A Literature Review and New Data Supporting an Interactive Account of Letter-By-Letter Reading Marlene Behrmann Carnegie Mellon University

Carnegie Mellon University Research Showcase @ CMU

Department of Psychology Dietrich College of Humanities and Social Sciences

1998 A Literature Review and New Data Supporting an Interactive Account of Letter-by-Letter Reading Marlene Behrmann Carnegie Mellon University

David C. Plaut Carnegie Mellon University

James Nelson Carnegie Mellon University

Follow this and additional works at: http://repository.cmu.edu/psychology

This Article is brought to you for free and open access by the Dietrich College of Humanities and Social Sciences at Research Showcase @ CMU. It has been accepted for inclusion in Department of Psychology by an authorized administrator of Research Showcase @ CMU. For more information, please contact [email protected]. COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2), 7- 51

A LITERATURE REVIEW AND NEW DATA SUPPORTING AN INTERACTIVE ACCOUNT OF LETTER-BY-LETTER READING

Marlene Behrmann, David C. Plaut and James Nelson Carnegie Mellon University, Pittsburgh, USA

We present a theoretical account of letter-by-letter reading (LBL) that reconciles discrepant findings associated with this form of acquired dyslexia. We claim that LBL reading is caused by a deficit that affects the normal activation of the orthographic representation of the stimulus. In spite of this lower-level deficit, the degraded orthographic information may be processed further, and lexical, semantic, and higher-order orthographic information may still influence the reading patterns of thesepatients. In support of our position, wepresent a review of 57 published cases of LBL reading in which we demonstrate that a peripheral deficit was evident in almostall of thepatients and that, simultaneously, strong effects of lexical/ semantic variables were observed on reading performance. We then go on to report findings from an empirical analysis of seven LBL readers in whom we document the joint effects of lexical variables (word frequency and imageability) and word length on naming latency. We argue that the reading performance of these patients reflects the residual functioning of the same interactive system that supported normal reading premorbidly.

INTRODUCTION acquired in adulthood, take an abnormally long time to read even single words. This read- Letter-by-letter (LBL) reading is the term used ing deficit is typically associated with a lesion to define the reading pattern of premorbidly in the posterior portion of the dominant hemi- literate patients who, following brain damage sphere and sometimes, but not always, accom-

Requests for reprints to Marlene Behrmann, Department of Psychology, Carnegie Mellon University, Pittsburgh, PA 15213-3890, USA (E-mail: [email protected]). This work was supported by NIMH FIRST awards to MB (MH54246-01) and to DCP (MH55628-01). We thank our colleagues, Dr. Sandra Black, Dr. Graham Ratcliff, and Dr. Jason Barton for referring the patients to us. We also thank Max Coltheart, Martha Farah, Marcel Kinsbourne, Jay McClelland, Marie Montant,Tatjana Nazir, Eleanor Saffran, Tim Shallice, and an anonymous reviewer for their insightful and constructive suggestions.

Ó 1998 Psychology Press Ltd 7 BEHRMANN, PLAUT, NELSON

panied by a lesion of white matter tracts such extent to produce such higher-order or later as the splenium of the corpus callosum or for- effects on performance. Two classes of theory ceps major (Damasio & Damasio, 1983). The have emerged, each of which emphasises one reading performance of such patients is char- of these two paradoxical findings. One class acterised by a “word length effect”, a signifi- argues that the deficit occurs early in process- cant increase in naming latency as a function ing, prior to the activation of an orthographic of the number of letters in a string. Times in the representation, and the early visual deficit ob- order of 1–3 seconds per additional letter in a served in LBL readers is consistent with this string have been measured for some LBL read- view. We will refer to such views as peripheral, ers, although there is considerable variability consistent with the Shallice and Warrington in reading speed across patients (Hanley & distinction in which impairments that ad- Kay, 1996). When the reading deficit manifests versely affect the attainment of the visual in the absence of other reading, writing, or word-form are considered to be peripheral spelling deficits, it is referred to as “pure (Shallice, 1988; Shallice & Warrington, 1980). alexia”. When other written language deficits By contrast, if the impairment is at a later stage, do accompany the LBL reading, they usually the dyslexia is classified as central. With respect consist of surface dyslexia (Bowers, Bub, & to LBLreading, central views maintain that the Arguin, 1996; Friedman & Hadley, 1992; Pat- deficit occurs after the activation of a well- terson & Kay, 1982) or surface dysgraphia specified orthographic description and thus (Behrmann & McLeod, 1995; Rapp & lexical and semantic information can still be Caramazza, 1991). Although less frequent, accessed. there are also reports of at least one case of What is probably evident, even from this deep dyslexia (Buxbaum & Coslett, 1996) and brief description, is that the peripheral and one case of phonological dyslexia (Friedman et central accounts of LBL readings are difficult al., 1993; Nitzberg-Lott, Friedman, & Line- to reconcile. Peripheral views cannot readily baugh, 1994) accompanying LBL reading. account for the lexical/ semantic findings and Two critical empirical findings concerning the central views do not explain the impaired LBL reading pose difficult challenges for theo- early visual processing in these patients. In the ries of this disorder. One finding is that these present paper, we will argue that both sets of patients are impaired at letter processing. A empirical findings can be accommodated second important finding is that some of these within a single, interactive reading system to patients have available to them lexical and which both hemispheres contribute. We first semantic information about the stimulus, as present in detail the findings for a low-level evidenced in their above-chance performance deficit in LBLreaders and then we consider the on lexical decision and semantic categorisation details of the lexical and semantic findings. tasks. This latter finding suggests that the vis- Thereafter, we examinenot only the peripheral ual stimulus has been processed to a sufficient and central accounts but also views that incor-

8 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

porate some aspects of both. Finally, we pre- Levine & Calvanio, 1978; Patterson & Kay, sent our account and, to substantiate it, we 1982; Schacter, Rapcsak, Rubens, Tharan, & review most of the published cases of LBL Laguna, 1990; see also Miozzo & Caramazza, reading and present new empirical data from this issue). Still others have maintained that seven LBL readers. LBL reading arises from insufficient attentional resources, thus forcing a serial, left–right strategy (Rapp & Caramazza, 1991), or from a problem in capacity or in switching visual at- KEY EMPIRICAL FINDINGS tention (Buxbaum & Coslett, 1996; Price & Humphreys, 1992). In a recent paper, one of us (Behrmann & Shallice, 1995) argued that the Impairment in Early Visual Processing core deficit is one of letter processing and that It is now well established that patients with the time to activate the representation for even LBL reading do not activate orthographic rep- a single letter is slow. Importantly, we argued resentations adequately; for example, many that the processing deficit is not spatially de- patients are adversely influenced by altera- termined, i.e. processing does not proceed tions of the surface characteristics of the stimu- from left to right of the array; rather, it has to lus (for example, poorer reading of script than do with serial order such that letters appearing print) and many make mostly visual errors in later in a string (even when all letters are pre- reading (for example, JAY ® “joy”). There are, sented at fixation in an RSVP paradigm) are however, several different explanations for the disadvantaged relative to letters appearing impairment in activating orthographic repre- earlier in the string. sentations. Some studies have claimed, for ex- It is important to note that many of these ample, that these patients suffer from a general peripheral views are not necessarily mutually perceptual deficit that impairs all forms of vis- exclusive. For example, although the critical ual processing (Behrmann, Nelson & Sekuler, deficit may be one of disordered perceptual 1998; Farah, 1991; Farah & Wallace, 1991; processing, an obvious consequence of such a Friedman & Alexander, 1984; Sekuler & deficit is an impairment in rapid and accurate Behrmann, 1996), although letter and word letter processing and identification (Behrmann recognition might perhaps be especially vul- & Shallice,1995; Sekuler & Behrmann, 1996). nerable (Farah, 1997). Others have claimed that What is central to all these peripheral views the impairment is specific to orthography and (labelled under the heading orthographic ac- impairs the identification of letters per se (Ar- cess view, [Bowers et al., 1996]), however, is guin & Bub, 1993; Karanth, 1985; Kay & Han- that the fundamental impairment is one of vis- ley, 1991; Reuter-Lorenz & Brunn, 1990), or ual processing, arising relatively early and pre- affects the rapid processing of multiple forms venting the derivation of an adequate in parallel (Kinsbourne & Warrington, 1962; orthographic representation.

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 9 BEHRMANN, PLAUT, NELSON

Numerous investigators have commented mented in several fairly early reports of LBL that problems in letter-processing tasks are so patients (Albert, Yamadori, Gardner, & common in LBL reading that impaired letter Howes, 1973; Caplan & Hedley-Whyte, 1974; identification is likely to underlie their reading Kreindler & Ionescu, 1961); even though the deficit (Coltheart, 1981; Patterson &Kay, 1982). patients in these studies could not identify a In support of this, Patterson & Kay showed written word overtly, they were still able to that all four of their patients made letter iden- match this target with a word spoken by the tification errors, albeit to varying degrees. Pa- investigator, or with a visually presented ob- tient CH, for example, identified correctly only ject. These initial observations of implicit or 16/ 26 upper-case and 10/ 26 lower-case letters covert reading abilities of LBL readers in the and chose the odd letter out (for example, f F absence of explicit identification have been up- E) correctly on only 75% of the trials. Patient held in a number of more recent studies. For TP, on the other hand, identified 25/ 26 lower- example, Shallice and Saffran (1986) reported case letters and made no errors on cross-case that their patient, ML, was above chance at matching (e.g. D d). Interestingly, the types of performing lexical decision and semantic cate- errors made by all the patients reflected the gorisation tasks with stimuli presented too visual similarity between the target and the briefly to permit overt identification. Several error, suggesting that visual feature overlap is other studies have similarly shown that their a major factor in letter misidentification (see patients can perform lexical decision (Bub & also Perri, Bartolomeo, &Silveri, 1996). Consis- Arguin, 1995) as well as semantic classification tent with the findings of letter misidentifica- of words (for example, living vs. nonliving) at tion, Behrmann and Shallice (1995) maintained exposure durations too brief for the patients to that there is no convincing evidence of normal have identified the target items explicitly (for letter processing in any LBL reader. They then example, Bub & Arguin, 1995; Howard, 1991). posed the challenge that, unless the hypothesis In the largest series, Coslett and colleagues of impaired letter processing was found want- (Coslett & Saffran, 1989a, 1994; Coslett, Saf- ing, this should constitute the default explana- fran, Greenbaum, & Schwartz, 1993) have de- tion for the reading impairment in patients scribed five patients who performed well with LBL reading. above chance on lexical decision and semantic categorisation tasks with the very same stimuli they could not identify explicitly (see also Saf- Lexical and Semantic Effects on Reading fran & Coslett, this volume). Coslett and col- Give this evidence, it is particularly puzzling leagues also described two additional patients that some LBL readers, even if they cannot who fit the definition of optic aphasia and who explicitly identify the stimulus, can nonethe- were completely unable to name letters (or any less demonstrate some lexical and semantic other stimuli from visual presentation). Even information about it. This result was docu- in the absence of letter naming, these two pa-

10 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

tients were fairly successful at lexical decision plicit tasks that do not require explicit process- and binary categorisation tasks (Coslett & Saf- ing, such as semantic categorisation or prim- fran, 1989b, 1992). ing, the contents of this system are not available for overt report. This view has probably fallen out of favour for a number of rea- THEORIESOF LETTER-BY-LETTER READING sons, including the fact that it cannot account for the existence of the early visual processing The two empirical findings of an early deficit impairment and that there is no explanation and the influence of later lexical and semantic for the hallmark feature of this problem, the properties of the stimulus on performance are letter-by-letter reading itself or the increase in now both well documented in the domain of naming latency with word length. LBL reading. As mentioned previously, there A second central view argues for a visual- are two main classes of explanation of LBL verbal disconnection, i.e., that the visual areas reading which differ in the extent to which involved in reading are anatomically and/ or they emphasise one or the other of these two functionally disconnected from the more se- findings. The peripheral view argues that the mantic/ conceptual areas. The best example of deficit occurs early in processing, consistent this view is from Déjerine who, in his famous with the early visual deficit observed in these 1891/ 1892 case studies (for overview, see Bub, readers. Proponents of this view have focused Arguin, & Lecours, 1993), interpreted the LBL on the letter-processing deficit and its under- syndrome as a disconnection of the visual ver- lying mechanism without paying much atten- bal input from “the visual memory centre for tion to the later, lexical and semantic effects. words”, which is located in the left angular The central view argues that the deficit occurs gyrus (Geschwind, 1965; Greenblatt, 1973; only after (or at least does not prevent) the Speedie, Rothi, & Heilman, 1982). Bowers et al. activation of a well-specified orthographic de- (1996a; also Arguin, Bub, & Bowers, this issue) scription and, thus, lexical and semantic infor- have adopted a similar perspective and main- mation can still be assessed. Within the central tained that the disconnection arises only after view there are two different accounts. One orthographic word representations have been account, although perhaps the less favoured at activated; thus the disconnection is between present, is that the patients have an intact read- orthographic representations (logogens) on ing system and can activate lexical and seman- the one hand and phonological codes on the tic knowledge normally, but that the output of other. This disconnection delays (or precludes) this intact system is disconnected from con- access to the phonological code for naming, sciousness (Schacter, McAndrews, & while leaving lexical decision, semantic cate- Moscovitch, 1988; Young & Haan, 1990). Thus, gorisation, and the word superiority effect in- although subjects can process the information, tact. In support of their disconnection account, and the results may be revealed through im- they showed that the reading performance of

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 11 BEHRMANN, PLAUT, NELSON

a LBL reader, IH, was facilitated by orthog- pretation, the lesion in pure alexia prevents raphically related primes (e.g. GATE-gate) but, visual information from both cortices from ac- unlike normal readers, not by homophonically cessing lexical and semantic representations in related primes (e.g. gait-gate) (Arguin, Bub, & the left hemisphere, but this information can Bowers, this issue). IH also showed no effect of still support the (albeit limited) reading capa- phonemic neighbourhood size on word recog- bilities of the right hemisphere. Consistent nition performance. Given that IH shows no with the right-hemisphere view, Coslett et al. evidence for covert phonological activation, (1993) showed that, as the LBLpatients became the functional site of the lesion, according to able to engage in covert reading, explicit rec- this account, lies between orthographic and ognition performance was influenced by word phonological processing (but see Montant, imageability and grammatical class,properties Behrmann, &Nazir, 1998, for demonstration of often associated with the reading skills of the phonological priming in an LBL reader). right hemisphere (Coltheart, 1980, 1983). In summary, both the peripheral and central Aparticularly appealing aspect of this right- explanations of LBL reading do well at ac- hemisphere view is that the later lexical and counting for the subset of empirical findings semantic effects observed in LBL readers are on which they focus. The limitations of most of similar to those generated by the right hemi- these accounts, however, is that they do not sphere of commissurotomy patients as well as simultaneously accommodate both theperiph- by left-hemispherectomy patients (Coltheart, eral and central aspects of LBL reading. One 1983). In fact, a similar explanation has been exception to this is the view of Coslett and proposed by some to account for the pattern of Saffran, which takes into account both of these reading in patients with deep dyslexia aspects (Buxbaum & Coslett, 1996; Coslett & (Coltheart, 1980, 1983; Saffran, Bogyo, Saffran, 1994; Saffran & Coslett, this issue). Schwartz, & Marin, 1980). These authors have suggested that there is a Although the right-hemisphere account of deficit (most likely peripheral and early) in the LBLreading takes into account both its periph- normal reading system in these patients and eral and central aspects, it makes the very spe- that the sequential output pattern typically as- cific assumption that these different aspects sociated with LBL reading is mediated by the reflect entirely distinct modes of reading and left hemisphere. The covert reading and later are mediated by different hemispheres. This lexical/ semantic effects do not depend on the assumption would seem to imply that the nor- inaccessible (or degraded) word form gener- mal modes of operation of the two hemi- ated by the left hemisphere. Rather, a separate spheres are altered in LBL reading, with the reading mechanism, subserved by the intact right hemisphere playing an increased role right hemisphere, is responsible for the find- relative to its normal contribution. However, ings of preserved lexical decision and semantic the exact relationship between the roles of the categorisation. On this right-hemisphere inter- left and right hemisphere in LBL reading and

12 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

their roles in normal reading has never been each other whereas inconsistent alternatives made explicit on the right-hemisphere ac- (e.g. the words MAKE and TAKE) mutually count. By contrast, we present an alternative inhibit each other. The critical properties of the theory in which both the early and late effects model for our purposes are that processing is in LBL reading arise as a result of a peripheral cascaded and interactive. Cascaded processing impairment to the normal word reading sys- means that partial results at each level, in the tem that is supported by both hemispheres. form of intermediate levels of activation, propagate to other levels immediately and continuously, rather than waiting until proc- AN INTERACTIVE ACCOUNTOF LETTER-BY-LETTER essing at lower levels is complete. Interactive READING processing means that activation not only propagates from lower to higher levels, but One possible reaction to the controversy be- that the activation at higher levels feeds back tween the peripheral and central accounts of to lower levels to provide additional support LBL reading is to argue that reconciliation is for those lower-level elements that are consis- unnecessary and that a host of different under- tent with the higher-level activation. Thus, cas- lying deficits may give rise to this problem. caded, interactive processing causes early Moreover, individual differences could also letter activation to feed forward and partially arise from the different compensatory strate- engage word representations, which in turn gies adopted by the patients (Price & Hum- feed back to the letter level to influence sub- phreys, 1992, 1995). On this view, no uniform sequent processing. account of letter-by-letter reading exists and We should point out that we are adopting none is needed. In this paper, however, we the IAM as a framework for explaining LBL argue that a reconciliation between these vari- reading not out of any commitment to the lo- ous positions is both desirable and possible calist word representations it contains, but be- and that LBL may be accounted for by a single, cause it provides a clear instantiation of the unifying view of the normal word reading sys- principles of cascaded, interactive processing. tem. The same principles apply within distributed Our account takes as its starting point the accounts of lexical processing (e.g. Plaut, framework of the Interactive Activation Model McClelland, Seidenberg, & Patterson, 1996; of letter and word perception (hereafter IAM; Seidenberg & McClelland, 1989; Van Orden, McClelland & Rumelhart, 1981; Rumelhart & Pennington, & Stone, 1990) and, as we expli- McClelland, 1982). The model contains three cate below, essentially the same account of LBL levels of processing units— letter features, let- reading holds within such models. ters, and words— such that elements and their We claim that the fundamental impairment components (e.g. the word MAKE and the let- in LBL reading, following an occipital lesion, is ter M in the first position) mutually support a general perceptual deficit that degrades the

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 13 BEHRMANN, PLAUT, NELSON

quality of visual input. In the IAM, this deficit LBL readers can improve perceptual process- can be conceptualised as damage to the letter ing by rescaling covert attention from the en- feature level or between this level and the letter tire word to apply successively to letters level. The impact of this perceptual impair- within the word1. ment on word recognition is that it permits Even though the word-level activation pro- only weak or partial parallel activation of the duced by the initial, weak, parallel letter acti- letters in a word. This weak activation does not vation is insufficient to support explicit suffice for explicit identification of the word identification, it nonetheless would be ex- (i.e. no word unit achieves a sufficiently high pected to activate the correct word to a greater level of activation to exceed the response extent than its competitors, and to produce threshold) and the system must resort to se- more activation overall than would be pro- quential processing to enhance the activation duced by a nonword (see McClelland & Ru- of individual letters. Critically, this type of melhart, 1981). Thus, assuming this lexical sequential processing is not an abnormal strat- activation propagates further into a semantic egy only employed following brain damage, system (not implemented in the IAM, but see but is the manifestation of the normal reading McClelland, 1987), during the course of the strategy of making additional fixations when sequential processing, activation from individ- encountering difficulty in reading text (Just & ual letters propagates into the system and adds Carpenter, 1987; Rayner &Pollatsek, 1987). For to the cumulative activation at the word level. example, normal subjects fixate more fre- Concurrently, this word-level activation feeds quently in a long word than in a short word in back to the letter level to facilitate subsequent order to enhance the quality of the stimulus recognition of the word’s letters. The strength (O’Regan & Levy-Schoen, 1989). LBL readers of this top-down support is a function of the also fixate more frequently; in fact, given the degree of partial world-level activation. We very poor quality of the visual input, they fix- assume that, in a more general system includ- ate almost every letter (Behrmann, Barton, & ing semantics, the degree of higher-level acti- Black, 1998), giving rise to the hallmark word vation would scale not only with frequency (as length effect. Presumably these fixations aid in the IAM) but also with imageability, such performance by permitting the increased spa- that words of higher frequency or imageability tial resolution of the fovea to be applied to would be more active, and hence provide multiple locations within the word. In fact, stronger top-down support for letter activa- even in the absence of overt saccades, a word tions, than would words of lower frequency or length effect would be expected, given that imageability. Consequently, the system would

1 Although the IAM does not incorporate mechanisms for overt or covert shifts of attention, the model could be extended to include such a mechanism (see, e.g. Hinton, 1990), and other word reading models that are consistent with cascaded, interactive processing do contain such mechanisms (e.g. Plaut, McClelland, & Seidenberg, 1995)

14 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

converge more quickly on responses for high- down support that facilitates subsequent imageability and high-frequency items lower-level processing. This account provides compared with low-frequency and low-im- a unified explanation of both the early visual ageability items. and later lexical and semantic findings in LBL A further determinant of the degree of the reading. strength of top-down support is the time over In fact, an interactive theory of this general which higher-level activation can accumulate, form has already been proposed to account for and it is this time factor that gives rise to an another type of acquired peripheral dyslexia, interaction between these lexical variables and neglect dyslexia (Behrmann, Moscovitch, word length. Given that longer words take Black, & Mozer, 1990; Mozer & Behrmann, longer to process in LBL reading, their higher- 1990). According to this theory, neglect dys- level representations have longer to integrate lexia arises from a low-level deficit in spatial bottom-up support and, therefore, produce attention that affects the bottom-up processing stronger top-down effects on performance. of one-side (typically the left) of a visual stimu- Thus, there is more opportunity for frequency lus. Given that reduced attention does not and imageability to influence the recognition completely filter out information, but only of seven-letter words compared with three-let- lowers the likelihood that it is propagated into ter words. Indeed, even in normal subjects, the system, the unattend ed information on the there is a significant interaction of word length neglected side of the stimulus may still engage and frequency; Weekes (1997) found that, higher-level processes, albeit to a lesser extent whereas, there was no effect of word length than the non-neglected information. Provided on the naming latency of high-frequency that the attentional deficit is not too severe, words, there was a small effect for low-fre- stimulus information from the neglected side quency words and a more marked effect for may be processed sufficiently to engage nonwords. higher-level (lexical / semantic) repre- The essence of our account, then, is that LBL sentations. Thus, factors such as lexical status readers make use of the same cascaded, inter- and morphological composition can still influ- active system as normal readers but are ence performance, leading to, for example, bet- prompted to resort to sequential processing ter reading of words compared with nonwords more often (manifest either as multiple eye and better reading of compound words com- movements or shift of covert attention) to compared with two single, unrelated words pensate for the degradation in visual input (Behrmann et al., 1990; Brunn & Farah, 1991; following the left occipital lesion. Nonetheless, Làdavas, Umiltà, & Mapelli, 1997; Sieroff, Pol- the weak, parallel activation from this input latsek, & Posner, 1988). Thus, as in LBL read- propagates to higher levels of the system to ing, both “early” effects (e.g. influence of engage lexical/ semantic representations par- stimulus size, horizontal position, and word tially, and these representations provide top- length) and “late” effects (e.g. influences of

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 15 BEHRMANN, PLAUT, NELSON

lexical status and morphology) may be ob- (thereby reflecting its lexical properties to a served in the same patient. degree) but may not settle it completely (which To this point, we have cast our views in would correspond to explicit recognition). terms of the IAM, in which the critical lexical Thus, the partial activation and competition effects are mediated by word-specific (localist) among word units in the localist IAM corre- processing units. However, as alluded to ear- sponds to the partial movement of activity lier, this choice is to a large extent merely ex- patterns towards word attractors within dis- pository. The same lexical effects can be recast tributed systems. Yet both types of systems within models of lexical processing employing employ cascaded, interactive processing such distributed representations (e.g. Seidenberg & that partially degraded orthographic input is McClelland, 1989; Plaut et al., 1996; Van Orden propagated throughout the system to engage et al., 1990). In a distributed representation, these “lexical” effects to varying degrees. Thus, words are encoded by distinct but overlapping we claim that our account of LBL within the patterns of activity, such that each word is framework of the IAM also holds for distrib- represented by the activity of many units and uted connectionist models of lexical process- each unit participates in representing many ing. words. Within a distributed system, lexical In fact, many of the effects that we ascribe to knowledge— the fact that certain patterns of top-down activation from word units in the activity but not others correspond to familiar IAM have been demonstrated in existing simu- stimuli— is reflected not in the structure of the lations of distributed attractor networks. For system but in the dynamics of how the units example, Hinton and Shallice (1991) demon- interact. Specifically, lexical knowledge in the strated relative sparing of categorisation perform of learned weights on the connections formance with poor explicit identification between units causes the activity pattern cor- following lower-level damage to an attractor responding to each word to form an attractor. network that was trained to map orthography What this means is that, when the system is in to semantics. Plaut and Shallice (1993) repli- an unfamiliar pattern of activity— one that cated this finding in an attractor network that does not correspond to a known word— inter- mapped orthography to phonology via seman- actions among units alter this pattern so that it tics, and also showed that, for the words the moves towards and ultimately settles to the network failed to read correctly following nearest (most similar) familiar attractor pat- damage near orthography, it was nonetheless tern. A critical property of attractor dynamics well above chance (d’ = 1.80)2 at lexical deci- for our purposes is that they can be partial; an sion for these words. Plaut and Shallice also activity pattern that is sufficiently far from the demonstrated in a similar network that high- nearest attractor may still be pulled towards it imageability words develop stronger semantic

2 This d’ value is reported incorrectly as 1.36 by Plaut and Shallice (1993, p. 445).

16 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

attractors than low-imageability words— as re- input to the same cascaded, interactive sys- flected by their relative robustness to dam- tems that supported normal reading premor- age— if imageability is instantiated in terms of bidly. Although we do not question the fact the relative richness of a word’s semantic rep- that the right hemisphere has some language resentations (i.e. how many semantic features and reading abilities (Beeman & Chiarello, are accessed consistently across contexts; also 1997; Michel, Henaff, & Intriligator, 1996; Var- see Plaut, 1995). The stronger semantic attrac- gha-Kadem et al., 1997), our claim is that the tors for high-imageability words would natu- lexical and semantic findings in LBL readers rally lead to stronger top-down effects on do not arise solely from the more primitive orthographic representations, although this right-hemisphere reading system. Rather, was not tested directly. And finally, in conjunc- these higher-level effects emerge from the re- tion with the interactive theory of neglect dys- sidual workings of the interactive reading sys- lexia mentioned earlier, Mozer and Behrmann tem that involves both hemispheres, governed (1990, 1992) simulated the co-occurrence of by the same computational principles. many of the lower-level perceptual effects and What type of evidence would support our higher-level lexical/ morphological effects ob- interactive account of LBL reading? First, all served in neglect dyslexia by damaging input patients should have an early deficit of some to the attentional mechanism within MORSEL form that prevents the rapid and reliable acti- (Mozer, 1991), a network model that imple- vation of orthographic information. Second, ments attractors for words using a “pull-out” variables considered to be diagnostic of later network over letter-cluster units. lexical processes should also influence the In contrast with the right-hemisphere ac- reading performance of these patients. In sup- count of LBL reading (Buxbaum & Coslett, port of our position, in this paper we present a 1996; Coslett & Saffran, 1994; Saffran &Coslett, review of the existing literature as well as new this issue), our view does not specifically im- empirical data obtained from seven LBL read- plicate a particular hemisphere as the locus of ers. As the findings will indicate, both the lit- the lexical and semantic effects. We have ar- erature review and the empirical data are gued that the simultaneous presence of visual consistent with our position, showing that and lexical/ semantic findings arise from a sin- early visual deficits are identifiable in almost gle interactive system, involving both hemi- every patient (and not demonstrably absent in spheres, and that there is no compelling reason the remaining cases) and that, when they have to invoke the separate right hemisphere as the been investigated carefully, strong effects of sole(or even primary)mediator of the postlexi- lexical variables are also observed. That a par- cal effects. Importantly, this view entails that, tial low-level deficit can permit partial higher- in LBL readers, the reading system per se is level activation (supporting lexical decision unchanged from its premorbid state; reading and categorisation performance) that further behaviour arises as a consequence of degraded influences lower-level processing (producing

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 17 BEHRMANN, PLAUT, NELSON

frequency and imageability effects on reading Hedley-White, 1974; Kreindler & Ionescu, latencies) is, on our view, a direct consequence 1961), although we have generally made refer- of the cascaded, interactive nature of the nor- ence to these papers elsewhere in the text. We mal reading system. are also aware of posters of LBL readers presented at conferences and have not included those (with the exception of the Vigliocco, Se- REVIEW OF PUBLISHEDCASESOF menza, & Neglia, 1992, because the description LETTER-BY-LETTER READING of the patients is sufficiently detailed). We have also come across a few papers published This review of the published studies of 57 LBL in other languages (El Alaoui-Faris et al., 1994), readers was undertaken both to document the but have restricted our analysis to published existence of early deficits and to examine the English papers. Finally, to make the table as extent to which there is higher-level processing comprehensive as possible, we have included in these patients. Although this review is in- our own patients at the end. To do so, we have tended to be as comprehensive as possible (see combined the findings from the empirical Table 1), we have deliberately excluded sev- analyses we describe in the latter part of this eral papers. Many of the excluded papers have paper with previous descriptions of these pa- a different focus and do not provide sufficient tients’ reading performance. detail for our analysis; for example, some papers focus on the anatomical aspects of the case Procedure rather than on the reading performance per se (Ajax, 1967; Damasio & Damasio, 1983; Green- For this analysis, we reviewed the published blatt, 1973), others describe aspects of the pa- papers to determine whether there was any tient’s performance such as the intact reading evidence for a peripheral deficit in the patient. of stenography (Regard, Landis, & Hess, 1985), If the authors of a paper classified the patient’s an associated deficit in music reading (Horik- deficit as occurring early (peripherally or oshi et al., 1997), or associated colour deficits prelexically), this was counted as positive evi- (Freedman &Costa, 1992), which are unrelated dence. We also took as positive evidence a to the issue at hand, and yet other papers re- reported perceptual difficulty in single-letter port a rehabilitation procedure for the patient processing or identification or a more general without including much detail on the patients’ perceptual deficit in which other visuopercep- pre-therapy reading performance (Kashiwagi tual abilities are impaired, even if the authors & Kashiwagi, 1989; Tuomainen & Laine, 1991). did not classify the patient in the peripheral There are also a small number of papers that group per se. We also determined whether are not included simply because the descrip- there was evidence for lexical and semantic tion of the patient’s reading is unclear or insuf- effects in these same LBL readers. This con- ficiently detailed for our purposes (Caplan & sisted of reviewing the reported papers and

18 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

determining whether there was positive evi- Although these other variables are relevant to dence for: (1) covert reading under brief expo- the issue under investigation, reports of this sure (either in lexical decision or semantic were too few to warrant a separate category. categorisation tasks), (2) a word superiority Because this analysis is retrospective, there effect, (3) effects of frequency, imageability, are obviously a number of problems. In many regularity, and part-of-speech on naming cases, the critical variables are not tested (or words, and (4) effects of frequency and image- perhaps tested but not reported or analysed ability on lexical decision. With regard to the statistically). Even when they are tested sys- word superiority effect, we simply noted the tematically, the dependent variable is often a better report of items in words over nonwords measure of accuracy rather than reaction time. in any experimental paradigm (not solely the Yet the effects in LBL reading are typically standard Wheeler-Reicher type task) and did more robust (and perhaps only evident) in re- not take special note of a pseudoword effect action time, given the high degree of accuracy (better report of items in legal pseudowords in many cases. We have not differentiated be- over illegal nonwords). tween the various dependent measures and In analysing the effects of lexical variables simply note the presence of the main effect such as frequency, imageability etc., we con- using either metric. sidered their impact on naming and on lexical decision under both brief and prolonged expo- Results sure duration. We have only documented whether these variables significantly influence Evidence for a Peripheral Deficit the patient’s reading performance (i.e. as a Many researchers have previously observed main effect) in either reaction time or accuracy. that LBL readers invariably perform poorly on Very few studies present the interaction be- tests of letter recognition, although, to date, tween these different variables and word there has been no substantiation of this claim. length (see Doctor, Sartori, & Saling, 1990, for Almost all the patients we reviewed had some an exception) so we do not consider the inter- noticeable problem with letter processing, as is actions in our tabulation, although we consider evident from the first column of Table 1. In- it to be crucial for our interactive view. Finally, deed, 50 of the 57 patients have a positivecheck some studies (for example, Buxbaum & mark in the first column. Although some of Coslett, 1996; Coslett & Monsul, 1994; Coslett these patients were highly accurate on naming & Saffran, 1989a; Coslett et al., 1993; Doctor et single letters for an unlimited exposure, a find- al., 1990; Friedman et al., 1993; Shallice & Saf- ing which might suggest no obvious impair- fran, 1986) assess the reading of other higher- ment, the time to name the individual letters is order lexical, orthographic, or semantic often abnormally long. For example, DR(Doc- variables, for example a comparison of per- tor et al., 1990) correctly identified 24/ 26 formance on suffixed/ pseudosuffixed words. lower-case and 25/ 26 upper-case letters, but

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 19 2 BEHRM 0 Table 1. Review of Published Cases of Letter-by-letter Reading CO

Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð A NN,

G Brief Exp Naming LD NITIV

——————————————— —————————— –————— PLA E Patient Study PL LD SC WSE Fr Im Rg PS Fr Im UT, NEURO

———————————————————————————————————————————————————————————————— NELSO DM Arguin & Bub (1993, 1994a,b) ° ° ° ° ° – – – ° – PSY Bub & Arguin (1995) N

CHO EL Behrmann et al. (1998) ° ± ± – ° ° – – – –

LO Montant et al. (1998)

G IH Bowers, Arguin, & Bub (1996) ° – – ° ° – ° – – – Y ,

1998, Bowers, Bub, & Arguin (1996) Arguin et al. (this issue)

15 15 JV Bub, Black, & Howell (1989) ° – – ° ± – – – – –

(1/ n/ a Buxbaum & Coslett, (1996) ° – – – ± ° ± ° – –

2) 1 Coslett & Saffran (1989a) ° ° ° – + ° – ° ° ± 2 Coslett & Saffran (1989a) ° ° ° – – – – + ° ± 3 Coslett & Saffran (1989a) ° ° ° – ° ° – ° ° ± 4 Coslett & Saffran (1989a) ° ° ° – – – – – ° – JWC Coslett et al. (1993) ° ° ° – – – – – ° ° DR Doctor et al. (1990) ° ° ° – – ° – – – – RE Feinberg et al. (1995) ° ° ° – ± ± – – ° ° n/ a Friedman & Alexander (1984) ° – ± – – – – ± – – BL Friedman & Hadley (1992) ? – – ° ° ± ° ± – – TL Friedman et al. (1992) ° – – – – – – ° – – n/ a Grossi et al. (1984) ° – + – – – – – – – DC* Hanley & Kay (1996) ° – – ° – – – – – – n/ a Henderson et al. (1995) ° – – – – – ± – – – PM Howard (1991) ° ± ± – – ± ± – – – KW Howard (1991) ° ± ° + ° ° ± – – – PD Kay & Hanley (1991) ° – – ± ° ° – – – – Hanley & Kay (1992, 1996) L Kinsbourne & Warrington (1962) ° – – – + – – – – – P Kinsbourne & Warrington (1962) ° – – – + – – – – – S Kinsbourne & Warrington (1962) ° – – – + – – – – – C Kinsbourne & Warrington (1962) ° – – – + – – – – – n/ a Landis et al. (1980) ± – + – – – ° ° – – ON Lazar & Scarisbrick (1993) ° – – – – – – – – – CC Levine & Calvanio (1978) ° – – ° – – – – – – RT Levine & Calvanio (1978) ° – – ° – – – – – – GC Levine & Calvanio (1978) ° ± # – – – – – – – – CP Montant, Nazir, & Poncet (this issue) ° – – – – – – – – – MW Patterson & Kay (1982) ° ± ± – – – – – – – CH Patterson & Kay (1982) ° ± – – – – – – – – TP Patterson & Kay (1982) ° ± ± – – – – – – – KC Patterson & Kay (1982) ° – ± – – – ° – – – SP Perri et al. (1996) ° – – – ° – ° – – – EW Price & Humphreys (1992) ° ± ± – ° – + – – – HT Price & Humphreys (1992) ° ± ± – + – ± – – – SA Price & Humprheys (1995) ° ± – – – – – – – – WH* Price & Humphreys (1995) ° ± – – – – – – – – HR* Rapp & Caramazza (1991) ° – ± – ° ± – – – – PT Rapcsak et al. (1990) ? – – – – – ± ± – – Schacter et al. (1990) – – – – – – – – WL Reuter-Lorenz & Brunn (1990) ° – – ° – – ± – – – ML Shallice & Saffran (1986) ° ° ° – – – – – ° – n/ a Stachowiak & Poeck (1976) ° – – – – – – – – – BY Staller et al. (1978) ° – – – – – – – ° ° BD Vigliocco et al. (1992) ? ± – ° – – – – – – RG Vigliocco et al. (1992) ? ± – ° – – – – – – RAV Warrington & Shallice (1980) ? – ± ± – ± – – – – CO JDC Warrington & Shallice (1980) ?

G – – – – – – – – – NITIV Our Patients E

NEURO DS Behrmann, Black & Bub (1990) ° – – ° ° ° – – – – Behrmann & Shallice (1995)

PSY Sekuler & Behrmann (1996) PC ° ° ±

CHO – – – – – – – MW Sekuler & Behrmann (1996) ° – – – ° ° – – – – LETTER-BY LO DK ° + + – ° ° – – – – G Y

, MA Sekuler & Behrmann (1996) ° – – – ° ° – – – – 1998, IS Behrmann & McLeod (1995) ° – – ° ° ± – – – – TU Farah & Wallace (1991) ° – – – ° ° ± – – – -LETTER 15 Sekuler & Behrmann (1996) (1/ —————————————————————————–—————————————————————————————————————–— 2) REA PL = Prelexical; LD = Lexical Decision; SC = Semantic Category; WSE= Word Superiority Effect; Fr = Frequency; Im = Imageability; Rg = Regularity;

# DING 2 PS = Part of Speech; ° = evidence for; ± = evidence against; + = some evidence for; ? = unclear; –= not tested; * = left-handed; = ambidextrous. 1 BEHRMANN, PLAUT, NELSON

the time to do so was 2.03 sec and 1.75 sec, at 100% on word stimuli), performance respectively. Patients PD and DC (Hanley and dropped below 50% for items in position 4 of Kay, 1996; Kay & Hanley, 1991) also made a words (and close to floor for items in position relatively small number of letter misidentifica- 4 of nonwords). A similar account of preserved tions (26/ 26 and 25/ 26 lower-case letters, re- accuracy but abnormal speed of processing spectively) but their speed of processing was may account for the patient reported by Landis dramatically slowed relative to a control sub- et al. (1980). Although this young man accu- ject. This slowing was particularly evident rately identified single letters, this was done so when the two patients performed a letter- hesitantly, especially when the letters were in- matching task and made same/ different termixed with numbers (see similar observa- judgements on two letters (e.g. aR) presented tion by Polk and Farah, reported in Farah, simultaneously or sequentially (500msec 1997). Both BD and GR (Vigliocco et al., 1992) SOA). The control subject showed a slight re- performed similarly to normal control subjects action time advantage for the sequential over on a letter-matching task under simultaneous the simultaneous condition. PD’s reaction and sequential conditions, suggesting that times were similar to the control in the sequen- speed is normal for them. We do not, however, tial condition but he took roughly 400msec know what their accuracy was on this task. longer when the letters were presented simul- When presented with single letters for 2 sec on taneously. DC was even slower than PD on the a separate task, a duration that is extremely simultaneous condition but, in addition, was long for normal subjects, BD identified 88% far slower than PD or the control subject on the correctly and GR 90%. The errors produced sequential condition (see Hanley & Kay, 1996, were all visual confusions with the target (for Fig. 4). These findings suggest that neither of example, d/ b, v/ u). It appears then that even these patients process letters normally. these two subjects might not have normal letter There remain seven contentious cases for processing and that the apparently normal re- whom a letter processing deficit is not clearly action times may be due to a speed-accuracy apparent. We consider each in turn. Patient BL trade-off. scored 25/ 26 correct on upper-case naming Both RAV and JDC (Warrington & Shallice, but we have no indication of the time required 1980) appear to identify single letters particu- to do so (Friedman & Hadley, 1992). There is larly well, even when separated by either al- also no report of other letter processing tests phanumeric characters (Expt. 1) or when nor tests of general visual processing for BL. flanked by distractors (Expt. 2). Nonetheless, We do know, however, that BL showed a very both patients are more affected than normal steep serial position curve in reporting letters readers by items presented in script compared from four-item strings under tachistoscopic with print. Moreover, RAV appear to show an presentation. Whereas performance was rea- interaction in reaction times between sonably accurate for items in position 1 (almost script/ print and word length. For three-letter

22 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

words, the difference between script/ print is deficit for reading. We have assumed that the 2.4 sec, whereas the difference increases to 5 letter processing deficit plays a causal role in sec for seven-letter words. Both the differences LBL reading and that, because the weakened between script and print and the interaction activation is insufficient, subjects make both with word length have been taken to be strong overt and covert gaze shifts to enhance the indicators of an early visual processing deficit letter activation. A correlation between read- (Farah & Wallace, 1991). Lastly, PT (Rapcsak, ing speed and the accuracy of single-letter Rubens, & Laguna, 1990; Schacter et al., 1990) identification, suggestive of a strong relation- performed well on tasks of letter discrimina- ship between the two, was described by Shal- tion (letters vs. nonletters or mirror-reversed lice (1988) in a small group of eight LBL letters; 100%), cross-case matching (95%), and readers. This correlation, however, does not pointing to letters (100%). Although accuracy seem to be perfectly upheld, as a recent study was low for letter naming, with a score of only described two patients with fairly similar letter 15/ 26 for upper-case and 15/ 26 for lower-case recognition patterns but with very different letters, this might possibly be a result of an performance in word recognition (Hanley & anomia rather than of a letter recognition defi- Kay, 1996). The exact relationship between cit per se. Once again, however, we have no poor letter and word recognition processes is indication that PT performs these simple letter beyond the scope of this paper. Also beyond tasks at normal speeds. Whether her letter the scope of this paper is whether the poor processing is indeed normal, therefore, re- letter processing derives from an even more mains unans wered at present. fundamental perceptual problem, as we The findings from the review thus far sug- (amongst others) have argued elsewhere gests that there is no single subject for whom (Behrmann et al., 1998; Sekuler & Behrmann, letter recognition is definitively normal. As is 1996; see also Chialant & Caramazza, this is- evident, patients may be impaired in speed, sue), or whether it is restricted to alphanu- even if not in accuracy, and sometimes the meric stimuli. For the present purposes, we are converse is true. Additionally, as noted by Pat- only interested in establishing that, across the terson and Kay (1982), even high accuracy population of LBL readers, there is strong evi- might not be a satisfactory indicator of letter dence for a peripheral deficit that adversely processing skill; an accuracy score of 85% in affects letter processing. We now turn to the letter naming may not seem too serious an question of whether variables associated with impairment but this might become very debili- later stages of word recognition are also ob- tating in a more taxing task when the subject is served in this population. required to represent letters for the purposes of word recognition. Evidence for Lexical and Semantic Effects An important question that remains, then, The major problem we encountered in this is what are the consequences of this peripheral review is that very few studies systematically

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 23 BEHRMANN, PLAUT, NELSON

document the higher-level effects in the read- 1991; Patterson & Kay, 1982; Price & Hum- ing behaviour of LBL readers (especially in phreys, 1992, 1995; Vigliocco et al., 1992; War- cases who are a priori diagnosed as having a rington & Shallice, 1980) and in some of our peripheral impairment). Because the interest own studies (Behrmann, Black, & Bub, 1990; has traditionally been on the effect of word Behrmann & McLeod, 1995; Behrmann & Shal- length, unconfounded by other variables, lice,1995). Why there are such marked individ- stimuli are standardly matched on frequency ual differences is taken up further in the and imageability, for example, and length General Discussion. alone is manipulated. In those few cases where With respect to the word superiority effect, these other variables are tested and reported, as with the covert reading, only a small subset performance is measured in terms of accuracy of the population has been tested and so strong rather than the more sensitive measures of claims about the effect of orthographic context reaction time and the statistical analyses of should be made cautiously. Of the 16 reported both the main effects and their interaction are instances, 12 patients show a word superiority often omitted. Nevertheless, we have re- effect, one shows a trend in that direction, and viewed the literature and present the findings the remaining 3 are not influenced by ortho- in Table 1. We discuss covert or tacit reading, graphic context. One of these 16 patients is DS the word superiority effect, and the influence who, in our initial testing in the first year post- of lexical variables separately. stroke, did not show a word superiority effect Positive evidence for preserved implicit (Behrmann et al., 1990) on a Wheeler-Reicher reading under brief exposure is found in only type task. In subsequent testing, roughly 5 a small number of cases. Aside from the stud- years post-stroke, DS did show a word supe- ies by Coslett and Saffran and colleagues, who riority effect on one task, reporting the begin- document covert effects in five LBL readers, ning and end letters more accurately from definitive covert processing is reported only in words than from nonwords. DS, however, did patients ML (Shallice & Saffran, 1986), KW not show an advantage for words in a second (Howard, 1991) (in semantic but not lexical task that required a decision on whether two decision tasks), DR(Doctor et al., 1990), and RE strings (both words or both nonwords) placed (Feinberg, Duckes-Berke, Miner, & Roane, one above the other for an unlimited duration 1995). A positive trend in this direction is also were the same or different (Behrmann & Shal- noted in Grossi, Fragassi, Orsini, Falco, & Sepe lice, 1995). (1984) and in Landis, Regard, and Serrat (1980), A suggestion for why the word superiority and we have some preliminary evidence of effect is only observed in some but not other relatively good semantic categorisation under cases has been made by Farah and Wallace brief exposure in one of our own patients, DK. (1992; see also Farah, 1997). They observed that By contrast, the absence of covert abilities has at least two of the patients who showed the been documented in several studies (Howard, word superiority effect had a letter recogni-

24 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

tion profile that was similar to normal subjects tistics are not always available) and only three (referred to as an “ends-in” profile). At least subjects are not obviously influenced by two of the cases who did not show a word frequency. One of the 3 subjects who does not superiority effect showed a gradient of letter show a frequency effect is patient RE recognition, with best performance on the in- (Feinberg et al., 1995), whose performance is so itial letter and decreasing accuracy across se- poor that he is unable to read even a single rial position. They attribute the presence of a word aloud. The absence of a frequency effect word superiority effect to the presence vs. ab- then might simply result from a floor effect in sence of the letter-by-letter gradient. In fact, this patient. even within a single subject, one can see this at Of the 19 subjects tested for imageability in work: Bub, Black, and Howell (1989) found naming, again in reaction time or in accuracy, that their patient showed a word superiority 12 show a positive result (5 of our 7 patients). effect in only one experiment and it was in this Again, one of the subjects who does not is single experiment that the patient did not patient RE, and this may again result from a show a left–right gradient in accuracy across floor effect. Five out of 14 subjects show an letter positions. The suggestion that strategy effect of regularity on naming and 5 out of 9 determines the presence of a word superiority subjects show an effect of part-of-speech on effect may well explain the variance in the data their performance. Few subjects are tested on on this effect in LBL readers. In fact, even in these lexical variables in lexical decision, as is normal subjects, focusing sequentially on indi- evident in the final two columns of Table 1. All vidual letters affects top-down effects; specifi- eight patients for whom frequency data are cally, the word superiority effect is reduced available for lexical decision are affected by when normal subjects attempt to read letters in frequency. There are imageability data for only particular positions rather than distributing five patients in lexical decision, and two of their attention across the entire stimulus array these show a positive finding. (Johnston &McClelland, 1980). Unfortunately, Taken together, the review of the later the patients who show covert reading are gen- lexical and semantic effects is not strongly erally not tested for a word superiority effect conclusive. Too few patients are tested on the and vice versa, making it difficult to reach higher-order variables and when they are, they conclusions about the relationship between are usually not tested on all of the different implicit reading and higher-order ortho- tasks or different variables, making graphic processing. comparisons difficult. One finding worth not- Of the 26 subjects tested for effects of fre- ing, however, is that across the population of quency on word naming, 17 are influenced by 57 subjects, there are only 13 subjects who do frequency either in reaction time or in accu- not show any lexical or semantic effects on any racy, including all 7 of our patients. A further of the measures. A further four patients have six subjects show some effect of frequency (sta- not been tested on any of the measures and so

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 25 BEHRMANN, PLAUT, NELSON

the presence of the later effects for them re- EMPIRICAL DATAFROMLETTER-BY-LETTER mains indeterminate (Lazar & Scarisbrick, READERS 1973; Montant, Nazir, & Poncet, this issue; Stachowiak & Poeck, 1976; Warrington & Given the paucity of data on the lexical and Shallice, 1980; patient JDC). A central finding semantic effects in LBL patients, we have un- of the review, then, is that the great majority of dertaken a retrospective analysis of data col- patients show some form of lexical/ semantic lected from seven who have participated in our effects on reading performance. studies over the last decade. In this analysis, we specifically explore whether both a peripheral visual processing deficit and the effect of lexical and semantic variables are observable across this group. All of these patients are Summary of Review highly accurate in single-letter identification The evidence for a peripheral deficit is strongly under unlimited exposure duration although, supported by the review and there is no con- as pointed out previously, this in no way indi- vincing counter-evidence. In those few cases cates that letter recognition is normal and, in where accuracy or even reaction times appear some instances, we know that it is not (for to be within normal limits, it is often the case example, patient DS; see Behrmann & Shallice, that this could result from a speed–accuracy 1995). It is also the case, though, that these trade-off and poorer performance is generally patients are much more impaired than their seen on the complementary metric. These re- control counterparts in identifying two or sults substantiate the initial part of our account three random letters presented under brief ex- and endorse the previous observation of a defi- posure (for example, patient DS, see cit in letter processing that is common to all Behrmann, Black, & Bub, 1990; patient IS, LBL readers. Unfortunately, strong conclu- Behrmann & McLeod, 1995). sions about later lexical and semantic effects Our view of LBL reading predicts that, even are not as obvious from this review, as there in those patients with a peripheral deficit, one are too many empty cells in the database. An would still see effects of word frequency and important finding, though, is that these later imageability on performance. On this account, effects are not restricted to a small subgroup of we therefore predict significant main effects of the population and there is at least partial evi- word frequency and imageability that will re- dence for them in more than two-thirds of the flect the strength of the top-down contribution. patients. We also predict that as word length increases, the difference between high- and low-frequency and the difference between high- and low-imageability items will increase, given that for longer words there is more opportu-

26 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

nity for top-down effects to influence perform- example, did well with single letters even at ance. the presentation limits of the computer (17msec; no masking) but performance was poor on two- and three-item arrays even when Subjects exposure duration was more than tripled Seven LBL readers are included in this analy- (Kinsbourne & Warrington, 1962; Levine & sis. Five of them have been described in detail Calvinio, 1978). Relative to normal subjects, in other publications (see final seven rows of who can identify letters at 10msec with mask- Table 1 for citations), whereas the remaining ing (Sperling & Melcher, 1978), these patients two, PC and DK, have only been tested fairly are much slower at letter identification. recently. The subjects will only be described Four of the seven subjects participated in the briefly in this paper and the reader is referred study by Sekuler and Behrmann (1996), which to the more detailed publications for further documented the ability of LBLpatients to proc- information. All subjects are right-handed and ess visual stimuli that are not alphanumeric. In native speakers of English. Biographical infor- that study, patients completed three different mation and anatomical details of their lesions experiments: they performed perceptual flu- as well as some reading data (which will be- ency judgements under time pressure, made come relevant later) are provided in Table 2. same/ different judgements to contour features that appeared on objects, and searched for a target in visual displays that increased in Evidence for an Early,Peripheral Deficit the number of distractor items. In all three As mentioned previously, all of these patients experiments, these patients performed signifi- do reasonably well, although not perfectly, on cantly more poorly than their control counter- single-letter identification under unlimited ex- parts. These findings suggest that all four of posure duration, although performance is these patients suffered from a general percep- markedly poorer than normal subjects when tual deficit, which extended beyond their abil- brief exposure is used. For example, patient ity to deal with orthographic material. It is this DK performed perfectly on single-letter iden- more fundamental visual perceptual problem, tification when the letters were presented for we claim, that gives rise to their LBL pattern in an unlimited duration. He made 30% errors the first instance. when the letters were presented for 250msec Five of these same seven subjects (excluding (even without a mask)and 40 errors at 100msec PC and TU) have also participated in a recent exposure duration. Similar patterns are seen in study examining their accuracy and reaction other patients. Those patients who perform time to name black-and-white line drawings of well on single-letter identification even at brief high and low visual complexity, taken from exposure duration may then do poorly when the Snodgrass and Vanderwart (1980) set of two or three letters are present; patient IS, for pictures (Behrmann et al., 1998). Whereas age-

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 27 2 BEHRM 8 CO A NN, G NITIV PLA E UT, NEURO NELSO Table 2. Biographical and Lesion Details of the Seven Letter-by-letter Readers PSY

Ð Ð Ð Ð Ð Ð Ð Ð aÐ Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð bÐ Ð Ð Ð Ð Ð Ð c N CHO Patient Age Etiology CT Scan Results Other Relevant Behaviours Freq Image

LO — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — G

Y DS 37 Posterior cerebral artery L occipital infarction upper quadrantanopia 397** 72** , 1998, occlusion; migrainous PC 63 Resected meningioma L occipitotemporal R homonymous hemianopia 1651* 3039**

15 15 (area 19/ 37 boundary) (1/ MW 67 Infarction L occipital lobe infarction R homonymous hemianopia; 570* 292** 2) ensuing depression DK 65 Posterior cerebral artery L occipital lobe infarction R homonymous hemianopia 588** 443 infarction & mass effect MS 37 Closed head injury No focal CT lesion; R homonymous hemianopia; 1552** 1629* bilateral frontal slowing EEG surface dysgraphia IS 46 Posterior cerebral L occipital-temporal region R upper quadrantanopia; 540** 99* artery infarction including hippocampus, mild memory deficit; fusiform and lingual gyri surface dysgraphia TU 56 Resected arteriovenous L occipital haemorrhage; R homonymous hemianopia; 968** 388 malformation additional L temporal damage R hemiparesis; marked anomia ———————————————————————————————————————————————————————————————— a Age refers to the age at which the initial testing took place. Some patients have participated in subsequent follow-up studies and the age of testing is then obviously different. b Frequency score in msec. c Imageability score in msec. *P < .05; P < .01. LETTER-BY-LETTER READING

matched normal control subjects showed a of higher-order processing in a naming latency 152msec reaction time advantage for low-com- task, including effects of word frequency, dif- plexity items over high-complexity items, the ferences between word and nonword reading LBL readers (with the possible exception of (word superiority), part-of-speech effects, and patient MW) showed a disproportionate in- an effect of imageability or concreteness crease in reaction times to name the high-com- (Coltheart, 1980; Warrington & Shallice, 1980). plexity items. Averaged across the patients, We have chosen to use word frequency and low-complexity items were named 530msec imageability, both of which are assumed to be faster than their high-complexity counterparts. reliable indicators of access to a preserved Taken together, these data suggest that these word form system (Kay & Hanley, 1991), and patients have a deficit that affects their proc- for which there is likely to be sufficient data for essing of several classes of visual stimuli and analysis. These two variables have also been that, under rigorous testing conditions, this studied extensively in normal readers and more widespread perceptual deficit may be there are several models detailing their specific uncovered. effects in word recognition. A recent analysis Based on these findings, we can definitively of the effects of frequency, for example, shows conclude that there is an early deficit in six of that word frequency is a critical determinant of the seven LBL readers. We do not have suffi- the speed of lexical access (what Monsell et al. cient information about PC, but the limited referred to as “lexical identification”)(Monsell, data obtained from her single-letter processing Doyle, & Haggard, 1989). Frequency effects suggest that she too may be impaired in early can be explained, depending on whether one’s stages of processing prior to activating an or- model is localist or distributed, as arising from thographic representation. Whereas she is able higher resting levels of activation, lower to identify accurately all three letters from a thresholds, or increased bias on the weights for random letter string (e.g. YFS) presented for an high-frequency compared with low-frequency unlimited exposure duration, she only identi- words (McClelland & Rumelhart, 1981; Mor- fies about 70%of the letters when the exposure ton, 1979; Plaut et al., 1996; Seidenberg & duration is around 1sec, suggesting that her McClelland, 1989). Thus, the speed by which a letter identification threshold is probably ab- word is recognised is determined by the time normally high. Unfortunately, PC died before to reach threshold or for the network to settle we were able to obtain any further data. on an interpretation of the stimulus; the time will thus be less for high-frequency than for low-frequency items. Evidence for Lexical and Semantic Influences on Imageability effects have not received as Word Recognition much press as frequency effects, although There are a number of different lexical vari- there is little doubt that the influence of image- ables that might potentially serve as markers ability on performance arises at later stages

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 29 BEHRMANN, PLAUT, NELSON

and is indicative of semantic processing. Strain from those lists in which the words were pre- et al. (Strain, Patterson, & Seidenberg, 1995), sented in script or cursive font. Frequency and for example, showed that normal readers are imageability are orthogonally crossed with slower and make more errors on low-image- word length in each list. The cut-off for fre- ability than high-imageability words but only quency was 20 per million, with items below for low-frequency exception words. This that classified as low in frequency and items three-way interaction between imageability, above that classified as high in frequency frequency, and regularity is explained with (Kucera & Francis, 1967). Imageability is reference to a system in which the time to standardly hand-coded but, for the purpose of translate orthography to phonology varies. this analysis, word imageability was taken When this translation process is slow or noisy, from the MRC Database; items which did not as is the case for low-frequency exceptions, have an imageability rating in this listing were words with rich semantic representations (i.e. excluded. Again, imageability was converted high-imageability words) are most likely to to a categorical variable with items exceeding benefit from this interaction. This interpreta- 525 being classified as high in imageability and tion of how later effects (semantic repre- those below this cut-off as being low in image- sentations in this case) interact with somewhat ability. earlier processes (translating orthography to In all cases, the words were presented on a phonology in this case) is consistent with the computer screen to the left of central fixation interactive account of LBL reading that we to circumvent the right visual field defect that have proposed. was present in all patients. The words were right-justified so that the final letter of each Stimuli and Procedure word appeared in the character space just to To examine the effects of frequency and image- the left of fixation. The standard procedure ability on the reading behaviour of LBL read- was as follows: a central fixation point ap- ers, we analysed naming latency data for single peared for 500msec. After a 1 sec delay, the words collected from these seven patients. At word appeared and remained on the screen for some point during their testing, these patients an unlimited duration until a response was read aloud lists of 60 words presented in a made. The visual angle subtended by stimuli variety of fonts and sizes containing 20 items of three, five- and seven-characters in length each of 3-, 5-, and 7-letter words. These lists were 1.5°, 2.4° and 3.6° respectively. Words and forms of presentation are used standardly were presented in black font against a white in our laboratory to measure the naming la- background. Both reaction time (via a voice-re- tency of LBL readers and patients typically lay-key) and accuracy (recorded by the exam- read three of four such lists during their pre- iner) was measured. In all patients, except in liminary testing. We have included as many the case of DK, presentation and timing was word lists as possible for each subject apart controlled by Psychlab (Bub & Gum, 1988) and

30 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) a Macintosh Classic or IIci was used, whereas line is the value of the slope calculated for that for DK, presentation was controlled by Psy- subject in a linear regression analysis, with RT Scope (Cohen, MacWhinney, Flatt, & Provost, set against word length. As is evident from this 1993) on a Powerbook 540C. figure, subjects differ fairly dramatically in their base reaction time [F(6,821) = 118.7, P < .0001]. Also evident from this figure is the sig- Results nificant main effect of word length [F(2,821) = Reading responses to a total of 24 word lists (N 167.3, P < .0001], with a mean across all sub- = 1440, cumulative across subjects) were col- jects of 1442, 2465, and 3520msec for 3-, 5-, and lected from the subjects. A number of trials, 7-letter words, respectively, indicating that however, were excluded for a variety of rea- roughly an extra 500msec is required to proc- sons: imageability ratings were unavailablefor ess each additional letter. 315 trials, patients made errors on 168 trials, Although every subject shows the hallmark the microphone was not triggered or the sub- monotonic relationship between reaction time ject coughed on a further 50 trials and 2 trials and word length, subjects differ in the slope or were extreme outliers from the subjects’ reac- increase in RT per-each additional letter tion time distribution (more than four SDs [F(12,821) = 18.3, P < .0001]. At the highest ex- from the mean). A repeated -measures treme is MA, who shows an increase of ANOVA with word length crossed with word 1409.3msec per letter. At the bottom extreme, frequency and word imageability was per- the mildest LBL subject, DS, who had a stroke formed on the remaining 905 trials. Subject roughly 10 years ago (Behrmann, Black, et al. was included as a further between-subject 1990), shows an increase of 97msec per letter. variable so that profiles for the individual sub- Although this latter slope is rather flat in com- jects could be drawn up. We note that the parison with the population of LBL readers, it amount of data per subject is not equivalent must be noted that this increase is roughly (given that different subjects read differing three times greater than the maximum of numbers of lists and error rates differed) and 30msec per letter needed by normal subjects in so the estimates of the effects are better for naming words presented to the left visual field some subjects than for others; subject PC has (Henderson, 1982; Young & Ellis, 1985). In- the least data and her performance is the most deed, some studies have demonstrated that variable (see all Figs). As mentioned earlier, normal subjects show no effect at all of word unfortunately PC died before we were able to length on naming latency for words (Schiep- collect further data. The number of trials per ers, 1980; Weekes, 1997). Any reliable increase subject is included in both Figs 2 and 3. in latency as a function of string length might, The reaction time (RT) to name a word as a therefore, be considered abnormal. On the function of string length is plotted for each same word lists as those read by our LBL read- subject individually in Fig. 1. Along with each ers, normal subjects, who served as control

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 31 BEHRMANN, PLAUT, NELSON

Fig. 1. Mean reaction times for the seven letter-by-letter Importantly for the present purposes is that, readers as a function of word length and the slope of the across these seven subjects, word frequency regression function. significantly affected reaction time [F(1,821) = 41.8, P < .0001]; high-frequency words (N = 566) were named, on average, 884msec faster subjects for some of the patients (some of than low-frequency (N = 339) words. The dif- whom are elderly), typically showed very flat ferences between the mean RTs for high- and slopes; control subject BR (Behrmann & low-frequency words for each patient are in- McLeod, 1995), for example, showed an included in Table 2. A one-way ANOVA per- crease in RT of 15.3msec for each additional formed separately for each subject, comparing letter in a word, whereas control subjects AS RTs for high- and low-frequency words, and JD, averaged, show a slope of 9.4msec yielded significant values for every subject. (Behrmann et al., 1998). Relative to these val- The effect of frequency did not vary as a func- ues, all of our subjects are markedly abnormal, tion of word length, collapsed across subjects including the mildest subject, DS. [F(2,821) = 2.06, P < .13], and the frequency

32 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

effects remained relatively constant across tion of word length, although this held to a subjects [F(6,821) = 1.17, P < .32]. However, varying degree across subjects [F(12,821) = the three-way interaction of frequency by 1.76, P < .05]. The interaction is particularly word length by subject is significant. This is evident in patients DS, MW, DK, and IS and, to seen in Fig. 2 which shows the mean reaction a lesser degree, in MA. It is only in patient PC times for the individual patients for high- and (for whom the least data are available) that this low-frequency words as a function of word function does not follow the predicted pattern. length. It is important to note that the y-axis on Interestingly, there is a good correlation across this figure differs across subjects. As is evident the subjects between the severity of the LBL from this figure, the difference between high- reading (defined by the slope of the regression and low-frequency words increased as a func- curve on the naming latency data in Fig. 1) and

Fig. 2. Reaction times for the seven letter-by-letter readers, plotted individually, as a function of word length for high- and low-frequency words.

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 33 BEHRMANN, PLAUT, NELSON

the effect of frequency on performance (de- tern across subjects is less clear. The expected fined as the msec difference between naming interaction, however, is clearly evident for pa- high- and low-frequency words (r2 = .85, P < tients PC, MW, DK, MA, and TU and less clear .005), indicating a close relationship between in DS and IS (although the increase in 5-letter severity and frequency, as would be predicted words is perhaps a bit more evident for IS). It by our account. is also worth noting that PC’s data are particu- Word imageability also significantly influ- larly variable; PC has a total of only 43 data enced naming [F(1,821) = 52.3, P < .0001]; points, 5 of which are 7-letters long and low- high-imageability words (N = 455) were re- imageability, and thus the very long mean RT ported 696msec faster than low-imageability of 13049msec for this last cell should be inter- words (N = 450) on average, although the ex- preted cautiously. There was a positive corre- tent of this difference varied across subjects lation between the severity of the LBL deficit [F(1,821) = 10.6, P < .0001]. The differences in (again, using the value of the regression slope mean RT for high- and low-imageability in naming latency) and the imageability effect words for each subject are presented in Table (defined as the msec difference between nam- 2 along with the significance values obtained ing high- and low-imageability words) (r2 = from a one-way ANOVA performed sepa- .53, P < .06), again indicating a relationship rately for these data. Five of the seven patients between severity and imageability, although showed a significant difference between high- this correlation was not as strong as was the and low-imageability stimuli. The advantage correlation with frequency. for high- over low-imageability words was, however, influenced by word length [F(2,821) Summary of Empirical Data = 20.7, P < .0001]; the difference between high- and low-imageability words is 169msec for 3- The findings from the empirical study are letter words compared with an 852msec differ- fairly straightfor ward. In a group of LBL ence for 7-letter words. Again, this difference patients, all of whom show the characteristic as a function of word length varies across sub- effect of word length on reaction time, we see jects. This is evident in Fig. 3 in which the mean a difference in performance between high- and reaction times for the individual patients with low-frequency items in all seven patients and high- and low-imageability words are plotted a difference between high- and low-imageabil- as a function of word length. Although there is ity words in five of them. Moreover, both fre- a general trend in all subjects (except for sub- quency and imageability interact with the ject DS) for the difference between high- and length of the string such that the frequency low-imageability words to become increas- effect and the imageability effect are more ingly magnified as word length increases marked for long than for short words. The [F(12,821) = 7.6, P < .0001], these data are more exact discrepancy between the high and low variable than the frequency data and the pat- items as a function of word length is evident to

34 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

Fig. 3. Reaction times for the seven letter-by-letter time, as reflected in both naming and lexical readers, plotted indivdually, as a function of word length decision tasks. There are two major findings in for high- and low-imageability words. the LBL literature: The first is that patients have a deficit at a peripheral stage of process- a greater degree in some patients than in others ing. Evidence supporting a peripheral impair- and this interaction is more pronounced in the ment comes from the finding that many, if not frequency than in the imageability analysis. all, patients are unable to represent orthographic input normally and they show an impairment in letter processing. The second DISCUSSION finding is that the reading performance of many patients is influenced by lexical and se- Letter-by-letter reading is a form of acquired mantic variables, suggesting that the deficit dyslexia in which patients show a monotonic occurs only after an adequate orthographic relationship between word length and reaction representation is activated. Support for a cen-

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 35 BEHRMANN, PLAUT, NELSON

tral impairment comes from the finding that length of the word. Thus, covert or tacit read- many LBL readers appear to have processed ing in the form of semantic categorisation and the visual stimulus sufficiently to have derived lexical decision is possible despite poor stimu- semantic, lexical, and higher-order ortho- lus identification; lexical variables such as fre- graphic information from it, as indicated by quency and imageability may still influence their ability to perform semantic categorisa- word recognition, and lexical/ semantic repre- tion tasks or to make lexical decisions about sentations may feed back onto letter repre- stimuli that they have not identified explicitly. sentations and produce a word superiority In this paper, we put forward an account of effect. Thus, just as higher-level repre- LBL reading that reconciles these disparate sentations can compensate for or mitigate findings. We argue that there is a fundamental against the spatially degraded input in neglect peripheral deficit in LBL that adversely affects dyslexia (Mozer & Behrmann, 1990), so too can parallel letter processing. Despite this lower- such representations feed back and support level deficit, however, some information is still the weakened orthographic activation in LBL propagated to higher-level (lexical and seman- reading. tic) representations. In an attempt to enhance In support of this theoretical position, we orthographic activations, subjects employ the cite evidence from a review of published re- normal reading strategy of making additional ports of 57 patients with LBL reading. There fixations (or covert attention shifts) when en- was positive evidence in almost every case for countering difficulties in text. This sequential the presence of a peripheral deficit and, even letter processing adds further to the activation in those few cases for which no definitive evi- of lexical/ semantic representations. Due to the dence existed, there was no compelling interactive nature of processing, these higher- counter-evidence. The findings from this re- level representations feed back to provide sup- view regarding the higher-level lexical and se- port for subsequent orthographic processing. mantic effects were somewhat less conclusive The strength of this support depends on the but this was largely because of the dearth of degree of higher-level activation, which we data. Relatively few patients have been tested assume scales with word frequency and im- systematically for the influence of different ageability. Because patients take longer to variables on performance and, even when process words with more letters, the lexi- there was some suggestive evidence in a case cal/ semantic activation has a longer time in report, the effects were not usually evaluated which to accumulate and influence the de- across a sufficient range of measures to pro- graded letter activations. The main point is vide a full description of the patient’s perform- that, in an interactive system, impoverished ance. Perhaps the most relevant finding from input can still activate lexical and semantic this review, however, was that more than two information and, moreover, that the strength thirds of the patients showed some evidence of of this activation over time is related to the higher-level effects, whether in implicit read-

36 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

ing under brief exposure, in the presence of a Moreover, the discrepancy between the two word superiority effect, or in the influence of types of items was greater for longer than lexical variables on reading performance. The shorter words: whereas there was no accuracy key point, then, is that in many of these pa- difference for high- and low-frequency four- tients with peripheral deficits, there is also a letter words, there was a difference for longer strong suggestion of some kind of lexical and words. Interactions of word length and image- semantic processing. ability have also been reported in a few cases. To substantiate our position further, we Patient DR (Doctor et al., 1990), for example, carried out a retrospective empirical analysis took longer to read low- than high-imageabil- of the reading performance of seven LBL read- ity printed words, particularly as word length ers we have tested over the last decade. We increased. A similar trend, albeit nonsignifi- have previously determined that these pa- cant, was also observed in this patient for tients all suffer from a peripheral impairment hand written words. Finally, JH (Buxbaum & of some form. In the current analysis, we dem- Coslett, 1996) showed a dramatic increase in onstrated a significant main effect of word fre- reaction time (and decrease in accuracy) as quency and word imageability on reaction word length increased, but this was especially time in almost all cases and, in most cases, true for low-imageability words. Thus, the dif- these effects interacted significantly with word ference in accuracy between high- and low-im- length (although to varying degrees and with ageability words was approximately 10% but more variance in imageability than in fre- increased to approximately 45% for eight-let- quency). We suggest that this interaction is ter words (see their Fig. 7). specifically predicted by an interactive ac- In addition to showing the predicted inter- count. In particular, we argue that, because of action between word length and imageability, the additional time required by LBLreaders for patient JH (Buxbaum &Coslett, 1996) provides longer words, there is more time in which additional support for our account. JH was higher-level activation can accumulate and af- clearly impaired at lower-level processing: He fect reading performance. was unable both to represent information on Our review of the literature has turned up a the right of a display and to distribute attention number of other cases in whom an interaction with sufficient resolution for the purposes of between a lexical variable and word length has letter identification. Despite this impaired pe- also been demonstrated. For example, Bowers, ripheral processing, however, JH’s reading Bub, & Arguin (1996) documented the effect of performance showed a part-of-speech effect word frequency on the accuracy and reaction (nouns were read better than functors), an ef- time of patient IH’s reading responses. Al- fect of lexical status (words were read better though a statistical analysis was not provided, than nonwords), and an effect of morphologi- high-frequency words were read better and cal composition (pseudosuffixed words were faster overall than were low-frequency words. read better than suffixed words). The two

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 37 BEHRMANN, PLAUT, NELSON

major findings, a peripheral deficit concurrent 1994), and even in normal subjects. Monsell et with lexical influences on reading, and an in- al. (1989), for example, have postulated a teraction between word length and lexical mechanism similar to ours to account for the variables, provide clear support for our uni- finding that normal subjects can perform lexi- tary view of letter-by-letter reading. cal decision well without being able to identify There are a number of suggestions in the the stimulus. They claimed that a subthreshold literature that bear some resemblance to, or signal from the target relative to the activity of even foreshadow, the view we have proposed neighbouring items may enable a legitimate here. More than a decade ago, Shallice and word to be distinguished from nonwords but Saffran (1986), in their explanation of patient may not suffice for identification. If the sub- ML’s tacit or covert reading abilities, sug- jects monitor the activation in the system, “the gested that weak input from an impaired familiarity assessment process”, then high-fre- word-form system might still allow sufficient quency words which have stronger repre- activation of a semantic representation that sentation and are more familiar can bias the would suffice for a task like semantic categori- subject towards a positive response before sation. They argued, however, that inhibition unique identification has occurred (see Plaut & might be inadequate in the system. Because it Shallice, 1993, for connectionist simulations is necessary to converge on a single correct consistent with this idea). output for stimulus identification, the alternatives must be inhibited and patient ML was unable to do so. Thus, weakened but sufficient MODELSOF LETTER-BY-LETTER READING activation in combination with inadequate inhibition might give rise to above-chance se- There are no existing models of reading that mantic categorisation in the absence of fully instantiate our theory of the mechanism stimulus identification (see Friedman et al., underlying the performance of LBL readers. 1993, for a similar account and Hinton & Shal- However, there are a number of models which, lice, 1991, for an interpretation of ML’s data in their behaviour under damage, exhibit that is consistent with this). Along similar many of the critical properties on which our lines, Bub et al. (1989) have suggested that, if account is based. As mentioned in the Intro- activation is sufficient, even if not of normal duction, the fact that a partial peripheral im- strength, lexical decision and semantic prim- pairment may give rise to both lower- and ing may still be possible. Similar ideas about higher-level effects has already been demon- weakened activation and/ or inhibition have strated with respect to lexical influences on also been suggested to explain phenomena of reading performance in neglect dyslexia tacit recognition in other neuropsychological (Mozer &Behrmann, 1990), in the preservation disorders, such as prosopagnosia or im- of semantic categorisation and lexical decision plicit/ explicit memory differences (Farah, with impaired overt recognition, and in the

38 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

occurrence of semantic errors following visual reading. More recently, Plaut (1998) has dem- damage in deep dyslexia (Hinton & Shallice, onstrated properties of LBL reading following 1991; Plaut & Shallice, 1993). In the specific peripheral damage to the “refixation” model context of LBL reading, support for co-occur- first described by Plaut et al. (1995). The model rence of lower- and higher-level effects follow- generates a sequence of phonemes as output in ing peripheral damage has recently been response to orthographic input presented over provided by Mayall and Humphreys (Mayall position-specific letter units. A critical aspect & Humphreys, 1996) who implemented a con- of the model is that, if it encounters difficulty nectionist model whose architecture incorpo- in generating a particular phoneme in the rated some of the properties of the IAM. In a course of pronouncing a word, it can refixate single architecture, reflecting the normal word the input (using an internally generated atten- processing system, Mayall and Humphreys tional signal) to bring the corresponding pe- were able to reproduce many of the covert ripheral orthographic segment to “fixation”, abilities of LBLpatients when they lesioned the where performance is better. Early on in train- network. They tested both early lesions (near ing, the model fixated virtually every graph- the input layer) and more central lesions (to eme but, by the end of training, it read correctly hidden units). Of primary concern to us is the 99.3% of the 2998 monosyllabic words on pattern of results obtained with the former, which it was trained, producing an average of peripheral type of disturbance. When a per- only 1.3 fixations per word. When letter activa- centage of the input units were disconnected tions were corrupted by noise, correct per- from the hidden units, letter naming, although formance dropped to 90.1% correct. Using a poor (35% correct) was significantly better median split on frequency, accuracy was than word naming (9% correct). The errors greater on high- versus low-frequency words produced were mostly visual rather than se- (92.1% vs. 88.6%, respectively) and short ver- mantic. In addition, a greater number of cor- sus long words (e.g. 92.0% for 4-letter words rect semantic categorisation and lexical vs. 85.2%for 6-letter words). Critically, among decision responses were observed compared words pronounced correctly, the average with naming responses. Finally, effects of number of fixations per word increased from word frequency and imageability were noted 1.3 to 1.93. Moreover, the number of fixa- in naming, categorisation, and lexical decision. tions— a loose analogue to naming la- This ensembleof findings is precisely what one tency— was strongly influenced by the length would expect with a cascaded, interactive of the word. For example, the model made an model. average of 1.76 fixations for 4-letter words but Asignificant limitation of Mayall and Hum- 2.92 fixations for 6-letters words. The model preys’ (1996) simulation, however, is that it also made fewer fixations to high- versus low- does not address the hallmark characteristic of frequency words (means 1.86 vs. 2.00, respec- LBL readers— namely, their letter-by-letter tively). Finally, and most importantly for the

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 39 BEHRMANN, PLAUT, NELSON

current account of LBL reading, there was a imageability). Exactly what the parameters of clear interaction of frequency and length: The this degradation are remains unclear and difference in the number of fixations for high- needs to be determined. There are, however, a versus low-frequency words was 0.04 for 4- few studies that have examined this issue in letter words but 0.20 for 6-letter words. Thus, normal subjects. An early study by Terry, Sa- under peripheral damage, the model exhibited muels and LaBerge (1976; Expt 1) had normal the hallmark word-length effect characteristic subjects press a response button when the pre- of LBL reading, combined with the appropri- sented item was an animal word (two thirds of ate higher-level effects: A word frequency ef- all words were animal words and were the fect that was greater for long compared with highest-frequency animal words). The words short words. However, the model contains no varied in length from three to eight letters and semantic representations and, thus, is unable the letters could be degraded or not. Unfortu- to account for the effects of imageability on nately, conclusions from this study are tenu- LBL reading, nor the relatively preserved lexi- ous for our purposes. First, it appears that the cal decision and semantic categorisation per- degradation had no effect on accuracy, as evi- formance of these patients. dent in the error data (op. cit. p. 580). Given that this is a potential ceiling effect, it is not surprising that there is no interaction between ADDITIONALISSUES degradation and word length even in reaction times. Second, the results for the longer words There are three outstanding issues that require (seven and eight letters) could not be calcu- discussion in the context of our theory of LBL lated as there were too few observations in reading. The first concerns effects of stimulus these cells. Finally, because the task is more degradation in normal readers, the second con- like a decision task than a naming latency task, cerns the heterogeneity in the performance of the effect of word length, if it exists, is likely to LBL readers, and the last concerns the sub- be smaller than is the case in standard naming strate mediating the lexical and semantic ef- latency tasks (see Henderson, 1982, for review fects. We take up these issues in turn. of relevant findings). In a more recent study, Snodgrass and Min- tzer (1993), have examined the effect of stimu- Effects of Stimulus Degradation in Normal lus degradation on word length in the context Readers of studying how orthographic neighbourhood A straightfor ward prediction that one might and word frequency influence the speed of make from our account is that normal subjects word recognition. They presented degraded should reveal a word length effect when tested word stimuli of varying lengths to normal under conditions of stimulus degradation (and readers and gradually improved the quality of that this should interact with frequency and the stimulus until it could be identified. The

40 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

threshold of time to identification did not vary ditions, subjects are unlikely to make an initial as a function of word length nor did it vary as attempt at parallel identification, as we pro- a function of neighbourhood or frequency, as pose is the case for LBL readers. Taken to- one might have expected (see Snodgrass & gether, the findings from these three studies Mintzer, 1993; Table1, p. 251). The absence of a fail to show the expected interaction of word correlation between the threshold of identifica- length and stimulus degradation. However, tion and any of these other variables, as ac- none of these studies appears to be a good test knowledged by the authors, might be a of the hypothesis given the particular meth- function of the fact that frequency and neigh- odological choices the experimenters have bourhood size were not particularly carefully made. Our prediction still holds then: If one controlled. In light of this, the authors go on to can simulate the nature of the peripheral im- a second experiment in which these variables pairment in LBL readers in normal subjects, are controlled and the expected results ob- then interactions with word length as well as tained. Unfortunately, length is not controlled the other lexical variables should be observed. in this second experiment and so conclusions Preliminary data are encouraging in that an about the effect of degradation of word length interaction between word length with fre- in normal readers remains to be determined. quency and with imageability is more evident A final experiment, perhaps the most with increased stimulus degradation (Nelson, closely related to the focus of this paper, is that Behrmann, & Plaut, 1998). of Farah and Wallace (1991) who plotted naming latency as a function of word length for Individual Differences in Letter-by-letter Reading stimuli that were either masked (as a proxy for stimulus degradation) or not. Whereas latency A particularly striking finding that seems to differed significantly as a function of word challenge to our account concerns the range of length and also of visual quality, there was no individual differences observed in the LBL interaction between them. On the surface, this readers. As is evident from Table 1, whereas seems to be a challenge to our account. Impor- some patients show striking lexical and se- tant to note, however, is that Farah and Wal- mantic effects, others do not. There are several lace (1991) instructed their subjects to read possible explanations for this variability. One letter-by-letter as they were particularly inter- explanation might be that this simply reflects ested in the visual impairment hypothesis of limitations in the available data. Patients are pure alexia. Unfortunately, for our sake, this not routinely and systematically tested and so does not constitute a relevant test of our ac- the sparse data may be exaggerating the differ- count, given that strategic effects can signifi- ences between patients. Although it is true that cantly alter the nature of the reading additional data are sorely needed, this is un- processing (see following section on individ- likely to be the sole explanation. It is also un- ual differences) and also that under these con- likely that individual differences in literacy or

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 41 BEHRMANN, PLAUT, NELSON

reading skill can fully explain the range of a 3-month period and showed imageability variability in the LBL reading profiles (Hanley and part-of-speech effects in reading (Coslett & Kay, 1996). & Monsul, 1994). Perhaps a more persuasive explanation for These findings suggest that there may be an differences among LBL readers concerns the inverted-Ufunction relating the severity of im- severity of the deficit. In Mozer and pairment to the strength of higher-level effects Behrmann’s (1990) interactive theory of ne- observed in LBL readers. When the visual in- glect dyslexia, the difference between the two put is well processed, top-down support is patients is explained by the severity of the largely unnecessary, and when the deficit is early attentional deficit. Thus, Mozer and too severe, higher-order representations are Behrmann suggested that, in the case of the not strongly engaged . It is only in the middle more severe neglect patient, the orthographic range, when the orthographic input is still suf- input was far too degraded to activate the cor- ficiently intact, that the later lexical effects be- responding higher-order lexical and semantic come apparent. At this point, then, when there representations. In the case of the more mild is sufficient activation, the exact magnitude of neglect patient, there was sufficient bottom-up the higher-ord er effects depends on the sever- activation of higher-order representations, and ity of the deficit. The slower the reader (or the effects of lexical status and morphological more marked the word length effect), the composition were observed. We suggest that longer the time for the top-down processes to differences in severity of impairments may play a role. This is well illustrated in our group also underlie the individual differences ob- of seven patients in whom there is a correlation served among LBL readers. For example, between the severity of LBL reading (defined Behrmann, Black, et al. (1990) showed that, in by the slope of the regression curve on the the early stages post-stroke, DS did not show a naming latency data; Fig. 1) and the frequency word superiority effect in her reading. When effect (difference in ms between high- and low- she was retested several years later, her read- frequency words). There is also a positive cor- ing had improved (as measured in the slope relation between slope values and the and intercept of the latency-per-letter func- imageability effect, although this correlation is tion), although she was still an LBL reader. more modest than the frequency effects. Thus, Importantly, at this later stage, she now the more severely affected the LBL reader showed a marked difference in reading of (within the range permitting higher-level acti- words versus nonwords. Consistent with this, vation), the greater the effect of the lexical vari- Coslett and Saffran (1989a) found that the pos- ables. tlexical effects in their patients became more The claim that severity, defined this way, apparent after partial recovery had taken might determine the extent of the lexical/ se- place. For example, patient JG gradually re- mantic effects, is interesting but is likely to be gained the ability to perform tacit reading over too simple. The slope of the reading function,

42 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

in our account, is determined by the severity of from one patient to another. Differences in re- the lower-level deficit, but we know that the sponse threshold and confidence in respond- correlation between the degree of peripheral ing might also bring about differences in the impairment and reading performance is not overall pattern of reading performance across perfect. Hanley and Kay (1996) have described patients. fairly large differences in the reading perform- A final but important difference concerns ance of two LBL patients with roughly equiva- the strategies individual subjects might em- lent letter identification performance. Their ploy in compensating for their peripheral im- interpretation of the discrepancy between the pairment. It is now well recognised that patients echoes that of Patterson and Kay particular strategies can diminish and possibly (1982), who invoked a second lesion (in addi- even eliminate lexical/ semantic effects (also, tion to the peripheral one) to account for the Howard, 1991). Farah and Wallace (1991) differences between patients. Specifically, they pointed this out in relation to the presence/ ab- proposed that an additional impairment at the sence of the word superiority effect and the level of the word form system is present in type of serial letter strategy used, whether some but not other LBL readers. “ends-in” or left-to-right. But perhaps the best Some authors have suggested that yet an- illustration of the importance of considering other potential source of individual differences strategy effects comes from patient JWC is parametric variation in the strength of over- (Coslett et al., 1993), who appeared to be able all inhibition within the lexical system. As to use two distinct strategies, only one of which mentioned previously, according to Shallice produced covert reading. When JWC was in- and Saffran (1986) it is the strength of inhibi- structed to name words, he employed a labo- tion that is critical for the system to converge rious, serial LBL strategy that eventually on a single response for identification. Suffi- resulted in explicit word identification. In con- cient (even if not normal) activation, on the trast, when instructed to make lexical decisions other hand, might be adequate for semantic on semantic categorisations on briefly pre- categorisation. One might imagine that differ- sented stimuli, he adopted a “whole-word” ing degrees of inhibition and activation might strategy that was fast and less effortful but lead to different patterns of performance, only which failed to provide explicit word identifi- some of which produce results consistent with cation. Based on these findings, Coslett and higher-level effects on performance. Arguin colleagues argued that differences in task de- and Bub (1993), in their demonstration of letter mands or instructions may be a critical factor priming in LBL reading in a connectionist that affects the emergence of the higher-level model, have made a similarpoint. They argued effects and, presumably, in their account, the that the crucial parameters are the amounts of involvement of the right hemisphere in read- inhibition and activation and the balance being. They recommended that patients be dis- tween them. It is this balance that may vary couraged from employing the slow and

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 43 BEHRMANN, PLAUT, NELSON

inefficient left-hemisphere LBL procedure ciples and the final manifestation of LBL read- during rehabilitation and should instead be ing is the result of the dynamic processing prompted to adopt the more parallel whole- within and between both hemispheres. word strategy (but see Chialant & Caramazza, What data, then, might compel us to aban- this issue. don this view and invoke a stronger right- hemisphere account, or a greater division of labour between the hemispheres? One source The Locus of the Lexical/Semantic Effects of evidence often cited in support of the right- The last remaining topic for discussion con- hemisphere account is the apparent similarity cerns the debate between the right-hemisphere in the pattern of performance in LBL patients view and our own. The former, two-system and in the right hemisphere of commissuro- account argues that the LBL reading and the tomy and left-hemispherectomy patients. word length effect are a consequence of the These last two groups of patients can make processing of the impaired left hemisphere, semantic judgements but are poor at process- whereas covert reading and later lexical/ se- ing low-imageability words, grammatical mantic effects are mediated by the right hemi- morphemes, and functors— precisely the pat- sphere (Coslett & Saffran, 1993; Saffran & tern reported for LBL readers (Coslett & Saf- Coslett, this issue). We have maintained in- fran, 1994). The qualitative similarity between stead that the empirical data and profiles of the patient groups suggests that the LBL read- performance associated with LBL readers do ers might well be accessing a limited-capabil- not require such a dichotomy. Rather, we have ity right-hemisphere reading system. There argued that the full range of effects in LBL are, however, some important differences be- reading can be explained in terms of properties tween these v arious populations both qualita- of the normal interactive lexical processing tively and quantitatively (Baynes, 1990). For system located in both hemispheres. Although one, unlike deep dyslexic patients, who are the right hemisphere undoubtedly contributes thought to be reading primarily with the right to performance, as does the left hemisphere, hemisphere given the large extent of their left- the observed LBL pattern is a reflection of the hemisphere damage, LBL readers almost combined processing of both hemispheres, never make semantic paralexias. A possible both for the sequential reading pattern and the response to this discrepancy is to claim that higher-level effects. After damage to the left LBL readers are able to access the phonology hemisphere, we would argue, the dynamic co- for the initial letter via the left hemisphere and operation between the hemispheres and their this prevents the production of semantic errors relative contribution to the output continues as (a similar view is cited by Marshall & New- was the case premorbidly. Importantly, on our combe, 1973, in the case of deep dyslexic pa- account, the two hemispheres are governed by tients who make fewer semantic paralexias the same computational and interactive prin- over time). Even if this were the case, this

44 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

would amount to the same claim as we are alexia after suffering small, exclusively subcor- making: Both hemispheres contribute to the tical lesions to the splenium of the corpus cal- final output. Setting the semantic errors aside, losum and in the region of the left lateral however, if it were the case that LBL readers geniculate nucleus (he also showed a dilatation were engaging the right hemisphere to such an of the left occipital lobe on subsequent CT extent, we might expect that they would read scans, perhaps reflecting degeneration secon- words as well as patients with deep dyslexia. dary to the lateral geniculate damage). The Some deep dyslexia patients read concrete subcortical nature of the damage is different words extremely well, and so the expectation from that in the overwhelming majority of LBL is that this should also be the case for LBL patients, whose lesions are cortical (Black & readers. This, however, is not the case. Behrmann, 1994). Thus, although the findings Perhaps the evidence to distinguish be- from JG are indeed provocative, we do not tween these accounts will only come from think that they are sufficiently firm on their physiological studies that directly investigate own just yet to motivate a strong right-hemi- the contribution of the two hemispheres to sphere account. reading in letter-by-letter patients. One such Our conclusion, then, is that there are no study is that by Coslett and Monsul (1994), data that compel us to accept the view that the which used transcranial magnetic stimulation right hemisphere solely or even primarily sub- (TMS) applied separately to the left and right serves the lexical and semantic effects in LBL hemisphere of their LBL patient, JG. Interest- reading. Given this, we maintain that the nor- ingly, JG’s reading was significantly disrupted mal premorbid reading system, degraded by by TMS to the right (from 71% to 21%, and brain damage, continues to function, and, be- showed an interaction with frequency), but not cause of its interactive nature, gives rise to the to the left hemisphere (68% to 61%). This find- expected patterns of lower- and higher-level ing is strongly indicative of the contribution of effects. Perhaps the final adjudication will the right hemisphere to reading. Interestingly, come from neuroimaging studies of LBL read- JG’s reading was not totally disrupted by right- ers in which brain activation of each hemi- hemisphere TMS; this might indicate that sphere is calculated separately and correlated either the TMS only partially disrupted the with the lexical/ semantic effects. right-hemisphere function or it totally disrupted right-hemisphere function and the residual reading was mediated by the left CONCLUSION hemisphere. Although this finding does seem, on the face of it, to suggest strongly that it is the We have presented a unitary account of LBL right-hemisphere that is mediating reading, it reading that, we have argued, reconciles pre- should be pointed out that JG is a somewhat viously discrepant findings. We have claimed anomalous LBL patient. He exhibited pure that a deficit in letter processing (perhaps at-

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 45 BEHRMANN, PLAUT, NELSON

tributable to an even more fundamental per- Arguin, M., & Bub, D. (1993). Single character processing in a caseof purealexia. Neuropsycholo- ceptual impairment) is common to all LBL gia, 31(5), 435–458. readers, and that this deficit prevents the nor- Arguin, M.E., & Bub, D. (1994a). Functional mal activation of orthographic repre- mechanisms in pure alexia: Evidence from letter sentations. It is the severity of this impairment processing. In M.J. Farah and G. Ratcliff (Eds.), The neuropsychology of high-level vision (pp. that determines the extent to which top-down 149–171). Hillsdale, NJ: Lawrence Erlbaum As- activation from higher-order lexical and se- sociates Inc. mantic representations can feed back and miti- Arguin, M.E., & Bub, D. (1994b). Pure alexia: At- gate the lower-level deficit. When the deficit is tempted rehabilitation and its implications for the interpretation of the deficit. Brain and Lan- too severe, little top-down influence will be guage, 47, 233–268. observed. Once the activation is sufficient, Arguin, M.E., Bub, D., & Bowers, J.S. (this issue). however, the more degraded the input repre- Extent and limits of covert lexical activation in letter-by-letter reading. Cognitive Neuropsychol- sentation and the longer the time required for ogy, 15, (1/ 2). processing, the more time is available for top- Baynes, K. (1990). Language and reading in the down support to accumulate. We have sug- right hemisphere: Highways or biways or the gested that, in the context of a cascaded, brain. Journal of Cognitive Neuroscience, 2, 159–179. interactive system, it is possible to observe Beeman, M., & Chiarello, C. (1997). Right-hemi- lexical and semantic effects simultaneously sphere language comprehension: Perspectives from with a peripheral lesion that affects lower-level cognitive neuroscience. Hillsdale, NJ: Lawrence processing. We have also proposed that the Erlbaum Associates Inc. Behrmann, M., Barton, J.J., & Black, S.E. (1998). observed reading pattern, including the se- Eye movements reveal the sequential processing in quential LBL reading itself, arises from the letter-by-letter reading. Manuscript in prepara- residual function of the normal reading system tion. Behrmann, M., Black, S.E., & Bub, D. (1990). The that probably involves both the left and right evolution of pure alexia: A longitudinal study of hemispheres and that there is no reason to recovery. Brain and Language, 39, 405–427. invoke the right-hemisphere reading system as Behrmann, M., & McLeod, J (1995). Rehabilitation the primary mediator of the lexical and seman- for pure alexia: Efficacy of therapy and implications for models of normal word tic effects. recognition. Neuropsychological Rehabilitation, 5, 149–180. Behrmann, M., Moscovitch, M., Black, S.E., & Mozer, M. (1990). Perceptual and conceptual factors in neglect dyslexia: Two contrasting case REFERENCES studies. Brain 113(4), 1163–1183. Behrmann, M.,Nelson, J.,&Sekuler, E. (1998). Vis- Ajax, E.T. (1967). Dyslexia without agraphia. Ar- ual complexity in letter-by-letter reading: “Pure” chives of Neurology, 17, 645–652. alexia is not sopure. Neuropsychologia, in press. Albert, M.L., Yamadori, A., Gardner, H., & Howes, Behrmann, M., & Shallice, T. (1995). Pure alexia: D. (1973). Comprehension in alexia. Brain, 96, An orthographic not spatial disorder. Cognitive 317–328. Neuropsychology, 12(4), 409–454.

46 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

Black, S.E., & Behrmann, M. (1994). Localization Coltheart, M. (1981). Disorders of reading and in alexia. In A. Kertesz (Ed.), Localization and their implications for models of normal reading. neuroimaging in neuropsychology (pp. 331–376). Visible Language, 15, 245–286. San Diego, CA: Academic Press. Coltheart, M. (1983). The right hemisphere and Bowers, J.S., Arguin, M., & Bub, D.N. (1996). Fast disorders of reading. In A.W. Young (Ed.), Func- and specificaccess to orthographic knowledge in tions of the right cerebral hemisphere, London: Aca- a case of letter-by-letter reading. Cognitive demic Press. Neuropsychology, 13(4), 525–567. Coslett, H.B., & Monsul, N. (1994). Reading with Bowers, J.S., Bub, D., & Arguin, M. (1966). A char- therighthemisphere: Evidence from transcranial acterization of the word superiority effect in pure magnetic stimulation. Brain and Language, 46, alexia. Cognitive Neuropsychology, 13, 415–441. 198–211. Brunn, J.L., & Farah, M. (1991). The relation be- Coslett, H.B., &Saffran, E.M. (1989a). Evidence for tween spatial attention and reading: Evidence preserved reading in “pure” alexia. Brain, 112, from the neglect syndrome. Cognitive Neuropsy- 327–359. chology, 8(1), 59–75. Coslett, H.B., & Saffran, E.M. (1989b). Preserved Bub, D., & Arguin, M. (1995). Visual word activa- object recognition and reading comprehension in tion in pure alexia. Brain and Language, 49, optic aphasia. Brain, 112, 1091–1110 77–103. Coslett, H.B. &Saffran, E.M. (1992). Optic aphasia Bub, D.N., Arguin, M., & Lecours, A.R. (1993). and the right hemisphere: A replication and ex- Jules Déjerine and his interpretation of pure tension, Brain and Language, 43, 148–161. alexia. Brain and Language, 45, 531–559. Coslett, H.B., & Saffran, E.M. (1994). Mechanisms Bub, D., Black, S.E., & Howell, J. (1989). Word rec- of implicit reading in alexia. In M.J. Farah & G. ognition and orthographic context effects in a Ratcliff (Eds.), The neuropsychology of high-level letter-by-letter reader. Brain and Language, 36, vision (pp. 299–330). Hillsdale, NJ: Lawrence 357–376. Erlbaum Associates Inc. Bub, D., & Gum, T. (1988). Psychlab experimental Coslett, H.B., Saffran, E.M., Greenbaum, S., & soft ware. Montreal: Montreal Neurological Insti- Schwartz, H. (1993). Reading in pure alexia. tute. Brain, 116, 327–359. Buxbaum, L., &Coslett,H.B. (1996). Deep dyslexic Damasia, A., & Damasio, H. (1983). The anatomic phenomenon in a letter-by-letter reader. Brain basis of pure alexia. Neurology, 33, 1573–1583. and Language, 54, 136–167. Doctor, E.A., Sartori, G., & Saling, M.M. (1990). A Caplan, L.R., & Hedley-Whyte, T. (1974). Cuing letter-by-letter reader who could not read non- and memory dysfunction in alexia without words. Cortex, 26, 247–262. agraphia. Brain, 97, 251–262. El Alaoui-Faris, M.,Bienbelaid, F., Alaoui, C., Tahiri, Chialant, D. & Caramazza, A. (this issue). Percep- L., Jiddane, M., Amarti, A., & Chkili, T. (1994). tual and lexical factors in a case of letter-by-letter Alexie sans agraphie en langue arabie etude reading. Cognitive Neuropsychology, 15(1/ 2). neurolinguistique et IRM. Revue Neurologique Cohen, J.D., MacWhinney, B., Flatt, M., & Provost, (Paris), 150, 771–775. J. (1993). PsyScope: A new graphic environ- Farah, M.J. (1991). Patterns of co-occurrence ment for designing psychology experiments. Be- among the associative agnosias: Implications for havioural Research Methods, Instruments and visual object recognition. Cognitive Neuropsychol- Computers, 25, 257–271. ogy, 8(1), 1–19. Coltheart, M. (1980). Deep dyslexia: A right hemi- Farah, M.J. (1994). Visual perception and visual sphere hypothesis. In M. Coltheart, K.E. Patter- awareness after brain damage: A tutorial review. son, & J.C. Marshall (Eds.), Deep dyslexia (pp. In C. Umiltà & M. Moscovitch (Eds.), Attention 326–380). London: Routledge. and performance XV: Conscious and nonconscious

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 47 BEHRMANN, PLAUT, NELSON

information processing (pp. 37–76). Cambridge, Inferences from purealexia without hemianopia. MA: Bradford Books, MIT Press. Neurology, 35, 962–968. Farah, M.J. (1997). Are there orthography-specific Hinton, G.E. (1990). Mapping part-whole hierar- brain regions? Neuropsychological and compu- chies in connectionist networks. Artificial Intelli- tational investigations. In R.M. Klein & P.A. gence, 46, 47–75. McMullen (Eds.), Converging methods for under- Hinton, G.E., & Shallice, T. (1991). Lesioning an standing reading and dyslexia. Cambridge, MA: attractor network: Investigations of acquired MIT Press. dyslexia. Psychological Review, 98(1), 74–95. Farah, M.J., & Wallace, M. (1991). Pure alexia as a Horikoshi, T., Asari, Y., Watanabe, A., Nagaseki, Y., visual impairment: A reconsideration. Cognitive Nukui, H., Sasaki, H., & Komiya, K. (1997). Neuropsychology, 8(3/4), 313–334. Music alexia in a patient with mild pure alexia: Feinberg, T.E., Dyckes-Berke, D., Miner, C.R., & Disturbed meaningful perception of nonverbal Roane, D.M. (1995). Knowledge, implicit figures. Cortex, 33, 187–194. knowledge and metaknowledge in visual ag- Howard, D. (1991). Letter-by-letter readers: Evi- nosia and pure alexia. Brain, 118, 789–800. dence for parallel processing. In D. Besner and Basic processes in read- Freedman, L., & Costa, L. (1982). Pure alexia and G.W. Humphreys (Eds.), ing: Visual word recognition right hemiachromatopsia in posterior dementia. (pp. 34–76). Hillsdale, Journal of Neurology, Neurosurgery and Psychiatry, NJ: Lawrence Erlbaum Associates Inc. Johnston, J.C., & McClelland, J.L. (1980). Experi- 55(6), 500–502. mental tests of a hierarchical model of word Friedman, R.B., & Alexander, M.P. (1984). Pic- identification. Journal of Verbal Learning and Ver- tures, images, and pure alexia: A case study. bal Behavior, 19, 503–524. Cognitive Neuropsychology, 1(1), 9 23. – Just, M.A., & Carpenter, P. (1987). The psychology Friedman, R.B., Beeman, M., Lott, S.N., Link, K., of reading and language comprehension. Boston, Grafman, J., & Robinson, S. (1993). Modality- MA: Allyn & Bacon. specific phonological alexia. Cognitive Neuropsy- Karanth, P. (1985). Dyslexia in a Dravidian chology, 10 , 549–568. language. In K.E. Patterson, J.C. Marshall, & M. Friedman, R.B., & Hadley, J.A. (1982). Letter-by- Coltheart (Eds.), Surface dyslexia (pp. 251–260). letter surfacealexia. Cognitive Neuropsychology, 9, Hove, UK: Lawrence Erlbaum Associates Ltd. 1–23. Kashiwagi, T., & Kashiwagi, A. (1989). Recovery Geschwind, N. (1965). Disconnection syndromes process of a Japanese alexic without agraphia. in animals and man. Brain, 88, 237–294. Aphasiology, 3, 75–91. Greenblatt, S.H. (1973). Alexia without agraphia. Kay, J., & Hanley, R. (1991). Simultaneous form Brain, 96, 307–316. perception and serial letter recognition in a case Grossi, D., Fragassi, N.A., Orsini, A., Falco, F.A.D., of letter-by-letter reading. Cognitive Neuropsy- & Sepe, O. (1984). Residual reading capability chology, 8(3/4), 249–273. in a patient with alexia without agraphia. Brain Kinsbourne, M. & Warrington, E.K. (1962). A dis- and Language, 23, 337–348. order of simultaneous form perception. Brain, 85, Hanley, J.R., & Kay, J. (1992). Does letter-by-letter 461–486. reading involve the spelling system? Neuropsy- Kreindler, A., & Ionescu, Y. (1961). A case of chologia, 30, 237–256. “pure” word blindness. Journal of Neurology, Hanley, J.R., & Kay, J. (1996). Reading speed in Neurosurgery and Psychiatry, 24, 275–280. pure alexia. Neuropsychologia, 34(12), 1165–1174. Kucera, F., & Francis, W.N. (1967). Computational Henderson, L. (1982). Orthography and word recog- analysis of present-day American English. Provi- nition in reading. London: Academic Press. dence, RI: Brown University Press. Henderson, V., Friedman, R., Teng, E., & Weiner, J. LaBerge, D. (1983). Spatial extent of attention and (1985). Left hemisphere pathways in reading: letters and words. Journal ofExperimental Psychol-

48 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

ogy: Human Perception and Performance, 9, tasks: Where are they? Journal of Experimental 371–379. Psychology: General, 118(1), 43–71. Làdavas, E., Umiltà, C., & Mapelli, D. (1997). Lexi- Montant, M., Behrmann, M., & Nazir, T. (1998). cal and semantic processing in the absence of Lexical activation in pure alexia. Manuscript in word reading: Evidence from neglect dyslexia. preparation. Neuropsychologia, 35(8), 1075–1085. Montant, M., Nazir, T.A., & Poncet, M. (this issue). Landis, T., Regard, M., & Serrat, A. (1980). Iconic Pure alexia and the viewing position effect in reading in a case of alexia without agraphic printed words. Cognitive Neuropsychology, caused by a brain tumour. A tachistoscopic 15(1/ 2). study. Brain and Language, 11, 45–53. Morton, J. (1969). The interaction of information in Lazar, R.M., & Scarisbrick, D. (1973). Alexia with- word recognition. Psychological Review, 76, out agraphia: A functional assessment of behav- 165–178. iour in focal neurologic disease. The Psychological Morton, J. (1979). Word recognition. In J. Morton Record, 43, 639–650. & J.C. Marshall (Eds.), Psycholinguistic Series 2. Levine, D.M., & Calvanio, R.A. (1978). A study of London: Elek. the visual defect in verbal alexia-simultanag- Mozer, M.C.. (1991). The perception of multiple ob- nosia. Brain, 101, 65–81. jects: A connectionist approach. Cambridge, MA: Marshall, J.C., & Newcombe, F. (1973). Patterns of MIT Press. paralexia: A psycholinguistic approach. Journal Mozer, M.C., &Behrmann, M. (1990). On theinter- of Psycholinguistic Research, 2, 175–199. action of selective attention and lexical knowl- Mayall, K., & Humpreys, G.W. (1996). A connec- edge: A connectionist account of neglect tionist model of alexia: Covert recognition and dyslexia. Journal of Cognitive Neuroscience, 2(2), case mixing effects: British Journal of Psychology, 96–123. 87, 355–402. Mozer, M.C., & Behrmann, M. (1992). Reading McClelland, J.L. (1979). On the time relations of with attentional impairments: A brain-damaged mental processes: An examination of systems of model of neglect and attentional dyslexia. In R. processes in cascade. Psychological Review, 86, Reilly & N.E. Sharkey (Eds.), Connectionist ap- 287–300. proaches to natural language processing (pp. McClelland, J.L. (1987). The case for interaction- 409–460). Hove, UK: Lawrence Erlbaum. ism in language processing. In M. Coltheart Nelson, J., Behrmann, M., &Plaut, D.C. (1998). The (Ed.), Attention and Performance XII (pp. 3–36). interaction of stimulus degradation and word length Hove, UK: Lawrence Erlbaum Associates Ltd. with frequency and with imageability. Manuscript McClelland, J.L., & Rumelhart, D.E. (1981). An in- in preparation. teractive activation model of context effects in Nitzberg-Lott, S., Friedman, R.B., & Linebaugh, letter perception: Part 1. An account of basic C.W. (1994). Rationale and efficacy of a tactile- findings. Psychological Review, 88(5), 375–407. kinaesthetic treatment for alexia. Aphasiology, 8, Michel, F., Henaff, M.A., & Intriligator, J. (1996). 181 195. Two different readers in the same brain after O’Reagan, K., & Levy-Schoen, A. (1989). Eye- a posterior callosal lesion. NeuroReport, 7, movement strategy and tactics in word 768–788. recognition and reading. In K. Rayner & A. Miozzo, M., & Caramazza, A. (this issue). Varie- Pollatsek (Eds.), The psychology of reading (pp. ties of pure alexia: The case of failure to access 361–383). Englewood Cliffs, NJ: Prentice Hall. graphemic representations. Cognitive Neuropsy- Patterson, K.E., & Kay, J. (1982). Letter-by-letter chology, 15(1/ 2). reading: Psychological descriptions of a neuro- Monsell, S., Doyle, M.C., & Haggard, P.N. (1989). logical syndrome. Quarterly Journal of Experimen- Effects of frequency on visual word recognition tal Psychology, 34A, 411–441.

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 49 BEHRMANN, PLAUT, NELSON

Perri, R., Bartolomeo, P., & Silveri, M.C. (1996). Regard, M., Landis, T., & Hess, K. (1985). Pre- Letter dyslexia in a letter-by-letter reader. Brain served stenography reading in a patient with and Language, 53, 390–407. pure alexia. Archives of Neurology, 42, 400–402. Plaut, D.C. (1995). Double dissociation without Reuter-Lorenz, P., & Brunn, J. (1990). A prelexical modularity: Evidence from connectionist basis for letter-by-letter reading: A case study. neuropsychology. Journal of Clinical and Experi- Cognitive Neuropsychology, 7, 1–20. mental Neuropsychology, 17(2), 291–321. Rumelheart, D.E., & McClelland, J.L. (1982). An Plaut, D.C. (1998). Connectionist modelling of interactive activation model of context effects in normal and impaired word reading. Manuscript letter perception: Part 2. The contextual enhance- in preparation. ment effect and some tests and extensions of the Plaut, D.C., McClelland, J.L., Seidenberg, M.S. model. Psychological Review, 89, 60–94. (1995). Reading exception words and Saffran, E.M., Bogyo, L.C., Schwartz, M.F., &Marin, pseudowords: Are two routes really necessary? O.S.M. (1980). Does deep dyslexia reflect right In J.P. Levy, D. Bairaktaris, J.A. Bullinaria, & hemisphere reading? In M. Coltheart, K.E. Pat- P. Cairns (Eds.), Connectionist models of mem- terson, & J.C. Marshall (Eds.), Deep dyslexia (pp. ory and language (pp. 145–159). London: UCL 381–406). London: Routledge. Press. Saffran, E.M., & Coslett, H.B. (this issue). Implicit Plaut, D.C., McClelland, J.L., Seidenberg, M., & Pat- vs. letter-by-letter reading in pure alexia: A tale terson, K.E. (1996). Understanding normal and of two systems. Cognitive Neuropsychology, impaired word reading: Computational princi- 15(1/ 2). ples in quasi-regular domains. Psychological Re- Schacter, D., McAndrews, M.P., & Moscovitch, M. view, 103, 56 115. (1988). Access to consciousness: Dissociations Plaut, D.C., & Shallice, T. (1993). Deep dyslexia: A between explicitknowledge in neuropsychologi- case study of connectionist neuropsychology. cal syndromes: In L. Weiskrantz (Ed.), Thought Cognitive Neuropsychology, 10(5), 377–500. without language. Oxford: Oxford University Price, C.J., & Humphreys, G.W. (1992). Letter-by- Press. letter reading? Functional deficits and compen- Schacter, D., Rapcsak, S.Z., Rubens, A.B., Tharan, satory strategies. Cognitive Neuropsychology, 9(5), M., & Laguna, J. (1990). Priming effects in a 427–457. letter-by-letter reader depend upon access to the Price, C.J., & Humphreys, G.W. (1995). Contrast- word form system. Neuropsychologia, 28(10), ing effects of letter-spacing in alexia: Further 1079–1094. evidence that different strategies generate w ord Schiepers, C. (1980). Response latency and accu- length effects in reading. Quarterly Journal of Ex- racy in visual word recognition. Perception and perimental Psychology, 48A, 573–597. Psychophysics, 27, 71–81. Rapcsak, S.Z., Rubens, A.B., & Laguna, J.F. (1990). Seidenberg, M., & McClelland, J.L. (1989). A dis- From letters to words: Procedures for word rec- tributed, developmental model of word recogni- ognition in letter-by-letter reading. Brain and Lan- tion and naming. Psychological Review, 96, guage, 38, 504–514. 523–568. Rapp, B.C., & Caramazza, A. (1991). Spatially de- Sekuler, E., & Behrmann, M. (1996). Perceptual termined deficits in letter and word processing. cues in pure alexia. Cognitive Neuropsychology, Cognitive Neuropsychology, 8(3/4), 275–312 13(7), 941–974. Rayner, K., &Pollatsek, A. (1987). Eye movements Shallice, T. (1988). From neuropsychology to mental in reading: A tutorial review. In M. Coltheart structure. Cambridge: Cambridge University (Ed.), Attention and performance XII: The psychol- Press. ogy ofreading (pp. 327–362). Hove, UK: Lawrence Shallice, T., & Saffran, E. (1986). Lexical process- Erlbaum Associates Ltd. ing in the absence of explicit word identification:

50 COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) LETTER-BY-LETTER READING

Evidence from a letter-by-letter reader. Cognitive Sternberg, S. (1969). The discovery of processing Neuropsychology, 3(4), 429–458. stages: Extensions of the Donders method. Acta Shallice, T., & Warrington, E.K. (1980). Single Psychologia, 30, 276–315. and multiple component central dyslexic syn- Strain, E., Patterson, K., & Seidenberg, M.S. (1995). dromes. In M. Coltheart, K.E. Patterson, & J.C. Semantic effects in single-word naming. Journal Marshall (Eds.), Deep dyslexia. London: Rout- of Experimental Psychology: Learning, Memory & ledge. Cognition, 21(5), 1140–1154. Sieroff, E., Pollatsek, A., & Posner, M.I. (1988). Terry, P., Samuels, S.J., & LaBerge, D. (1976). The effects of letter degradation and letter spacing on Recognition of visual letter strings following word recognition. Journal of Verbal Learning and injury to the posterior visual spatial atten- Verbal Behavior, 15, 577–585. tion system. Cognitive Neuropsychology, 5(4), Townsend, J.T. (1990). Serial vs. parallel process- 427–449. ing. Psychological Science, 1, 46–54. Snodgrass, J.G., & Mintzer, M. (1993). Neighbour- Tuomainen, J., & Laine, M. (1991). Multiple oral hood effects in visual word recognition: Facilita- re-reading technique in rehabilitation of pure tory or inhibitory? Memory and Cognition, 21(2), alexia. Aphasiology, 5, 401–409. 247–266. Van Orden, G.C., Pennington, B.F., & Stone, G.O. Snodgrass, J.G., & Vanderwart, M.A. (1980). A (1990). Word identification in reading and the standardised set of 260 pictures:Norms for name promise of subsymbolic psycholinguistics. Psy- agreement, image agreement, familiarity and cholinguistic Review, 97(4), 488–522. visual complexity. Journal of Experimental Psy- Vargha-Kadem, F., Carr, L.J., Isaacs, E., Brett, F., chology: Learning, Memory and Cognition, 6, Adams, C., & Mishkin, M. (1997). Onset of 174–215. speech after left hemispherectomy in a 9-year- Speedie, L., Rothi, L.J., & Heilman, K.M. (1982). old boy. Brain, 120, 159–182. Spelling dyslexia: A form of cross-cuing. Brain Vigliocco, G., Semenz a, C., & Neglia, M. (1992). and Language, 15, 340–352. Different deficits and different strategies in let- Sperling, G., & Melcher, M.J. (1978). Visual ter-by-letter reading. Paper presented at the search, visual attention and the attention Academy of Aphasia, Toronto. Warrington, E.K., & Shallice, T. (1980). Word- operating characteristic.In J. Requin (Ed.), Atten- form dyslexia. Brain, 103, 99 112. tion and performance, Vol. 7. Hillsdale, NJ: – Weekes, B.S. (1997). Differential effects of number Erlbaum. of letters on word and nonword naming latency. Stachowiak, F.L., & Poeck, K. (1976). Functional Quarterly Journal of Experimental Psychology, disconnection in pure alexia and color naming 50A(2), 439–456. deficit demonstrated by facilitation methods. Young, A.W., &Ellis,A.W. (1985). Different meth- Brain and Language, 3, 135–143. ods of lexical access for words presented in the Staller, L., Buchanan, D., Singer, M., Lappin, J., & left and right visual hemifields. Brain and Lan- Webb, W. (1978). Alexia without agraphia: An guage, 24, 326–358. experimental case study. Brain and Language, 5, Young, A.W., & De Haan, E. (1990). Impairments 378–387. of visual awareness. Mindand Language, 5, 29–48.

COGNITIVE NEUROPSYCHOLOGY, 1998, 15 (1/2) 51