 (72%) (e.g.
aquarium [akwariOm]–parfum [parfœ̃]); (2) 27/29 (93%) (e.g. gâteau [gato]–
tiroir [tirwar]); (3) 21/29 (72%) (e.g. souper [supe]–laver [lave]); and (4) 16/
29 (55%) (e.g. balai [bal1]–sommet [sOm1]). Mean scores of the three bilingual
(Arabic/French) controls were (1) 28.33/29 (98%), (2) 28.65/29 (99%), (3) 24.45
(88%), and (4) 21.33/29 (73%).
5.6.2. Rhyme judgment from visual input
The stimuli set comprised 108 word pairs taken from the set of pairs used in the
auditory rhyme judgment task. The distribution of the number of pairs for each
category was the following: 26 pairs in category (1), 27 pairs in category (2), 29
pairs in category (3), and 26 pairs in category (4). The patient was requested to
answer yes if the word endings in the pairs rhymed and no if they did not.
ZT obtained the following scores: (1) 9/26 (35%); (2) 22/27 (81%); (3) 15/29
(52%); and (4) 3/26 (12%). The mean scores of the three bilingual (Arabic/French)
controls were (1) 22.11/26 (85%), (2) 27/27 (100%), (3) 28.03/29 (97%) and (4)
18.22/26 (70%).
5.6.3. Summary
ZT’s overall performance in rhyme judgment was significantly worse in the visual
input modality (49=108 ¼ 45%) than in the auditory modality (85=116 ¼ 73%),
suggesting an impaired sub-lexical access to phonology from orthography. This
could result from his impairment in verbal working memory. Note that his performance in rhyme judgment from both auditory and visual input is worse than that of
the controls even though it shows a similar trend with the control subjects (highest
score on category (2) and lowest score on category (4)). However, the discrepancy
observed in the results across categories, which was also observed in bilingual
control subjects, is an indication that orthography has an effect on performance in
auditory as well as visual rhyme judgment.
Results from reading aloud and visual rhyme judgment tasks indicate an impaired
access to phonology from orthography at both lexical and sub-lexical levels. The
question that arises now is whether access to phonology from the semantic system is
impaired (pathway F-D in Fig. 3). A picture-naming task was designed to this effect.

98

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Table 5
Correct responses and error distribution in oral naming of 100 picture stimuli in Arabic and in French
Response type

Arabic (n ¼ 100)

French (n ¼ 100)

Correct
No response
Circumlocution
Gestural
Morphological
Other
Phonological
Semantic
Semantic-and-phonological
Semantic-then-translation
Translation

34
1
1
0
2
6
1
22
0
11
22

62
0
0
1
0
2
2
28
1
2
2

5.7. Oral picture naming
A set of 100 pictures was used in this task. As illustrated in Table 5, the patient
made semantic errors in both languages, an indication that pathway F is impaired.
5.8. Repetition
Semantic errors produced in reading and those produced in picture naming seem
to result from an impaired retrieval of the phonological representation subsequent to
a correct activation of the concept in the semantic system. However, the presence of
phonological errors observed in both reading and picture naming is an indication
that there might be impairment to the phonemic output buffer. To further investigate
this possibility, repetition and delayed repetition tasks were designed.
5.8.1. Immediate repetition
ZT was tested in immediate repetition for the concreteness effect (concrete vs.
abstract words), the function word effect (function words vs. open class items), the
grammatical class effect (verbs, nouns, and adjectives) and the lexicality effect
(words vs. nonwords). The results summarized in Table 6 show that ZT’s error
rate is low in both Arabic (21=396 ¼ 5:3%) and French (15=240 ¼ 6:25%) in
word repetition; the difference is not significant. The errors collected in Arabic
and French were either morphological, phonological or phonological/lexical.
The results show no effect of concreteness, function word or grammatical class in
any of the two languages. However, a significant lexicality effect was observed in
both languages. The percentage of errors was significantly higher on nonwords than
on words in both languages (x2 ¼ 36:42, P , 0:001 for Arabic; x2 ¼ 8:98, P ,
0:01 for French).
According to the model in Fig. 3, this discrepancy is accounted for by refreshment
of the phonological trace through pathway D-G for words. The presence of permanent representations for words in the phonological lexicon, which are not available

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

99

Table 6
Number and percentage of errors for different word categories in Arabic and French in repetition
Arabic

Concreteness effect
Concrete
Abstract
Function word effect
Open class items
Closed class items
Grammatical class effect
Nouns
Verbs
Adjectives
Lexicality effect
Words
Nonwords

French

No.

% errors

No.

% errors

0/24
1/24

0
4

1/30
5/30

3
17

0/35
0/35

0
0

1/30
2/30

3
7

13/160
5/74
2/44

8
7
4

1/40
3/40
2/40

3
8
5

13/160
29/67

8
43

26/139
23/57

19
40

for nonwords, contributes to the phonological output processing by allowing a
refreshment of the acoustic input trace through pathway G. A refreshment of the
trace through pathway D-G should be even more important in a delayed repetition
task, and consequently, the resulting phonological activation could propagate to the
semantic system through pathway H. Since the semantic output pathway F is
impaired, semantic errors are expected to occur in delayed repetition. Furthermore,
an increase in the delay should produce an increase of semantic errors.

5.8.2. Delayed repetition
Two sets of 120 and 124 word stimuli were used in Arabic and French, respectively. The patient was requested to repeat a word only after he had finished counting
silently from one to five in the 5 s repetition delay and from one to ten in the 10 s
repetition delay. In Arabic, ZT produced 36 errors out of 120 stimuli (30%) in the 5 s
delay and 26 errors (22%) in the 10 s delay. In French, ZT produced 27 errors out of
124 stimuli (22%) in the 5 s delay and 24 (19%) in the 10 s delay. The difference in
scores between the two delays is not significant in either of the languages. The error
distribution for both delays is presented in Figs. 4 and 5.
Semantic errors were produced in both languages in the two tasks. The increased
delay results in a lower rate of semantic errors in Arabic (Fig. 4) and in a higher rate
in French (Fig. 5). In contrast, phonological errors increase in Arabic but decrease in
French. Although these differences are not significant, mainly because of the low
number of errors, it is worth pointing out that there is a trend toward a double
dissociation. According to predictions by Knott, Patterson, and Hodges (1997) on
DTA patients with semantic dementia, the increase of semantic errors in French
following the increase in the delay would be an indication of a deficit in phonological access. On the other hand, the increase of phonological errors following the

100

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Fig. 4. Error distribution in delayed repetition in Arabic for short (5 s) and long (10 s) delays.

increase in the delay may be a result of trace degradation. An interpretation of this
double dissociation is presented later within the connectionist approach.
5.8.3. Summary
The presence of semantic errors in reading aloud, delayed repetition and oral
picture naming in the presence of a partially impaired semantic system suggests
an impairment in accessing the phonological lexicon from the semantic system
(pathway F in Fig. 6). Errors in auditory and visual LDTs indicate impairment to
pathways B and I.
ZT’s preserved comprehension of the words he read correctly and of those he read
with semantic errors is an indication that the patient is not relying on pathway C for
reading, a route generally assumed to be damaged in deep dyslexic readers. Errors in
nonword reading and in nonword repetition are an indication of impairment to the
phonemic output buffer. The above five lesions are indicated in Fig. 6 by large Xs.

Fig. 5. Error distribution in delayed repetition in French for short (5 s) and long (10 s) delays.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

101

Fig. 6. The functional lesions typically found in output deep dyslexic patients.

The shaded area in the semantic system represents the partial impairment
observed in abstract word reading and morpho-semantic judgment.
The lesions in pathways C and F and the lesion to the phonemic buffer correspond
to the typical lesions found in output deep dyslexic English-speaking patients (JA in
Katz & Lanzonni, 1992; GR in Hildebrandt & Sokol, 1993; JC in Buchanan et al.,
1994; and LW in Newton & Barry, 1997). In three of the four reported output
dyslexic cases, authors aimed to demonstrate that implicit (automatic) access to
the sub-lexical phonology of words (JA and CR) and nonwords (JC) is preserved
in output deep dyslexia. Buchanan et al. (1994) reported that their patient JC showed
significant phonological priming effects in a visual LDT. The patient’s reaction
times (RTs) were faster when the word stimuli were preceded by a homophonous
nonword than when they were preceded by a non-homophonous nonword. For
instance, JC showed a priming effect, that is, shorter RTs when ‘chair’ was preceded
by the pseudohomophone “taybul” than when the same target was preceded by the
orthographic control “tarble”. The authors interpreted the priming effect they

102

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

obtained as an indication that assembled phonological recoding through pathway A
may not be abolished in output deep dyslexia.
In the case of our patient, performance in nonword reading aloud was very poor,
indicating that explicit phonological recoding was not possible. However, ZT’s
implicit assembled phonological recoding from visual input may be preserved. A
well-suited task to investigate an implicit access to nonword phonological recoding
is a visual LDT within a cross-language phonological priming paradigm.
5.9. Cross-language phonological priming
As mentioned earlier, Arabic and French have highly contrastive writing systems
but their phonemic inventories share 14 consonants and three vowels. It was, therefore, possible to construct cross-language quasi-homophonous pairs (the phonemes
coming from different phoneme inventories, the pairs are not totally homophonous),
that is, pairs in which a nonword-prime and a word-target are written in different
languages and show phonological similarity without orthographic similarity. A
priming effect, that is, shorter RTs in homophonous pairs than in non-homophonous
pairs, could be obtained only if there is a phonological recoding of the visual
nonword-prime.
Two visual LDTs within a cross-language phonological priming paradigm were
designed. In LDT1, targets were written in French and primes in Arabic wo.
wo In
LDT2, targets were written in Arabic wo and primes in French. In both LDTs, the
priming effect was measured by comparing RTs in phonologically related pairs
(cross-language homophonous pairs: e.g. “calbée” –
‘dog’) with RTs in phonologically unrelated pairs (pairs in which primes from related pairs were replaced by
non-homophonous nonwords: e.g. “beurli” –
‘dog’). Related and unrelated pairs
were included in two separate testing conditions (experimental and control, respectively), and administered in two separate sessions within a 1 month interval. Stimuli
were presented wo except in three words where the vowel diacritics were provided
to avoid semantic confusion with homographic counterparts.
5.9.1. LDT1: Arabic nonword-prime/French word-target
5.9.1.1. Design and material. In the experimental condition, the stimuli set
consisted of 108 pairs in which the prime was an Arabic word or nonword and
the target was a French word or nonword. The 36 related pairs consisted of homophonous items (Arabic nonword/French word:
[S1z]–chaise [S1z] ‘chair’). The
remaining 72 pairs were Arabic/French non-homophonous fillers (18 nonword/
nonword pairs, 18 word/nonword pairs, and 36 word/word pairs). In the 108 trials,
72 corresponded to yes responses and 36 to no responses. Among the 72 yes
responses, 36 were preceded by a word-prime and 36 by a nonword-prime.
Among the 36 no responses, 18 were preceded by a nonword-prime and 18 by a
word-prime. Thus, the lexical decision (yes–no) could not be based on the lexical
status of the prime, since half of the nonword-primes were followed by a word-target
and half were followed by a nonword-target.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

103

In the control condition, the 36 related pairs were replaced by 36 unrelated pairs in
which the prime was a non-homophonous nonword. All other pairs remained the
same.
5.9.1.2. Procedure. The experiments were run on a computer using the Psychlab
software package (Bub & Gum, 1988). Stimuli were presented in the middle of the
display screen. For each trial, primes appeared for 500 ms. They were replaced by
the targets, which remained on the screen until the subject gave a response. The
stimuli interval was 50 ms. The patient was instructed to be fast and to press the right
shift key whenever he recognized a word and to press the left shift key otherwise. A
training session consisting of 12 trials, which were not part of the stimuli sets,
preceded the testing sessions.
5.9.1.3. Results. RTs for correct responses in the related and unrelated pairs appear
in Table 7. Three standard deviations on either side of the mean were removed prior
to analysis, and only responses that were correct in both the experimental and the
control conditions were considered. This resulted in the loss of nine pairs. An
ANOVA taking into account pair types (four different types) and testing
conditions (experimental vs. control) reveals a significant testing condition effect
(Fð1; 85Þ ¼ 6:53, P , 0:05) and a significant interaction between testing condition
and pair type (Fð3; 85Þ ¼ 7:20, P , 0:001). Decomposition of the interaction
reveals a significant testing condition effect in the word/nonword filler type only
(Fð1; 85Þ ¼ 18:25, P , 0:001). Mean RTs for filler types are significantly longer in
the control testing condition (mean RTs ¼ 1131 ms) than in the experimental testing
condition (mean RTs ¼ 899 ms). RTs for correct yes responses to nonword/word
pairs were entered in a paired t-test. As indicated in Table 7, ZT showed no
significant priming effect. The difference in mean RTs between related and
unrelated pairs was not significant (tð26Þ ¼ 0:88, P ¼ 0:18). These results do not
show implicit assembled phonological recoding (through pathway A in Fig. 6) for
Arabic nonwords, but they do not rule out the possibility of an implicit access for
French nonwords.
5.9.2. LDT2: French nonword-prime/Arabic word-target
5.9.2.1. Design and material. In the experimental condition, the stimuli set
consisted of 129 pairs in which the prime was a French word or nonword and the
target was an Arabic word or nonword. Thirty-four related pairs consisted of homoTable 7
Results of cross-language phonological priming in LDT1 and LDT2

Related pairs
Unrelated pairs
P value

LDT1 mean RTs (ms)

LDT2 mean RTs (ms)

1047
1014
. 0.05

1034
1198
, 0.05

104

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

phonous experimental items (French nonword/Arabic word: “calbée” [kalbe]/
[kalbe] ‘dog’). The remaining 95 pairs were French/Arabic non-homophonous
fillers (29 nonword/nonword pairs, 35 word/nonword pairs, and 31 word/word
pairs). In the 129 trials, 65 corresponded to yes responses and 64 to no responses.
Among the 65 yes responses, 34 were preceded by a word-prime and 31 by a
nonword-prime. Among the 64 no responses, 29 were preceded by a nonwordprime and 35 by a word-prime. Thus, the lexical decision (yes–no) could not be
based on the lexical status of the prime, since half of the nonword-primes were
followed by a word-target and half were followed by a nonword-target. In the
control condition, the 34 related pairs were replaced by 34 unrelated pairs. All
other pairs remained the same.
5.9.2.2. Procedure. The same procedure as in LDT1 above was followed.
5.9.2.3. Results. RTs for correct responses in the experimental and the control trials
appear in Table 7. Three standard deviations on either side of the mean were
removed prior to analysis, and only responses that were correct in both the
experimental and the control conditions were considered. This resulted in no loss
of responses. An ANOVA examining the effects of pair types (four different types)
and testing conditions (experimental vs. control) on RTs reveals a significant
interaction between testing condition and pair type (Fð3; 125Þ ¼ 5:46, P , 0:001).
Decomposition of the interaction reveals a significant testing condition effect for the
word/nonword filler type (Fð1; 125Þ ¼ 11:6, P , 0:001) and the nonword/word pair
type (Fð1; 125Þ ¼ 4:74, P , 0:05). Mean RTs for word/nonword fillers are shorter
(mean RTs ¼ 1029 ms) in the control than in the experimental condition (mean
RTs ¼ 1282 ms). RTs for correct yes responses to nonword/word pairs were entered
in a paired t-test. As indicated in Table 7, mean RTs in related pairs were shorter
than mean RTs in unrelated pairs (tð33Þ ¼ 22:20, P ¼ 0:017).
An absence of priming in LDT1 shows that there is no implicit assembled phonological recoding for Arabic nonwords. This may be a result of the difficulty ZT has in
establishing spelling-to-sound correspondence, or the failure of the phonological
activation in producing a priming effect. To account for the priming effect observed
between a visual French nonword and a visual homophonous Arabic word in LDT2,
given the functional lesions indicated in Fig. 6, two interpretations may be put
forward.
(1) ZT first assembles the phonology of the written French nonword through
pathway A (see Fig. 6). The resulting phonological activation via pathway G is
sufficient to activate the phonological representation of the homophonous Arabic
word, which implies that the two phonological representations are in some way
linked. Then, the semantic representation of the Arabic word is activated through
pathway H. Finally, the written Arabic target appears and is read via the semantic
pathway B-E. Pre-activation in the semantic system is responsible for reduced RTs.
(2) ZT first assembles the phonology of the written French nonword through
pathway A. The resulting phonological activation via pathway G is sufficient enough
to activate the phonological representation of the homophonous Arabic word, which

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

105

in turn activates the semantic representation of the Arabic word through pathway H,
then up to the orthographic lexicon via pathway K. When the written Arabic target
appears, it is read through mildly impaired pathway B. Pre-activation in the orthographic lexicon is responsible for reduced RTs.
5.10. Interpretation of ZT’s impairments within a three-route model
As reported in Table 3, ZT’s performance in reading shows most of the features of
deep dyslexia. Furthermore, 13 of the 16 error types were found in both languages.
Within the multi-impairment approach, ZT’s overall linguistic deficit can be
accounted for by positing two sets of lesions, each affecting separate monolingual
architectures.
A set of six lesions for each language is required to explain all of the impairments
observed. First, ZT’s various errors in reading, naming and delayed repetition could
be explained as resulting from lesions to the phonological output component itself,
to the activation of the phonological lexicon by the semantic system and to the nonsemantic lexical route. The concreteness effect in reading and the low performance
in the morpho-semantic judgment task are indications that the semantic system itself
is partially impaired. Finally, lesions to pathways B and I could be postulated to
account for the presence of errors in visual and auditory lexical decisions.
Furthermore, given that the qualitative analysis showed similar deficits in the two
languages of the patient, one should conclude to similar lesions sites in the functional architectures corresponding to Arabic and French. Taken together, the lesions
responsible for ZT’s impairments in the two languages amount to 12, a large number
indeed.
However, even if the phonological priming obtained in LDT2 leads us to propose
links between the two phonological lexicons, and this proposal reduces the number
of lesions responsible for deep dyslexia in bilingual subjects, this number would still
range between six and 12, depending on the quantity of shared components and
pathways among the two monolingual architectures. We find this explanation somehow unattractive.
A further weakness of the multi-impairment approach lies in its inability to
account for the discrepancy observed in error distribution across languages. As
indicated in Tables 3 and 5, the semantic error rate is higher in French, whereas
the translation error rate is higher in Arabic. To explain the higher semantic error
rate in picture naming and reading aloud in French compared to Arabic, one should
postulate that the lesion along pathway F is more severe in French than in Arabic. On
the other hand, the higher translation error rate in Arabic as opposed to French
should be explained by positing that pathway F linking access to the French phonological lexicon from the semantic system is less impaired than the corresponding
pathway for Arabic. ZT’s tendency to translate from Arabic to French is indeed a
clear indication that the mapping between the semantics and the French phonology
along pathway F is unimpaired.
Furthermore, since there was no significant difference in error rate and in error
distribution between the two languages in the task of immediate word repetition, the

106

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

dissociation across languages cannot be accounted for by positing a lesion to the
components located downstream of pathway F, namely in the phonological lexicon
itself or in the transfer to the phonemic buffer.
To conclude, the multi-impairment approach to bilingual deep dyslexia is unparsimonious and does not provide an explanation for the dissociation found across
languages in error distribution.
An alternative to the three-route model is the connectionist approach to deep
dyslexia. Plaut and Shallice (1993) have conducted simulations of deep dyslexia,
which have resulted in quantitative evaluations of error distribution, overall performance and error types according to lesion sites and lesion severity. In the following
section, we will attempt an interpretation of ZT’s reading impairment within the
Plaut and Shallice (1993) model, which we will refer to as the P & S model. We will
first propose accommodations to this model to account for bilingual deep dyslexic
patients and then compare ZT’s error distribution to that obtained by Plaut and
Shallice after lesioning the 40–60 network.

6. Interpretation of ZT’s impairment within a connectionist model
6.1. Accommodations to the Plaut and Shallice (1993) model to account for
translinguistic errors
As briefly stated, Plaut and Shallice (1993) accounted for the deep dyslexia
syndrome by lesioning a connectionist model which maps orthography to phonology
through semantics. Lesions to different parts of this network result in the same
overall syndrome. At the same time, aspects of the syndrome differ between lesion
sites. Therefore, this model might be the appropriate tool to account for the overall
syndrome of output deep dyslexia observed in both ZT’s languages, as well as to
explain the discrepancies observed across tasks and languages. Before comparing
ZT’s error pattern to that obtained by Plaut and Shallice after lesioning the 40–60
network, we propose extensions to their network in order to accommodate for deep
dyslexia in bilingual subjects.
An important concept from the Hinton and Shallice (1991) model that Plaut and
Shallice (1993) and Plaut (1997) reintroduced is that of the attractor, i.e. the final
stable pattern of activity resulting from interactions in the network between input
units and network units when processing a word, or as defined by Plaut (1997, p. 2),
“In processing an input unit, units interact until the network as a whole settles into a
stable pattern of activity – termed an attractor – corresponding to its interpretation
of the input.” The co-occurrence of visual and semantic errors in deep dyslexia is the
result of a damaged network that builds attractors in mapping two arbitrarily related
domains. As reproduced in Fig. 7, in Hinton and Shallice’s network the visual error
(CAT ! COT) is explained as follows: the words CAT and COT which are visually
similar build similar initial large basins of attraction in the semantics.
As indicated by the arrows in the semantics component, the effect of the learning
procedure is to pull the two distinct meanings apart. In a damaged network, the

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

107

Fig. 7. Illustration of the visual error CAT ! COT. The solid ovals depict the normal basins of attraction;
the dotted ones depict a basin after semantic damage. Plaut & Shallice (1993), Deep dyslexia: a case study
of connectionist neuropsychology. Cognitive Neuropsychology, 10 (5) p. 393. Reprinted by permission of
Psychology Press Ltd., Hove, UK.

original attractor basins (represented by solid ovals in Fig. 7) are distorted and
enlarged (dotted basin in Fig. 7) such that the semantic pattern of CAT is captured
into the basin of the visually similar word COT. Thus, within this approach, visual
and semantic errors co-occur because the layout of attractor basins is sensitive to
both visual and semantic similarity (Plaut & Shallice, 1993, p. 394). However, the
layout of attractor basins as developed in Hinton and Shallice (1991) and in Plaut
and Shallice (1993) does not account for bilingual subjects and hence, requires some
accommodations.
In both Arabic and French, the links between orthography and semantics are arbitrary. In addition, the two languages contain words that are visually similar but
semantically unrelated (e.g. French: matelas ‘mattress’ vs. matelot ‘sailor’; Arabic:
‘mail’ vs.
‘cold’) as well as words that are semantically related but visually
unrelated (French: rose ‘rose’ vs. fleur ‘flower’; Arabic:
‘rose’ vs.
‘flower’).
In French, words that are phonologically similar are not necessarily visually
similar (e.g. biais ‘bias’ vs. billet ‘ticket’). In Arabic, given the high regularity in
spelling-to-sound correspondence, words that are visually similar are phonologically similar as well (e.g. Arabic
‘rumor’ and
‘a watch’). Consequently,
it is impossible to determine whether the layout of attractor basins is sensitive to
visual and semantic similarity or to phonological and semantic similarity. Since the
Hinton and Shallice (1991) model does not include the mapping from semantics to
phonology, the accommodations we are proposing are limited to the links between
orthography and semantics (Fig. 7).
Our proposal is that the building of attractor basins for Arabic/French bilingual
subjects proceeds as proposed by Hinton and Shallice (1991) except that the training
procedure involves the building in semantics of attractor basins linked to two ortho-

108

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

graphy and two phonology components. The attractor basins in French represent a
disjunctive set of the Arabic attractor basins since words that are visually or phonologically similar in French (e.g. salade ‘salad’–sandale ‘sandal’) are not mandatorily visually or phonologically similar in Arabic (e.g.
‘salad’; ‘sandal’ ), and
as a matter of fact, an Arabic word cannot be visually similar to a French word since
the two languages have different alphabets.
The link between the two languages lies in the semantics: words that are semantically related in Arabic are also semantically related in French. This extension of
attractor basins for bilingual subjects allows us to account for a number of translinguistic errors produced by ZT. We distinguish four types of translinguistic errors: (1)
translation (the patient’s response is given in the other language rather than in the
language of testing); (2) visual-then-translation; (3) semantic-then-translation; and
(4) translation-then-visual/phonological (see Appendix A).
6.1.1. Translation errors
Translation errors are the most easily explained errors within the bilingual
network we are proposing. As illustrated in Fig. 8, a translation error (Arabic
‘glasses’ read as French lunettes ‘glasses’) occurs when the semantic representation of a word is erroneously captured into the attractor basins of the other
language. Thus, such errors, which occur because the two languages build attractor
basins that are semantically related, should be equally found in the two languages of
an early bilingual patient. Moreover, they should occur in any language task involving the semantic component.
6.1.2. Visual/phonological-then-translation errors
As illustrated in Fig. 9, the visual/phonological-then-translation type of errors,
such as French portillon ‘gate’ read as
‘butterfly’ in Arabic, involves first a

Fig. 8. Illustration of a translation error in reading aloud: Arabic
‘glasses’.

‘glasses’ read as French lunettes

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

109

Fig. 9. Illustration of a visual/phonological-then-translation error in reading aloud: French portillon ‘gate’
read as
‘butterfly’ in Arabic.

visual similarity in French between portillon and papillon ‘butterfly’. As in the
CAT–COT example, the word portillon is erroneously captured into the attractor
basin of papillon because of the enlargement of the attractor basin. From there, the
semantic representation of the word is erroneously captured into the visual/phonological representation of the Arabic word for papillon.
6.1.3. Semantic-then-translation errors
The semantic-then-translation error type (e.g. French éponge ‘sponge’ read as
‘sea’ in Arabic) illustrated in Fig. 10 is the semantic counterpart of the visual/
phonological-then-translation error. The word éponge in French is erroneously
captured into the attractor basin of French mer ‘sea’, which is, in turn, erroneously
captured into the visual/phonological representation of that word in Arabic.
6.1.4. Translation-then-visual/phonological errors
Translation-then-visual/phonological errors are, probably, the most difficult ones
to detect in that they involve the search for visual/phonological similarity in the
other language. Fig. 11 illustrates a hypothetical example of this type of error
predicted to occur in bilingual Arabic/French deep dyslexic patients. In this example, French rumeur ‘rumor’ is read as French montre ‘watch’. To account for this
error one should posit first that the word rumeur is correctly assigned to its correct
meaning. From then, it is erroneously captured into the attractor basin of the Arabic
equivalent
‘rumor’, which is phonologically and visually similar to the Arabic
word
‘a watch’. Then, the correct semantic representation of the word ‘watch’
in Arabic is, as in simple translation errors, associated to the phonological representation of that word in French resulting in the production of the French montre
‘watch’. The closest example found in our corpus corresponding to this type of error
is French photo ‘photo’ read as French montre ‘watch’; this example does not fit

110

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Fig. 10. Illustration of a semantic-then-translation error in reading aloud: French éponge ‘sponge’ read as
‘sea’ in Arabic.

quite well into the present category since the Arabic word
corresponding to
photo is phonologically but not visually similar to the Arabic equivalent
for
montre. We predict that this error type should logically occur in deep dyslexic
bilingual patients. For instance, in a French/English bilingual patient, we expect
errors such as the French word sourd read as mort given that the two words are close
in English (‘deaf’ vs. ‘death’) and the other way around, the English word “bread”
read as “pine” because the corresponding words pain and pin are close in French.
6.1.5. Summary
The above accommodations to the P & S model allow us to account for the

Fig. 11. Illustration of a hypothetical example of a translation-then-visual/phonological error in reading
aloud: French rumeur ‘rumor’ read as French montre ‘watch’.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

111

occurrence of ZT’s translinguistic errors. Let us now turn to ZT’s overall performance in French and Arabic and compare it to that obtained by Plaut and Shallice
(1993) after lesioning the 40–60 output network. Since the accommodations
proposed above have not yet been implemented and no simulation of the extended
network has been run, the following interpretation of the consequences on reading
performance after damage to such a bilingual network is purely hypothetical.
6.2. Lesion severity and lesion sites
ZT’s error distribution in reading aloud for word stimuli is given in Table 3. The
percentage of correct responses is very similar in Arabic wo (219=610 ¼ 35:90%)
and French (200=553 ¼ 36:17%). In the P & S model, the output form of deep
dyslexia results from lesions to the direct pathway of the output network (S ! Ip,
Ip ! P), more specifically, from lesions affecting the S ! Ip and the Ip ! P connections, i.e. the connections between the sememe units (S) and the intermediate units
(Ip) in the output pathway to phoneme units, and the connections from these intermediate units to the phoneme units (P) (see Fig. 13). According to the simulations
conducted (Plaut & Shallice, 1993, p. 418), a lesion with a severity of 0.3 affecting
either or both output pathways S ! Ip and Ip ! P results in an overall correct
performance between 30 and 40%, which is comparable to that observed in ZT’s
two languages. However, to determine whether pathway S ! Ip, Ip ! P or both is
lesioned in French and Arabic, ZT’s error distribution in reading aloud must be
compared to that obtained by Plaut and Shallice after selectively lesioning pathways
S ! Ip and Ip ! P. According to Plaut and Shallice (1993, p. 420) a higher semantic
error rate is associated with lesions affecting the S ! Ip connections, whereas a
lower semantic error rate is attributed to lesions affecting the Ip ! P connections.
ZT’s error distribution in reading word stimuli aloud in Arabic wo and in French is
illustrated in Fig. 12. As shown, visual/phonological error rates are similar in French
(55=553 ¼ 9:95%) and Arabic wo (58=610 ¼ 9:51%) (x2 ¼ 0:23, P . 0:05). The
semantic error rate (see Table 3) is higher in French (86=553 ¼ 15:55%) than in
Arabic wo (62=610 ¼ 10:16%); the difference is significant (x2 ¼ 7:10, P , 0:01).
It is thus possible, within a connectionist approach, to account for the difference in
error distribution in the patient’s two languages by positing lesions affecting different connections of the 40–60 output network. Lesions to the output pathways S ! Ip
and Ip ! P account for the comparable visual/phonological error rate observed in
the two languages. Lesions affecting mostly Ip ! P connections may be responsible
for a lower semantic error rate in Arabic whereas lesions affecting the S ! Ip
connections may explain the higher semantic error rate in French (illustrated in
Fig. 13 with dotted arrows).
The morphological error rate (see Table 3) is higher in Arabic wo
(81=610 ¼ 13:28%) than in French (24=553 ¼ 4:34%) and the difference is significant (x2 ¼ 27:14, P , 0:001). According to Plaut and Shallice (1993), morphological errors are a sub-variety of visual/phonological-and-semantic errors. In ZT’s
French error corpus, both morphological errors (e.g. goûter ‘to taste’ read as goût
‘taste’) and visual/phonological-and-semantic errors (e.g. canari ‘canary’ read as

112

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Fig. 12. Error distribution in reading aloud in Arabic wo (n ¼ 610) and in French (n ¼ 553). Error types
with error rates $10% in either language are plotted.

canard ‘duck’) are found. In Arabic, only morphological errors are found. This may
be accounted for by the difference in the morphological structure of the two
languages. In Arabic, given the root-based structure of the lexicon, words that are
visually, phonologically and semantically related are more likely to be morphologically related as well. According to this interpretation, ZT’s morphological error
rate is higher in Arabic because the links between morphophonology and semantics
are stronger in Arabic than in French.

Fig. 13. Representation of the bilingual network. Dotted arrows indicate the lesioned connections: O ! I
in Arabic and French, S ! Ip in French and Ip ! P in Arabic.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

113

6.3. Concreteness effect
As was reported earlier, when reading aloud ZT produced significantly fewer
errors in concrete than in abstract words in both languages. According to Plaut
and Shallice (1993), a concreteness effect results from lesions affecting the input
pathway of reading (O ! I or I ! S). Because concrete words have more semantic
features, their reading is less influenced by damage to the input pathway. Conversely, because abstract words have fewer semantic features, their reading is more
likely to rely on this damaged direct pathway, thus resulting in a higher error rate.
According to Plaut and Shallice (1993, Table 11, p. 457) (the relevant information
is reproduced in our Table 8), the difference in error rate between concrete and
abstract words varies as a function of both lesion severity and lesion location along
the direct O ! I and I ! S pathways. ZT’s performance on concrete and abstract
words is shown in Table 2. In Arabic wo,
wo the percentage of correct responses is
42.59% (23/54) on concrete words and 22.22% (12/54) on abstract words, a difference of 20.37 percentage points in favor of concrete words. In French, the percentage of correct responses is 56.66% (34/60) on concrete words and 26.66% (16/60)
on abstract words, a difference of 30 percentage points in favor of concrete words. If
we look at Plaut and Shallice’s data on lesion location and lesion severity affecting
performance on concrete and abstract words (see Table 8), ZT’s reading performance on abstract and concrete words corresponds to a lesion severity affecting the
O ! I connections of 0.25 for Arabic and of 0.20 for French (lesioned O ! I
connections are indicated by dotted arrows in Fig. 13).
With respect to error distribution, the model predicts more visual than semantic
errors on abstract words but more semantic than visual errors on concrete words. In
Plaut and Shallice (1993, p. 459) simulations, visual errors are higher for abstract
than concrete words; conversely, the proportion of semantic errors is higher for
concrete than abstract words. As reported in Fig. 14, ZT’s error distribution in
Arabic follows Plaut and Shallice’s prediction. Although the observed differences
are not significant, the patient produced more visual (12=42 ¼ 29%) than semantic
Table 8
Correct performance for concrete and abstract words for each lesion location as a function of lesion
severity. Plaut & Shallice (1993) Deep dyslexia: a case study of connectionist neuropsychology. Cognitive
Neuropychology 10 (5) p. 457. Reprinted by permission of Psychology Press Ltd., Hove, UK.
Lesion location

Word type

Lesion severity
0.05

0.10

0.15

0.20

0.25

0.30

0.40

O!I

Concrete
Abstract
Difference

88.6
69.0
19.6

75.0
50.8
24.2

67.1
40.0
27.1

52.7
25.4
27.3

44.2
21.3
22.9

38.3
16.1
22.2

23.0
10.4
12.6

I!S

Concrete
Abstract
Difference

75.1
53.9
21.2

54.8
26.6
28.2

38.2
16.4
21.8

28.2
10.0
18.2

19.9
6.1
13.8

14.1
3.0
11.1

6.3
1.2
5.1

114

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

errors (8=42 ¼ 19%) on abstract words and more semantic (8=42 ¼ 19%) than visual
errors (3=42 ¼ 7%) on concrete words.
However, as illustrated in Fig. 15, ZT showed a tendency to produce in French
more semantic (14=44 ¼ 32%) than visual errors (8=44 ¼ 18%) on abstract words
(x2 ¼ 1:51, P . 0:05), as well as more semantic (9=44 ¼ 20%) than visual errors
(2=44 ¼ 5%) on concrete words (x2 ¼ 3:74, P ¼ 0:053).
Given that the stimuli sets in the two languages refer to the same concepts (French
concrete and abstract words were translated into Arabic) a difference in error distribution between the two languages cannot be accounted for by a difference in their
degree of concreteness. A possible interpretation is that the higher semantic error
rate in reading both abstract and concrete words in French results from the lesions
affecting the S ! Ip connections which, as reported above, are more severe in
French than in Arabic. This interpretation is corroborated by the distribution of
visual/phonological-then-semantic errors. Examination of the error distribution in
Table 3 reveals that ZT produced 11 visual/phonological-then-semantic errors in
French but none in Arabic. Moreover, this error type in French was more frequent in
reading abstract than concrete words. According to Plaut and Shallice (1993),
visual-then-semantic errors result from confusion in the input pathway (O ! I)
followed by a semantic confusion in the output (S ! Ip) network. Given that the
reading of abstract words relies more on the direct pathway and that S ! Ip connections are more severely lesioned in French than in Arabic, visual/phonological-thensemantic errors are more likely to occur in French than in Arabic, and more likely to
occur on abstract than on concrete words.
6.4. Error distribution across tasks
As reported in the previous section, in both languages the semantic error rate was

Fig. 14. Number of correct responses and error distribution in reading aloud of concrete and abstract word
stimuli in Arabic wo.
wo Error types with a number of errors $5 in either stimulus type are plotted.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

115

Fig. 15. Number of correct responses and error distribution in reading aloud of concrete and abstract word
stimuli in French. Error types with a number of errors $5 in either stimulus type are plotted.

higher in oral picture naming (see Table 5; Arabic: 22% ¼ 22=100; French:
28% ¼ 28=100) than in oral reading (see Table 3; Arabic wo:
wo 10:16% ¼ 62=610;
Arabic w: 10:11% ¼ 28=277; French: 15:55% ¼ 86=553). These results may stem
from the use of distinctive stimuli in the two tasks. To eliminate this bias, we designed
an oral picture naming task and a reading task, which included identical stimuli. We
added written picture naming and writing to dictation, two other tasks in which output
deep dyslexics are expected to produce semantic errors. In the four tasks, sets of
stimuli composed of 55 and 60 items in Arabic and French, respectively, were administered to the patient. The sets were almost identical in that the Arabic set contained
55 of the 60 French items. The percentage of correct responses in Arabic was 51% (28/
55), 70% (39/55), 43% (24/55), and 45% (25/55) in oral picture naming, reading
aloud, written picture naming, and writing to dictation, respectively. In French, the
percentage of correct responses was 65% (39/60), 53% (32/60), 45% (27/60), and
63% (38/60), respectively. More interestingly, despite the variation in the performance across tasks and languages, semantic errors were found in all four tasks both
in Arabic and French. Moreover, as indicated in Figs. 16 and 17, the semantic error
rate shows an identical rank ordering in both languages (oral picture naming .
written picture naming . reading aloud . writing to dictation).
A possible interpretation of these results is that the production of semantic errors
depends on the location of the initial activation: the closer the initial activation to the
lesioned S ! Ip connections, the more likely semantic errors are to occur. Initial
activation is located on the semantic layer in oral and written picture naming,
resulting in higher semantic error rates. Conversely, in reading and writing to dictation, the initial activation is on the orthographic and the phonological layers respectively, resulting in lower semantic error rates. This interpretation is congruent with

116

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Fig. 16. Error distribution across tasks for the 55 Arabic stimuli. Error types with a number of errors $5 in
any of the four tasks are plotted.

the principle that early lesions in a cascade are more detrimental than late lesions
(Sitton, Mozer, & Farah, 2000).
6.5. Delayed repetition
Results in delayed repetition revealed a trend toward a double dissociation in the
two languages: the increased delay resulted in a lower rate of semantic errors in
Arabic and a higher rate in French whereas phonological errors increased in Arabic
but decreased in French (see Figs. 4 and 5). Lesions affecting S ! Ip connections for

Fig. 17. Error distribution across tasks for the 60 French stimuli. Error types with a number of errors $5 in
any of the four tasks are plotted.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

117

Fig. 18. Distribution of translinguistic errors across tasks in Arabic w and wo,
wo and in French.

French and Ip ! P connections for Arabic were posited earlier to account for the
dissociation in the semantic error rate. The difference in the location of the lesioned
connections may account for the double dissociation in delayed repetition. In both
languages, an increase in the delay taxes the semantic system heavily. The higher
semantic error rate in the 10 s delayed repetition in French is a result of lesions to the
S ! Ip connections whereas in the 10 s delayed repetition in Arabic the higher
phonological error rate results from lesions to Ip ! P connections.

6.6. Distribution of translinguistic errors
Fig. 18 shows that translinguistic errors are much more frequent in Arabic than in
French. We assumed earlier that translation errors occur because the two languages
build attractor basins that are connected via the semantics. More specifically, translation errors produced in tasks involving an oral output occur when S ! Ip connections are first activated in Arabic, then Ip ! P connections are activated in French
rather than in Arabic.
Some psycholinguistic models for the bilingual mental lexicon claim that the
links between concepts and the lexicon may differ in L1 and L2. According to the
Kroll and Stewart (1994) model, the links from concepts to the lexicon are stronger
in L1 than in L2. Within a connectionist approach, this would entail stronger and/or
more numerous connections between semantics (concepts) and orthography and
between semantics and phonology in Arabic, ZT’s dominant language. Therefore,
if one assumes that the S ! Ip connections are stronger and/or more numerous in
Arabic than in French, one should expect translation errors to occur more frequently
when the patient is tested in Arabic (L1) than in French (L2). Translation errors may
therefore be construed as an unconscious adaptive strategy on the part of the patient

118

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

when experiencing difficulties in activating S ! Ip connections in French and
Ip ! P connections in Arabic. 4

6.7. Nonword reading in Arabic and in French
The 40–60 network in Plaut and Shallice (1993) does not account for nonword
reading since it lacks direct non-semantic orthography-to-phonology connections.
Consequently, it cannot account for the phonological priming effect found in the
LDT2 task. However, the orthography-to-phonology connections are present in
more recent connectionist models (Plaut, 1997; Plaut, Mc Clelland, Seidenberg,
& Patterson, 1996), which allow us to account for error pattern differences found
across languages in ZT’s reading of nonwords.
Nonword reading is slightly better in French (score of 10=84 ¼ 11:90%) than in
Arabic (score of 3=63 ¼ 4:74% on nonwords wo and of 0=22 ¼ 0% on nonwords w),
but the difference is not significant. However, error type distribution in the two
languages (see Table 4) shows an important discrepancy in the no response rate,
which is much higher in Arabic (wo
wo: 32=63 ¼ 51%; w: 15=22 ¼ 68%) than in
French (11=84 ¼ 13%). This no response rate in nonword reading is, in fact, significantly higher in Arabic (wo
wo: x2 ¼ 22:93, P , 0:001; w: x2 ¼ 25:68, P , 0:001)
than in French. The structure of the Arabic language may account for the observed
discrepancy. A former study conducted with the same patient (Prunet et al., 2000)
revealed that the error corpus contained a large number (n ¼ 119) of metathesis
errors consisting of root consonant reversals (e.g.
[qird] ‘monkey’ !
[qidr] ‘cooking pot’) in Arabic word stimuli. This type of metathesis error was
not, however, observed in French. An exhaustive linguistic analysis of metathesis in
slips of the tongue, language games and aphasic errors reported in Arabic and other
Semitic languages led the authors to claim that the Arabic consonantal root constitutes a privileged unit of access to the mental Arabic lexicon.
In light of this, the higher no response rate in Arabic nonwords may be explained by
the fact that the nonwords used in our experiments did not contain an existing consonantal root. Results of the cross-linguistic priming experiment are consistent with this
interpretation. A significant priming effect was found when a French nonword (e.g.
“calbée” [kalbe]) was followed by a homophonous Arabic word
[kalbe]). If
one assumes that phonological priming involves the mapping of the consonants /k-l-b/
making up the French nonword with the consonantal root in Arabic /k-l-b/, then when
an Arabic nonword
[S1z] homophonous to the French target word chaise was
used as a prime, even if root extraction may have occurred, priming could not be
obtained since the root does not constitute a lexical unit in French.

4

One of the reviewers suggested that translation errors could be produced by normal bilingual subjects
as well. In fact, ZT’s behavior is quite similar to what has been reported on bilingual speakers in codeswitching situations.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

119

7. Concluding remarks
Within the three-route model of reading used in the initial stage of this study,
bilingual output deep dyslexia is interpreted as resulting from a number of distinct
functional lesions disrupting the non-lexical route, the semantic lexical route and the
lexical route. Even though this model provides an account for the patient’s deficits in
each language taken separately, it does not offer a comprehensive theory for bilingual
output deep dyslexia. When attempting to account for the dissociation across
languages in the distribution of errors, the model leads us to a paradox: a lesion
must be posited on the pathway from the semantic system to the French phonological
lexicon to account for the higher semantic error rate in French, but the presence of this
lesion is incompatible with the higher translation error rate in Arabic. Furthermore,
the linear and modular aspects of the model preclude an interpretation of visual errors
in the processing of concrete vs. abstract words. Indeed, in such a model, on-line
processing is influenced by the features of the words that are represented at each level.
Visual errors are presumed to stem from an impairment in visual analysis, a process
that is influenced only by the visual properties of the stimuli and not by their semantic
features. Given that the concreteness feature is represented in the semantic system, it
should not have any influence on the visual error rate.
In contrast to the three-route model, which is limited in its predictive power, the
proposed bilingual connectionist network allows us to make predictions with respect
to error distribution in the two languages of bilingual deep dyslexic patients.
A first prediction refers to the language dominance. In the extensions to the network
that we have proposed, semantics must be common to both languages in order to
account for semantic-then-translation and visual/phonological-then-translation
errors. To explain both error types, semantics must constitute the bridge towards
the second language, an observation that is consistent with models of the bilingual
lexicon, which claim a common semantic system. The higher proportion of translation
errors in Arabic testing as opposed to French testing was construed as an adaptive
strategy by the patient when experiencing difficulties in activating S ! Ip connections in French and Ip ! P connections in Arabic. According to the Kroll and Stewart
(1994) model, the links between the concepts and the lexicon are stronger in L1 than in
L2. We posited that S ! Ip connections were more severely lesioned in French (L2)
than in Arabic (L1). Within the P & S model, this may be due to the fact that these
connections were either more numerous or stronger in Arabic (L1) than in French (L2)
premorbidly. Therefore, we predict that output deep dyslexic patients showing deep
dyslexia in both languages with a similar degree of severity are more likely to produce
translation errors when tested in L1 than when tested in L2.
A second prediction issuing from the accommodations proposed for bilingual
patients is that the form of translinguistic errors should vary according to the
languages spoken by the bilingual patients. Arabic and French have absolutely no
resemblance in their orthography, so the occurrence of errors triggered by visual
similarity between the words in the two languages is precluded. In bilingual deep
dyslexic patients speaking languages written with the same orthographic script (e.g.
French, English, Spanish), one should expect the same types of errors found in ZT’s

120

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

performance as well as translinguistic errors triggered by visual and phonological
similarity between the two orthographies. For example, one may expect the French
word an ‘year’ to be confused with the English article “an” or the French word botte
[bOt] ‘boot’ to be mixed up with the English word [bt] ‘but’.
To conclude, in this case study the superiority of the connectionist approach over
the multi-impairment approach is demonstrated, first through the capacity of the
computational properties of the network to account for the co-occurrence of visual/
phonological, semantic, visual and semantic errors, and second in the capacity of the
model in predicting error patterns in bilingual deep dyslexia. The presence of crosslanguage phonological priming obtained in the LDT2 is an indication that activation
of the phonology in the other language may occur without semantic activation.
Therefore, the semantic layer may not be the only bridge for connections between
the two languages of a bilingual subject.
Acknowledgements
This research was supported by the Fonds de la Recherche en Santé du Québec
(Grant # 952342) and by the Conseil de la Recherche en Sciences Humaines du
Canada (Grant # 410-92-0015). We are grateful to ZT for his cheerful and unconditional participation. Our thanks go to ZT’s family for their support, to Suzanne
Danault, Speech Therapist and Diane Martineau, Neuropsychologist for their assistance in various aspects of this study, and to Isabelle Peretz, Sylvie Belleville,
Martin Arguin, and Daniel Fiset, and the anonymous reviewers for suggestions
and useful comments on an earlier version of this paper. Special thanks to Ali Idrissi,
Geneviève Duchesne, and Soumeya Sadou for their help in the testing in Arabic and
in French and to Benoı̂t Gagnon for designing various figures.
Appendix A. Error type definitions
No response: after a reasonable delay, the patient produced no response.
Circumlocution: the patient produced a sentence to describe the word or the
picture stimulus;
e.g.
Arabic: to the stimulus
‘life’ presented in reading aloud, the patient’s response
was
‘woman, man’.
French: to the stimulus état ‘state’ presented in reading aloud, the patient’s
response was malade par exemple pour une semaine ‘sick for instance for a
week’.
Gestural: the patient used hand or finger gestures to describe the shape of an
object or to mimic the action corresponding to a word or picture stimulus;
e.g.
Arabic: to the stimulus
‘basket’ in reading aloud, the patient produced a hand
gesture corresponding to the shape of a basket.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

121

French: to the stimulus serpent ‘snake’ in reading aloud, the patient produced a
hand gesture showing a moving snake.
Morphological: the patient’s response was morphologically related to the stimulus when corresponding to an inflected or derived form of the stimulus. Morphological errors are by definition semantically related to the stimulus. Note that in
French, but not in Arabic, morphologically related words are not mandatorily
visually and phonologically related (e.g. bouche ‘mouth’ vs. buccal ‘oral’);
e.g.
Arabic: the patient’s response to the noun
‘stupidity’ in reading aloud was the
derived form
‘stupid’. The patient’s response to
‘cold’ in reading aloud
was the feminine form of the derived verb:
‘she is cold’.
French: the patient’s response to the verb goûter ‘to taste’ in reading aloud was
the derived form goût ‘taste’. The patient’s response to laisser ‘to let’ in reading
aloud was an inflected form of the same verb, laissa ‘he let’ past tense.
Morphological-then-translation: the patient’s response was a translation (in
Arabic in the French testing and in French in the Arabic testing) of a word that is
morphologically related to the stimulus. Note that the morphological part of the
error does not necessarily precede the translation part. Since there exists no way to
distinguish “morphological-then-translation” from “translation-then-morphological” errors, any error not given in the language of testing and not agreeing in gender,
number, tense or grammatical class with the stimulus was categorized as a morphological-then-translation error;
e.g.
Arabic: to the stimulus
‘shyness’ in reading aloud, the patient’s response was
timide ‘shy’.
French: to the stimulus enfant ‘child’ in reading aloud, the patient’s response was
‘children’.
Other: the patient’s response is a word in the language of testing which has no
semantic, morphological, visual or phonological relationship with the stimulus;
e.g.
Arabic: in reading aloud, the patient’s response to the stimulus
‘symbol’ was
‘voice’.
French: in reading aloud, the patient’s response to the stimulus vanité ‘vanity’ was
poignée ‘handle’.
Perseveration: the patient repeated a response produced earlier in the testing
session;
e.g.
Arabic: no perseveration errors were found in Arabic.
French: in the 5 s delayed repetition, the patient’s response to the stimulus théière
‘tea pot’ was désir ‘desire’, a stimulus given earlier in the testing.

122

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Phonological: the patient’s response was a nonword sharing phonological and/or
visual similarity with the stimulus;
e.g.
Arabic: in reading aloud, the patient’s response to the word
‘shop’ was the
nonword [manzar].
French: in reading aloud, the patient’s response to the word banalité [banalite]
‘banality’ was the nonword [bananli].
Phonological/lexical: the patient’s response in repetition was a word sharing
phonological and/or visual similarity with the stimulus;
e.g.
Arabic: in delayed repetition, the patient’s response to the word
‘term’ was the
word
‘gas’. French: in delayed repetition, the patient’s response to the word
assister [asiste] ‘to attend’ was the word assis [asi] ‘seated’.
Semantic: the patient’s response was a word semantically but not morphologically and thus, not visually and/or phonologically related to the stimulus;
e.g.
Arabic: in reading aloud, the patient’s response to the stimulus
‘roses’ was
‘flowers’.
French: in reading aloud, the patient’s response to the stimulus gain ‘earning’ was
argent ‘money’.
Semantic-and-phonological: the patient’s response was a nonword sharing
phonological and/or visual similarity with a word semantically related to the stimulus;
e.g.
Arabic: in written picture naming the patient’s response to the stimulus
‘newspaper’ was
, a nonword orthographically related to
‘news’,
which is semantically related to the stimulus.
French: in oral picture naming the patient’s response to the stimulus thermomètre
‘thermometer’ was [sãperatyr], a nonword phonologically related to température ‘temperature’, which is semantically related to the stimulus.
Semantic-then-translation: the patient’s response was a translation (in Arabic in
the French testing and in French in the Arabic testing) of a word that is semantically
related to the stimulus. Note that the semantic part of the error does not necessarily
precede the translation part. Since there exists no way to distinguish “semantic-thentranslation” from “translation-then-semantic” errors, any error not given in the
language of testing and semantically related to the stimulus was categorized as a
semantic-then-translation error;
e.g.
Arabic: in reading aloud, the patient’s response to the word
‘law’ was juge
‘judge’.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

123

French: in reading aloud, the patient’s response to the word éponge ‘sponge’ was
‘sea’.
Translation: the patient’s response was a translation of the stimulus in Arabic
into French testing and a translation into French in Arabic testing;
e.g.
Arabic: the patient’s response to the stimulus
‘Thursday’ was jeudi ‘Thursday’.
French: the patient’s response to the stimulus poisson ‘fish’ was
‘fish’.
Visual/phonological: the patient’s response was a word visually and/or phonologically but neither semantically nor morphologically related to the stimulus. At
least 50% of the letters (phonemes) of the response were the same and in the same
order as in the stimulus word;
e.g.
Arabic: in reading aloud, the patient’s response to the stimulus
‘to hate’ was
‘ball’.
French: in reading aloud, the patient’s response to the stimulus concept ‘concept’
was correct ‘correct’.
Visual/phonological (root): the patient’s response was a word visually and
phonologically but not semantically related to the stimulus, which contained the
same Arabic root;
e.g.
Arabic: in reading aloud, the patient’s response to the stimulus
‘to fill up’ was
‘dictation’.
Visual/phonological-and-semantic: the patient’s response was a word visually
and/or phonologically and semantically related to the stimulus;
e.g.
Arabic: no visual/phonological and semantic errors were found in Arabic.
French: in reading aloud, the patient’s response to the stimulus hangar ‘shed’ was
grange ‘barn’.
Visual/phonological-then-morphological: the patient’s response was a word
that is morphologically related to a bridge word, which is in turn visually and/or
phonologically related to the stimulus;
e.g.
Arabic: no visual/phonological-then-morphological errors were found in Arabic.
French: in reading aloud, the patient’s response to the stimulus bocal ‘jar’ was
bouche ‘mouth’, a word morphologically related to buccal ‘oral’, which in turn is
visually and phonologically close to the stimulus bocal.
Visual/phonological-then-semantic: the patient’s response was a word that is
semantically related to a bridge word, which is in turn visually and/or phonologi-

124

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

cally related to the stimulus;
e.g.
Arabic: in reading aloud, the patient’s response to the stimulus
‘preacher’
was
‘roof’, a word semantically related to the bridge word
‘ceiling’,
which is in turn visually and phonologically similar to
.
French: in reading aloud, the patient’s response to the stimulus lingot ‘ingot’ was
chemise ‘shirt’, a word semantically related to the bridge word linge ‘cloth’,
which is in turn visually and phonologically similar to lingot.
Visual/phonological-then-translation: the patient produced a translation (into
French in the Arabic testing and into Arabic in the French testing) of a bridge word,
which is visually and/or phonologically close to the stimulus;
e.g.
Arabic: in reading aloud, the patient’s response to the stimulus
‘fork’ was
chocolat ‘chocolate’ through the Arabic bridge word
‘chocolate’, a word
visually and phonologically close to the stimulus
‘fork’.
French: in reading aloud, the patient’s response to the stimulus portillon ‘gate’
was
‘butterfly’ through the French bridge word papillon ‘butterfly’, a word
visually and phonologically close to the stimulus portillon ‘gate’.
References
Allport, D. A. (1984). Speech production and comprehension: one lexicon or two? In W. Prinz & A. F.
Sanders (Eds.), Cognition and motor processes (pp. 209–228). Berlin: Springer-Verlag.
Allport, D. A., & Funnell, E. (1981). Components of the mental lexicon. Philosophical Transactions of the
Royal Society of London, B295, 397–410.
Ardila, A. (1991). Errors resembling semantic paralexias in Spanish-speaking aphasics. Brain and
Language, 41, 437–445.
Béland, R., Bois, M., Seron, X., & Damien, B. (1999). Phonological spelling in a DAT patient: the role of
the segmentation subsystem in the phoneme-to-grapheme conversion. Cognitive Neuropsychology,
16, 115–155.
Béland, R., & Paradis, C. (1997). Principled syllabic dissolution in a primary progressive aphasia case.
Aphasiology, 11 (12), 1171–1196.
Bentin, S., & Frost, R. (1987). Processing lexical ambiguity and visual word recognition in a deep
orthography. Memory and Cognition, 15 (1), 15–23.
Bentin, S., & Ibrahim, R. (1996). New evidence of phonological processing during visual word recognition: the case of Arabic. Journal of Experimental Psychology: Learning Memory and Cognition, 22
(2), 309–323.
Birnboim, S. L., & Share, D. L. (1995). Surface dyslexia in Hebrew: a case study. Cognitive Neuropsychology, 12 (8), 825–846.
Bub, D. N., & Gum, T. (1988). Psychlab software. Unpublished manuscript.
Buchanan, L., & Besner, D. (1993). Reading aloud: evidence for the use of a whole word nonsemantic
pathway. Canadian Journal of Experimental Psychology, 47 (2), 133–152.
Buchanan, L., Hildebrandt, N., & MacKinnon, G. E. (1994). Phonological processing of nonwords by a
deep dyslexic patient: a rowse is implicitly a rose. Journal of Neurolinguistics, 8 (3), 163–181.
Byng, S., Coltheart, M., Masterson, J., Prior, M., & Riddoch, J. (1984). Bilingual biscriptal deep dyslexia.
The Quarterly Journal of Experimental Psychology, 36A, 417–433.
Caplan, D., & Bub, D. N. (1990). Psycholinguistic assessment of aphasia. Paper presented at the meeting
of the American Speech and Hearing Association Conference, Seattle, WA.

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

125

Caramazza, A., & Hillis, A. E. (1990). Where do semantic errors come from? Cortex, 26, 95–122.
Cardebat, D., Puel, M., Démonet, J. F., & Nespoulous, J. L. (1991). Les différentes “boı̂tes” de la
répétition: analyse d’un cas de dysphasie profonde. Revue de Neuropsychologie, 1 (3), 215–232.
Coltheart, M. (1980a). Deep dyslexia: a review of the syndrome. In M. Coltheart, K. E. Patterson & J. C.
Marshall (Eds.), Deep dyslexia (pp. 22–48). London: Routledge & Kegan Paul.
Coltheart, M. (1980b). Deep dyslexia: a right-hemisphere hypothesis. In M. Coltheart, K. E. Patterson & J.
C. Marshall (Eds.), Deep dyslexia (pp. 326–380). London: Routledge & Kegan Paul.
Coltheart, M. (1983). The right hemisphere and disorders of reading. In A. Young (Ed.), Functions of the
right cerebral hemisphere (pp. 171–201). London: Academic Press.
Coltheart, M. (2000). Deep dyslexia is right-hemisphere reading. Brain and Language, 71, 299–309.
Coltheart, M., Patterson, K. E., & Marshall, J. C. (1987). Deep dyslexia since 1980. In M. Coltheart, K. E.
Patterson & J. C. Marshall (Eds.), Deep dyslexia (pp. 326–380). London: Routledge & Kegan Paul.
Coslett, H. B., & Saffran, E. M. (1989). Evidence for preserved reading in pure alexia. Brain, 112, 327–
359.
Cuetos, F., Valle-Arroyo, F., & Suárez, M-P. (1996). A case of phonological dyslexia in Spanish.
Cognitive Neuropsychology, 13 (1), 1–24.
Delgado, A. P. (1998). Implications de l’étude d’un cas de dyslexie profonde chez un sujet lusophone.
M.A. dissertation, Montreal University, Montreal.
Derouesné, J., & Beauvois, M. F. (1985). The “phonemic” stage in the non-lexical reading process:
evidence from a case of phonological alexia. In K. E. Patterson, J. C. Marshall & M. Coltheart
(Eds.), Surface dyslexia (pp. 399–457). Hillsdale, NJ: Laurence Erlbaum.
El Alaoui-Faris, F. M., Benbelaid, F., Alaoui, C. H., Tahiri, L., Jiddane, M., Amarti, A., & Chkili, T.
(1994). Alexie sans agraphie en langue arabe étude neurolinguistique et IRM. Revue Neurologique,
150 (11), 771–775.
Ferreres, A. R., & Miravalles, G. (1995). The production of semantic paralexias in a Spanish-speaking
aphasic. Brain and Language, 49, 153–172.
Frost, R. (1994). Prelexical and postlexical strategies in reading: evidence form a deep and a shallow
orthography. Journal of Experimental Psychology: Learning Memory and Cognition, 20 (1), 116–129.
Gagnon, J. (1988). Activation automatique et contrôlée du savoir lexico-sémantique chez les cérébrolésés
droits et gauches. Unpublished M.A. dissertation, Montreal University, Montreal.
Gardye, F., Béland, R., & Nespoulous, J. -L. (1990). Lexical access in agrammatism: lexical decision task
versus naming. Brain and Language, 39 (4), 593.
Glosser, G., & Friedman, R. B. (1990). The continuum of deep/phonological alexia. Cortex, 26, 343–359.
Hildebrandt, N., & Sokol, S. M. (1993). Implicit sublexical processing in an acquired dyslexic patient.
Reading and Writing: an Interdisciplinary Journal, 5, 43–68.
Hinton, G. E., & Shallice, T. (1991). Lesioning an attractor network: investigations of acquired dyslexia.
Psychological Review, 98 (1), 74–95.
Katz, L., & Feldman, L. B. (1983). Relation between pronunciation and recognition of printed words in
deep and shallow orthographies. Journal of Experimental Psychology: Learning, Memory and Cognition, 9 (1), 157–166.
Katz, L., & Frost, R. (1992). Reading in different orthographies: the orthographic depth hypothesis. In R.
Frost & L. Katz (Eds.), Orthography, phonology, morphology, and meaning (pp. 67–84). Amsterdam:
Elsevier.
Katz, R. B., & Lanzonni, S. M. (1992). Automatic activation of word phonology from print in deep
dyslexia. The Quarterly Journal of Experimental Psychology, 45A (4), 575–608.
Knott, R., Patterson, K., & Hodges, J. R. (1997). Lexical and semantic binding effects in short-term
memory: evidence from semantic dementia. Cognitive Neuropsychology, 14 (8), 1165–1216.
Kroll, J. F., & Stewart, E. (1994). Category interference in translation and picture naming: evidence for
asymmetric connections between bilingual memory representations. Journal of Memory and
Language, 33, 149–174.
Lecours, A. R., Lupien, S., & Bub, D. (1990). Semic extraction behavior in deep dyslexia: morphological
errors. In J. L. Nespoulous & P. Villiard (Eds.), Morphology, phonology and aphasia (pp. 60–71).
New York: Springer-Verlag.

126

R. Béland, Z. Mimouni / Cognition 82 (2001) 77–126

Marshall, J. C., & Newcombe, F. (1973). Patterns of paraplexia: a psycholinguistic approach. Journal of
Psycholinguistic Research, 2, 175–199.
McCarthy, J. J. (1981). A prosodic theory of nonconcatenative morphology. Linguistic Inquiry, 12, 373–
418.
McCarthy, J. J. (1982). Prosodic templates, morphemic templates, and morphemic tiers. In H. Van der
Hulst & N. Smith (Eds.), The structure of phonological representations (pp. 191–223). , 2. Dordrecht:
Foris.
Michel, F., Hénaff, M. A., & Intriligator, J. (1996). Two different readers in the same brain after a
posterior callosal lesion. NeuroReport, 7, 786–788.
Mimouni, Z., & Jarema, G. (1997). Agrammatic aphasia in Arabic. Aphasiology, 11 (2), 125–144.
Morton, J., & Patterson, K. (1980). A new attempt at an interpretation, or, an attempt at a new interpretation. In M. Coltheart, K. E. Patterson & J. C. Marshall (Eds.), Deep dyslexia (pp. 91–118). London:
Routledge & Kegan Paul.
Newton, P. K., & Barry, C. (1997). Concreteness effects in word production but not word comprehension
in deep dyslexia. Cognitive Neuropsychology, 14 (4), 481–509.
Paradis, M. (1991). Maghrebian version of the bilingual aphasia test, Mahwah, NJ: Lawrence Erlbaum.
Patterson, K. E. (1979). What is right with “deep” dyslexics? Brain and Language, 8, 111–129.
Patterson, K. E., & Besner, D. (1984). Is the right hemisphere literate? Cognitive Neuropsychology, 1 (4),
315–341.
Plaut, D. C. (1997). Structure and function in the lexical system: insights from distributed models of word
reading and lexical decision. Language and Cognitive Processes, 12 (5/6), 1–19.
Plaut, D. C., Mc Clelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and
impaired word reading: computational principles in quasi-regular domains. Psychological Review,
103 (10), 56–115.
Plaut, D. C., & Shallice, T. (1993). Deep dyslexia: a case study of connectionist neuropsychology.
Cognitive Neuropsychology, 10 (5), 377–500.
Price, C. J., Howard, D., Patterson, K., Warburton, E. A., Friston, K. J., & Frackowiak, R. S. J. (1998). A
functional neuroimaging description of two deep dyslexic patients. Journal of Cognitive
Neuroscience, 10 (3), 303–315.
Prunet, J. F., Béland, R., & Idrissi, A. (2000). The mental representation of Semitic words. Linguistic
Inquiry, 31 (4), 609–648.
Roeltgen, D. P. (1987). Loss of deep dyslexic reading ability from a second left hemisphere lesion.
Archives of Neurology, 44, 346–348.
Ruiz, A., Ansaldo, A. I., & Lecours, A. R. (1994). Two cases of deep dyslexia in unilingual hispanophone
aphasics. Brain and Language, 46, 245–256.
Saffran, E. M., Bogyo, L. C., Schwartz, M. F., & Marin, O. S. M. (1980). Does deep dyslexia reflect righthemisphere reading? In M. Coltheart, K. E. Patterson & J. C. Marshall (Eds.), Deep dyslexia (pp. 381–
406). London: Routledge & Kegan Paul.
Seymour, P. H. K., & Elder, L. (1986). Beginning reading without phonology. Cognitive Neuropsychology, 1, 315–341.
Shallice, T., & Warrington, E. (1980). Single and multiple component central dyslexic syndromes. In M.
Coltheart, K. E. Patterson & J. C. Marshall (Eds.), Deep dyslexia (pp. 119–145). London: Routledge &
Kegan Paul.
Sitton, M., Mozer, M. C., & Farah, M. (2000). Superadditive effects of multiple lesions in a connectionist
architecture: implications for the neuropsychology of optic aphasia. Psychological Review, 107, 709–
734.
Valdois, S., Carbonnel, S., David, D., Rousset, S., & Pellat, J. (1995). Confrontation of PDP models and
dual-route models through the analysis of a case of deep dysphasia. Cognitive Neuropsychology, 12
(7), 681–724.
Wydell, T. N., & Butterworth, B. (1999). A case study of an English-Japanese bilingual with monolingual
dyslexia. Cognition, 70, 273–305.