COGNITIVE NEUROPSYCHOLOGY, 2002, 19 (7), 653– 671web.jhu.edu › cogsci › templates › images › rapp › ... · More recently, evidence for the interaction of lexical and sublexical

THE INTERACTION OF LEXICAL AND SUBLEXICALINFORMATION IN SPELLING WHATrsquoS THE POINT

Jocelyn R FolkKent State University OH USA

Brenda Rapp and Matthew GoldrickJohns Hopkins University Baltimore USA

Most theories of spelling propose two major processes for translating between orthography and phonol-ogy a lexical process for retrieving the spellings of familiar words and a sublexical process for assemblingthe spellings of unfamiliar letter strings based on knowledge of the systematic correspondences betweenphonemes and graphemes We investigated how the lexical and sublexical processes function and inter-act in spelling by selectively interfering with the sublexical process in a dysgraphic individual By com-paring spelling performance under normal conditions and under conditions of sublexical disruption wewere able to gain insight into the functioning and the unique contributions of the sublexical processThe results support the hypothesis that the sublexical process serves to strengthen a target word andprovide it with a competitive advantage over orthographically and phonologically similar word neigh-bours that are in competition with the target for selection

INTRODUCTION

Most accounts of spelling and reading propose twomajor routes or processes for translating betweenorthography and phonology a lexical and asublexical process The lexical process is assumed toretrieve the spellings or pronunciations for familiarletter strings The sublexical process assembles aspelling or pronunciation for unfamiliar letterstrings using knowledge of the systematic corre-spondences between phonemes and graphemes Inspelling processing in the lexical system involvesretrieving a wordrsquos spelling from a long-term mem-ory store referred to as the orthographic output lexi-con (eg Ellis 1982 Patterson 1988) whereasprocessing in the sublexical system involves theassembly of a plausible spelling from a phonological

code (eg Goodman amp Caramazza 1986) (seeFigure 1) In addition some theorists propose asecond lexical route (a third route) that directlyassociates representations in the phonological inputlexicon with representations in the orthographicoutput lexicon bypassing semantics (Patterson1986) Although the lexical and sublexical pro-cesses are hypothesised to be independent in thatneither requires the other and each can be lesionedindependently (Beauvois amp Deacuterouesne 1981Shallice 1981) there is also considerable evidencethat the two processes interact (Barry amp Seymour1988 Hillis amp Caramazza 1991 Hillis Rapp ampCaramazza 1999 Rapp Epstein amp Tainturier2002) The current study investigates the purposethat might be served by this lexicalsublexicalinteraction in spelling We present data from a

COGNITIVE NEUROPSYCHOLOGY 2002 19 (7) 653ndash671

Oacute 2002 Psychology Press Ltdhttpwwwtandfcoukjournalspp02643294html DOI10108002643290244000184

653

Requests for reprints should be addressed to Jocelyn R Folk Department of Psychology PO Box 5190 Kent State UniversityKent Ohio 44242-0001 USA (Tel 330-672-4095 Fax 330-672-3786 Email jfolkkentedu)

We very much appreciate the support of NIMH grant R29MH55758Amy Kanersquos many contributions to the investigation andMMDrsquos cheerful patience

Q0120ndashCN2901 Sep 24 02 (Tue) [19 pages 2 tables 4 figures 0 footnotes 3 Appendices] ndash S endings PDF File Read

dysgraphic subject which specifically supports thehypothesis that lexical and sublexical interactionserves to strengthen the representations of targetwords and their constituent graphemes

Evidence of interaction

In research on reading much of the evidence forinteraction between lexical and sublexical processescomes from regularity effects whereby irregularwords (words with at least one low-frequencygrapheme-phoneme correspondence) take longerto read aloud than regular words (words containingonly high-frequency grapheme-phoneme corre-

spondences) (eg Baron amp Strawson 1976Stanovich amp Bauer 1978) Regularity effects arethought to arise because for irregular words the lex-ical process will generate the correct pronunciationfor a word stimulus (eg BROAD reg b r d)whereas the sublexical process will regularise thepronunciation generating a pronunciation that isconsistent with the most common way of pro-nouncing the graphemes of the word (egBROAD reg b r o d) In this way conflicting out-puts will be generated for irregular but not regularwords Longer pronunciation times for irregularwords are assumed to result from the time it takesto reconcile the conflicting outputs of the twoprocesses (eg Paap amp Noel 1991) In effectregularity effects arise from the interaction oflexical and sublexical information (eg ColtheartRastle Perry Langdon amp Ziegler 2001 ZorziHoughton amp Butterworth 1998)1

More recently evidence for the interaction oflexical and sublexical processes in spelling hasbegun to emerge Evidence has come both fromstudies involving the priming of nonword spellingsin intact subjects and from case studies ofdysgraphic subjects Various studies with intactsubjects have examined the influence of the spell-ings of previously heard rhyming words (ldquolexicalprimingrdquo) on nonword spellings (Barry amp Seymour1988 Campbell 1983 Cuetos 1993) For ex-ample Barry and Seymour compared the spellingsthat subjects produced for nonwords on lexicalpriming trials where a nonword was preceded by arhyming word with the spellings produced for thesame nonwords when they were not preceded bywords Barry and Seymour found lexical primingeffects such that nonword spellings were influencedby the spellings of previously heard rhyming wordsFor example although the most common spellingof the nonword n ^ T was N-U-T-C-H (Hanna

654 COGNITIVE NEUROPSYCHOLOGY 2002 19 (7)

FOLK RAPP GOLDRICK

Figure 1 The functional architecture of the spelling system

1 Certain connectionist accounts of reading do not posit a separate lexical and sublexical process (Plaut McClelland Seidenbergamp Patterson 1996) but instead posit phonological and semantic processes The phonological process maps orthographic tophonological representations whereas the semantic process maps semantic to orthographic representations neither involving whatwould be traditionally considered to be lexical look-up Within this type of architecture it is the sublexical process that encodes thefrequency of orthography to phonology relationships giving rise to frequency times regularity interaction effects Moreover within thissort of architecture phonological and semantic processes are assumed to interact in that they both activate a common level oforthographic representation

Hanna Hodges amp Rudorf 1966) it was morelikely to be spelled N-O-U-C-H when preceded bythe oral presentation of the rhyming word ldquotouchrdquothan when there was no word prime Barry andSeymour (1988) interpreted this as evidence thatthe sublexical process is open to lexical influencemdashlexicalsublexical interaction However no specificmechanism for this interaction was proposed

A number of case studies of dysgraphic subjectsalso provide support for the proposal that lexicaland sublexical processes interact in spelling Forexample JJ (Hillis amp Caramazza 1991) madesemantic errors in the written and spoken namingof picture stimuli and in the comprehension ofitems from all semantic categories except for ani-mals This pattern of errors indicates a deficit at thelevel of the lexical semantic system (see Figure 1)Surprisingly JJ made no semantic errors in a task ofwriting to dictation Rather he made frequent pho-nologically plausible errors (PPEs) such as spellingldquoyachtrdquo as y-o-t (more than 90 of his errors) andthere was an effect of regularity on spelling accu-racy indicating that the sublexical process wasinvolved in his spelling of words Interestinglyhowever JJrsquos accuracy in spelling to dictation wasaffected by his comprehension (as assessed by defi-nition and picture-word verification tasks) Forthose irregular words that he at least partiallyunderstood his spelling accuracy was 72 in con-trast he spelled incorrectly all irregular words thathe did not understand This pattern of spellingerrors suggests that JJ relied neither on the lexicalsystem nor on the sublexical system If JJ relied onlyon the lexical system for spelling to dictation heshould have produced semantic errors as he did inwritten naming if he relied only on the sublexicalprocess then he should have produced PPEs for allwords with low-frequency phoneme-graphemecorrespondences (PGCs) regardless of compre-hension Hillis and Caramazza argued that thispattern of performance could be accounted for byassuming that JJ combined the outputs of the lexicaland sublexical processes to eliminate semanticerrors They reasoned that because of damage to thesemantic system a picture of a pear might have gen-erated an impoverished semantic representationsuch as [yellow fruit] This impoverished represen-

tation would in turn have activated multiple lexicalcandidates in the orthographic output lexicon suchas BANANA LEMON PEAR etc In writtenpicture naming an incorrect candidate wouldsometimes be selected from this set However inthe task of spelling to dictation the phonologicalinput p ae r would be processed by the sublexicalsystem that would generate a plausible spelling suchas P-A-I-R Though incorrect this output wouldbe sufficient to select correctly among the multipleactive lexical candidates (BANANA LEMONPEAR) because it shares more letters with the tar-get word (PEAR) than with the other candidates

Mechanism for interaction

Despite the growing body of evidence that sublexicaland lexical processes interact in spelling little atten-tion has been given to the question of the specificmechanism by which this interaction takes placeRapp et al (2002 Rapp 2002) proposed a mecha-nism of interaction that serves among other thingsto explain the reduction in semantic errors in spellingto dictation in cases such as JJrsquos (Hillis amp Caramazza1991 see also Hillis Rapp amp Caramazza 1999)The proposal consists of two elements (1) the inte-gration of lexical and sublexical information at thegrapheme level and (2) the feedback of informationfrom the graphemic level to the lexical level (seeFigure 2) With regard to integration it was sug-gested that both lexical and sublexical systems sendactivation to candidate graphemic elements at thelevel of the graphemic buffer Although activationfrom both systems is combined (integrated)activation from the lexical system normally domi-nates the process Subsequently via feedback linksbetween graphemes and lexemes the activatedgraphemes send activation to those words at thelexical level that contain them An iterative feed-forwardfeed-back process ensues until a point intime at which a spelling is selected for output

In support of the integrative component of thisproposal Rapp et al (2002) described the case of adysgraphic individual LAT who had difficultiesspelling words correctly (83ndash93 correct)although his sublexical and semantic processingwere relatively intact (90ndash98 accuracy in spelling

COGNITIVE NEUROPSYCHOLOGY 2002 19 (7) 655

LEXICAL AND SUBLEXICAL INTERACTION

nonwords 95 correct in defining words) Hisword spelling exhibited effects of both lexicalfrequency and phoneme-grapheme probabilityand his misspellings were almost always phonologi-cally plausible errors (PPEs) This performancepattern indicated that the orthographic output lexi-con was compromised and that LAT often relied onthe sublexical system to spell word stimuli Mostinteresting was the fact that LATrsquos PPEs oftenincluded very low probability PGCs that werecontained in the target word (eg ldquobouquetrdquo regBOUKET) While K is the most common spellingfor the phoneme k and it is thus not surprising tofind k reg K in a PPE ET is a very low-probabilityspelling for the phoneme ei Such low-probabilityspellings should be produced only very rarely by asublexical process that produces spellings accordingto the frequency of PGCs in the language (Good-man amp Caramazza 1986 Sanders amp Caramazza1990) Additionally LAT did not produce just any

low-frequency PGCsmdashhe often produced onesthat were lexically correct (eg ET in BOUKET)Rapp et al argued that PPEs containing low-probability lexically correct elements might beexpected if we assume a cognitive architecture inwhich the output of the sublexical system is com-bined with information originating from the lexicalsystem Although in the intact system the activa-tion process would be dominated by the lexicalprocess in certain individuals (such as LAT) whosethe lexical system has sustained damage activationfrom the lexical system may be reduced (but noteliminated) yielding the types of responsesobserved in LATrsquos case For example if the lexicalsystem only moderately activates the correctgraphemic elements (B-O-U-Q- U-E-T) whilethe sublexical system strongly activates a set ofplausible elements (B-U-K-A-Y) the outcomemay contain elements of both responses(BOUKET)

FOLK RAPP GOLDRICK


Figure 2 An example of the integration of lexical and sublexical information at the grapheme level and the feedback of information fromthe graphemic level to the lexical level in the spelling system

McCloskey Macaruso and Rapp (1999) pro-vided empirical support for the feedback compo-nent of the proposal with their report on thedysgraphic subject CM CMrsquos spelling errors con-tained numerous intrusions of letters that were notin the target word (eg ldquoskirtrdquo reg SKIN T) and heproduced lexical substitutions (eg ldquotoolrdquo regTOOK) Interestingly the letter intrusions tendedto be letters that appeared in CMrsquos preceding spell-ing responses and McCloskey et al argued that theletter intrusions reflected an impairment in whichelevated levels of grapheme activation persistedeven after a response had been made Thus whenspelling a word CM would encounter interferencefrom graphemes that had remained active from pre-vious responses This typically resulted in the pro-duction of nonwords however approximately onethird of CMrsquos errors were lexical substitutions (ldquofitrdquoreg FILTER)mdasha rate greater than expected fromthe chance intrusion of persisting lettersMcCloskey et al explained this greater than chancelevel of lexical errors by assuming feedback connec-tions from graphemes to lexical entries in the ortho-graphic output lexicon For example if the targetword ldquochainrdquo had been preceded by ldquodeskrdquo thenwhen the graphemes of C H A I N were acti-vated some of the letters of D E S K might stillbe active The total set of active letters (from bothldquochainrdquo and ldquodeskrdquo) would feed activation back tothe lexemes containing those letters (eg DEANSKI etc) Under these circumstances even thoughDESK was the target a word such as SKI couldenter into competition with it and sometimes beproduced

Within the proposed framework activationfrom the sublexical system provides support forhigh-probability graphemes The fact that bothregular and irregular words are largely made up ofhigh-probability graphemes (Houghton amp Zorzi1998) implies that the sublexical system will typi-cally contribute to the activation of the target wordand its constituent graphemes This should providethe target with additional capacity to withstand anynoise that might arise in the course of processingand allow it to compete successfully with otherlexemes that might also be active Within thisframework the reduction of semantic errors in

spelling to dictation vs written naming observed incases such as JJrsquos can be understood by assumingthat the phonologically plausible graphemes acti-vated by the sublexical system (eg P-A-I-R forthe target ldquopearrdquo) feed activation back to the lexemelevel contributing more to the activation of the tar-get lexeme (PEAR) than to the other lexical candi-dates (BANANA LEMON etc)

A proposal by Houghton and Zorzi (19982002) can also account for certain of the findingsdiscussed here without positing a feedback mech-anism According to their proposal lexical activa-tion is reinforced by input from the sublexicalsystem but the interaction between the lexical andsublexical process occurs only at the graphemic out-put stage without feedback from the grapheme tothe orthographic lexeme level A computer simula-tion of their proposal has successfully simulatedLATrsquos data pattern However although both theRapp et al (2002) and Houghton and Zorzi (1998)proposals posit that sublexical information canreinforce lexical activation the proposal by Rapp etal also accounts for the McCloskey et al (1999)data which indicates that there is feedback fromthe grapheme to the lexeme level in spelling Fur-thermore it remains to be seen whether or not thereduction in the number of semantic errors in spell-ing to dictation versus written picture naming (asexhibited by JJ Hillis amp Caramazza 1991) can beachieved in a system that lacks feedback connec-tions A determination of the role of feedback inaccounting for certain phenomena is likely torequire further computational work

The sublexical process and lexicalcompetition

Lexicalsublexical interaction in the reading systemis often discussed in terms of the conflicts it gener-ates for irregular words in reading (eg ColtheartRastle Perry Langdon amp Ziegler 2001) althoughconflict generation is unlikely to be its primaryfunction Recent computational work in spellingsuggests that although interaction between lexicaland sublexical processes may sometimes generateconflict most of the time the outputs from the twosystems reinforce each other (Houghton amp Zorzi



1998 2002 see Zorzi Houghton amp Butterworth1998 see Plaut et al 1996 for a similar account inreading) The hypothesis we wish to consider is thatlexicalsublexical interaction in spelling serves tostrengthen the target so that among other things itwill be able to compete successfully with other acti-vated word neighbours Other activated words mayeither be semantic neighbours or form-related wordneighbours2 In the proposed framework semanticneighbours may be activated as a result of a cascad-ing feedforward process by which words sharing thesemantic features of the target are activated at somepoint in the process With regard to form-relatedneighbours there are multiple possible sources ofactivation cascading feed-forward activation fromwords in the phonological input lexicon that arephonologically related to the target feedback acti-vation from the graphemes of the target itself andfeedback activation from the graphemes activatedby the sublexical process

In an intact system few errors are produced andso we are unlikely to easily ldquoseerdquo evidence of thesecompetitors However individual cases of acquireddysgraphia may provide us with opportunities towitness at least certain aspects of this competitionIn cases such as JJrsquos where there has been semanticlevel damage and specifically in the task of writtenpicture naming (where the sublexical system may beof little help) we see evidence of the semantic com-petitors In other cases where there is no semanticdamage but instead there is a weakened ortho-graphic output lexicon we might expect form-related neighbours to be more prominent

If this hypothesis regarding the functional roleof lexicalsublexical interaction is correct thenunder conditions where the sublexical system iscompromised (either as a result of damage or exper-imental manipulation) a target wordrsquos graphemeswill be less active and presumably more susceptibleto interference from the most highly active lexicalcompetitors

In the current study we investigated this predic-tion by examining the consequences of selectivelyinterfering with the sublexical process in an indi-vidual whose orthographic lexicon was compro-

mised by neurological damage and who producednumerous lexical substitution errors in spelling Weused articulatory suppression to interfere with thesublexical process assuming that by selectivelyldquodeactivatingrdquo the sublexical system we couldobserve the functioning of an isolated and weak-ened lexical process By comparing spelling perfor-mance with and without articulatory suppressionwe were able to gain information regarding the spe-cific contributions of the sublexical process to spell-ing In particular we found support for thehypothesis that the sublexical process contributesimportantly to the competition between a targetword and its form-related neighbours

CASE STUDY

MMD was a right-handed woman who suffered acerebrovascular accident (CVA) at the age of 65 6months prior to the onset of this investigationMMD had a high-school education and worked ina clerical position until retirement CT scanningindicated a left posterior parietal and superior tem-poral lesion (Figure 3) The CVA resulted in mildspoken language difficulties and a significant spell-ing impairment

Spoken language comprehension andproduction

MMDrsquos auditory comprehension of single wordswas very good as evidenced by her relatively normalperformance on the tasks of auditory discrimina-tion auditory lexical decision and wordpictureverification MMD was 98 correct (4546) in anauditory discrimination task in which she was askedto respond whetherword pairs were the same or dif-ferent On the combined Imageability and Mor-phology auditory lexical decision tasks from thePsycholinguistic Assessments of Language Pro-cessing in Aphasia (PALPA) tests (Kay Lesser ampColtheart 1992) MMD was correct on 95 ofword trials (105110) and 91 (100110) ofnonword trials (scores generally within normal

FOLK RAPP GOLDRICK


2 Form-related word neighbours are any real words that are not semantically or morphologically related to the target



Figure 3 CT scans for MMD indicating a left-hemisphere parietal lesion extending to the superior temporal gyrus The images reveal theinferior-superior extension of the lesions The right side of each image corresponds to the left hemisphere

range) Finally MMD was 95 correct (247260)on an auditory wordpicture verification task whereshe had to indicate whether a spoken namematched a picture (based on 260 line drawings ofobjects from Snodgrass amp Vanderwart 1980) Thisscore was at the low end of the normal range

MMD exhibited mild spoken production diffi-culties in single word reading picture naming andrepetition Given her good comprehension hermild difficulties on these tasks presumably repre-sent some deficits specifically affecting mechan-isms of spoken production To test her single wordreading abilities MMD was given 231 words typedin lower-case script and presented one at a timeMMD made errors on 13 of trials (31231) Themajority (55 1731) of these errors consisted ofwords that were visually andor phonologicallysimilar to the target (eg pierce reg ldquofiercerdquo) Inaddition she also produced morphologicalsemantic and nonword errors (see Table 1)MMDrsquos error rate on nonword reading was 64(68107) The majority of these errors (3568)were phonologically similar nonwords (eg kroidreg k Aring d) suggesting a deficit in grapheme-to-phoneme conversion The remaining errors (3368) consisted of form-related words (eg reesh regldquobeachrdquo)

MMDrsquos picture naming accuracy level and errortypes were generally similar to those of her singleword reading MMD was given pictures fromSnodgrass and Vanderwart (1980) for oral namingand she made errors on only 8 of trials (20253)Her errors primarily consisted of responses thatwere phonologically or visually similar to the target(see Table 1)3

MMDrsquos mild spoken production difficultieswere also evident in her repetition where she madeerrors on only 1 of trials with word targets (3218ldquoironrdquo reg ironing ldquoboughtrdquo reg fought ldquobuglerdquo regf u g l) but on 15 of trials with nonword targets(748) Three of these errors involved producingphonologicallysimilar words (eg s A r k l reg s Aringk lcircle) and the remaining four errors were pho-nologically similar nonwords (eg b r u T reg b ru f) Interestingly MMD made a number of mor-phological errors when asked to repeat a list of 71morphologically complex words On this listMMD made errors on 20 of trials (1471) and 12(86) of these errors were morphological (egturned reg ldquoturnrdquo) The remaining two errors werephonologically similar word errors (eg tried regldquotiredrdquo) This relatively large number of morpho-logical errors in repetition suggests some type ofmorphological deficit It is not clear whether these

FOLK RAPP GOLDRICK


Table 1 Examples of MMDrsquos picture naming errors and single word reading errors

Picture naming Single word readingmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdash mdashmdashmdashmdashmdashmdashmdashmdashmdashmdash

Error type errors Example errors Example

Form-related word 5 fox reg f l A k s 55 pierce reg fierceMorphological 10 carrot reg carrots 23 angry reg angerPhonologically related non-word 15 onion reg I Icirc ^ n 16 rust reg r ^ s

Semantic 20 pitcher reg cup 3 above reg belowOther 10 violin reg n I k 3 eats reg sitsVisually related word 15 cherry reg moonCircumlocution 15 watering can reg ldquowater your flowers

and stuff with itrdquoDonrsquot know response 10 needle reg ldquoI donrsquot know what it

looks likerdquo

Form-related word errors were either orthographically similar to the target phonologically similar or both

3Although the distribution of errors was somewhat different for the two tasks the low error rate and the fact that different sets of

items were administered makes a detailed comparison difficult

difficulties occurred on input output or centrallyWe did not however explore her morphologicaldifficulties any further and given the possibility ofthis additional morphological deficit we excludedmorphological errors from the analyses of her spell-ing errors

Written production

MMD wrote with her dominant right hand print-ing easily and legibly in lower case On each trial inspelling to dictation the target word was presentedto MMD aurally and she was asked to repeat thetarget word correctly before spelling it either inwriting or orally As noted earlier her repetitionwas excellent for words though she sometimesmade mistakes repeating pseudowords On thosefew occasions on which she made a repetition errorshe did not begin spelling until she correctlyrepeated the target stimulus

Across the course of the investigation MMDwas administered a total of 2430 words for writtenspelling to dictation She made errors on 867 ofthose wordsmdasha 36 error rate Of those errors28 (239867) were phonologically plausible errors(ldquocopyrdquo reg C-O-P-P-I-E) 59 were nonworderrors (eg ldquocoachrdquo reg C-H-O-A-H) and theremaining 14 (118867) were lexical substitutionsor ldquoform-related word errorsrdquo (eg ldquoloafrdquo reg L-E-A-F) On some trials MMD reported that anotherword was interfering with her trying to spell the tar-get Sometimes this resulted in a lexical substitu-tion and sometimes she ultimately spelled thetarget word correctly despite the reportedinterference

MMD was also administered 186 items for writ-ten picture naming Her accuracy level and errortypes were generally similar to those observed inspelling to dictation She had a 24 error rate onthis task and her errors were distributed as follows27 were phonologically plausible 33 werenonword errors 20 were lexical substitutions13 were semantic errors and 2 were ldquodonrsquotknowrdquo responses It is worth noting that the reduc-tion in semantic errors in spelling to dictation ver-sus written picture naming is consistent with theobservation of Hillis and Caramazza (1991) that

formed the basis of their proposal for interactionbetween lexical and sublexical processes in spellingHowever more detailed predictions regarding thedistribution of errors in the two tasks depends on anumber of assumptions regarding the strength ofvarious processes including MMDrsquos ability to self-generate the phonological forms of the target wordsin picture naming Since a comparison between thetwo tasks will not play a central role in the argu-ments we will present we will not discuss this com-parison any further

MMD was also administered 132 nonwords forwritten spelling to dictation A correct nonwordspelling response was scored as one that was phono-logically plausible according to the phoneme-grapheme correspondences of English (Hanna etal 1966) MMD spelled 32 of the 132 (24)nonwords correctly Her errors consisted of ortho-graphically legal nonwords (93 h r p reg H-A-P-E) orthographically illegal nonwords (3 k l ik reg K-A-C-L-K) and word errors (4 p E S reg P-E-A-C-H) A further analysis of thenonword errors revealed that MMD correctlyspelled78 (613790) of letters and 79 (417529)of phonological segments in the nonwords

Premorbid spelling

It is crucial to establish that MMDrsquos spelling diffi-culties are a result of neurological damage and not apremorbid lack of knowledge of word spellings Toinvestigate this we examined several pages of notesthat MMD had taken as part of a class at herchurch and a handwritten paper that she had writ-ten for the class prior to the CVA These premorbidwriting samples revealed less than 02 errors outof approximately 1650 words Of the fivepremorbid spelling errors three were letterdeletions (ldquoprimerdquo reg P-R-I-M ldquoChristianityrdquo regC-H-R-I-S-T-I-A-N-T-Y ldquosanctuariesrdquo reg S-A-N-C-T-U-A-R-E-S) one a letter transposition(ldquofirstrdquo reg F-R-I-S-T) and one a possible PPE(ldquotonsrdquo reg T-O-N-N-E-S) Her very low pre-morbid error rate is a testament to her strong intactpremorbid spelling abilities

To further ensure that MMDrsquos spelling errorsresulted from the CVA we asked MMD to spell a



subset of the words from the premorbid corpusShe was asked to write to dictation 51 words thatshe had spelled correctly premorbidly She spelledonly 63 of these words correctly (3251) a clearindication of the acquired nature of her spellingdeficit

Loci of the impairment in spellingMMDrsquos difficulties in spelling nonwords in the faceof relatively intact auditory discrimination and lexi-cal decision clearly indicate a moderate deficit tothe sublexical process However her difficulties inword spelling point to an additional deficit some-where within the lexical system In this regardMMDrsquos good repetition of monomorphemicwords good performance in auditory discrimina-tion and lexical decision and her excellent repeti-tion of target words on spelling trials indicated thather spelling difficulties cannot be attributed to adeficit in processing the input or in semantic pro-cessing To examine if difficulties arose frommodality-specific components of letter shape orname conversion we compared her performance inwritten and oral spelling administering four lists ofwords from the Johns Hopkins University (JHU)Dysgraphia Battery (Goodman amp Caramazza1985) for both written and oral spelling to dictation(N = 262) MMD showed no significant differencein spelling accuracy on the two tasks with 66accuracy in written spelling and 64 accuracy inoral spelling c2(1 N = 524) = 0008 p gt 05 Inaddition similar error types were produced in thetwo tasks PPEs form-related word errors andnonword errors These results clearly indicate nosignificant peripheral difficulties in either lettershape or letter name selection On the basis of thesimilarity between the two tasks we combine thedata from the two in all subsequent analyses

Having ruled out a peripheral deficit and deficitsaffecting her comprehension abilities we consid-ered whether the spelling deficit affected the ortho-graphic output lexicon the graphemic buffer orboth We concluded that MMD suffered from adeficit affecting her ability to activate representa-tions in the orthographic output lexicon success-fully on the basis of the following very clearevidence First a deficit to the graphemic buffer

was unlikely as there was no effect of word length asassessed by the Length List from the JHUDysgraphia Battery (Goodman amp Caramazza1985) with both oral and written spelling to dicta-tion This list contains 70 words 14 each in lengths4 through 8 with word frequency matched acrosseach word length Length effects were absentwhether evaluated by word accuracy c2(4 N = 140)= 411p gt 05 letter accuracy c2(4 N = 840)= 124p gt 05 or in a direct comparison of accuracy with 4-letter vs 8-letter words (88 vs 85) c2(1 N =336) = 0384 p gt 05 (see Table 2) This findingstrongly suggests that damage at the level of thegraphemic buffer is unlikely (Goodman ampCaramazza 1986) A deficit at the level of thegraphemic buffer is expected to result in a lengtheffect given the temporary nature of the informa-tion that is held in the buffer Presumably thememory trace decays rapidly The buffer must holdthe letters of a long word for more time than thoseof a short word it takes more time to serially outputmore letters making longer words more susceptibleto errors

Second there was a significant effect of wordfrequency 78 correct with high-frequency words(mean frequency = 150) (Kuccedilera amp Francis 1967)versus 53 with low-frequency words (mean fre-quency = 17) c2(1 N = 408) = 2938 p lt 05 (high-and low-frequency words were equated for length)It is generally assumed that at the level of the ortho-graphic output lexicon high-frequency words arerepresented more robustly than low-frequencywords and are therefore less susceptible to disrup-tion than low frequency words For this reason fre-quency effects are assumed to index deficitsaffecting the orthographic output lexicon

A third finding consistent with damage to theorthographic output lexicon was that MMD dis-

FOLK RAPP GOLDRICK


Table 2 MMDrsquos performance with words of different lengths inwritten and oral spelling to dictation combined

Word length Letters correct Words correct

4 88 615 87 646 87 617 89 578 85 39

played effects of phoneme-grapheme probabilityHer spelling performance on a list from the JHUDysgraphia Battery (Goodman amp Caramazza1985) indicated a small albeit nonsignificant effectof phoneme-grapheme mapping probability with83 correct for regular words versus 71 correct forirregular words c2(1 N = 216) = 335 p gt 05MMD was also administered for both oral andwritten spelling to dictation a more closelymatched list of 15 words containing high P-Gprobability segments and 15 words with low P-Gprobability segments (Hanna et al 1966) Thewords were matched on length consonant-vowelstructure word frequency and the position of thecritical low- or high-probability segment On thislist MMD spelled significantly more of the high-probability (85) than the low-probability (60)segments correctly c2(1 N = 108)= 842p lt 05

The final piece of evidence indicating an ortho-graphic output lexicon locus of impairment was thefact that 28 (239867) of MMDrsquos spelling errorswere PPEs (eg ldquotraderdquo reg T-R-A-I-D) theseerrors are assumed to arise in the context of a failureto activate a wordrsquos spelling from the orthographicoutput lexicon

In sum MMD suffered from two deficits withinthe spelling systemmdashone to the sublexical processanother to the orthographic output lexicon How-ever we do not make a distinction between whetherthe orthographic output lexicon itself or the processinvolved in activating the lexical representations isaffected (see Rapp amp Caramazza 1993 for a dis-cussion of the difficulties involved in making suchdistinctions)

Lexical substitution errorsAs indicated earlier a common error type forMMD (approximately 15 of her errors) were lexi-cal substitution errors (ldquopoiserdquo reg P-A-U-S-E)Given that we will be specifically concerned withmonitoring the rate of such errors in the experi-mental task we describe them in some detail at thispoint We take up the question of their origins inthe General Discussion

Throughout the course of the investigationMMD produced 91 word errors4 that were neithersemantically nor morphologically related to the tar-get word To document the form-similaritybetween targets and errors specifically we evaluatedthe phonological and orthographic overlap betweentargets and errors on various measures initial seg-ment overlap length overlap and overall segmentoverlap We found that 92 of the errors over-lapped in initial letter with the target and 95shared the first phoneme with the target We alsofound that 96 of the errors were within plusmn1 letterofthe target letter length and 97 were within plusmn1phoneme of the target phoneme length Finallyoverall letter overlap was 77 and overall phono-logical overlap was 69

EXPERIMENTAL STUDYCONTRIBUTIONS OF THESUBLEXICAL PROCESS

MMDrsquos spelling performance indicates that shesuffered damage to both the lexical and sublexicalspelling processes As we indicated earlier she pro-duced numerous form-related word errors or lexicalsubstitutions (eg trip reg T-R-I-B-E) which sug-gested that ldquoneighboursrdquo of the target word some-times successfully competed with the target Toinvestigate the contribution of the sublexical systemto this apparent competition among word neigh-bours we used a concurrent secondary taskmdasharticulatory suppressionmdashin order to attempt tointerfere with MMDrsquos sublexical system and inthis way isolate her already weakened lexical sys-tem We hypothesised that if lexical-sublexicalinteraction serves to strengthen correct lexicalresponses then disruption to the sublexical systemshould lead to an increase in the number of form-related word errors We chose articulatory suppres-sion as a secondary task because we wanted a taskthat would not be highly stressful to MMD and yetwould be somewhat demanding of attention Paapand Noel (1991) in an attempt to disrupt the



4We excluded 27 of MMDrsquos form-related word errors from the analysis because they were also morphologically related to the

target word perhaps arising because of a morphological deficit

sublexical process in reading argued that thesublexical process is more resource demanding thanthe lexical process and thus more likely to beaffected by increased attentional demands In addi-tion we hoped that the recruitment of phonologicalprocesses by the articulatory suppression task mightalso be more disruptive to the sublexical systemthan to the lexical one

Given the hypothesis under investigation wemade three predictions regarding the consequencesof articulatory suppression (1) There should be adecrease in the number of phonologically plausibleerrors produced This would provide independentevidence of the successful disruption of thesublexical system (2) There should be a decrease inthe regularity effect reflecting again the successfuldisruption of the sublexical system (3) If thesublexical system acts to strengthen correct lexicalresponses then when the contribution of that sys-tem is disrupted there should be an increase in therate of form-related word errors from MMDrsquosalready weakened lexical system

Procedure

MMD was administered 320 words under two con-ditionsmdash ldquonormalrdquo written spelling to dictation andwritten spelling to dictation under articulatory sup-pression In both conditions MMD had to repeatthe target word before spelling it additionally inthe articulatory suppression condition after theexperimenter said the target word aloud MMD hadto repeat the word then begin to say ldquoba ba bardquoand continue saying this while writing the wordThe presentation of the words in the two conditionswas counterbalanced such that half of the wordswere first presented for normal spelling followed byspelling under suppression at a later testing datethe other half were presented for spelling undersuppression first followed by normal spelling at alater testing date Thus in a given testing sessionMMD was asked to perform both normal writtenspelling to dictation and written spelling under sup-pression in blocked presentation

Results

MMDrsquos overall spelling accuracy decreased signifi-cantly under suppressionmdashshe spelled 77 (246320) of the words correctly in normal spelling todictation versus 67 (213320) correctly with sup-pression c2(1 N = 640) = 790 p lt 05 However itis not surprising that articulatory suppression inter-feres with normal spelling More to the pointMMDrsquos performance matched the predictions out-lined earlier

First articulatory suppression produced a signifi-cant decrease in phonologically plausible errorsWithout suppression 31 (2374) of MMDrsquos errorswere phonologically plausible whereas with sup-pression only 14 (15107) were PPEs (see Figure4) c2(1 N = 181) = 676 p lt 05 This provides evi-dence that the articulatory suppression successfullyinterrupted the sublexical process thereby making itmore difficult for the sublexical system to generatephonologically plausible spellings

Second articulatory suppression produced a sig-nificant decrease in the size of the regularity effectdemonstrated by MMD Included in the stimuliwere the stimuli from the P-G list of the JHUDysgraphia Battery and a smaller list comprised of30 words that contained a high-frequency P-Gmapping segment and 30 with a low-frequency P-G mapping segment Without suppression MMDmade significantly more errors on words that con-tained low-frequency P-G segments (23mdash29

FOLK RAPP GOLDRICK


Figure 4 Percentage of spelling errors for normal spelling versusspelling under articulatory suppression

128) than on words that contained high-frequencyP-G segments (8mdash680) c2(1 N = 208) = 711 plt 05 This effect disappeared under articulatorysuppression as there was no difference in thenumber of errors made on words that containedlow-frequency P-G segments (18mdash23128)compared to words that contained high-frequencyP-G segments (20mdash1680) c2(1 N = 208) =003 p gt 05

Third the proportion of form-related worderrors increased under articulatory suppressionUnder normal conditions 7 (574) of MMDrsquoserrors were other words (ldquoloaf rdquo reg L-E-A-F) butunder suppression this rate increased significantly to20 (21107) c2(1 N = 181) = 490 p lt 05Although the rate of form-related errors increasedunder suppression the rate of nonword errors didnot change significantly under suppression Sixty-two percent of MMDrsquos errors were nonwords (4674) in the normal spelling condition whereas 66 ofher errors were nonwords (71107) under suppres-sion This difference was not significant c2(1 N =180) = 020 p gt 05 Thus the change in error pat-tern was very specific and matched the predictionsprecisely

The data from the articulatory suppression studysupport the prediction that if lexical-sublexicalinteraction serves to strengthen correct lexicalresponses then disruption to the sublexical systemshould lead to an increase in the production ofform-related word errors The decrease in the rateof PPEs and the elimination of the regularity effectunder suppression relative to normal spelling indi-cates that we were successful in disrupting thesublexical system and the increase in the rate ofform-related word errors produced under suppres-sion is consistent with the hypothesis that when thesublexical process is disrupted form-related lexicalneighbours can compete more successfully with thetarget This results in an increase in the number ofform-related lexical neighbours produced in placeof the target spelling The fact that the rate of form-related word errors increased under suppressionwhereas the rate of nonword errors did not increasesignificantly is consistent with the claim thatMMDrsquos form-related word errors do not arise from

chance substitutions deletions and transpositionsIf they did we would have expected to see anincrease in the rate of nonword errors

GENERAL DISCUSSION

We found that articulatory suppression can be usedeffectively to interrupt the sublexical process inwritten spelling A comparison of error patternsproduced under normal spelling conditions versusarticulatory suppression revealed an increase inform-related lexical substitutions under conditionsof articulatory suppression This indicates that thesublexical process plays an important role instrengthening a target wordrsquos constituentgraphemes in spelling creating a competitiveadvantage for the target over orthographically andor phonologically similar word neighbours thatcompete with it for output These data are consis-tent with the cognitive architecture of the spellingsystem proposed by Rapp et al (2002) andHoughton and Zorzi (1998 2002) in which thereis interaction between lexical and sublexical pro-cesses in spelling Simply stated on the basis of thefindings reported here the point of this integrationis to allow sublexical information to strengthentargeted lexical representations Thus models ofspelling must have a mechanism by which sub-lexical activation can interact with and strengthencorrect lexical responses (eg Houghton amp Zorzi1998 2002 Rapp et al 2002)

On the activation of form-related lexicalneighbours

Presumably MMD produced lexical substitutionsunder normal spelling conditions for two reasons(1) weakened lexical activation of the target word asa result of lexical impairment and (2) impoverishedinput from her impaired sublexical system That iswhen a lexical substitution was produced undernormal spelling conditions the target was not fullyactivated in the impaired lexical system and thedamaged sublexical system was not able to provideenough support for the targetrsquos graphemes to help it



compete successfully with its form-related wordneighbours However although MMDrsquos sublexicalsystem was impaired it was at least partially intactas evidenced by the fact that she was actually able tospell 79 of segments in nonwords correctly and toproduce PPEs in spelling words Apparently thesublexical system was sufficiently intact to providesupport for at least some of the targetrsquos graphemesat least some of the time The articulatory suppres-sion task reveals that when the sublexical input wasfurther reduced target words were even more vul-nerable to competition and form-related lexicalsubstitutions were produced at higher rates

This leads quite naturally to the question Howdid the form-related lexical competitors becomeactive in the first place As mentioned briefly in theintroduction within the framework depicted inFigure 2 there are three potential sources of activa-tion for form-related lexical entries in the ortho-graphic output lexicon (1) cascading feedforwardactivation from words in the phonological inputlexicon that are activated upon hearing the target(2) feedback activation from the graphemes of thetarget itself and (3) feedback activation from thegraphemes activated by the sublexical process

Each of these sources of activation has some-what different consequences If upon hearing thetarget word activation cascades from the phono-logical input lexicon through the system and on tothe orthographic output lexicon5 we would expectactivation of lexical competitors that are phonolog-ically related to the target For example the targetldquophonerdquo might activate the orthographic represen-tation FOAM In contrast feedback activationfrom the graphemes of the target itself (P-H-O-N-E) should lead to the activation of words that sharegraphemes with the target (eg PONY) Finallyfeedback from graphemes activated by thesublexical process could result in the activation ofeither orthographically or phonologically relatedneighbours depending on the nature and frequencyof the PGCs that are involved For example for thetarget ldquophonerdquo the PGC f reg F has a higher fre-quency than f reg PH (8472 vs 1117) (Hanna etal 1966) therefore the sublexical process might

favour phonologically related neighbours spelledwith F (ldquophonerdquo reg FOAM) over orthographicallyrelated neighbours containing P or H In othercases the reverse would be expected

Specifically with regard to MMDrsquos form-relatederrors we noted earlier that they were both ortho-graphically and phonologically similar to theirtargets This substantial similarity along bothdimensions prevents an easy determination ofwhether the source of activation of the lexical com-petitors was predominantly orthographic or pho-nological This is because despite the number ofldquoirregularitiesrdquo in the mapping of phonemes tographemes in English the relationship is largely adirect one and as a result words that are similar intheir orthography will necessarily be similar in theirphonology and vice versa Therefore in order todetermine if there were independent effects of pho-nological and orthographic relatedness we wouldneed to evaluate the orthographic overlap of the tar-get-error pairs while controlling for phonologicaloverlap and vice versa

Thus with respect to orthographic overlap thequestion is if MMDrsquos targets and errors were moresimilar orthographically than would be predictedpurely on the basis of their phonological related-ness In order to answer this question one mightassume that if only phonological factors were at playin activating the word neighbours and there was noindependent orthographic contribution then thereshould be as much orthographic overlap in MMDrsquosactual target-error pairs as in word pairs randomlyselected from the CELEX database (BaayenPiepenbrock amp Gulikers 1995) to be as similarphonologically as MMDrsquos actual target-error pairsTo examine this possibility we took MMDrsquos tar-get-error pairs (eg POISE reg PAUSE) and foundthe set of words in the CELEX lexicon thatmatched each error (eg PAUSE) in terms of itsphonological similarity with its target (egPOISE) We then examined the degree of ortho-graphic overlap in these target-ldquopseudoerrorrdquo pairs(eg POISE reg PURSE) and compared it to theorthographic overlap observed in MMDrsquos actualtarget-error pairs We also performed a similar

FOLK RAPP GOLDRICK


5 There is no strong evidence that cascading of this sort takes place however it cannot be ruled out

analysis to evaluate the possibility of phonologicalsimilarity effects above and beyond what would bepredicted if only orthographic factors were at playin activating the word neighbours (see Appendix Afor details of these analyses)

With regard to orthographic similarity wefound that the mean orthographic overlap value of765 for MMDrsquos actual target-error pairs was sig-nificantly greater than expected by chance giventhe degree of phonological overlap between heractual targets and errors (p lt 001) (see Appendix Bfor more detailed results) This was also true whenwe carried out an additional (and perhaps morestringent) analysis in which the pseudoerrors werenot only matched to MMDrsquos target-error pairs interms of phonological overlap but also in phonemelength In terms of phonological overlap we alsofound that the mean phonological overlap value of685 for MMDrsquos actual target-error pairs was sig-nificantly greater than expected by chance given thedegree of orthographic overlap between her actualtargets and errors (p lt 001) (see Appendix C)Again this was also true when we carried out theadditional analysis involving pseudoerrors thatwere not only matched to MMDrsquos target-errorpairs in terms of orthographic overlap but also inletter length

In sum we found evidence of independenteffects of both phonological and orthographic simi-larity in MMDrsquos form-related word errors Theseresults do not allow us to localise the source of acti-vation of the lexical form-related neighbours pre-cisely but they do at least indicate that feedbackfrom the targetrsquos graphemes is unlikely to be theonly source of this activation This feedback sourcealone would not account for the independent pho-nological similarity effects The lexical neighboursactive during the spelling process and which arerevealed through MMDrsquos form-related errors mayhave received activation from the various sourcessuggested by the architecture depicted in Figure 2

Further implications of lexicalsublexicalinteraction

The integration architecture that is proposed pre-dicts that when the graphemes activated by the

sublexical system are consistent with those of thetarget word sublexical activation helps tostrengthen correct lexical responses Howeveranother prediction that arises from the proposedarchitecture is that feedback will sometimes resultin sublexically based activation of not only the tar-get but also of other words that share letters acti-vated by the sublexical system When thesublexically activated graphemes are inconsistentwith the target (eg F in the case of the targetldquophonerdquo) then neighbours will be strengthenedwhose constituent graphemes are at odds withthose of the target Thus integration provides boththe opportunity to strengthen the target as well asthe opportunity to activate competing neighboursEvidence consistent with both of these conse-quencesmdashsublexical information aiding correctlexical responses and sometimes generating compe-titionmdashis illustrated in the case of RCM (Hillis etal 1999)

RCM like MMD also produced form-relatedword errors in spelling However unlike MMDRCMrsquos predominant error type in both written pic-ture naming and spelling to dictation was semanticerrors Strikingly however she did not produce anysemantic errors in either oral picture naming or oralreading What then was the source of the semanticerrors in written production A semantic-levelimpairment is ruled out by the fact that she did notproduce semantic errors in oral picture naming andoral reading Thus RCM was unlike JJ (Hillis ampCaramazza 1991) in that her semantic system wasintact but also unlike JJ (as well as MMD) in thather sublexical process was very severely damagedHillis et al (1999) proposed that the semanticerrors in written production arose from semanticactivation of semantically related lexical neighboursin a compromised orthographic output lexiconcoupled with very little sublexical support for thetarget Due to these factors the correct lexicalresponse was sometimes unavailable leading to theselection of a competing lexical representation thatshared semantic features with the target

When evaluated in the first week after herstroke 56 of RCMrsquos spelling errors in writing todictation and written picture naming were seman-tic In addition the damage to her sublexical system



was very severe as evidenced by the fact that shespelled no nonwords correctly (only 42 of theindividual phonological segments of nonwordswere spelled correctly) and she produced no PPEsin spelling Interestingly when RCM was evalu-ated again 2 weeks later the rate of semantic errorsdropped dramatically to only 10 of her spellingerrors This dramatic decrease was attributed toincreased sublexical contribution for the followingreasons First her spelling of nonwords improvedsuch that 3 of nonwords and 67 of the individ-ual phonological segments were spelled correctlyFurthermore she had begun to produce PPEs inspelling to dictation Under the integrationaccount the improved sublexical system was able toactivate some of the graphemes that were consistentwith the ldquocorrectrdquo lexical response and throughfeedback contribute to the correct selection amongthe multiple semantically related lexical candidatesThis aspect of the case constitutes another exampleof sublexical information supporting the correctlexical candidate

Interestingly however although the rate ofRCMrsquos semantic errors decreased with increasedsublexical input her rate of form-related worderrors (eg ldquomythrdquo reg METHOD) increased from33 of errors to 54 Hillis et al (1999) attributedthis increase in form-related errors to RCMrsquosimproved albeit still impoverished sublexical pro-cess For example the activation generated by thesublexical process may have been sufficient tofavour PEAR over BANANA (for the targetldquopearrdquo) but in doing so it might have contributedto the activation of words such as PAIN PEACEetc This aspect of the case exemplifies the way inwhich the sublexical process might contribute tothe activation of form-related competitors

There are alternative accounts that can explainthe present data without positing a feedback mech-anism between graphemes and orthographiclexemes6 According to one such account a targetword such as LOAF activates a word neighbour

such as LEAF as a result of cascading activationfrom the phonological input lexicon This mayresult in competition between target graphemesand the graphemes of its lexical neighbours at thegrapheme level We have shown that in theabsence of support from the sublexical route thegraphemes in target words will be less activatedthan under normal circumstances If additionallydue to disruption in the lexical system the probabil-ity of selecting O decreases it is likely that the prob-ability of selecting E will be relatively greater Inthat case a form-related word error may be pro-duced without assuming any feedback The com-petitive process may involve lateral inhibitionamong the graphemes of the target word and acompeting lexical neighbour (Houghton amp Zorzi1998) In other words in an undamaged systemactivation from the sublexical system biases thecompetition in favour of the target thus giving anadvantage to the targetrsquos graphemes over those ofthe competing lexical neighbour Introducing dis-ruption to the lexical system may then decrease theadvantage of the target wordrsquos graphemes over thecompetitorrsquos

Although our data do not require a feedbackmechanism we favour an architecture that includesfeedback given the independent evidence of a feed-back mechanism reported by McCloskey et al(1999) Clearly however more research is neededto adjudicate among alternative accounts of howform-related word errors occur

Conclusions

Through an examination of the changes in errorsthat are produced in spelling under normal condi-tions versus spelling under conditions that selec-tively disrupt the sublexical process we havecollected evidence regarding the role of lexicalsublexical interaction in the spelling process Wehave argued that this integration contributesimportantly to the strengthening of a target word in

FOLK RAPP GOLDRICK


6 Connectionist architectures of the spelling process modelled after the reading architecture of Plaut et al (1996) also assumeinteraction between two major spelling processes (semantic and phonological) (Graham Patterson amp Hodges 1997 2000)Presumably the effects we have reported here can be accommodated within such an architecture although it is not straightforward todetermine the specific predictions of this proposal regarding the consequences of disruption to the phonological process

the face of competition from its lexical neighboursIn addition these findings begin to reveal the com-plex nature of the interactive processes involved inspelling and help us to understand detailed aspectsof a number of cases of dysgraphia as well as find-ings from neurologically intact subjects that havebeen reported in the literature

Manuscript received 13 September 2001Revised manuscript received 20 June 2002Revised manuscript accepted 26 June 2002

REFERENCES

Baayen RH Piepenbrock R amp Gulikers L (1995)The CELEX Lexical Database (Release 2) [CD-ROM] Philadelphia PA Linguistics DataConsortium

Baron J amp Strawson D (1976) Use of orthographicand word specific knowledge in reading words aloudJournal of Experimental Psychology Human Perceptionand Performance 4 207ndash214

Barry C amp Seymour PHK (1988) Lexical primingand sound-to-spelling contingency effects innonword spelling The Quarterly Journal of Experi-mental Psychology 40 5ndash40

Beauvois MF amp Deacuterousne J (1981)Lexical or ortho-graphic agraphia Brain 104 21ndash49

Campbell R (1983) Writing nonwords to dictationBrain and Language 19 153ndash178

Coltheart M Rastle K Perry C Langdon R ampZiegler J (2001)DRC A dual route cascaded modelof visual word recognition and reading aloud Psycho-logical Review 108 204ndash256

Cuetos F (1993) Writing processes in shallow orthog-raphy Reading and Writing 5 17ndash28

Ellis AW (1982) Spelling and writing (and readingand speaking) In AW Ellis (Ed) Normality andpathology in cognitive functions New York AcademicPress

Goodman RA amp Caramazza A (1985) The JohnsHopkins University Dysgraphia Battery BaltimoreMD Johns Hopkins University

Goodman RA amp Caramazza A (1986) Aspects ofthe spelling process Evidence from a case of acquireddysgraphia Language and Cognitive Processes 1 263ndash296

Graham NL Patterson K amp Hodges JR (1997)Progressive dysgraphia Co-occurrence of central and

peripheral impairments Cognitive Neuropsychology14 975ndash1005

Graham NL Patterson K amp Hodges JR (2000)The impact of semantic memory impairment onspelling Evidence from semantic dementiaNeuropsychologia 38 143ndash163

Hanna PR Hanna JS Hodges RR amp RudorfEH Jr (1966) Phoneme-grapheme correspondences ascues to spelling improvement Washington DC UnitedStates Department of Health Education andWelfare

Hillis AE amp Caramazza A (1991) Mechanisms foraccessing lexical representations for output Evidencefrom a category-specific semantic deficit Brain andLanguage 40 106ndash144

Hillis AE Rapp B amp Caramazza A (1999)When arose is a rose in speech but a tulip in writing Cortex35 337ndash356

Houghton G amp Zorzi M (1998) A model of thesound-spelling mapping in English and its role inword and nonword spelling In MA Gernsbacher(Ed) Proceedings of the Twentieth Annual Conferenceof the Cognitive Science Society (pp 490ndash501)Mahwah NJ Lawrence Erlbaum Associates Inc

Houghton G amp Zorzi M (2002) A model of the soundspelling mapping in English and its role in word andnonword spelling Manuscript submitted forpublication

Kay J Lesser R amp Coltheart M (1992)Psycholinguistic Assessments of Language Processing inAphasia Hove UK Lawrence Erlbaum AssociatesLtd

Kuccedilera H amp Francis WN (1967) Computationalanalysis of presentndashday American English ProvidenceRI Brown University Press

McCloskey M Macaruso P amp Rapp B (1999)Graphemendashtondashlexeme feedback in the spelling systemPaper presented at the Annual Academy of AphasiaVenice Italy

Paap KR amp Noel RW (1991) Dual-route models ofprint to sound Still a good horse race PsychologicalResearch 53 13ndash24

Patterson KE (1986) Lexical but nonsemantic spell-ing Cognitive Neuropsychology 3 341ndash367

Patterson KE (1988) Acquired disorders of spellingIn G Denes C Semenza amp P Bisiacchi (Eds)Perspectives on cognitive neuropsychology (pp 213ndash229)Hove UK Lawrence Erlbaum Associates Ltd

Plaut DC McClelland JL Seidenberg MS ampPatterson K (1996) Understanding normal andimpaired word reading Computational principles in



quasi-regular domains Psychological Review 103 56ndash115

Rapp B (2002) Uncovering the cognitive architectureof spelling In A E Hillis (Ed) Handbook on adultlanguage disorders Integrating cognitive neuro-psychology neurology and rehabilitation New YorkPsychology Press

Rapp B amp Caramazza A (1993) On the distinctionbetween deficits of access and deficits of storage Aquestion of theory Cognitive Neuropsychology 10113ndash141

Rapp B Epstein C amp Tainturier MJ (2002) Theintegration of information across lexical andsublexical processes in spelling Cognitive Neuro-psychology 19 1ndash29

Sanders RJ amp Caramazza A (1990) Operation ofphoneme-to-grapheme conversion mechanism in abrain injured patient Reading and Writing 2 61ndash82

Shallice T (1981)Phonologicalagraphia and the lexicalroute in writing Brain 104 413ndash429

Snodgrass JG amp Vanderwart M (1980) A standard-ized set of 260 pictures Norms for name agreementimage agreement familiarity and visual complexityJournal of Experimental Psychology Human Learningand Memory 6 174ndash215

Stanovich KE amp Bauer D (1978)Experiments of thespelling-to-sound regularity effect in word recogni-tion Memory and Cognition 6 410ndash415

Zorzi M Houghton G amp Butterworth B (1998)Two routes or one in reading aloud A connectionistdual-process model Journal of Experimental Psychol-ogy Human Perception and Performance 24 1131ndash1161

FOLK RAPP GOLDRICK


APPENDIX A

Analysis of form-related word errorsOrthographic overlapStep 1 We calculated orthographic and phonological similarityfor each of MMDrsquos target-error pairs using the followingorthographic overlap (OOI) and phonological overlap (POI)indices

OOI = of shared letters in any position in target and error 2letter length of target + letter length of error

POI= of shared phonemes in anyposition in target and error 2phonemic length of target + phonemic length of error

Step 2 We developed a computer program that for each ofMMDrsquos target-error pairs (eg POISE-PAUSE) found the setof words in the CELEX lexical database (Baayen et al 1995)that matched the error (eg PAUSE) in terms of itsphonological similarity with the target (eg POISE) We usedas a measure of phonological similarity the target-error pairrsquosPOI plusmn10 These were called pseudoerrors For example thephonological similarity (POI plusmn10) for the target-error pairPOISE-PAUSE was 66 plusmn10 and among the pseudoerrorswithin this phonological overlap range were NOISE POINTPEAS This procedure was carried out for each of MMDrsquos 91target-error pairsStep 3 The program randomly selected one item from each set ofpseudoerrors and calculated the mean OOI value of the targetsand selected pseudoerrors This was repeated 1000 times toderive a distribution of chance values of orthographic overlap

between word pairs whose phonological overlap was comparableto that observed in MMDrsquos actual target-error pairsStep 4 We compared MMDrsquos actual mean target-error OOIwith the distribution of 1000 mean OOI values derived from thesets of target-pseudoerror pairs In addition we carried outanalyses involving pseudoerrors that were not only matched toMMDrsquos target-error pairs in terms of OOI plusmn10 but were alsomatched in terms of letter lengthResults As shown in Appendix B the mean orthographicoverlap observed in MMDrsquos actual target-error pairs wassignificantly greater than expected by either chance measure(p lt 001)

Analysis of form-related word errorsPhonological overlapPrecisely the same procedure was followed to determine if thephonological similarity between targets and errors was greaterthan would be expected purely on the basis of their orthographicsimilarity As with orthographic overlap we carried outadditional analyses involving pseudoerrors that were not onlymatched to MMDrsquos target error pairs in terms of POIplusmn10 butthat were also matched in phoneme lengthResults As shown in Appendix C the mean phonologicaloverlap observed in MMDrsquos actual target-error pairs wassignificantly greater than expected by either chance measure(p lt 001)



APPENDIX C

Likelihood of the observed phonologicaloverlap between target words and form-related word errors occurring as a result ofchance given their orthographic overlap andletter length overlap

OOI amp letterControl factor(s) OOI length overlap

Observed POI value 685 685p-value lt001 lt001Chance distributionMean 464 515Max 512 564Min 416 462SD 16 16

OOI and POI see Appendix B Footnote

APPENDIX B

Likelihood of the observed orthographicoverlap between target words and form-related word errors occurring as a result ofchance given their phonological overlap andphonemic length overlap

POI amp phonemeControl factor(s) POI length overlap

Observed OOI Value 765 765p-value lt001 lt001Chance distributionMean 546 568Max 590 615Min 503 523SD 14 14

OOI = orthographic overlap index a measure of orthographicoverlap between target words and their form-related worderrors POI = phonological overlap index a measure ofphonological overlap between target words and their form-related word errors

dysgraphic subject which specifically supports thehypothesis that lexical and sublexical interactionserves to strengthen the representations of targetwords and their constituent graphemes

Evidence of interaction

In research on reading much of the evidence forinteraction between lexical and sublexical processescomes from regularity effects whereby irregularwords (words with at least one low-frequencygrapheme-phoneme correspondence) take longerto read aloud than regular words (words containingonly high-frequency grapheme-phoneme corre-

spondences) (eg Baron amp Strawson 1976Stanovich amp Bauer 1978) Regularity effects arethought to arise because for irregular words the lex-ical process will generate the correct pronunciationfor a word stimulus (eg BROAD reg b r d)whereas the sublexical process will regularise thepronunciation generating a pronunciation that isconsistent with the most common way of pro-nouncing the graphemes of the word (egBROAD reg b r o d) In this way conflicting out-puts will be generated for irregular but not regularwords Longer pronunciation times for irregularwords are assumed to result from the time it takesto reconcile the conflicting outputs of the twoprocesses (eg Paap amp Noel 1991) In effectregularity effects arise from the interaction oflexical and sublexical information (eg ColtheartRastle Perry Langdon amp Ziegler 2001 ZorziHoughton amp Butterworth 1998)1

More recently evidence for the interaction oflexical and sublexical processes in spelling hasbegun to emerge Evidence has come both fromstudies involving the priming of nonword spellingsin intact subjects and from case studies ofdysgraphic subjects Various studies with intactsubjects have examined the influence of the spell-ings of previously heard rhyming words (ldquolexicalprimingrdquo) on nonword spellings (Barry amp Seymour1988 Campbell 1983 Cuetos 1993) For ex-ample Barry and Seymour compared the spellingsthat subjects produced for nonwords on lexicalpriming trials where a nonword was preceded by arhyming word with the spellings produced for thesame nonwords when they were not preceded bywords Barry and Seymour found lexical primingeffects such that nonword spellings were influencedby the spellings of previously heard rhyming wordsFor example although the most common spellingof the nonword n ^ T was N-U-T-C-H (Hanna


FOLK RAPP GOLDRICK

Figure 1 The functional architecture of the spelling system

1 Certain connectionist accounts of reading do not posit a separate lexical and sublexical process (Plaut McClelland Seidenbergamp Patterson 1996) but instead posit phonological and semantic processes The phonological process maps orthographic tophonological representations whereas the semantic process maps semantic to orthographic representations neither involving whatwould be traditionally considered to be lexical look-up Within this type of architecture it is the sublexical process that encodes thefrequency of orthography to phonology relationships giving rise to frequency times regularity interaction effects Moreover within thissort of architecture phonological and semantic processes are assumed to interact in that they both activate a common level oforthographic representation











FOLK RAPP GOLDRICK
















CASE STUDY




FOLK RAPP GOLDRICK










FOLK RAPP GOLDRICK








looks likerdquo





Written production






Premorbid spelling











FOLK RAPP GOLDRICK




4 88 615 87 646 87 617 89 578 85 39














Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B















FOLK RAPP GOLDRICK
















CASE STUDY




FOLK RAPP GOLDRICK










FOLK RAPP GOLDRICK








looks likerdquo





Written production






Premorbid spelling











FOLK RAPP GOLDRICK




4 88 615 87 646 87 617 89 578 85 39














Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B







FOLK RAPP GOLDRICK
















CASE STUDY




FOLK RAPP GOLDRICK










FOLK RAPP GOLDRICK








looks likerdquo





Written production






Premorbid spelling











FOLK RAPP GOLDRICK




4 88 615 87 646 87 617 89 578 85 39














Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B


















CASE STUDY




FOLK RAPP GOLDRICK










FOLK RAPP GOLDRICK








looks likerdquo





Written production






Premorbid spelling











FOLK RAPP GOLDRICK




4 88 615 87 646 87 617 89 578 85 39














Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B










CASE STUDY




FOLK RAPP GOLDRICK










FOLK RAPP GOLDRICK








looks likerdquo





Written production






Premorbid spelling











FOLK RAPP GOLDRICK




4 88 615 87 646 87 617 89 578 85 39














Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B












FOLK RAPP GOLDRICK








looks likerdquo





Written production






Premorbid spelling











FOLK RAPP GOLDRICK




4 88 615 87 646 87 617 89 578 85 39














Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B









FOLK RAPP GOLDRICK








looks likerdquo





Written production






Premorbid spelling











FOLK RAPP GOLDRICK




4 88 615 87 646 87 617 89 578 85 39














Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B






Written production






Premorbid spelling











FOLK RAPP GOLDRICK




4 88 615 87 646 87 617 89 578 85 39














Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B











FOLK RAPP GOLDRICK




4 88 615 87 646 87 617 89 578 85 39














Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B


















Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B







Procedure


Results




FOLK RAPP GOLDRICK







GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B









GENERAL DISCUSSION












FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B











FOLK RAPP GOLDRICK


















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B




















Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B










Conclusions


FOLK RAPP GOLDRICK





REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B







REFERENCES





































FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B














FOLK RAPP GOLDRICK


APPENDIX A









APPENDIX C





APPENDIX B







APPENDIX C





APPENDIX B





Documents

COGNITIVE NEUROPSYCHOLOGY, 2002, 19 (7), 653– 671web.jhu.edu › cogsci › templates › images › rapp › ... · More recently, evidence for the interaction of lexical and sublexical