Letter Position Dyslexia

COGNITIVE NEUROPSY CHOLOGY, 2001, 18 (8), 673–696

LETTER POSITION DYSLEXIA Naama Friedmann Tel Aviv University, Israel

Aviah Gvion Loewenstein Rehabilitation Center, Ra’anana, Israel

Many word-reading models assume that the early stages of reading involve a separate process of letter position encoding. However, neuropsychological evidence for the existence and selectivity of this function has been rather indirect, coming mainly from position preservation in migrations between words in attentional dyslexia, and from nonselective reading deficits. No pure demonstration of selective impairment of letter position function has yet been made. In this paper two Hebrew-speaking acquired dyslexic patients with occipito-parietal lesions are presented who suffer from a highly selective deficit to letter position encoding. As a result of this deficit, they predominantly make errors of letter migration within words (such as reading “broad” for “board”) in a wide variety of tasks: oral reading, lexical decision, same-different decision, and letter location. The deficit is specific to orthographic material, and is manifested mainly in medial letter positions. The implications of the findings to models of reading and attention are discussed.

INTRODUCTION In recent years several distinct types of acquired dyslexia have been identified. These discoveries owe much to a fruitful interaction between the study of reading disorders and information processing models. The models of normal single word reading have been constructed and shaped by the identification of selective deficits, each indicative of failures in different parts and stages of the reading process (e.g., Coltheart, 1981; Patterson, 1981). On the other hand, new dyslexic patterns have been identified following predictions derived from information processing models and other, already known, types of dyslexia have become better understood through their use. Most of the components of these word-reading models have been found to have correlates in selective reading deficits. The later stages of reading, such as the orthographic

lexicon, the grapheme to phoneme converter, and the connections between them, have been found to be selectively impaired in various central dyslexias (Shallice & Warrington, 1980). However, for some of the theoretically postulated functions of the earlier stages of visual analysis, dyslexic correlates have not yet been identified. In this paper we present a new type of peripheral dyslexia that, though predicted by the model, has not previously been demonstrated clinically. This dyslexia is a result of a highly selective deficit to the visual analysis system. According to Ellis and Young (1988), the visual analyser has three distinct functions: 1. Letter identification. 2. Letter-to-word binding: allocation of letters to the word they belong to (or an attenuation filter that reduces input from words outside the appropriate window: Shallice, 1988).

Requests for reprints (and any other correspondence) should be addressed to Naama Friedmann, School of Education, Tel Aviv University, Tel Aviv 69978, Israel (http://www.tau.ac.il/~naamafr/. Email: [email protected]). We thank Uri Hadar, David Swinney, and Karalyn Patterson for their helpful comments and discussion. Ó 2001 Psychology Press Ltd http://www.tandf.co.uk/journals/pp/02643294.html

673

DOI:10.1080/02643290143000051

FRIEDMANN AND GVION

3. Encoding of letter position within a word (Ellis, 1993; Ellis, Flude, & Young, 1987), or its position relative to exterior letters (Humphreys, Evett, & Quinlan, 1990; Peressotti & Grainger, 1995). Each of these three functions is predicted to be susceptible to a selective deficit, causing completely different patterns of errors. When the first function is impaired, the result should be a failure to identify letters correctly. This has been shown to be the case in letter agnosia, which is a deficit located in the letter identification function of the visual analyser, in which patients fail to recognise letters, even in isolation. Another type of letter identification error occurs when patients make visual paralexias—reading n instead of m, or b instead of d (termed “visual dyslexia” by Marshall & Newcombe, 1973; Newcombe & Marshall, 1981; see also Lambon Ralph & Ellis, 1997). A different type of deficit in processing letters or part-of-theword appears in neglect dyslexia and its close relative, positional dyslexia. In these dyslexias, patients fail to read and report letters in a specific side of the word—either left or right, or in a specific position in the word (Arguin & Bub, 1997; Caramazza & Hillis, 1990; Ellis et al., 1987; Ellis, Young, & Flude, 1993; Katz & Sevush, 1989; Kinsbourne & Warrington, 1962; Riddoch, Humphreys, Cleton, & Fery, 1990; Warrington, 1991; Young, Newcombe, & Ellis, 1991). Interestingly, these patients successfully encode the position of letters, as indicated both by the consistent omission of letters in a specific spatial position, and by letter substitutions at the impaired side, which usually preserve letter position and word length. (Caramazza & Hillis, 1990; Ellis et al., 1987; Young et al., 1991). When the second function of the visual analyser fails, the reader should fail to allocate letters to the words they belong to. In normal skilled readers, this causes occasional letter migrations between words when words are presented briefly (Allport, 1977; McClelland & Mozer, 1986; Mozer, 1983; Shallice 1

& McGill, 1978; Treisman & Souther, 1986; Van der Velde, 1992). A more permanent failure of the letter-to-word binding function causes attentional dyslexia, which is a similar phenomenon, but with a much higher error rate: patients fail to relate letters to the words they belong to1, and the result is letters that “migrate” from one word to the other even in the absence of a time limit (Shallice & Warrington, 1977; Warrington, Cipolotti, & McNeil, 1993). A selective impairment to the third function— identifying the position of letters within a word— has never been demonstrated. The only evidence from neuropsychology in favour of the withinword-position function of the visual analyser has been indirectly deduced from the pattern of letter migration in attentional dyslexia, where letters migrate primarily to corresponding positions in the other word, and preserve their within-word positions (Saffran & Coslett, 1996; Shallice & Warrington, 1977). The current study offers direct evidence for this function of the visual analyser from two Hebrewspeaking individuals who present a selective deficit to positioning of letters within a word, with intact letter identification and intact binding of letters to words (henceforth: Letter Position Dyslexia, LPD). The special nature of Hebrew orthography offers an interesting test case for the study of reading. Hebrew is a Semitic language which is written from right to left in Hebrew letters. Vowels are usually not represented in Hebrew orthography, and many words comprise only consonant letters. For example, the vowels /a/ and /e/ are almost never represented (except for at the end of words), and therefore words that sound completely different are written exactly the same. For example, /kerex/ (volume), /karax/ (bound), and /krax/ (metropolis) are all written the same: (KRK). The vowels /i/, /o/, and /u/ are only represented in some of the words. Furthermore, even when a vowel is represented orthographically by a letter, this letter is usually ambiguous between several vowels and consonants (the letter “ ” for example can be read as

Or fail to set the attentional window (Humphreys & Riddoch, 1992) or to attenuate irrelevant neighbouring words (Shallice, 1988).

674

COGNITIVE NEUROPSYCHOLOGY , 2001, 18 (8)

LETTER POSITION DYSLEXIA

/o/, /u/, /w/, or /v/; and some consonants are ambiguous too). This leads to numerous reading possibilities for almost every Hebrew letter sequence, some of these readings being existing words (a 4-letter word, when represented by consonants only, can actually be read in more than 63 different ways!). As a result, when a letter erroneously changes position in a sequence, there are many possible ways to read the new sequence. This increases the probability that at least one of them will be an existing word, and therefore lexical knowledge cannot always compensate for letters that are perceived in a wrong position. These properties of the Hebrew orthography (together with the Semitic morphological system, see Discussion) make a selective deficit of letter migration within words easier to detect in Hebrew. Two Hebrew-speaking dyslexic patients with Letter Position Dyslexia are presented: They identify letters correctly, but are impaired in assigning letters to their proper position within a word. Thus, they predominantly make letter-order errors such as reading for (BSLNIT =/bashlanit/ =a woman who likes cooking, for BLSNIT = /balshanit/ = female linguist). This pattern suggests that letter position encoding is a separate function of the visual analysis system.

SUBJECTS Two right-handed Hebrew-speaking patients, BS and PY, participated in the study. Both were referred to the clinic with reading difficulties following a left-hemisphere lesion. Both had no prior reading disorder. BS was a 75-year-old right-handed man, who worked prior to his impairment as a graphic editor and a calligraphy artist and had a bachelor degree from the academy of arts. He was admitted to the aphasia clinic for language therapy. Three months prior to his admission BS underwent left parietooccipital tumour removal craniotomy, which was complicated 5 days later with haematoma at the bed of the tumour, which resulted in a mild right hemiparesis, right field hemianopsia, and language disorders. Upon admission, neurological analysis of

the CT scan reported left parieto-occipital lesion; BS was oriented in time and place and was attentive and cooperative. His main complaint was difficulties in reading. Language assessment using the Hebrew version of the WAB (Kertesz, 1982; Hebrew version by Soroker, 1997) revealed fluent speech with very mild nominal difficulties in spontaneous speech (WAB Spontaneous speech = 18/ 20) as well as mild nominal difficulties in confrontation naming (WAB Object naming = 50/60) with no phonological or verbal paraphasias (only circumlocutions and “Don’t know” responses), good repetition (WAB Repetition = 93/100), and fairly good auditory comprehension (WAB Auditory Word Recognition = 58/60; WAB Sequential Commands = 68/80). His writing to dictation was much better than his reading, with only nine letter position errors and two homophone substitutions in writing 155 words. PY was a 70-year-old man, a right-handed army veteran with high school education. He was tested 4 months post onset of a left ischaemic parieto-occipital infarct, which resulted in a right hemiparesis, which was already resolved at the time of the language assessment, and with no visual field deficits. Upon admission, neurological analysis of the CT scan reported parieto-occipital infarct in the left hemisphere. Language assessment revealed nearly intact performance in all language modalities except reading: spontaneous speech (WAB Spontaneous Speech = 20/20), sentence repetition (WAB Repetition = 93/100), naming (WAB Object Naming = 57/60, Fluency = 14/20), and auditory comprehension (WAB Auditory Word Recognition = 60/60, WAB Sequential Commands = 80/80). No phonological paraphasias were present in his spontaneous speech and naming. No visuospatial neglect was found for either BS or PY when tested with the Behavioral Inattention Test neglect battery (BIT—Wilson, Cockburn, & Halligan, 1987) all subtests including picture copying and human figure drawing were intact and included no sign of spatial or configural deficit. Simultanagnosia was also ruled out by the patients’ performance on complex picture description tasks, which was completely fluent and normal (the cookie theft from the BDAE, Goodglass & COGNITIVE NEUROPSYCHOLOGY, 2001, 18 (8)

675

FRIEDMANN AND GVION

Kaplan, 1983, and a sequence of four detailed pictures). In addition, no deficit in object perception and identification was observed in context of several objects appearing simultaneously in the visual field. Reading of single letters was good for both patients, even in short presentation: In single letter naming tasks, BS correctly read 52 of 54 letters (10 items with no time limit, 22 presented on computer screen for 1 s, and 22 for 0.7 s); BS could not perform the task in shorter presentation times. PY read 44/44 letters correctly in 0.1 s computerised presentation.

Error types and effects on reading The reading pattern presented by the patients was unique in both in the error types and in the effects on reading (namely, which word stimuli induced more errors). The predominant error was letter migrations within words. Unlike in the central dyslexias, no regularisation, semantic paralexias, or lexicalisations (except for “migratable nonwords”, see Experiment 2) were observed. In addition, there were no visual paralexias, no errors typical to neglect dyslexia, no letter-by-letter reading, and only a few letters migrated between words. In order to assess the effects on reading performance such as regularity, imageability and grammatical category, the relevant subtests of the Hebrew PALPA (Kay, Lesser, & Coltheart, 1992; Hebrew version by Gil & Edelstein, 1999) were devised. The patients did not show effects of regularity on PALPA 35 (Table 1; no effect for each patient separately using c2, p > .1, and no effect for both subjects combined, using Mantel Haenszel test for collection of 2 × 2 tables, c2 = 0.071, p > .5). The patients also showed no effect of imageability on PALPA 31 (Table 2; no effect for each patient separately using c2, p > .1, and no effect for both subjects together using Mantel Haenszel test, c2 = 0.65, p > .1). The relatively small number of items in the PALPA imageability subtest is not responsible for the lack of imageability effect. 2

Table 1. Reading of regular and irregular words: % correct (number correct/total) Subject #

BS # PY Total #

#

Regular

Irregular

75% 100%

(24/32) (32/32)

75% 94%

(24/32) (30/32)

88%

(56/64)

84%

(54/64)

n.s., p > .1.

When we added to the PALPA results a reanalysis by imageability of all the words read in the oral reading task in Experiment 1 (400 items for BS, 356 items for PY), there still was no significant effect of imageability (BS performed 80% in lowimageability words and 82% in high-imageability words, using c2, p > .05; PY performed 88% correct in low-imageability words and 92% correct in highimageability words, again, using c2, p > .05). Word length was not a critical factor for reading performance either—increasing length did not cause increase in error rate (Table 3)2. No linear trend was found using logistic regression for PY or for BS (without the three-item cell), p > .1. Long words of 5 and 6 letters that do not have lexical anagrams (and, as will be shown soon, are therefore less prone to letter migration errors) were read better than short 4-letter words that do have lexical anagrams (93% correct in 5–6-letter no-anagram words vs. 66% in 4-letter words with anagrams). The grammatical category effect that was found using Hebrew PALPA 32 was very different from the “classical” grammatical class effect (as witnessed for example in the reading of many patients with deep dyslexia, see Coltheart, 1980; Marshall & Newcombe, 1980; Morton & Patterson, 1980; Table 2. Reading of words of high and low imageability: % correct (number correct/total) Subject #

BS # PY Total #

#

Low imageability

High imageability

87% 93%

(26/30) (28/30)

77% 90%

(23/30) (27/30)

90%

(54/60)

83%

(50/60)

n.s., p > .1.

In order to allow a clear estimate of length effect we looked only at nonmigratable words, because in migratable words the migration potential depends on number of letters, and middle migration is only possible in words of four letters and longer.

676



Table 3. Reading of words as a function of number of letters: % correct (non-migratable words) 2 BS PY

3

100% (11/11) 100% (11/11)

97% (28/29) 92% (35/38)

4

5

83% (20/24) 93% (14/15)

Patterson, 1979). For example, function words, which are usually the hardest category, turned out to be the best category for our patients. Function words were significantly better than verbs for PY and for both subjects combined (using Mantel Haenszel test c2 =6.87, p =.009); all other comparisons of function words with other categories for each patient and for the sum were nonsignificant in the direction of better reading of function words, using c2 and Bonferroni adjustment (Table 4). Even after adding the results of a reanalysis by grammatical category of all the stimuli in the oral reading task in Experiment 1 to the PALPA results (400 items for BS, 356 items for PY), grammatical category effect was still in the same direction, with function words being not significantly different from nouns, and significantly better than verbs and adjectives. This unusual pattern of grammatical category effect, with best reading performance on function words, is probably only a by-product of the differential liability of these categories to letter migrations: function words were the least susceptible to migration errors due to their small number of letters, and verbs and adjectives were the most vulnerable due to the nature of Semitic word templates, which makes verbs and adverbs interchangeable by changing the location of a single letter (see Discussion for further detail on the effects of word length and Semitic templates on liability to migration errors). These two factors were probably responsible for the unusual effect of grammatical category on reading.

90% (38/42) 98% (41/42)

6 85% (11/13) 93% (13/14)

>6 33% (1/3) 100% (8/8)

It seems, then, on analysis of both error types and various effects on reading, that this deficit is not consistent with any of the known dyslexias. We therefore proceeded to examine the nature of the positional deficit in reading in greater detail.

Control subjects To obtain data on normal performance in the following experiments, all tests were also administered to 10 control subjects without reading deficits aged 32–75, 5 men and 5 women. All subjects were native speakers of Hebrew, without language or reading disorders. Two of the control subjects were matched in age, gender, and education level to the two dyslexic patients, and their data were analysed separately. The performance of all control subjects on all tasks was above 95% correct; the exact performance rate for each test is given in Appendix A.

EXPERIMENTAL INVESTIGATIONS Experiment 1: Reading “migratable” words The first experiment examined reading of migratable words, namely, words that consist of letters which, in a different order, could assemble at least one additional word (a relevant example in English is the letter set [b,r,e,a,d] which could be read as beard, bared, bread, and debar). Our assumption was that if letter position information is not available to the patients, they might

Table 4. Reading words of different grammatical categories: % correct (number correct/ total) Subject

Verb

Noun

Adjective

Function word

BS PY

70% (14/20) 70% (14/20)

75% (15/20) 80% (16/20)

80% (16/20) 90% (18/20)

90% (18/20) 100% (19/19)

Total

70% (28/40)

78% (31/40)

85% (34/40)

95% (37/39)

COGNITIVE NEUROPSYCHOLOGY, 2001, 18 (8)

677

FRIEDMANN AND GVION

rely on lexical knowledge, either as a part of the automatic reading process or as a compensatory strategy to limit themselves to existing words. For many words, these lexical constraints will lead to a single possible reading: the only word that could be constructed from the letters identified. For example, for the English letter set [b,u,t,t,e,r], only one reading possibility is allowed by the lexicon— the word “butter”. In these words, then, a deficit in letter location will mostly be compensated. The deficit will become more apparent in letter sets which have more than one lexical reading. In these cases, the lexicon’s help is not enough, and the patients are bound to make more errors of letter migration within a word (reading “bread” as “beard”, for example). We therefore compared words for which the letters could be ordered to form a different word (migratable words) with words that are built from letter sets that have a unique reading (nonmigratable words). We used oral reading and same-different decision tasks. Oral reading of migratable words Method. Migratable and nonmigratable Hebrew words were presented in large print (18-point font) without time limitation (54 of the nonmigratable words presented to PY were presented on a computer screen for 0.1 s. His performance on the time limited task was similar to that of the untimed task, and the results were collapsed together). The migratable words included words with middle letter migration potential (to form another existing word, see example 1), and words with a potential of exterior letter migration (see example 2). The patients were asked to read the words aloud, and no response-contingent feedback was given during the test. 1. Migratable word pair, middle migration: (/hispik/ = managed /hifsik/ = stopped)

HSPIK-HPSIK

3

2. Migratable word pair, exterior migration: (/smixa/ = blanket /meshixa/ = attraction, withdrawal, pull)

SMIXH-MSIXH

Migratable and nonmigratable stimuli were 2–8 letters long (average length for migratable words was 4.1, average length for nonmigratable words was 4.2 letters) and were balanced for lexical category (6:2:1 ratio of nouns, verbs, and adjectives respectively for both the migratable and nonmigratable words)3. Results. The results presented in Table 5 show that BS is more impaired in reading than PY. However, they share the same error pattern. For both patients, migratable word reading was more impaired than reading of non-migratable words (the difference was significant for each individual patient using c2, p < .005, and for the sum using Mantel Haenszel test c2 = 20.82, p< .0001). Within-word migrations were by far the most frequent error in the reading of migratable words, accounting for 87% (93/107) of the errors in migratable words. In nonmigratable words, errors were mainly mixed errors of letter addition and letter migration in middle positions, all producing other existing words. Lexical knowledge was probably responsible for the more preserved reading of nonmigratable words. Frequency was not responsible for the difference in performance between migratable and nonmigratable words. In a post hoc assessment of the frequency of words in this test, we obtained frequency estimation ratings from 50 native speakers Table 5. Reading of migratable and non-migratable words: % correct (number correct/total) Subject #

BS # PY Total #

#

Migratable words

Non-migratable words

74% (207/278 ) 84% (192/228)

89% (109/122) 95% (122/128)

79% (399/506)

92% (231/250)

p < .005.

In tests in which some items are missing (when the patients did not complete the whole test), the relevant Method sections include length and grammatical category data for the items that were actually read by the patients and were included in the results, and not for all the pre-planned items.

678



of Hebrew, who rated the words in this test on a scale of 1–7, 7 being “very frequent”. In addition, we used words from the frequency estimation database for Hebrew words by Ram Frost. The average frequency estimations of migratable and nonmigratable words did not differ significantly (t < 1). Average frequency of nonmigratable words was 4.45 (SD = 1.38), average frequency for migratable words was 4.43 (SD = 1.39). Words with exterior migration potential had an average frequency of 4.43 (SD = 1.37), and words with middle letter migration 4.44 (SD = 1.41). Most of the migration errors involved middleletter migration to another middle position. Only three migrations from exterior positions occurred for both patients together. For this reason, migratable words in the following tests were based only upon medial letter migration (see Table 14 for a full analysis of exterior vs. medial position pure migration errors in this test). Note that the good reading of migratable words with only exterior letter migration potential means that Table 5 actually presents an overestimation of the patients’ ability to read migratable words with middle-letter migration potential. Same-different decision Method. One hundred and twenty Hebrew word pairs were presented in random order. The test included 40 pairs of words differing in the relative order of middle letters ( TIRS-TRIS = cornshutter), 40 pairs of words differing in the identity of a single middle letter ( AFOR-ASOR = gray-prohibited), and 40 pairs of identical words ( MXSV -MXSV = computer). Word stimuli were balanced for length and grammatical category between conditions. Each condition included 48 nouns, 24 verbs, and 8 adjectives. Words were 47 letters long, with an average length of 5.0 for each condition. The word pairs were presented visually without time limitation, and the patients were asked to determine whether the words in the pair were identical or not, without reading them aloud. The test was administered twice to PY: once without time limit, and once with 1-s exposure for each pair. The timed test was administered 1 month after the

untimed test. PY’s performance on the two tasks was identical (except for the “same” pairs, in which the unspeeded presentation yielded 35 items correct, and the speeded yielded 32 correct items) and therefore the results were lumped together in Table 6. Results. The patients’ performance is presented in Table 6. The results were striking: A chance performance was evinced in same-different decision of words that differed in letter order (not significantly different from chance for each patient and for the sum, p > .1, binomial test). In marked contrast, the performance on same-different decision for words that differed in letter identity was relatively intact and differed significantly from the different-order pairs (for each individual patient p < .0001, and for the sum using Mantel Haenszel test, c2 = 50.09, p < .0001), and from chance (for each patient and for the sum, p < .0001, using the binomial test), and did not differ from the performance in the “same” condition (for each subject p > .1, and for the sum, Mantel Haenszel c2 = 2.27, p > .1). Thus, in this task too, the patients’ errors were mainly errors of letter migration within words. Summary From both experiments it is evident that letter migration within words is the predominant error type in the reading of these two patients. The scarcity of letter substitutions in the various reading tasks, and the good performance in same-different decision of words that differ in one letter, suggest that the deficit is selective to letter location, and that letter identification is relatively unimpaired. Table 6. Same-different decision: Different order and different letter pairs: % correct (number correct/total) Subject

Different order

Different letter

Same

BS PY

60% (24/40) 48% (38/80)

95% (38/40) 93% (74/80)

93% (37/40) 84% (67/80)

Total

52% (62/120)

93% (112/120)

87% (104/120)


679

FRIEDMANN AND GVION

Experiment 2: Nonword reading Reliance on lexical knowledge in the absence of letter position information is useful when reading nonmigratable words. Nevertheless, exclusive reliance on the lexicon could become an obstacle in reading nonwords that are derived from existing words by changing the letter order: It could lead to misreading of nonwords as words. For example, in the absence of letter position information, the nonword “pincel” is prone to be read as “pencil”. It is therefore interesting to examine the nonword reading performance of patients who have impaired position encoding, and rely on lexical information. Nonword reading was examined by means of two types of tests: reading aloud and lexical decision (both without time limitation). Oral reading of nonwords Method. We compared nonwords that can be rearranged as Hebrew words following middle-letter rearrangement (migratable nonwords) with nonwords that cannot (nonmigratable nonwords). Nonwords were constructed by changing a single letter in existing words. All nonwords were 4–6 letters long, with an average of 5.0 letters for each condition. Results. The patients’ performance in this task is presented in Table 7. Both patients showed a tendency toward reading migratable nonwords worse than nonmigratable nonwords. For PY, it was significant, c2 = 10.80, p = .001; BS showed only nonsignificant tendency, c2 = 0.74, p > .1, and both of them together showed significant difference using Mantel Haenszel test, c2 = 6.81, p < .01. Unlike the nonmigratable nonword reading, only 4% of the errors in the migratable nonwords formed

another nonword: 23 of the 24 errors in the migratable nonwords were migration errors that formed an existing word, and only one was an addition error that formed another nonword. In the nonmigratable nonword reading, errors were migrations that formed another nonword, and letter substitutions that formed a word. Lexical decision Method. Twenty eight migratable nonwords and 28 Hebrew words were presented individually to the patients, printed on paper (18 point font). Sequences were 5–6 letters long, with an average of 5.3 in both conditions. The patients were asked to determine whether the sequence was an existing word or not. Results. As shown in Table 8, migratable nonwords were judged as words almost half of the time (using the binomial test, not significantly different from chance, for each patient and for the sum, p > .05). Words were judged as nonwords only 2% of the time, and were significantly better than nonwords (for each patient p < .0005, and for the sum using Mantel Haenszel test, c2 = 26.21, p < .0001). The results indicate that the patients rely on lexical knowledge, and this tendency leads them to read migratable nonwords as words with the same letters, but different letter order. The pattern that emerges from the experiments presented so far regarding reliance on the lexicon is that lexical knowledge helps to confine reading to the right word in nonmigratable words, is less helpful in migratable words, induces errors in migratable nonwords, and does not make a difference in (nonmigratable) nonword reading.

Table 7. Reading aloud—migratable and non-migratable nonwords: % correct (number correct/total)

Table 8. Lexical decision task

Subject

Migratable nonwords

Nonmigratable nonwords

Subject

BS PY *

#

52% (15/29) 66% (19/29)

65% (11/17) 97% (33/34)

Total*

59% (34/58)

86% (44/51)

#

n.s., *p = .01.

680


Migratable nonwords judged as words

Words judged as nonwords

BS* PY *

50% (14/28) 39% (11/28)

4% (1/28) 0% (0/28)

Total*

45% (25/56)

2% (1/56)

*p < .0005.


Experiment 3: Letter location tasks In order to focus on the letter location deficit we explored the locational ability directly, by asking the subjects to name a letter according to its serial position in the word, and to determine the position of a letter in a presented sequence. Naming letters according to position in a word: Visual presentation Method. One hundred and fifteen nonmigratable Hebrew words of 3–6 letters (average length 4.3) were visually presented to one of the patients (BS). In each word, the patient was asked to name a letter according to its serial position provided by the experimenter, without time limitation (“What is the second letter in this word?”). Results. The results again indicate a deficit in letter location, which is most evident in middle positions, c2 = 5.59, p < .05 (see Table 9). Errors were naming letters that appeared in a different serial position in the word. A demonstration of the difficulty BS experienced in these letter location tasks could be seen in his response when trying to name the middle letter in the five-letter word MTRIA (/mitria/ –umbrella): ‘tee. no . . . there isn’t really a middle letter . . . ar?’ Interestingly, in all 20 cases in which the patient failed to name middle letters according to their position, he succeeded in oral spelling of the whole word (letter by letter). Naming letters according to position in a word: Auditory presentation Method. Letter naming by position was tested in auditory presentation as well. In this task, the experimenter said a Hebrew word, and the patient had to use his mental imagery of the orthographic Table 9. Naming letters according to position—visual presentation: % correct (number correct/total) Subject BS* *p < .05.

Middle position

First / Last position

74% (56/76)

92% (36/39)

representation of the word in order to name a letter according to its position (“What is the second letter in the word broad?”). Words were 3–6 letters long with an average length of 4.3 letters. All words had a potential of middle-letter migration. Words in the first/last letter naming condition had exterior letter migration potential as well, such as the word = voter, chooses, which has a middle-letter migration potential for =guy, but also exterior migration potential for , , (= runsaway, carob, street). Results. The results of the auditory presentation were similar to those observed in the visual presentation: naming middle letters was poor, whereas first- and last-letter naming was intact (significant difference between middle and exterior position for each subject, p < .05, and for the sum using Mantel Haenszel test, c2 = 9.6, p < .005). Although in Hebrew there are several homophone letters, and the phoneme-to-grapheme relation is highly irregular, the patients did not make letter selection errors (homophone letter selection), only migration errors. This indicates that they did not lean on phonological analysis for this task, but rather retrieved the items from the orthographic lexicon. Determining letter position within a symbol sequence Method. Sequences were presented on a computer screen. Each sequence consisted of four pound symbols, and a Hebrew letter in one of the five possible positions within the sequence (# # # #). Each sequence was presented for 0.1 s to PY and for 1 s (30 sequences) or 0.7 s (10 sequences) to BS. The patients were asked to locate the position of the letter (say the serial position from right to left). Table 10. Naming letters according to position—auditory presentation: % correct (number correct/total) Subject

Middle position

First / Last position

BS* PY *

67% (51/76) 69% (36/52)

100% (9/9) 100% (18/18)

Total *

68% (87/128)

100% (27/27)

*p < .05. COGNITIVE NEUROPSYCHOLOGY, 2001, 18 (8)

681

FRIEDMANN AND GVION

Table 11. Determining letter position in a sequence: % correct (number correct/total)

Subject

Letter position —————————————— Middle First/last

BS* # PY

67% (16/24) 61% (11/18)

100% (16/16) 83% (10/12)

64% (27/42)

93% (26/28)

Total* *p < .05, n.s. #

Results. In this task, again, both patients showed impairment in letter location, primarily in medial positions, as shown in Table 11. The difference between middle and first/last positions was significant only for BS, c2 = 6.67, p = .01, and not significant for PY, c2 = 1.69, p > 0.1, significant for both using Mantel Haenszel test, c2 = 5.89, p < .05. Summing up, direct assessment of the patients’ ability to locate letters within words and sequences indicates a deficit in this function. The vulnerability of the middle letters, observed in the previous reading tasks, was also detected in the letter location tasks.

Experiment 4: “Classical” attentional dyslexia The results indicate that the patients suffer a problem with encoding the position of each letter within the word, which causes migrations within words. Do they also evince letter migration between words, like the patients reported in Shallice and Warrington’s (1977) classical study? The patients described by Shallice and Warrington (and the Alzheimer’s patient in Saffran & Coslett, 1996) showed migration of letters between words, which in most cases preserved their within-word position (the first letter in one word migrated to the first position in the other word). A patient with letter migrations within words will not necessarily suffer from migrations between words as well (see, for example, Riddoch et al., 1990, and Saffran & Coslett, 1996, for a related discussion of attentional deficit between and within

682


words; and Duncan, 1987, for a comparison of between- and within-word migration in normal reading). However, in case letters do migrate between words, we surmised that the pattern of migration between words would be different for LPD patients, since the patients’ ability to encode letter position is impaired. For these patients, letters are also expected to migrate to non-corresponding positions, without preserving the original position within the word. In order to check whether migrations between words occur in these patients’ reading and to examine position preservation, semantically unrelated Hebrew word pairs were presented for oral reading, without time limitation. Words were 4–6 letters long, and shared 2–3 letters in the beginning, the end, or the middle. They were constructed in such a way that two letters, separately or combined, could migrate to the second word and create two other words (see examples 3 and 4 below). The distance between words was 0.2 cm—the regular spacing between 20-point letters in font David. 3.

ALIM KDIM

= gods,violent-jars

/elim/-/kadim/

possible migrations: KLIM (/kelim/ ADIM (/edim/ = vapour) = tools); 4. XOLBT XOSFT /xolevet/-/xosefet/ = milks-exposes possible migrations: XOLFT (/xolefet/ XOSBT (/xoshevet/ =thinks) =passes); Results. Few migration errors between words were found for both patients. The errors were not consistent with the position preservation reported for “classical” attentional dyslexia: 55% of the migrations between words did not preserve the original position (Table 12; no significant difference was found between position preserving and nonpreserving errors for each subject, p > .1, and for both of them together using Mantel Haenszel test, c2 = 0.25, p > .1). Furthermore, letter position preservation in the remaining 45% of the migration errors does not necessarily indicate preserved knowledge of letter position. It may be the case that it is merely the lexical constraints that prevented the letters from migrating to a different position: in all


Table 12. Letter migrations between words Migrations between words/ word pairs

Error type ———————————– Preserving Not preserving position position

BS PY

16% (16/100) 12% (15/124)

38% (6/16) 53% (8/15)

63% (10/16) 47% (7/15)

Total

14% (31/224)

45% (14/31)

55% (17/31)

the position-preserving cases, there was no other position to insert the migrating letter and still keep the sequence a lexical item.

Experiment 5: Reading nonlinguistic material—numbers and icons In order to assess whether the reading deficit was limited to linguistic-orthographic material, reading of numbers and icons was examined. Oral reading of numbers The patients were asked to read aloud 28 numbers. The numbers were 3–6 digits long. Results. BS’s number reading was 71% correct (20/ 28); PY’s reading was 86% correct (24/28). In number reading there were no errors of migration of the type evinced in word reading (e.g., reading 1423 instead of 1243). The 12 errors in the oral reading task were digit substitutions only, 9 of them were doubling of digits: replacing a digit with a different existing digit in the same number (reading 1223 instead of 1243)4. It is hard to determine whether the doubling errors were a result of random digit substitution or migration of a digit without deletion of the digit from the original position. Same-different decision—numbers The patients were asked to make same-different decisions on 3–6 digit number pairs. Half of the pairs were identical, and half differed in digit order,

but consisted of the same digits. The same and different pairs were randomly ordered. Results. As seen in Table 13, the performance of both patients in the same-different task was good. The errors in same-different decision consisted of four false alarms (calling a same pair “different”), and one misdetection (calling a different pair “same”). The results of this test were significantly different from the equivalent task in word reading: Unlike in words, where the patients could not detect a difference in letter position and performed at chance in same-different decision task, in the number task they performed significantly better than chance (for each subject, and for the sum using binomial test, p < .0001), and significantly different from the same-different order decision in the word task (for each patient, and for the sum, c2 = 33.28, p < .0001). Digit position within a number Thirty-five numbers of 4–6 digits were presented on paper (in 20-point font). The patient (BS) was asked to name a digit according to its position in the number. (The experimenter asked “What is the second digit in this number?” pointing to the number). Only digits in medial positions were included. Results. BS scored 89% correct (31/35) on this task. All four errors involved naming of a digit from a different position. Icon position Twelve sequences of five icons were presented separately to BS. The size of the icons was the same as Table 13. Same-different decision—numbers: % correct (number correct/total) Same-different decision BS PY

91% (29/32) 95% (35/37)

Total

93% (64/69)

4

In addition, BS had six “decimal” errors (such as naming 200 as 2000): three pure decimal, and three mixed with substitutions , that were also included in Table 13. COGNITIVE NEUROPSYCHOLOGY, 2001, 18 (8)

683

FRIEDMANN AND GVION

the size of letters in the other tests. The patient was asked to name an icon according to its position within a sequence (“What is the second icon in ?”). Since Hebrew is read from right to left, the patient was instructed to refer to the rightmost icon as the first. Only medial positions (2nd, 3rd, and 4th icons) were tested. A short training session preceded this test in which the patient was asked to name each icon separately. Results. BS performed perfectly on this test: He detected 12/12 correct icon positions5. Although more data are needed, especially regarding icon reading, it seems that the patients’ deficit manifests mainly in words. In icon-sequence naming the deficit was not observed at all, and in number reading the error pattern was different, and smaller in extent. This adds to the data that indicate that the locational deficit does not extend to objects and pictures from the patients’ good performance on complex pictures, which they flawlessly drew, copied and described (see Subject description, p. 675).

FURTHER ANALY SES: TOWARD A CHARACTERISATION OF LPD Medial vs. first and last letter migration— Exterior letter advantage A comparison of migration errors in different positions within words revealed that not all letter positions were equally impaired. A reanalysis of the oral reading of migratable words in Experiment 1 by position of potential migration, summarised in Table 14 (only pure migration errors, words with both middle and exterior migration potential

Table 14. Word reading: Medial vs. exterior letter migration: errors of a certain type/total words with potential migration of this type Subject

Medial

Exterior

BS * PY *

20% (38/194) 12% (21/174)

1% (1/195) 1% (2/171)

Total *

16% (59/368)

1% (3/366)

*p < .0001.

counted twice), showed that medial letters were 10 to 38 times more vulnerable to migration than letters in first and last positions (for each individual patient, p < .0001, and for the sum using Mantel Haenszel test, c2 = 55.10, p < .0001). Migration errors occurred mainly in medial positions, whereas first and last letters were relatively migrationproof 6. (Average frequency estimations for words with middle—and for words with exterior—migration potential did not differ significantly, t < 1. Frequency estimations for words with only exterior migration potential was 4.43, SD = 1.37, and for words with middle letter migration 4.44, SD = 1.41). Both consonants and vowels migrated. In addition, both template and root letters migrated to the same degree when they were in middle position, and both were migration-proof in first and last positions. The data about word reading (Table 14) as well as about letter position tasks (Tables 9, 10, and 11) indicate that migration errors occur mainly in medial positions, and almost never in first and last positions. The privileged status of end letters is in accordance with numerous previous findings on normal reading and impaired reading, obtained in different experimental tasks. Bradshaw, Bradley, Gates, and Patterson (1977); Bradshaw and Mapp (1982); Grainger and Segui (1990); Humphreys et al., (1990); Humphreys, Evett, Quinlan, and

5

It might be that this task was easier due to the lower similarity between these items compared to letters. Future studies might use more similar nonorthographic stimuli to avoid this problem. 6

A tendency toward exterior position advantage was also found in number reading: In the oral number reading, 3 out of 4 migration errors PY made were in medial position, and 7 out of 11 errors (including doubling) of BS were in medial position. This parallelism between letters and numbers with regard to exterior position advantage in normals was reported by Hammond and Green (1982) and Mason (1982).

684



Besner (1987); Mason (1982); Merikle and Coltheart (1972); Perea (1998); and others have shown that in normal reading, letters at the beginning and end of words are processed differently from and faster than letters within. The aforementioned studies dealt mainly with letter identification. Humphreys et al. (1990) have studied letter position processing as well, and have found lack of position specificity for medial letters: End letters were found to be more accurately tied to their relative position than were internal letters in normal word reading. Estes (1975) directly examined transpositions in normal reading, and found that twice as many occur in middle than in end positions in words, and in a 4:1 ratio in nonwords. Why are exterior letters identified and located better than middle letters? Some researchers have suggested that exterior letters are the ones that are used to access the subset of word candidates, and that these are the positions that provide more activation to the lexicon (Forster, 1976; Grainger & Segui, 1990)7. This importance of exterior letters might be the reason that in normal reading, word processing begins at the ends, namely that exterior letters are accessed faster than interior ones (Bradshaw et al., 1977; Bradshaw & Mapp, 1982; Mason, 1982; Merikle & Coltheart, 1972). If initial and final letters indeed have priority in word processing, it might be the case that our patients, with limited attentional capacity (for reading), allocate attention to these positions, while interior letters are not attended. As a result, illusory conjunctions between letters and positions occur in medial positions. In this respect, LPD readers are somewhat like participants in an attention task of the kind used by Treisman and Schmidt (1982) in which subjects were required to report the two digits that appear before and after a letter string. There, too, participants experience illusory conjunctions in the unattended letter string appearing between the two attended exterior digits. Another contribution to the better preservation of position in exterior letters might be that they

have fewer neighbours, and therefore a smaller number of transposition opportunities (or competing positions) compared to middle letters. However, we need to account somehow for the magnitude of difference between errors on middle and exterior letters. As Estes (1975) noted, an error ratio of 2:1 of middle compared to end letters would be expected, but our findings show a much higher ratio of 38:1 for one patient, and 10:1 for the other. Mozer (personal communication), suggests that the observed ratio can be explained by some sort of nonlinear lateral interference, in which having two neighbours causes uncertainty far greater than twice the uncertainty caused by having a single neighbour. Note that the witnessed pattern of letter location errors can result either from complete loss of location information or from an enhanced uncertainty with respect to letter location, with letters being encoded in a probabilistic way, high probability on the letter’s actual position and a distribution of lower probability around this position. More empirical data regarding the rate of transpositions between neighbouring positions compared to nonneighbouring positions might help determine which of the possibilities describes the deficit better.

Word frequency An analysis of the migration errors was made that examined the directionality of errors with respect to word frequency. We examined whether errors were more likely to occur from less frequent to more frequent words, or whether words were read and errors were made regardless of their relative frequency. Since Hebrew does not have an updated frequency list for all lexical items, relative frequency of migratable words that the patients read with migration errors was determined by 25 normal readers, native speakers of Hebrew. Judges were asked to decide, for each written word pair, which word was more frequent. The 30 word pairs that were agreed

7

If the access code to the input lexicon is the initial letter and the word’s length, then it is evident why last-letter position in essential for lexical access: A word’s length is determined by the serial position of the terminal letter. (See Mason, 1982, for a similar account regarding last item in words and numbers.) COGNITIVE NEUROPSYCHOLOGY, 2001, 18 (8)

685

FRIEDMANN AND GVION

upon by at least 80% of the judges were included in the analysis. Analysis of the reading of these 60 words shows a directional pattern of migration errors, with a significant difference between frequent and nonfrequent words. The patients tended to read nonfrequent words as their frequent counterparts, and not vice versa (Table 15). For instance, the word “ ” (Teflon) was read by both patients as “ ” (telephone), whereas the other direction was not witnessed—“telephone” was not read as “Teflon” (100% of the judges rated the word “telephone” as the more frequent word). This difference was significant for each patient, p < .05; and for the sum using Mantel Haenszel, c2 = 13.17, p < .0005. In a further test that required reading of 24 word pairs of frequent and nonfrequent words (each word was separately presented, in random order), the same tendency toward the frequent reading was observed, although in a relatively small number of errors. BS had four out of five migration errors toward the frequent reading; PY had two out of three. These results show preference for the more familiar or frequent word, and a tendency to read the more frequent anagram in cases of failure to encode letter position, thus supporting the contribution of lexical knowledge to word reading in Letter Position Dyslexia8.

Table 15. The effect of target word frequency on letter migration Non-frequent to frequent 17/30 11/30

8/30 1/30

47% (28/60)

15% (9/60)

BS* PY** Total**

Frequent to non-frequent

*p < .05; **p < .001.

Is this an orthographic input deficit or a phonological output deficit? Until now we have ascribed letter migration in our patients’ reading to orthographic-visual input deficit. Theoretically, however, an input deficit is not the only possible explanation, as incorrect reading aloud can also be due to phonological output deficit. The data show, however, that a phonological deficit is not the source of our patients’ letter migrations. If there were an output deficit, we would expect the deficit to be manifest also in speech contexts that do not involve reading, such as spontaneous speech, confrontation naming, and word repetition. In addition, we would not expect to find the deficit in reading tasks that do not require reading aloud. The data were just the opposite. Neither patient showed any sign of phonological deficit in speech contexts: They repeated words normally, and there were no phonological paraphasias in their spontaneous speech or in oral confrontation naming. In addition, their migration errors affected word comprehension: Words that were read with a migration error were assigned the meaning of the response word and not the target word. Had the problem been a phonological output deficit, they would have read the word aloud incorrectly, but understood it as the target word. Furthermore, the results of the same-different decision, the lexical decision, and the letter location tasks indicate that the same errors occurred when the patients were not required (and even asked not) to read aloud9. The error pattern also favours an orthographicinputover aphonological outputexplanation:Errors preserved the letters but not the sounds of the target word. Namely, the migrated units were graphemes rather than phonemes. For example, the two-syllable mishpat became mefashet ( ¬ ). These words share the same letters, but they do not have any syllable in common. Furthermore, the dissociation found between orthographically

8

This finding is more consistent with position uncertainty at the first perceptual stages of position encoding, which is later assisted by the lexical level, either by feedback from the lexical level to the visual analysis stage or by some type of guessing procedure, than with wrong position perception at the very first perceptual stages, which are not by themselves sensitive to word frequency. 9

Still the possibility of “inner speech” exists. A possible way to prevent inner speech in future studies would be to include “articulatory suppression” (Baddeley, 1990) in same-different and lexical decision tasks.

686



migratable and nonmigratable words cannot be explained by a phonological output deficit. In light of all these facts, we can confidently conclude that the deficit of our patients lies in the early stages of input analysis rather than in the late phonological output processes.

DISCUSSION We have reported a series of experiments examining the word reading of two individuals with acquired dyslexia. The main characteristic of their reading deficit was within-word migrations of middle letters that accounted for approximately 90% of the errors. The main effects on reading accuracy were the factors that determined the string’s liability to migration: the lexical status and frequency of the possible migration outcomes compared to the lexical status and frequency of the target. Reading was the poorest for nonwords that had existing word anagrams, and for words that had high-frequency anagrams. Target words with no lexical anagram were read with fewer errors. Regularity, semantic content, length, and grammatical class effects were not exhibited by our patients. The deficit proved highly selective. Letter identification was intact, and letter-to-word binding was relatively spared. Very few “classical attentional dyslexia” errors in which letters migrated between words occurred, and when they did, letters did not keep their within-word position This pattern of results indicates a novel type of peripheral dyslexia that stems from a deficit in the visual analysis stage. Only one function of the visual analyser is impaired here: the ability to locate letters within a word. Letter identification and letter-toword binding are relatively spared. Such a dissociation immediately reflects on the different separate functions of the normal visual analyser. Specifically, it offers direct evidence in favour of the single word reading model suggested by Ellis and Young (1988; Ellis et al., 1987), which postulates letter location as

one of the functions of the visual analyser that can be separately impaired. Until now, the existence of a separate function of letter location has been inferred from indirect evidence from patients with attentional dyslexia, who present letter migrations between words, but preserve letter position within words (Shallice & Warrington, 1977), and from patients with neglect dyslexia who tend to keep letter position and word length when substituting letters in one part of the word, thus showing successful assignment of letters to word positions (Caramazza & Hillis, 1990; Ellis et al., 1987; Young et al., 1991). However, although predicted from the model, a selective impairment to the letter location function has not been reported until now, and it is presented here for the first time.

Why was it easier to detect LPD in Hebrew? This selective deficit of letter migration within words has not been identified until now, probably due to the nature of the languages in which dyslexia was studied. In Hebrew it is much easier to detect such a deficit because, due to the nature of Hebrew’s “deep orthography” (Frost, 1992; Frost, Katz, & Bentin, 1987) and Semitic morphology, more letter migrations form legal sequences, and these sequences have a higher probability to be read as another existing word. Thus, migration errors cannot be avoided by reliance on lexical information, and a patient who suffers from letter position deficit is liable to produce more migration errors in reading. In Hebrew, vowels are usually not represented in the orthography 10. Consequently, the Hebrew reader has to tackle a complicated reading task, equipped only with partial knowledge about the phonological structure of the word. Without vowels, the reader has six possible readings for every consonant (a reading for every vowel). For example, the consonant can be read in the context of a word as /pa/, /pi/, /po/, /pu/, /pe/, and p with schwa (as well as /f/ with all these vowels). Some of these readings within a letter sequence form existing

10

Hebrew has two orthography systems: pointed and unpointed. In the pointed system, diacritics are added to the consonant letters and convey vowel information. The unpointed system does not contain the diacritics, and vowels are usually not represented in it. The majority of adult reading uses the unpointed system, whereas pointed writing is used mainly in children’s books and in poetry. COGNITIVE NEUROPSYCHOLOGY, 2001, 18 (8)

687

FRIEDMANN AND GVION

words11. For instance, the sequence (SPR) can be read as sefer (book), safar (counted), siper (told), sapar (barber), sfar (frontier), saper (tell-imperative), supar (tell-passive), and sper (colloquial for ‘spare wheel’). As if to make things more complicated, even when vowels are represented, they are ambiguous. The Hebrew letter “ ” can be read both as the vowels /o/ or /u/ and as the consonants /w/ and /v/; the letter “ ” can be read as /i/ and /ei/ and sometimes even as /a/, or as the consonant /y/. As a result, even when vowels are represented, the degree of freedom in word reading is still very large. Due to this underrepresentation of vowels, almost all letter sequences that are formed by migration are legal sequences (possible written words). The lack of vowel representation and, therefore, lack of markers for syllable nuclei allows almost every reading and every syllable partition, and increases the number of possible readings per sequence. Even a sequence of four or five consonants can be a word, as opposed to English, for example, where consonant-vowel alternation is usually required. Furthermore, the very large degree of freedom in reading a Hebrew letter sequence makes it more probable that at least one of the possible readings of a sequence that was created by migration will be an existing word (whereas in English, for example, a migration of one letter usually forms only one reading and therefore rarely forms an existing word12). The second factor that contributes to the high probability that a letter position error will result in another existing word is the Semitic derivational morphology. Most of the words in Hebrew consist of a three-consonant root and a template (Bat-El, 1989; McCarthy, 1979). The fact that many roots differ only in consonant order, and that there are templates that differ in the position of one vowel only, increases the probability that once the target word is a possible word in Hebrew that is built from a root and a template, letter migration would also

create a morphologically well-formed sequence, and possibly an existing word. For example, active and passive participles differ only in the location of one vowel; adjectives and adjective-derived nouns also differ in vowel position only. A migration of this vowel, for many words in these templates, would thus result in another existing word. If a consonant migrates, then if other roots exist with the same letters in different order, we might again end up with an existing word because the new root will be inserted in a legal template. For example: XoSeV-XoVeS-XaVuS (thinks-bandages, paramedic-bandaged, quince), DaGuL-GaDoL-GoDeLladDuG (distinguished-big-size-to fish). Our data have shown that migrations occur mainly when the sequence formed by migration is an existing word. Since the combination of orthography and derivational morphology in Hebrew causes more migrations to result in existing words, more migration errors were produced, and this made it easier to detect letter position deficit in Hebrew.

Were any other cases of letter position deficit reported in the literature? Careful perusal of the literature reveals some scattered evidence for letter migration within strings in English as well. The patients reported in Price and Humphreys (1993); Shallice and Warrington (1977), and in Warrington et al. (1993), who suffered from attentional dyslexia between words, and the patient in Katz and Sevush (1989), who suffered from positional dyslexia, had letter migration errors within letter arrays. When asked to name a specific letter within a letter sequence, these patients made migration errors—naming a letter from a different position within the same sequence. This accounted for 25% of the errors of Price and Humphreys’

11

In addition to the underspecification of vowels, the orthographic representation of some of the consonants in Hebrew is also ambiguous. For example, the letter can be read both as /p/ and as /f/. 12

The reader is invited to try and find pairs of English words that differ only in the position of letters in middle positions (such as “could” and “cloud”). In our experience, they are very hard to find in English, whereas in Hebrew such pairs are abundant, especially in verbs.

688



patient, 34% of the errors of Warrington et al.’s patient, 36% and 77% of Shallice and Warrington’s patients’ errors, and for 100% of Katz and Sevush’s patient13. Thus, these patients too were able in some cases to identify the letters in the sequence but failed to attach them to their positions. Note that these patients’ attentional deficit was noticed within arrays of letters, not within words14. Due to the character of English, and to the reliance on lexical knowledge, letter position deficit in words, unlike in letter-arrays, could go undetected. All these reported dyslexic patients had letter position location deficit on top of other reading or general visual attentional deficits: Some of them had hemineglect, some had letter identification deficits, and some evinced the attentional deficit also in nonverbal pictorial material. The current study presented two individuals with a pure letter location deficit, and thus enabled a closer look at the exact characteristics of the Letter Position Dyslexia and the letter position function15.

Is the location deficit of our patients specific to orthographic material? It is hard to say at this point. First, given the patients’ flawless drawing and copying, and their

good description of complex pictures, as well as their good performance in the BIT, it seems that the deficit does not extend to objects and pictures. The good performance on icon-sequence naming goes in the same direction, although more items are needed as well as information from stimuli that are more similar to each other. What about numbers? Here the picture is less clear. Data indicated that number reading was less impaired than word reading, and, crucially, the error pattern was completely different—whereas in words the predominant error was migration, in number reading there were only substitutions and doublings. There are two possible ways to interpret the observed difference between words and numbers. It might be that the position deficit is indeed specific to letters, and that orthographic material is processed separately from digits. However, it might also be that the deficit affects both letters and digits, but the different nature of words and numbers causes the deficit to be manifested in words but not in numbers. The two main differences between words and numbers that come to mind are, first, that whereas letters in words are processed in parallel as a group, digits in numbers are processed independently. If the attentional deficit is evident whenever several letters have to be processed in parallel, this predicts a deficit in real words but much better performance

13

Transposition errors in words were reported in writing, though not as the main error type (Caramazza, Miceli, Villa, & Romani, 1987; Caramazza & Miceli, 1990; Hillis & Caramazza, 1989; McCloskey, Badecker, Goodman-Schulman, & Aliminosa, 1994). In addition, LB (described in Caramazza, Capasso, & Miceli, 1996; Caramazza et al., 1987) showed transposition errors both in spelling and in nonword reading. Unlike our patients, transpositions were not their main error type: letter substitution s and insertions were more frequent than transpositions. 14

Price and Humphreys (1993) report that their patient read five-letter words with accuracy of only 55–58%, but they do not analyse the types of errors she made in word reading. 15

One reviewer has suggested that our patients might have suffered from word-form dyslexia (Warrington & Shallice, 1980, termed by others “letter-by-letter reading” or “pure alexia”). However, the reading pattern did not conform with word-form deficit. Word-form dyslexics frequently use letter-by-letter reading, have difficulties reading long whole words of any type, their performance deteriorates considerably in short presentation, and their reading performance is greatly affected by length. However, in our patients, no letter-by-letter reading was presented, and their reading of non-migratable words was good (see Tables 5 and 6), even with short presentation time of 0.1 s (for PY). No length effect was found for non-migratable words: long words were not more difficult to read than short words if they did not contain a potential of within-word letter migration (Table 3). Furthermore, migratability was a far more important factor for reading than length. Long non-migratable words of 5 and 6 letters were read better than short 4 letter migratable words (93% correct in 5-6 letter non-migratable words vs. 66% in 4-letter migratable words). In addition, some individuals with word-form deficit (pure alexia) show deficits in single letter identification in naming (Arguin & Bub, 1994; Lott & Friedman, 1999; Rapp & Caramazza, 1991). The patients in the current study had good single letter reading even in short presentation, with no deficit in letter form identification or naming, and no errors based on form similarity between letters. Thus, the patients presented in the current study were not word-form dyslexics. COGNITIVE NEUROPSYCHOLOGY, 2001, 18 (8)

689

FRIEDMANN AND GVION

in items that do not require parallel processing, such as numbers (and nonwords). A second important difference is that lexical cues are available for words but not for numbers (except for significant numbers like 1968, 2000, etc.; see Patterson & Wilson, 1990). These two factors might also explain the similar performance of oral reading of nonmigratable nonwords (Experiment 2) and numbers (Experiment 5). Another factor that might have influenced the different performance on numbers and words, especially given that the deficit seems attentional in nature, is that in Hebrew words are read from right to left whereas numbers are read from left to right. (See Seron & Noël, 1992 for a review of contradictory results regarding the parallelism between letters and digits.)16 In what follows we discuss the implications of the reading pattern of the described dyslexic patients for three issues: first, we consider the contribution of lexical knowledge to reading in the absence of letter position information. Then, we discuss the theoretical implications of LPD: We show that most connectionist models for reading are not consistent with our data, and point to the constraints that the data pose on these models. We then suggest an attentional account for the findings that will account both for within word migrations and for the relative resistance to migration of first and last letters.

LPD and lexical knowledge In LPD, very much like in other known dyslexias, there was “an attempt to make lexical sense” (Marcel, 1980). This was demonstrated in several ways in the reading performance of our patients. First, exactly as was found for migration between words (McClelland & Mozer, 1986) and for transpositions within words in normal reading (Estes, 1975), the lexical status of the target string compared to the lexical status of the potential migration responses 16

played a crucial role in string reading: When the string was an existing word without a lexical anagram (nonmigratable word), it was read relatively well. When the string was a nonword with a lexical anagram, performance deteriorated to around 50% correct: When the patients were asked to read a nonword with lexical anagram (like puprle) there was a strong tendency, both in reading aloud and in lexical decision tasks, toward lexicalisation (purple). This reliance on lexical knowledge is manifest in the error pattern in an additional way: The few substitution and addition errors that occurred beside letter migrations mostly created existing words. This ended up in lexicalisations in the case of nonwords, and in word substitutions in the case of words. The relative frequency of the target word and its anagrams affected reading accuracy as well: Patients preferred the more frequent word, and as a result read frequent words better, and tended to read a more frequent transposed word instead of a low-frequency target. Thus, although the letter position deficit probably affects all words to the same degree, letter position deficit, together with the reliance on lexical knowledge, gives rise to a hierarchy of liability to migration errors: The least liable to migration are nonmigratable words, namely words whose letters do not combine to form any other existing word; migratable words are more prone to migrations, especially those with high-frequency anagrams, and migratable nonwords (nonwords whose letters compose an existing word) are the most susceptible to errors. One additional finding that should be accounted for in this regard is that although in nonwords patients did not have lexical knowledge on their side, one of the patients read (nonmigratable) nonwords at the same good level he read existing (nonmigratable) words. This fact can be explained by reference to Prinzmetal’s (1981) and Prinzmetal and Millis-Wright’s (1984) findings regarding illusory conjunctions and perceptual groups.

Similar findings regarding material-specific attentional deficit were also reported for neglect. Costello and Warrington (1987) and Patterson and Wilson (1990) described patients who had neglect (or positional) dyslexia for orthographic material in the presence of general neglect to the opposite side or with no general attentional deficit at all.

690



Prinzmetal and Millis-Wright found that subjects made significantly more illusory conjunction errors (incorrectly integrating a letter form and colour) within words than within nonwords. They accounted for this by Prinzmetal’s principle, which stated that illusory conjunctions are more likely to occur within a perceptual group. Because words are parsed into multiletter perceptual units, and are processed by perceptual units larger than individual letters, and nonwords are processed by individual letter units, illusory conjunctions occur more often within words than within nonwords. Our patient’s good performance in nonword reading can be explained in a similar vein: while his attentional deficit made him fail when he had to correctly integrate several positions to several letters within words, such a failure was prevented in nonwords, since in nonwords attention is allocated separately for every letter, and each time only a single position has to be combined with a single letter.

LPD and connectionist models of reading McClelland and Rumelhart (1981; Rumelhart & McClelland, 1982) proposed an interactive-activation model for word reading. According to this model, at the letter-nodes, letters are already encoded separately in each position within the word. Namely, for each position within a word, there is a full separate set of letter detectors. This model was utilised to account for several dyslexias. Katz and Sevush (1989), for example, suggested that positional dyslexia is caused by selective damage to the activation of specific letter position nodes, in their case the first letter node. Is it possible that the LPD as described here is a deficit to middle-letter nodes? We believe not, for two reasons: First, the absolute position did not play a role here, only the relative position: the third letter could be completely spared when it was the last letter of a three-letter word, but impaired when it was medial in a longer word. Second, unlike in positional dys-

lexia, there were no letter identification errors, only wrong positioning of medial letters. Actually, there is no way a model with positionspecific letter detectors can account for correct perception of letter in a wrong position. The deficit that causes letter migration within words cannot be ascribed to any of the stages of this type of model, because the first stage of letter identification is already position specific. Similarly, McClelland and Rumelhart’s models do not appear to offer any way to accommodate the finding that a word activates a similar word which is a result of medial letter migration more than a similar word which is a result of medial letter substitution. For example, these models cannot account for the fact that the target broad yields more board responses than blind responses, since according to these models a letter in a certain position does not contribute to the activation of a word that contains the same letter in different position. Such a model can only handle within-word migrations by introducing cross-talk between neighbouring letter positions (Peressotti & Grainger, 1995). Some later connectionist models represent words and letter order using letter clusters, or trigrams (BLIRNET: Mozer, 1987; Seidenberg & McClelland, 1989). The relative position within a word is encoded within each trigram and in the trigram combinations. Transposition errors in these models are explained in terms of “spurious activations which involve clusters whose letters are present in the display but in a slightly rearranged order”. However, for a trigram representation model to account for single transposition of adjacent letters, many spurious clusters are required: for example, as CALM and CLAM share only two out of six trigrams, for CLAM to become CALM, four spurious trigrams have to be activated: two different order and two different letter trigrams. This makes trigrams a less natural representation, compared to single letters, to explain within-word migrations17. Furthermore, one of the interesting features of these models (Mozer, 1987, 1991) is that they

17

One exception is PABLO, a programmable blackboard model (McClelland, 1986) that was designed to activate anagrams of the target word especially in medial positions, by encoding medial letters in pairs rather than in trigrams: letters are represented in PABLO as following and preceding another letter (xA, Ax). COGNITIVE NEUROPSYCHOLOGY, 2001, 18 (8)

691

FRIEDMANN AND GVION

produce transposition errors within nonword sequences, but not within existing words. Recall that the LPD readers exhibited a quite different pattern, with impaired position within words and less impairment in nonwords. Thus, additional assumptions must be added to these models in order to account for the pattern manifest by the current findings of transpositions within words. A later proposal of Mozer is perhaps more suitable to account for transposition errors: According to the spatial uncertainty hypothesis (Mozer, 1989), parallel processing encodes only letter identities, whereas focal attention is required to register the position of these letters and bind letters to their location attribute. When focal attention is prevented in normal readers (for example by short exposure duration), letter position errors occur. It therefore might very well be that LPD reading disorder is attention related18.

An attentional account for LPD An account for the deficit in LPD should be one that explains letter migration within words, taking into account the difference between medial and exterior positions, lexicality effects, and frequency effects. We suggest that the impairment lies in the early stage of word processing—the visual analysis of orthographic input, and more specifically in the letter location function of the visual analyser, which is responsible for encoding the relative position of letters within words. Although different views regarding the role of attention in word reading have been suggested, spatial attention appears to play a role in letter localisation (see McCann, Folk, & Johnston, 1992, for a review). According to Treisman and Souther (1986), the location of letters sometimes fails to be registered when attention is overloaded in normal readers. This might also be

the explanation for the deficit of letter location in our dyslexic patients. These patients may suffer an attentional deficit (possibly specific to orthographic material) that prevents them from locating letters within words, or from integrating letters with their relative position-within-word features. As we have suggested, since first and last letters are accessed separately from the rest of the letters, they are correctly located. On the other hand, middle letters are processed together and therefore they are not integrated correctly with their relative within-word position. In the absence of attention to the integration of middle letters with their correct positions, two theoretical possibilities exist for the next stage of reading: illusory position perception or accessing lexical information with partial information. The first possibility is that illusory conjunctions between letters and their positions occur, as a result of which letters are actually perceived in a wrong position. This possibility is somewhat problematic given the lexicality and frequency effects on these conjunctions. (Treisman & Souther, 1986, who also found lexicality effects on migration between words, suggested that lexical status and top-down processes can have an effect on perception and sensory evidence.) The second possibility is that the patients are left with the partial information of free-floating middle letters. With this partial information, they consult the lexicon, and come up with a lexical entry that is adequate for the information—namely a word that starts and ends with the correct letters, and that includes the same middle letters19. This can easily account for both lexicality and frequency effects: The first appropriate item to be retrieved from the lexicon will probably be a frequent, existing word. Manuscript received 14 June 1999 Revised manuscript received 3 August 2000 Revised manuscript accepted 13 April 2001

18

Our findings regarding different liability to migrations of different letter positions pose another type of problem to McClelland and Rumelhart’s models, and actually to every model that assumes equal activation to all letter positions in visual word recognition. These findings join the large body of results that suggest that different letter positions have different weight in word processing (Humphreys et al., 1990; Perea, 1998). 19

See Caramazza and Hillis, 1990; Patterson and Wilson, 1990; and Riddoch, 1990, for related discussions of compensation strategies and the contribution of top-down processing to successful identification of words in neglect dyslexia.

692



REFERENCES Allport, D.A. (1977). On knowing the meaning of words we are unable to report: The effects of visual masking. In S. Dornic (Ed.), Attention and performance VI (pp. 503–533). Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Arguin, M., & Bub, D. (1994). Pure alexia: Attempted rehabilitation and its implications for interpretation of the deficit. Brain and Language, 47, 233–268. Arguin, M., & Bub, D. (1997). Lexical constraints on reading accuracy in neglect dyslexia. Cognitive Neuropsychology, 14, 765–800. Baddeley, A.D. (1990). Human memory: Theory and practice. Boston: Allyn & Bacon. Bat-El, O. (1989). Phonology and word structure in Modern Hebrew. Unpublished PhD. dissertation, UCLA. Bradshaw, J.L., Bradley, D., Gates, A., & Patterson, K. (1977). Serial, parallel, or holistic identification of single words in the two visual fields? Perception and Psychophysics, 21, 431–438. Bradshaw, J.L., & Mapp, A. (1982). Laterally presented words: Orthographic analysis and serial, parallel or holistic modes of processing. Australian Journal of Psychology, 34, 71–90. Caramazza, A., Capasso, R., & Miceli, G. (1996). The role of the graphemic buffer in reading, Cognitive Neuropsychology, 13, 673–698. Caramazza, A., & Hillis, A.E. (1990). Levels of representation, co-ordinate frames, and unilateral neglect. Cognitive Neuropsychology, 7, 391–445. Caramazza, A., & Miceli, G. (1990). The structure of graphemic representations. Cognition, 37, 243–297. Caramazza, A., Miceli, G., Villa, G., & Romani, C. (1987). The role of the graphemic buffer in spelling: Evidence from a case of acquired dysgraphia. Cognition, 26, 59–85. Coltheart, M. (1980). Deep dyslexia: A review of the syndrome. In M. Coltheart, K. Patterson, & J.C. Marshall (Eds.), Deep dyslexia (pp. 22–47). New York: Routledge & Kegan Paul. Coltheart, M. (1981). Disorders of reading and their implications for models of normal reading. Visible Language, 15, 245–286. Costello, A. de L., & Warrington, E.K. (1987). The dissociation of visuo spatial neglect and neglect dyslexia. Journal of Neurology, Neurosurgery and Psychiatry, 50, 1110–1116. Duncan, J. (1987). Attention and reading: Wholes and parts in shape recognition—a tutorial review. In M.

Coltheart (Ed.), The psychology of reading. Attention and performance XII (pp. 39–61). Hove, UK: Lawrence Erlbaum Associates Ltd. Ellis, A.W. (1993). Reading, writing and dyslexia: A cognitive analysis. Hove, UK: Lawrence Erlbaum Associates Ltd. Ellis, A., Flude, B.M., & Young, A.W. (1987). “Neglect dyslexia” and the early visual processing of letters in words and nonwords. Cognitive Neuropsychology, 4, 439–363. Ellis, A.W., & Young, A.W. (1988). Human cognitive neuropsychology. Hove, UK: Lawrence Erlbaum Associates Ltd. Ellis, A.W., Young A.W., & Flude, B.M. (1993). Neglect and visual language. In I.H. Robertson & J. C. Marshall (Eds.), Unilateral neglect: Clinical and experimental studies. Hove, UK: Lawrence Erlbaum Associates Ltd. Estes, W. (1975). The locus of inferential and perceptual processes in letter identification. Journal of Experimental Psychology: General, 104, 122–145. Forster, K.I. (1976). Accessing the mental lexicon. In E.W. Walker & J. Wales (Eds.), New approaches to language mechanisms. Amsterdam: North-Holland. Frost, R. (1992). Orthography and phonology: The psychological reality of orthographic depth. In P. Downing, S.D. Lima, & M. Noonan (Eds.), The linguistics of literacy. Amsterdam/Philadelphia: John Benjamins. Frost, R., Katz, L., & Bentin, S. (1987). Strategies for visual word recognition and orthographical depth: A multilingual comparison. Journal of Experimental Psychology: Human Perception and Performance, 13, 104– 114. Gil, M., & Edelstein, C. (1999). Hebrew version of the PALPA. Ra’anana, Israel: Loewenstein Hospital Rehabilitation Center. Goodglass, H., & Kaplan, E. (1983). Boston Diagnostic Aphasia Examination. Malvern, PA: Lea & Febinger. Grainger, J., & Segui, J. (1990). Neighborhood frequency effects in visual word recognition: A comparison of lexical decision and masked identification latencies. Perception and Psychophysics, 47, 191– 198. Hammond, E.J., & Green, D.W. (1982). Detecting targets in letter and non-letter arrays. Canadian Journal of Psychology, 36, 67–82. Hillis, A.E., & Caramazza, A. (1989). The graphemic buffer and attentional mechanisms. Brain and Language, 36, 208–235. COGNITIVE NEUROPSYCHOLOGY, 2001, 18 (8)

693

FRIEDMANN AND GVION

Humphreys, G.W., Evett, L.J., & Quinlan, P.T. (1990). Orthographic processing in visual word identification. Cognitive Psychology, 22, 517–560. Humphreys, G.W., Evett, L.J., Quinlan, P.T., & Besner, D. (1987). Orthographic priming: Qualitative differences between priming from identified and unidentified primes. In M. Coltheart (Ed.), The psychology of reading. Attention and performance XII (pp. 105–125). Hove, UK: Lawrence Erlbaum Associates Ltd. Humphreys, G.W., & Riddoch, M.J. (1992). Interactions between object and space vision revealed by neuropsychology. In D.E. Meyer & S. Kornblum (Eds.), Synergies in experimental psychology, artificial intelligence, and cognitive neuroscience. Attention and performance XIV. Cambridge, MA: MIT Press. Katz, R.B., & Sevush, S. (1989). Positional dyslexia. Brain and Language, 37, 266–289. Kay, J., Lesser, R., & Coltheart, M. (1992). PALPA: Psycholinguistic Assessments of Language Processing in Aphasia. Hove, UK: Lawrence Erlbaum Associates Ltd. Kertesz, A. (1982). Western Aphasia Battery. Orlando, FL: Grune & Stratton. Kinsbourne, M., & Warrington, E.K. (1962). A variety of reading disability associated with right hemisphere lesions. Journal of Neurology, Neurosurgery and Psychiatry, 25, 339–344. Lambon Ralph, M.A., & Ellis, A.W. (1997). “Patterns of paralexia” revisited: Report of a case of visual dyslexia. Cognitive Neuropsychology, 14, 953–974. Lott, S.N., &. Friedman, R.B. (1999). Can treatment for Pure alexia improve letter-by-letter reading speed without sacrificing accuracy? Brain and Language, 67, 188–201. Marcel, T. (1980). Surface dyslexia and beginning reading: A revised hypothesis of the pronunciation of print and its impairments. In M. Coltheart, K. Patterson, & J.C. Marshall (Eds.), Deep dyslexia. New York: Routledge & Kegan Paul. Marshall, J.C., & Newcombe, F. (1973). Patterns of paralexia: A psycholinguistic approach. Journal of Psycholinguistic Research, 2, 179–199. Marshall, J.C., & Newcombe, F. (1980). The conceptual status of deep dyslexia. In M. Coltheart, K. Patterson, & J.C. Marshall (Eds.), Deep dyslexia (pp. 1–23). New York: Routledge & Kegan Paul. Mason, M. (1982). Recognition time for letters and nonletters: Effects of serial position, array size, and processing order. Journal of Experimental Psychology: Human Perception and Performance, 8, 724–738.

694


McCann, R.S., Folk, C.L., & Johnston, J.C. (1992). The role of spatial attention in visual word processing. Journal of Experimental Psychology: Human Perception and Performance, 18, 1015–1029. McCarthy, J. (1979). Formal problems in Semitic phonology and morphology. PhD dissertation. MIT, Cambridge, MA. McClelland, J.L. (1986). The programmable blackboard model of reading. In J.L. McClelland & D.E. Rumelhart (Eds.), Parallel Distributed Processing: Explorations in the microstructure of cognition, Vol. 2. Psychological and biological models. Cambridge, MA: MIT Press. McClelland, J.L., & Mozer, M.C. (1986). Perceptual interactions in two-word displays: Familiarity and similarity effects. Journal of Experimental Psychology: Human Perception and Performance, 12, 18–35. McClelland, J.L., & Rumelhart, D.E. (1981). An interactive activation model of context effects in letter perception: Part 1, An account of basic findings. Psychological Review, 88, 375–407. McCloskey, M., Badecker, W., Goodman-Schulman, R.A., & Aliminosa, D. (1994). The structure of graphemic representations in spelling: Evidence from a case of acquired dysgraphia. Cognitive Neuropsychology, 11, 341–392. Merikle, P.M., & Coltheart, M. (1972). Selective forward masking. Canadian Journal of Psychology, 26, 296–302. Morton, J., & Patterson, K.E. (1980). ‘Little words— No!’ In M. Coltheart, K. Patterson, & J.C. Marshall (Eds.), Deep dyslexia (pp. 270–285). New York: Routledge & Kegan Paul. Mozer, M.C. (1983). Letter migration in word perception. Journal of Experimental Psychology: Human Perception and Performance, 9, 531–546. Mozer, M.C. (1987). Early parallel processing in reading: A connectionist approach. In M. Coltheart (Ed.), The psychology of reading. Attention and performance XII (pp. 83–104). Hove, UK: Lawrence Erlbaum Associates Ltd. Mozer, M.C. (1989). Types and tokens in visual letter perception. Journal of Experimental Psychology: Human Perception and Performance, 15, 287–303. Mozer, M.C. (1991). The perception of multiple objects: A connectionist approach. Cambridge, MA: MIT Press/ Bradford books. Newcombe, F., & Marshall, J.C. (1981). On psycholinguistic classifications of the acquired dyslexias. Bulletin of the Orton Society, 31, 29–46.


Patterson, K.E. (1979). What is right with “deep” dyslexic patients? Brain and Language, 8, 111–129. Patterson, K.E. (1981). Neuropsychological approaches to the study of reading. British Journal of Psychology, 72, 151–174. Patterson, K.E., & Wilson, B. (1990). A ROSE is ROSE or a NOSE. A deficit in initial letter identification. Cognitive Neuropsychology,7, 447–477. Perea, M. (1998). Orthographic neighbours are not all equal. Language and Cognitive Processes, 13, 77–90. Peressotti, F., & Grainger, J. (1995). Letter position coding in random consonant arrays. Perception and Psychophysics, 37, 875–890. Price, C.J., & Humphreys, G.W. (1993). Attentional dyslexia: The effect of co-occurring deficits. Cognitive Neuropsychology, 10, 569–592. Prinzmetal, W. (1981). Principles of feature integration in visual perception. Perception and Psychophysics, 30, 330–340. Prinzmetal, W., & Millis-Wright, M. (1984). Cognitive and linguistic factors affect visual feature integration. Cognitive Psychology, 16, 305–340. Rapp, B.C., & Caramazza, A. (1991). Spatially determined deficits in letter and word processing. Cognitive Neuropsychology, 8, 275–311. Riddoch, J. (1990). Neglect and peripheral dyslexias. Cognitive Neuropsychology, 7, 369–389. Riddoch, J., Humphreys, G., Cleton, P., & Fery, P. (1990). Interaction of attentional and lexical processes in neglect dyslexia. Cognitive Neuropsychology, 7, 479–517. Rumelhart, D.E., & McClelland, J.L. (1982). An interactive activation model of context effects in letter perception: Part II. The contextual enhancement effect and some tests and extensions of the model. Psychological Review, 89, 60–84. Saffran, E.M., & Coslett, H.B. (1996). “Attentional dyslexia” in Alzheimer’s Disease: A case study. Cognitive Neuropsychology, 13, 205–228. Seidenberg, M.S., & McClelland, J.L. (1989). A distributed, developmental model of word recognition and naming. Psychological Review, 96, 523–568. Seron, X., & Noël, M.-P. (1992). Language and numerical disorders: A neuropsychological approach. In J. Algeria, D. Holender, J. de Morais, & M. Radeau

(Eds.), Analytic approaches to human cognition. Amsterdam: Elsevier. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge, UK: Cambridge University Press. Shallice, T., & McGill, J. (1978). The origins of mixed errors. In J. Requin (Ed.), Attention and performance VII (pp. 193–208). Hillsdale, NJ: Lawrence Erlbaum Associates Inc. Shallice, T., & Warrington, E.K. (1977). The possible role of selective attention in acquired dyslexia. Neuropsychologia, 15, 31–41. Shallice, T., & Warrington, E.K. (1980). Single and multiple component central dyslexic syndromes. In M. Coltheart, K.E. Patterson, & J.C. Marshall (Eds.), Deep dyslexia. London: Routledge & Kegan Paul. Soroker, N. (1997). Hebrew Western Aphasia Battery. Ra’anana, Israel: Loewenstein Hospital Rehabilitation Center. Treisman, A., & Schmidt, H. (1982). Illusory conjunctions in the perception of objects. Cognitive Psychology, 14, 107–141. Treisman, A., & Souther, J. (1986). Illusory words: The roles of attention and of top-down constraints in conjoining letters to form words. Journal of Experimental Psychology: Human Perception and Performance, 12, 3– 17. Van der Velde, F. (1992). Effect of orthographic preactivation on letter migration. Journal of Experimental Psychology: Human Perception and Performance, 18, 449–459. Warrington, E.K. (1991). Right neglect dyslexia: A single case study. Cognitive Neuropsychology, 8, 193– 212. Warrington, E.K., Cipolotti, L., & McNeil, J. (1993). Attentional dyslexia: A single case study. Neuropsychologia, 31, 871–885. Warrington, E.K., & Shallice, T. (1980). Word form dyslexia. Brain, 103, 99–112. Wilson, B., Cockburn, C., & Halligan, P. (1987). The Behavioral Inattention Test. Bury St Edmunds, UK: Thames Valley Test Company. Young, A.W., Newcombe, F., & Ellis, A.W. (1991). Different impairments contribute to neglect dyslexia. Cognitive Neuropsychology, 8, 177–191.


695

FRIEDMANN AND GVION

APPENDIX A Control subjects’ performance Experiment Oral reading migratable words Same-different decision Nonword reading Lexical decision Visual letter location Auditory letter location Position within a sequence Word pair reading

Matched subjects 97% 95% 100% 100% 100% 100% 100% 96%

All control subjects 99% 99% 100% 100% 100% 99% 99% 98%

The control subjects’ errors were migration of medial letters in migratable words. In the word pair reading, errors were position preserving.

696
