Cognitive Neuropsychologybrain.bnu.edu.cn/home/yanchaobi/recent papers/Han et al 2007CN.pdf · writing to dictation, written naming), or the output modalities (e.g., written spelling,

This article was downloaded by:[Beijing Normal University]On: 3 August 2007Access Details: [subscription number 775699731]Publisher: Psychology PressInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Cognitive NeuropsychologyPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t713659042

The orthographic buffer in writing Chinese characters:Evidence from a dysgraphic patient

Online Publication Date: 01 June 2007To cite this Article: Han, Zaizhu, Zhang, Yumei, Shu, Hua and Bi, Yanchao (2007)'The orthographic buffer in writing Chinese characters: Evidence from a dysgraphicpatient', Cognitive Neuropsychology, 24:4, 431 - 450To link to this article: DOI: 10.1080/02643290701381853URL: http://dx.doi.org/10.1080/02643290701381853

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expresslyforbidden.

The publisher does not give any warranty express or implied or make any representation that the contents will becomplete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should beindependently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with orarising out of the use of this material.

© Taylor and Francis 2007

http://www.informaworld.com/smpp/title~content=t713659042http://dx.doi.org/10.1080/02643290701381853http://www.informaworld.com/terms-and-conditions-of-access.pdf

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

The orthographic buffer in writing Chinese characters:Evidence from a dysgraphic patient

Zaizhu HanState Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, China

Yumei ZhangNeurological Department, Beijing Tiantan Hospital, Beijing, China

Hua Shu and Yanchao BiState Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, China

We investigated the postlexical processes in writing Chinese characters by studying the delayedcopying performance of a Chinese dysgraphic patient, W.L.Z. His delayed copying difficulty couldnot be attributed to peripheral motor deficit and could not be readily explained by lexical or semanticfactors. Instead, the copying performance was sensitive to a word length variable (number of logogra-phemes), and the most prevalent errors were logographeme substitutions. Furthermore, in the substi-tution errors, the target logographemes and responses tended to share visual/motoric attributes. Wepropose that the delayed copying difficulty reflects a deficit to the buffering component in writing(coined “logographeme output buffer”), and the universality and language-specific features of theoutput buffer in writing are discussed.

Chinese characters are complex things. Writingthem involves spatial arrangement of strokes intoa two-dimensional square in complicated ways.Take the character (brain, /nao3/1) forexample, strokes are connected in various direc-tions placed in various relationships

, and so on. What guides thewriting of these structures? Is there a closed setof stroke combinations in Chinese that are

comparable to letters or graphemes in alphabeticscripts? Importantly, can models developed forwriting alphabetic words be applied to writingChinese characters? The current article tries toinvestigate the postlexical processes in writing bystudying a Chinese dysgraphic patient.

Research on writing in alphabetic languages hasrevealed that writing involves multiple stages (seeFigure 1). First, the orthographic properties of a

1Within the slashes are the phonetic transcripts of the Chinese words, following the pinyin system. The number denotes the tonefor the preceding syllable. There are four tones in Mandarin. The number 0 represents an unstressed syllable.

Correspondence should be addressed to Yanchao Bi, State Key Laboratory of Cognitive Neuroscience and Learning, BeijingNormal University, Beijing 100875, China (E-mail: [email protected]).

This research was supported by grants from the Pangdeng project (95-special-09), National Natural Science Foundation ofChina (30470574), and the Beijing Natural Science Foundation (7052035). We would like to thank Alfonso Caramazza andBrenda Rapp for extensive discussions and helpful comments on earlier drafts of the paper and Sam-Po Law and Jianfeng Yangfor their suggestions on the data analyses. We are especially grateful to W.L.Z. for his kind collaboration.

# 2007 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business 431http://www.psypress.com/cogneuropsychology DOI:10.1080/02643290701381853

COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (4), 431–450

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

word could either be retrieved from memorydirectly, or could be computed from thephoneme–grapheme conversion mechanism.Once the orthographic information is retrieved,it is held in an amodal “graphemic output buffer”awaiting further processing. In written spelling,the “graphemes” in this buffer are converted intoletter shapes (e.g., allographic representation andgraphic motor pattern), including the correctcase and font properties, after which the corre-sponding effector-specific motor system isemployed (e.g., Ellis, 1982, 1988; Margolin,1984; Rapp & Caramazza, 1997). In oral spelling,the “graphemes” in the buffer are converted intoletter names and then are produced orally (e.g.,Bub & Kertesz, 1982).

The proposal specifying these representationsand processes is based not only on computationalneeds but also on empirical evidence, especiallythe observations of dysgraphic patients. Take thegraphemic output buffer for example. Given itsposition within the writing architecture depictedin Figure 1, Caramazza, Miceli, Villa, andRomani (1987) reasoned that a series of beha-vioural patterns should associate with the selective

disruption of this element. Because it is a postlex-ical constituent, it should not be influenced bylexical and semantic factors (such as frequency,concreteness, or grammatical class), lexicality(words vs. nonwords), the input modalities (e.g.,writing to dictation, written naming), or theoutput modalities (e.g., written spelling, oral spel-ling, or typing). The number of writing errorsshould increase as a function of the word length,due to the greater difficulty in retaining longerletter sequences. Furthermore, because letters (orgraphemes) are basic processing units in thisbuffer, the writing errors should occur on letters.Patients fitting this profile have indeed beenreported in various alphabetic languages, includingItalian, English, and French (e.g., Annoni, Lemay,de Mattos Pimenta, & Lecours, 1998; Caramazza& Miceli, 1990; Caramazza, et al., 1987;Cipolotti, Bird, Glasspool, & Shallice, 2004;Hillis & Caramazza, 1989; McCloskey,Badecker, Goodman-Schulman, & Ajiminosa,1994; Rapp & Kong, 2002). Furthermore, detailedanalyses of the errors produced by patients withselective graphemic output buffer impairmenthave shown that the structure of representations

Figure 1. A model of writing in alphabetic language (Adapted from Rapp & Caramazza, 1997).

432 COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (4)

HAN ET AL.

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

held in the buffer is rather rich, including the iden-tity and order of letters, the consonant/vowelstatus of letters (Buchwald & Rapp, 2006;Caramazza & Miceli, 1990; Cotelli, Abutalebi,Zorzi, & Cappa, 2003; Ward & Romani, 2000),the morphemic structure (Badecker, Hillis, &Caramazza, 1990; Orliaguet & Boë, 1993;Schiller, Greenhall, Shelton, & Caramazza,2001), the graphosyllabic structure (Caramazza& Miceli, 1990; Zesiger, Orliaguet, Boë, &Moundoud, 1994), letter doubling (Caramazza& Miceli, 1990; Tainturier & Caramazza, 1996),and digraphs (Tainturier & Rapp, 2004).

Dysgraphic patients have also been reportedwho are assumed to have deficits located at proces-sing stages beyond the graphemic output bufferand prior to the effector-specific peripheralmotor systems (see the “allographic represen-tation” and “graphic motor pattern” in Figure 1;e.g., Ellis, 1982, 1988; Margolin, 1984). Errorsproduced by these patients are similar to graphe-mic output buffer patients in that they all makewell-formed letter substitution errors and thattheir writing performance is not affected bylexical factors or input modalities. However,different from the buffer deficit patients, they donot show any length effect, the impairment isspecific to written production leaving oral spellingintact, and their substitutions are sensitive not toconsonant/vowel categories but to “stroke fea-tures” (e.g., Rapp & Caramazza, 1997).Although controversies remain regardingwhether properties like case and font are capturedby having different representations or are realizedthrough some dynamic mechanism (e.g., case con-version; Ellis, 1982, 1988; Margolin, 1984; Rapp& Caramazza, 1997; Zesiger, Martory, & Mayer,1997), and whether letter shapes are representedby visual-spatial properties or stroke features prop-erties, the idea of this kind of “letter shape” rep-resentation is both theoretically and empiricallyjustified.

By contrast, we know far less about the cogni-tive processes involved in writing logographiclanguages, such as Chinese. By studying differentwriting systems, not only could we learn abouthow language-specific features shape the structure

of each representation, we could also examine whatcognitive processes in writing are universal. Inother words, it is an open question whether par-ticular cognitive components in writing alphabeticlanguages are theoretical and empirically justifiedfor writing Chinese. For example, one character-istic of a selective deficit to the output buffer inalphabetic languages is that the error patterns arehighly similar in oral spelling and written spellingbecause the buffer is shared by these two modal-ities. However, oral spelling of a Chinese characteris impossible for almost all characters due to thelack of an exhaustive set of pronounceable sub-character components. On the other hand, in thewriting process of all languages, the output unitsof the orthographic lexicon are usually of largersize than the units that the more peripheralmotor system employs. Therefore, a buffer-likecomponent to hold the to-be-processed infor-mation should be universal. It should be notedthat there are debates about the lexical processesof orthographic representation retrieval inwriting Chinese words, such as whether tworoutes (a semantic pathway and a nonsemanticpathway) are involved, the nature of the nonse-mantic pathway, and the interaction between thetwo pathways (e.g., Graham, Patterson, &Hodges, 1997; Law & Or, 2001; Reich, Chou,& Patterson, 2003; Weekes, Yin, Su, & Chen,2006). These issues are beyond the scope of thecurrent article, and the focus here is on the pro-cesses after the orthographic representation isretrieved.

The approach we take here is to study aChinese dysgraphic patient whose writing errorsare not due to a peripheral motor deficit and arenot merely attributable to lexical factors. Westart from the generic framework in Figure 1 andassume that our patient’s errors originated fromthe stage(s) that are comparable to the levelsbetween the orthographic lexicon and the neuro-muscular execution—the graphemic outputbuffer, the allographic store, and the graphicmotor pattern—which are referred to here as the“graphic pathway” for the sake of brevity. Thenwe investigate the detailed origins of our patient’swriting deficits within this pathway and explore

COGNITIVE NEUROPSYCHOLOGY, 2007, 24 (4) 433

ORTHOGRAPHIC BUFFER IN WRITING CHINESE

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

the characteristics of that impaired component.The answers to these questions will inform uswhether particular components in Figure 1 areuniversal and how language-specific parametersin Chinese are realized in such a cognitivetheory. Before elaborating on our case study, wefirst briefly introduce the characteristics ofChinese writing scripts.2

Characteristics of Chinese scripts

Chinese is a logographic language, and the basicwriting units are characters (e.g., Wang, 1973).Each character corresponds to a syllable in soundand almost always a morpheme. While somehighly frequent words are monosyllabic, 88% ofChinese words are compounds that are composedof multiple morphemes (characters), and themajority (74%) are two-morpheme/charactercompounds (ILTR, 1986). Within a written com-pound word, the characters are linearly arranged ina left-to-right fashion, each occupying a space-independent square. For example, the word,

(psychology, /xin1 li3 xue2/) is composedof three characters, (heart, /xin1/), (reason,/li3/) and (research, /xue2/).

There are more than 20,000 characters inmodern Chinese language, including about 3,000commonly used characters. A character can bespatially analysed into a hierarchical structureinvolving several different-size units, convention-ally including the radical layer, the logographemelayer, and the strokes (see Standards Press ofChina, 1994; State Language Commission,1998). Strokes are combined in rich spatialrelationships to form characters, but their combi-nation is not random. For instance, a neveroccurs right below a .

More than 80% of characters are so-calledsemantic-phonetic composite characters (Shu,

2003). A composite character (e.g., , locust,/huang2/) includes two parts: the semanticradical ( , insect, /chong2/), which providesclues to the meaning of the character, and thephonetic radical ( , emperor, /huang2/), whichmay give clues to the pronunciation of the charac-ter. Most, but not all, phonetic radicals are alsoexisting characters when they stand alone. Asmaller percentage of semantic radicals are alsoused as independent characters, and when theyare, they often undergo slight form alternation(e.g., ! ). Neither the semantic radicals northe phonetic radicals are very reliable indexes butthey have been shown to affect the processing ofChinese characters in comprehension and reading(e.g., Bi, Han, Weekes, & Shu, in press; Law,2004; Law, Yeung, Wong, & Chiu, 2005). Theirroles in writing characters are less certain.

Some linguists and psycholinguists (e.g., Fu,1991; Law & Leung, 2000; Su, 1994; Zhang,1984) have proposed an intermediate levelbetween strokes and radicals in visual Chinesecharacters—logographemes—based on spatial-visual principles. Logographemes,3 a term coinedby Law and Leung (2000), are the smallest unitsin a character that are spatially separated. Forinstance, the three parts ( and ) inare spatially separate from each other (as opposedto being crossed) and are therefore considered asdifferent logographemes. Such visual units areproductive in that they appear in many characters.For example, the part is found in characters

andso on. Only the smallest blocks that cannot befurther disassembled into other logographemesare considered logographemes. Hence, manyphonetic radicals (e.g., the in ) could befurther analysed into two or more logographemes( and ). Based on these principals, TheChinese Character Component Standard of

2 While Chinese is rich in spoken dialects that are different by various degrees, there are currently two kinds of Chinese writtenscripts used—the traditional fonts used in Taiwan and Hong Kong regions and the simplified fonts used in mainland China. Thetraditional fonts are usually more complex than simplified fonts, while the structural principals are comparable. Here in our paper wefocus primarily on the simplified fonts.

3 The same concept has been addressed as , /bu4jian4/ (components or subcomponents) in earlier linguistic references(e.g., CCCSGCSIP, State Language Commission, 1998).


HAN ET AL.

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

GB13000.1 Character Set for Information Processing(CCCSGCSIP; State Language Commission,1998) listed 560 logographemes that constructedthe 20,902 characters in the UCS ChineseCharacter Database (Standards Press of China,1994). This logographeme database will be theone used throughout the article.

Previous research on writing in logographiclanguages

The above description shows that Chinese charac-ters could be analysed into various levels: strokes,logographemes, radicals, and whole characters.What then are the basic functional units inwriting Chinese characters? How are they rep-resented and retrieved? Insights on these issueshave mainly come from the errors that patientswith brain damage make. Kokubo, Suzuki,Yamadori, and Satou (2001) reported a Japanesebrain-damaged patient who suffered from selectiveimpairment in the “orthographic output buffer” inwriting Kana (syllabogram) characters and wasnormal in writing Kanji (logogram) characters.Based on such observations the authors proposedthat there exist two separate graphemic buffersfor Kanji and Kana words in writing Japaneseand that their patient had selective impairmentto the buffer used for Kana words.

Law and colleagues (Law, 1994; 2004; Law &Caramazza, 1995; Law & Leung, 2000; Lawet al., 2005) reported a series of case studies onthe writing performance of Chinese dysgraphicpatients who were Cantonese speakers using tra-ditional characters. One group of patients (Law,1994; 2004; Law & Caramazza, 1995; Lawet al., 2005) made many writing errors on theradical level, where semantic and phonetic radicalswere replaced (e.g., ! ), deleted (e.g., !), or added (e.g., ! ), suggesting that the

semantic/phonetic radical might be a processinglevel that could be impaired selectively in characterwriting. One particularly relevant case (S.F.T.)had a preponderance of writing errors on the

logographeme level in a delayed copy task,leading the authors to conclude that logogra-phemes are functional units in writing as well(Law & Leung, 2000). However, this conclusionis premature because logographemes and radicalswere confounded in a large proportion of theerrors—that is, the errors could be classifiedeither as a logographeme error or as a radicalerror. Furthermore, many of the errors resultedin real characters, and therefore lexical factorsmight be at play. Most critically, while S.F.T.was poor at delayed copying (40%4), she was alsoimpaired with direct copying (53%), raising thepossibility that her copying errors may actuallylie in peripheral visual or motor systems.

Our article reports a case that has a disruptedability in delayed copying with preserved directcopying ability. His overall profile in delayedcopy shared similarities in certain aspects with pre-vious patients in alphabetic languages with deficitsascribed to “the graphic pathway”. We presentdetailed analyses of his writing errors to addressthe following questions: (a) What deficit causesthe delayed copying difficulties? (b) What are thefunctional units in the impaired representation?(c) What structural characteristics does that rep-resentation have?

Case background

W.L.Z. is a 36-year-old, right-handed,Mandarin-speaking male with a college education.Prior to a stroke, he worked in a foreign clothingcompany and had normal language abilities. InMay 2004, he was admitted to a hospital due toa severe headache and difficulty in speaking. Acomputed tomography (CT) scan at the acutestage revealed a haemorrhage at the left temporallobe. Evaluations from the Chinese version ofWestern Aphasia Battery (Gao et al., 1993;Kertesz, 1982) categorized him as suffering fromsensory aphasia. A magnetic resonance imaging(MRI) scan performed in June 2004 indicatedthe assimilation stage of left tempo-occipital

4 Unless otherwise noted, the numbers given are correct percentages.



Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

haematoma, with the possibility of hidden vascularmalformation (see Figure 2). In that same month,a Transcranial Doppler (TCD) techniquedisclosed a weak blood signal in the temporalwindow and decreased blood flow of the leftvertebrobasilar artery.

W.L.Z. was first administered the MandarinClinical Language Screening Battery that wasadapted by author YB from the Harvard CNLabLanguage Screening Battery. Control data were col-lected from normal participants who were matchedto W.L.Z. on gender (male), education (college),handedness (right), and age (mean: 26 years;range: 22–36). The control group were perfecton all tasks unless reported below. Worth notingis that control participants who had the same occu-pation and age were on the high end of the controlgroup performance.

W.L.Z. was perfect at the bucco-facial apraxiatask (15/15), picture copying (2/2), phoneme dis-crimination (40/40), and repetition (words with syl-lable number varying from one to four, 35/35;nonwords, 5/5). He was moderately impaired atauditory digit span (forward, 4; backward, 2;control groupmean: forward, 8.25; backward, 6.25).

W.L.Z.’s lexical recognition and comprehen-sion skills were impaired. He scored 12/20 on anauditory-word/visual-word matching task wherehe needed to match one spoken compound wordto one of three visual words including one targetand two related foils (semantic, orthographic, or

phonological); 41/50 on an auditory-word/picture matching task where he matched onespoken word to one of two pictures (a target anda foil that was either semantically or visuallyrelated or unrelated to target); 41/50 on thevisual version of the word/picture matching task;and 17/20 on an auditory lexical decision taskwhere nonwords were created by combining twocharacters/syllables (e.g., “tea–row”; controlgroup mean, 19.5/20).

W.L.Z.’s oral and written production wereseverely impaired. He was almost unable to readwords (0/57, all no responses) or name pictures(11/82, mostly circumlocution errors; controlgroup mean, 95%). He was unable to perform thewriting to dictation task (0/10) and the writtenpicture-naming task (0/11; control group mean,95%), where all the writing errors were noresponses. Nevertheless he had no difficultycopying the characters (20/20, for an example seeFigure 3), where he had the target character infront of him during the copying. If the target char-acters were taken away for 2 s before he wasinstructed to write down what he saw (a delayedcopy task), he was correct for 19/30 of the items(for an example see Figure 3). The responses herewere all well formed, but only 2 out of the 11errors he made resulted in another real character:

(land, /lu4/) ! (to accompany, /pei2/) and(building, /lou2/) ! (to hug, /lou3/). The

remaining 9 erroneous responses were all

Figure 2. MRI scans for W.L.Z.


HAN ET AL.

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

noncharacters, all of which involve errors onthe logographeme level—for example, (bowl,/wan3/) ! (manuscript, /gao3/) !

(add, /tian2/) ! (apprentice,/tu2/) ! .

In summary, the screening tests revealed thatW.L.Z. was impaired on a range of cognitive func-tions, including word and sentence comprehen-sion, reading, and oral and written naming. Ourstudy focuses primarily on his written productionpatterns using the delayed copy task, because ofthe observations that W.L.Z.’s delayed copyresponse patterns seemingly revealed a deficit

along the “graphic pathway”, and W.L.Z. wasunable to perform written production frommemory tasks (e.g., writing to dictation, writtenpicture naming) and because oral spelling ishardly feasible in Chinese in most cases. In orderto examine whether and how a deficit or deficitsto these functional components along thispathway could account forW.L.Z.’s copying beha-viour, and to investigate the structure of that com-ponent, we designed the following delayedcopying experiment.

EXPERIMENT: DELAYED COPY

Wementioned in the Introduction that if a patienthas a selective deficit along the “graphic pathway”for writing (see Figure 1), certain predictions couldbe derived about the patient’s writing perform-ance: It should not be influenced by lexical-semantic factors (frequency, concreteness, orgrammatical class) and lexicality (words vs. non-words) and should present the same patternacross different modalities. In an alphabeticlanguage the errors should occur on the letterlevel, such as letter substitutions, deletions, inser-tions, or transpositions. If the deficit is located ata buffer-like component, the performance shouldbe sensitive to word length (amount of infor-mation to be held in the buffer). If the errors arerelated to the targets by visual-spatial or strokefeatures, the origin of the errors is likely to be ata level that is more comparable to “allographicstore” and/or “graphic motor pattern”.

It is obvious that our patient is not onlyimpaired with the graphic pathway, but also hasproblems in the semantic system, the orthographicoutput lexicon, and the orthographic input lexiconamong others. Our rationale here is to firstexamine whether the difficulties in this particulartask—delayed copy—are due to impairmentalong the graphic pathway, and, if so, by lookingat the error patterns we could gain insights intothe structural characteristics of the represen-tation(s) of interest.

The rationale of the experimental design anderror analyses is as follows. First, to establish

Figure 3. Samples of W.L.Z.’s writing performance (erroneouspart/s being circled).



Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

whether his delayed copying errors come from adeficit on the graphic pathway, we manipulatedfactors including lexical frequency, orthography–phonology regularity, concreteness, and grammaticalclass of the test characters and compared hiscopying performance on words to his performanceon nonwords. If the errors in delayed copyingindeed originate from the graphic pathway, thesefactors should not affect the copying performance.The word-length factors of the test characters(number of strokes and number of logographemes)were also manipulated to explore whether theimpaired component has buffer-like properties.Then, based on the error corpus obtained in theexperiment, we examined at what orthographiclevels the errors occurred on (strokes, logogra-phemes, or radicals), what potential factorsmight predict the performance (e.g., positioninside of the test character), and what relationshipsthe target and the erroneous units have (e.g.,visual-spatial or stroke features). The resultsfrom these analyses would potentially answerthese following interrelated questions: (a) Whatis an exact deficit point along “the graphicpathway” that best explains the error patterns—“the graphemic output buffer”, the “allographicrepresentation”, or the “graphic motor pattern”?(b) What are the functional units in the impairedcomponent? (c) What are the structural character-istics of that impaired component?

Method

MaterialsTo avoid further complications such as mor-phology, we only selected single-character (alsosingle-syllable) words as test material.

Frequency and phonetic regularity. A total of 160“composite” characters were subdivided into four40-character lists: 2 (frequency: high, low) � 2(regularity: regular, irregular). A regular compositecharacter (e.g., , horror, /bu4/) shares identicalpronunciation with its phonetic radical (e.g., ,

cloth, /bu4/), whereas an irregular character(e.g., , wrong, /cuo4/) has a completely differentpronunciation from that of its phonetic radical(e.g., , past, /xi1/). The character frequencycounts are from the Frequency Dictionary ofModern Chinese (ILTR, 1986). We matched thefour kinds of characters on visual complexitymeasures including number of strokes andnumber of logographemes. The statistics formean frequency, mean logographeme number,and mean stroke number in each category are thefollowing: high-frequency regular (353 + 171,5

2.93 + 0.89, 10.25 + 2.92); high-frequencyirregular (353 + 240, 2.63 + 0.67, 10.25 +2.23), low-frequency regular (13 + 9, 2.73 +0.64, 10.25 + 2.38), and low-frequency irregular(13 + 9, 2.83 + 0.90, 10.25 + 2.23).

Concreteness. A total of 44 characters were selected,half of which were semantically concrete (e.g., ,lamp), and the other half were semanticallyabstract (e.g., , similar). The selected characterswere given to 10 normal participants to rate theirsemantic concreteness/abstractness on a 7-pointscale, with 1 being most concrete and 7 beingmost abstract. The mean ratings were 2.03 forthe concrete characters and 5.17 for the abstractcharacters, and the difference was highly signifi-cant, t(42) ¼ –11.275, p , .0001. These twolists of characters were matched on frequency,number of logographemes, and number ofstrokes (concrete, 76 + 123, 2.82 + 1.01,9.54 + 3.47; abstract, 76 + 125, 2.86 +1.04, 10.00 + 3.00, respectively).

Grammatical word class. The following three 33-character lists were chosen: concrete nouns (e.g.,, wolf), abstract nouns (e.g., , disaster), and

abstract verbs (e.g., , forget). They werematched on frequency and number of strokes(concrete nouns, 396 + 504, 9.64 + 2.87;abstract nouns, 405 + 404, 9.70 + 2.72;abstract verbs, 414 + 592, 9.61 + 3.47). Carewas also taken that equal number of items ineach list were regular composite characters.

5 The first value is the condition mean and the second the standard deviation.


HAN ET AL.

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

Lexicality. A total of 20 real characters wereselected, and 20 (legal) noncharacters were gener-ated by randomly pairing semantic radicalsand phonetic radicals together. The two listswere matched on the number of strokesand logographemes (real characters, 8.1 + 2.22and 2.45 + 0.60; noncharacters, 8.1 + 2.43and 2.45 + 0.60, respectively).

Number of strokes. A total of 148 items were splitequally into a “few-stroke” character list and a“many-stroke” character list (stroke number,9.51 + 0.73 vs. 14.38 + 1.63). They werebalanced on character frequency (56 + 69.34vs. 57 + 60.22) and logographeme number(3.00 + 1.01 vs. 3.03 + 1.04).

Number of logographemes. In total 385 characterswere selected including 140 two-, 140 three-,and 105 four-logographeme characters. Wematched these three sets of characters on numberof strokes (11.04 + 6.65, 10.91 + 6.25, 11.50+ 6.67, respectively) and character frequency(186 + 212, 151 + 186, 173 + 257,respectively).

ProcedureIn each trial of the testing, the experimenter pre-sented one visual character in the middle of asheet of paper, and W.L.Z. was allowed to lookat it for two seconds. The stimulus was thenremoved, and W.L.Z. was required to write itdown. In fact, although we encouraged W.L.Z.to look at the target character for two seconds,he often wrote it after a quick glance. In that situ-ation, we removed the target once he began towrite. The presentation time did not seem toaffect the copying performance. The testing wascompleted in nine sessions in 2004.

Results

The results section is organized according to thequestions that we laid out at the beginning ofthe experiment. First we looked at the variablesthat we manipulated on the target characters.Then all the errors were compiled to see

what the basic error units were, what character-istics of the units predicted his copying perform-ance, and what relationships existed between thetargets and responses.

1. Various lists of the target charactersTable 1 displays W.L.Z.’s correct percentage forvarious effects in the delayed copy task. Therewere no significant effects of lexical variables,including frequency, x2(1) ¼ 1.82, p ¼ .18, ortho-graphy–phonology regularity, x2(1) , 1, concre-teness, x2(1) ¼ 2.07, p ¼ .15, and grammaticalclass, x2(2) ¼ 1.22, p ¼ .54. His writing was notinfluenced by the lexicality factor either: Therewas no difference between the performance onreal characters and noncharacters, x2(1) , 1.

Table 1. Delayed copying performance as a function of theproperties of the target characters

Word type SignificanceaCorrect

responseb

Word frequency andregularity effect nsHigh frequency, regular 43 (17/40)High frequency, irregular 33 (13/40)Low frequency, regular 23 (9/40)Low frequency, irregular 33 (13/40)

Concreteness effect nsConcrete 14 (3/22)Abstract 32 (7/22)

Grammatical class effect nsConcrete noun 64 (21/33)Abstract noun 76 (25/33)Verb 67 (22/33)

Lexicality effect nsCharacter 55 (11/20)Noncharacter 45 (9/20)

Stroke number effect nsFew strokes 29 (15/74)More strokes 26 (19/74)

Logographeme number effect ��

Two-logographemecharacter

48 (67/140)

Three-logographemecharacter

31 (44/140)

Four-logographemecharacter

24 (24/105)

aChi-square test: ns: p . .05; ��p , .001. bIn percentages,numbers in parentheses.



Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

The picture with regard to the “word length”factors was more complicated. No effect of strokenumber on copying performance was observed,x2(1) , 1. There was a trend such that the morelogographemes a character has, the more likely itis for errors to occur on the character, but the sig-nificance of the effect depends on the way in whichit is evaluated. First, if the accuracy rate is calcu-lated on the character level, the correct percentagein two-, three-, and four-logographeme characterswere 48%, 31%, and 24%, respectively, x2(2) ¼17.75, p , .0001. To evaluate whether the prob-ability of getting a logographeme correct is deter-mined by the number of logographemes in theword, we calculated the logographeme error ratesby calculating the percentage of mistaken logogra-pheme instances divided by the total number oflogographemes in each group. The corrected logo-grapheme percentages in two-, three-, and four-logographeme characters were 77%, 68%, and67%, respectively, x2(2) ¼ 4.767, p ¼ .09. Thismarginally significant trend of the word lengtheffect was further examined using regressionanalysis, and is shown in Section 3 of the Resultssection.

2. Error analyses: Stroke vs. logographeme vs.radicalsIn the above analyses we looked at what factors ofthe test target affected the likelihood of errors;now we turn to the characteristics of the errorsthemselves. First, when an error is made, is thewhole character being mistakenly produced, is astroke miswritten, or are the logographemes/rad-icals substituted, deleted, added, or transposed?The answer to this question will provide infor-mation regarding the nature of the functionalunits in the impaired representation. To investi-gate this issue we compiled all the errors (557characters) from all subsets in the experiment(876 characters in total). We first classified theerroneous characters into two types according towhether or not the response was a real character.The character responses were scored as whetherthey had semantic or phonological/orthographicrelationship with the targets. The noncharacterresponses were further divided into four categories

according to type of the erroneous orthographicunits: logographeme errors, stroke errors, combi-nation errors, and unrecognizable errors (see Law& Leung, 2000). Errors were scored as logogra-pheme errors when a target logographeme wassubstituted, deleted, added, or transposed. Strokeerrors were those cases where only a stroke wasmiswritten. Combination errors were those errorswhen the responses contained both logographemeerrors and strokes errors. The possibility that someerrors should be categorized as radical errors is dis-cussed later. An unrecognizable error indicatedthat it was difficult to identify the response.Table 2 displays the examples and distributionsof various error types. We can see that the mostcommon errors were logographeme errors.

We further classified the logographeme errorsinto logographeme substitution, deletion, inser-tion, and transposition errors (see Law & Leung,2000). A total of 55% of the 499 characters withlogographeme errors contained only a single logo-grapheme error, 32% contained two logographemeerrors, 13% contained three, and 1% containedfour. When the errors involved multiple logogra-phemes, only in about 8% of the cases were theerroneous logographemes apart (e.g., being the2nd and the 4th in a four-logographeme charac-ter)—that is, the majority were adjacent to eachother. This pattern might be associated with thefact that W.L.Z. tended to make more errorstowards the end positions of the character (seethe Section below) and therefore is not exploredfurther. In total, we collected 796 individual logo-grapheme errors. Table 3 presents the distributionand examples of each error subtype.Logographeme substitution errors were found tobe the most prevalent type, indicating that thelogographemes are the functional units for theimpaired cognitive components.

One caveat that we need to consider is whetherthe logographeme errors are instead radical errorsor stroke errors. The relationship between radicalsand logographemes are not systematic. While rad-icals are linguistic units that bear various kinds oflexical and/or semantic information, logogra-phemes are visual-spatial/motoric units. Someradicals (most semantic radicals) correspond to


HAN ET AL.

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

one logographeme, and others correspond to twoor more logographemes (most phonetic radicals).If the errors involve single-logographeme radicals,it is impossible to tease apart these two kinds oferroneous units. If an error occurs on two ormore logographemes of a target character, wecould examine whether these multiple logogra-phemes belong to one radical and for the substi-tution cases whether the erroneous response alsocorresponds to a radical. We found that in thewhole set of 557 erroneously written characters,the majority (398) of the errors occurred on logo-graphemes that did not correspond to any radical,147 involved errors (target and/or response) thatcould be classified both as one logographeme andas one radical, and in only 12 cases were multiplelogographemes involved that corresponded to rad-icals. Therefore, real logographeme errors, asopposed to radical errors, were the most commontype of error. By the same token, it is highly

unlikely that these errors are stroke errors thathappened to constitute logographemes. In allerroneously written logographemes, 95% of theerrors were multiple-stroke logographemes.Furthermore, in the rare cases where errorsoccurred on a single-stroke logographeme, it wasalways substituted by a multiple-strokelogographeme.

3. Logographeme errors: A regression analysisWe have observed that the copying performancedid not seem to be affected by the lexical factorsof the target characters, and that the errors werealmost exclusively logographeme errors. In thissection, we carried out a logistic regression analysisto explore more potential variables that mightpredict the copying performance of a particularlogographeme. We analysed the whole set of2,931 logographemes that appeared in all thetested characters, out of which W.L.Z. correctly

Table 2. The percentage and examples of various error types

Examples

Error type % (N) Target Response

Character 4 (27) (axe, /fu3/) (father, /ba4/)Noncharacter

Logographemea 91 (499)Stroke 2 (13) (escape, /tao2/)

Logographeme& stroke

0.3 (2) (invite, /qing3/)

Unrecognizable 2 (16)Total 100 (557)

aSee Table 3 for further information about logographeme errors.

Table 3. The percentage and examples of various types of logographeme errors

Examples

Error type % (N) Target Response

Substitution 80 (639) (hoarse, /si1/)

Deletion 19 (151) (wither, /wei3/)

Insertion 1 (6) (escape, /tao2/)

Transposition 0 (0)

Total 100 (796)



Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

copied 68%. The dependent variable was the scoreofW.L.Z.’s copying result for a particular logogra-pheme (1 for correct and 0 for incorrect). The pre-dictors covered a range of various properties of thelogographemes and of the corresponding charac-ters, including number of logographemes in thecorresponding test character, log frequency of thecorresponding test character (Sun, 1998), strokenumber of the logographeme, log frequency ofthe logographeme (Standards Press of China,1994), and the temporal position of the logogra-pheme in the corresponding test character.

The correlation matrix among the predictors(see Table 4) shows some predicted correlations.For instance, position of logographeme in charac-ter and number of logographemes per characterwere positively correlated. The character frequencyand the number of logographemes per characterwere negatively correlated, indicating that thehigher the frequency was, the visually and/ormotorically simpler a character tended to be.Similarly, logographeme frequency and strokenumber in a logographeme were also negativelycorrelated. Some correlations were not necessarilypredicted but seemed reasonable and interesting—for example, the stroke number within a logogra-pheme and its position in the corresponding char-acter were negative correlated. It might indicatethat “simpler” logographemes tend to occur atthe end of characters. Some correlations—forexample, the negative correlation between the

logographeme frequency and the number of logo-grapheme in the target character—are hard tointerpret and may be due to unexplored factorsor chance.

We conducted a logistic regression using aforward (LR) stepwise method, where the vari-ables are automatically selected and entered intothe model in sequence of their weight of contri-bution on the dependent variable. The results aredisplayed in Table 5. We found that of the vari-ables, position in the test character was the mostsignificant predictor to the probability of the logo-grapheme copying score, x2(1) ¼ 546.2, p , .000,followed by the number of logographemes per testcharacter, x2(1) ¼ 82.7, p , .000, the log fre-quency of the logographeme, x2(1) ¼ 17.7, p ,.000, and the log frequency of the test character,x2(1) ¼ 13.0, p , .000. Stroke number withinthe logographeme did not make an independentcontribution and was not included in the model.

The regression analysis results partly replicatedthe previous findings that W.L.Z.’s accuracy inwriting logographemes was a function of howmany logographemes the corresponding test char-acter contained, and the number of strokes in thecharacter or the logographeme did not seem tomatter, indicating that character complexity orlength was better measured by logographemethan by stroke. The target frequency reached sig-nificance in predicting the scoring of a logogra-pheme in this analysis, contradicting the absenceof the target frequency effect in the previoussection (see Table 1). To confirm that it indeedhad an independent contribution, we entered allthe other variables into the equation first andthen entered the target frequency variable, andwe found that it significantly increased the predic-tive power (p , .001). This part of the resultssuggests either that lexical knowledge influencedW.L.Z.’s copying performance (e.g., see Sage &Ellis, 2004) or that his lexical deficit mightindeed play a role in the copying pattern. Weargue that it would not seriously challenge ourrationale of using W.L.Z.’s copying difficulties tostudy the postlexical “graphic pathway” forwriting, if we look at the overall pattern of hisbehaviour. In particular, in all analyses when

Table 4. R-values among the predictors for the regression analysis.

NLC FC SNL FL PLC

NLC 1FC 20.084�� 1SNL 20.377�� 20.049� 1FL 20.277�� 0.006 20.357�� 1PLC 0.414�� 20.035 20.164�� 0.018 1

Note: Labels: NLC ¼ number of logographemes percorresponding test character. FC ¼ log frequency of thecorresponding test character. SNL ¼ stroke number ofthe logographeme. FL¼ log frequency of the logographeme.PLC ¼ temporal position of the logographeme in thecorresponding test character.


HAN ET AL.

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

variables other than frequency were examined fre-quency was always controlled for.

One interesting observation that emerged fromthe regression analysis is that the performance on aparticular logographeme was highly affected by itsserial order position in character (p , .0001). Tofurther clarify this “position effect”, we split allthe characters with two to five logographemes indelayed copy according to the numbers of logogra-phemes in each character into 242 two-, 389three-, 300 four-, and 52 five-logographeme char-acters. We then scored each logographeme in eachcharacter. The scoring procedure approximatelyabided by the principle of analysis on incorrectresponse used in the case of patient L.B.(Caramazza & Miceli, 1990, p. 250).

Figure 4 shows that W.L.Z.’s error percentagefor each position within the characters variedaccording to the logographemes’ numbers.Collapsing all the characters regardless of thecharacter length, the first, second, third, fourth,and fifth logographemes in the characters wereincorrectly written 7% (71/983), 31% (304/983),51% (381/741), 63% (223/352), and 63% (33/52), respectively. W.L.Z.’s writing exhibited a sig-nificant gradual increase in error percentages fromthe initial logographeme to the final one in charac-ter, x2(4) ¼ 583.5, p , .0001. Moreover, a similarpattern presented in every stimuli group broken upby length: two-logographeme, x2(1) ¼ 95.8,

p , .0001; three-logographeme, x2(2) ¼ 215.7,p , .0001; four-logographeme, x2(3) ¼ 287.4,p , .0001; and five-logographeme, x2(4) ¼63.3, p , .0001, characters. For example, hewrote as , where the last logographeme wassubstituted with . He wrote as , where thefirst logographeme was accurately written, themiddle one was replaced by , and the lastone was deleted. In other words, a linear serialposition effect presented in W.L.Z.’s writing.

There are several possible explanations for thiseffect. First of all, it is possible that the logogra-phemes at the end position tend to be more diffi-cult (e.g., visually more complex or less frequent).Second, if the effect is real, it might originate

Table 5. Results of a logistic regression analysis of 2,931 items with W.L.Z.’s writing accuracy as the dependent variable

Step VariableModel loglikelihood

Change in 22 loglikelihood df p-value

1 Position of logo in char 21,838.2 546.2 1 ,.00012 Number of logo per char 21,565.1 82.7 1 ,.0001

Position of logo in char 21,834.0 620.6 1 ,.00013 Number of logo per char 21,542.2 54.7 1 ,.0001

Position of logo in char 21,812.7 595.8 1 ,.0001Freq of logo 21,523.7 17.7 1 ,.0001

4 Number of logo per char 21,537.8 58.9 1 ,.0001Freq of char 21,514.8 13.0 1 ,.0001Position of logo in char 21,807.6 598.5 1 ,.0001Freq of logo 21,516.4 16.1 1 ,.0001

Note: Position of logo in char¼ temporal position of the logographeme in the corresponding test character. Number of logo per char¼ number of logographemes per corresponding test character. Freq of logo ¼ log frequency of the logographeme.

Figure 4. Serial position effect of the logographemes in copyingcharacters.



Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

either from the input process (when visuallyencoding the stimuli) or from the output process(when reproducing the stimuli). Although therehas been evidence that the visual input in wordrecognition happens in a parallel fashion (e.g.,Coltheart, Rastle, Perry, Langdon, & Ziegler,2001), one can imagine that the input process ina copying task scans the target character in aleft-to-right/top-to-bottom fashion and thereforefavours the left/top logographemes over those onthe right/bottom. Indeed, in an study (Zhang &Sheng, 1999) where participants were asked toname logographemes embedded in different pos-ition of visually presented Chinese characters, itwas observed that the naming performance forthe same logographeme was better in the top pos-ition than in the bottom position and was better inthe left position than in the right position. Theauthors attributed such results to the viewinghabit of Chinese readers. By contrast, if the pos-itional effect originates from an output process, a“temporal” effect is predicted such that earlieroutput logographemes are better copied thanlater output logographemes. Many Chinese char-acters present an interesting discrepancy betweenthe “(input) spatial order” and the “(output) tem-poral order”. For example, the character hastwo logographemes: and . In a temporal per-spective, is its last written logographeme,whereas from the spatial left-to-right sequence,it is the first one. Given that an input deficitwould produce either no positional effect (as inparallel processing) or an effect favouring theleft/top over the right/bottom logographemes(in that scanning order), if the later outputelements in such characters are found to be moreprone to impairment, it should be due to a buffer-ing process for output. In other words, if a “tem-poral” positional effect was observed, it would bestrong evidence for the hypothesis that W.L.Z.’scopying errors result from the impairment of abuffering component in the output process thatserves a similar functional role as the “graphemicoutput buffer” in writing alphabetic languages.The following analyses and experiment weredone to test these three hypotheses regarding theposition effect.

3.1. Testing the difficulty difference account. To ruleout the possibility that the position effect is dueto the difficulty difference in different positions,30 pairs of characters were selected. The two char-acters in each pair share one identical logogra-pheme that are in different positions (e.g.,

, with the logographeme on the leftof and the right of ). The two groupsof characters were also matched on frequency(410 vs. 535; t , 1). W.L.Z. correctly copiedfewer critical logographemes when they were onthe right position than when they were on theleft position (22/30 vs. 13/30); x2(1) ¼ 5.55,p , .05. For example, he correctly wrote in

, but miswrote in as . Thispattern was also replicated by using 30 pairs ofnoncharacters where the same logographemeswere used both in the left and in the right positionson different trials, x2(1) ¼ 40.85, p , .0001.

3.2. Distinguishing the (input) spatial position vs. the(output) temporal position. To clarify whether thepositional effect is an (input) spatial effect or an(output) temporal effect, we inspected all charac-ters (N ¼ 28) from the tested set that showeddiscrepancy between the spatial position and thetemporal position—for example, (the lastspatial logographeme is , and the last logogra-pheme to be written is ). If it is the spatial pos-ition that matters we would predict thelogographeme on the left ( ) to be copied betterthan the one on the right ( ), and vice versa ifthe temporal position matters. The resultsshowed that the (relatively) right logographemes(19/28) were indeed copied more correctly thanthe (relatively) left logographemes (12/28) inthese characters, x2(1) ¼ 3.54, p ¼ .06.Furthermore, this difference is not attributable toany “difficulty” differences because the 28 rightlogographemes were “simpler” (stroke number:2.89 vs. 3.68, p , .05) and of higher frequency(112 vs. 38, p ¼ .07) than the 28 left logogra-phemes. Thus, although the “temporal” positionaltrend was only marginally significant for thesecharacters, it was in the reverse direction from allthe other cases where temporal position and


HAN ET AL.

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

spatial position were consistent, indicating that theposition effect indeed was temporal.

4. Target–response logographeme relationshipsWhen W.L.Z. failed to copy a logographeme cor-rectly, what did he write? We examined the visual/motoric relationship between the target logogra-pheme and the corresponding erroneous responseto study whether the errors maintained any par-ticular kind of properties of the target logogra-pheme. Such analyses would inform us whetherthere is more information retained in the impairedcomponent when the specific identity of the logo-grapheme is lost. To avoid any potential confound,we studied the characters in which W.L.Z. madeonly one substitution error and collected 209 logo-grapheme substitution errors. We compared eachtarget–response pair on visual/motoric dimen-sions, using measures on the overall visual con-figurations (structure, stroke relation) andindividual stroke element (first stroke shape).“Structure” and “stroke relation” provide measuresfor the visual arrangements of the elements (e.g.,structure—strokes are aligned in a left/rightfashion or in a top/down fashion or otherwise;stroke relation—strokes are aligned in a crossingmanner or a connecting manner, etc.). The“stroke shape” variable measures the motoric fea-tures and/or visual shapes of an individual strokeby categorizing strokes into five rough categories(e.g., horizontal or vertical). We used classificationcriteria for each category of these propertiesderived from the CCCSGCSIP (1998; see thelabels in Table 6 for detailed descriptions), calcu-lated the probability of W.L.Z.’s substitutionerrors to be within a same category for a givenproperty, and compared the observed within-cat-egory probability to a chance level.

Take the “stroke relation” property, forexample. It can be categorized into six subtypes(single strokes, crossing, separate, connecting,crossed-connecting, and crossed-separate). Wewant to know whether W.L.Z.’s within-categorysubstitution (i.e., a single-stroke logographemesubstituted for another single-stroke one, a separ-ate stroke logographeme substituted for anotherseparate stroke one, etc.) tended to occur more

frequently than the chance level. First, wecounted W.L.Z.’s observed value of within-subca-tegory substitution performance and found that36% (75/209) substitutions occurred within cat-egory. Then we calculated the chance level ofwithin-subcategory substitutions using a MonteCarlo simulation procedure, adapting themethod used to establish the chance levels ofvisual/motor similarity in Rapp and Caramazza(1997). Firstly we randomly re-paired the 209erroneous logographemes with the target logogra-phemes and then computed a within-categoryprobability on the result of this re-pairing. Atotal of 5,000 such random pairings of the itemsin the target-error list were carried, generating5,000 baseline within-category probability valuesfor this “stroke-relation” property. To obtainvalues reflecting how likely W.L.Z.’s observedwithin-category probability (36%) for thismeasure is due to chance level, we calculated thepercentage of instances with a value equal to orhigher than this observed value in the 5,000 base-line runs. We found that such instances were gen-erated only 43 times, and the percentage (43/5,000) was less than 0.01 (i.e., p , .01). Thesame procedure was also conducted for the proper-ties “structure” and “first stroke shape”. As shownin Table 6, the patient’s observed within-categoryprobability for structure and first stroke shape werenever observed in 5,000 random pairings(ps, .0002). Thus, W.L.Z.’s logographeme sub-stitution errors tended to occur within the samecategories for all three measures, indicating thepreservation of visual/motoric properties.

GENERAL DISCUSSION

We presented a Chinese patient on whom wefocused on the delayed copying performance tostudy the postlexical process involved in writingChinese characters. We found that the semanticcharacteristics of the test characters did not affectthe likelihood of making an error but the characterfrequency did, suggesting that the lexical deficit ofthe patient might contribute to the copying per-formance. After we controlled for the lexical



Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

frequency contribution, the following findingsregarding his delayed copying performanceemerged from our experiment:

1. A particular word length factor—number oflogographemes—affected delayed copyingperformance.

2. The erroneous units almost always corre-sponded to logographemes as opposed to rad-icals or strokes, with logographemesubstitutions being the most frequent type oferror.

3. The logographeme’s temporal position in thecorresponding character and the logographemefrequency were significant predictors of thecopying performance, but its stroke numberwas not.

4. When a logographeme substitution error wasmade, it was substituted by those havingsimilar visual/motoric properties.

We first infer the deficit locus of our patientfrom these findings based on the framework devel-oped with alphabetic writing, and then we discussin turn the universality and language specificity ofthe impaired representation based on the analysesof the patient.

Locus of the deficit

Because W.L.Z. was perfect in direct copying, hisdifficulty in delayed copying cannot be attributed

to any peripheral motor deficit. His errors indelayed copying did not originate from thesemantic or phonology–orthography–conversionsystems because of the absence of semantic, gram-matical, and phonetic regularity effects. Althoughthe significant contribution of the characterfrequency in the regression analyses suggests thathis lexical deficit played a role in the copyingtask, we controlled for this frequency variable invarious analyses and therefore the error patternswe reported should result from impairment inthe stages beyond lexical retrieval—that is, fromthe “the graphic pathway”.

Within “the graphic pathway”, Finding 1 andthe positional effect of Finding 3 suggest thatthe deficit originated in a buffer-like structuresince it had the characteristics of a working-memory component. The performance wassensitive to the amount of information (numberof logographemes) and the time being held inmemory. The more information the buffer con-tained, and the longer it remained in the buffer,the more likely errors were to occur. It is criticalto note that we observed a temporal and notspatial order effect, indicating that effect origi-nated at a buffering process during outputinstead of input in the delayed copy task.Therefore we propose that W.L.Z. was impairedat the level of a buffering component for output,and such impairment contributed to the delayedcopying errors. The brain damage seemed to

Table 6. Comparison of the observed versus the expected percentage of the within-category substitutions between the target logographemes andresponses

Observed value Expected value

Logographeme property % N % Range Instance p-value

1. Structure 43 89/209 28 + 3 19–37 0 ,.00022. Stroke relation 36 75/209 29 + 3 19–40 43 ,.013. First stroke shape 42 87/209 23 + 3 13–34 0 ,.0002

Note: Labels: 1. A total of 10 overall spatial categories, including unique (e.g., ), left-right (e.g., ), top-bottom (e.g., ), top-middle-bottom (e.g., ), surround upper left (e.g., ), surround upper right (e.g., ), surround below (e.g., ), surround three-quarters (e.g., ), surround full (e.g., ), and frame (e.g., ). 2. Six types of stroke relation including single strokes (e.g., ),crossing (e.g., ), separate (e.g., ), connecting (e.g., ), crossed-connecting (e.g., ), and crossed-separate (e.g., ). 3. Thephysical configuration of the strokes in the logographeme, including horizontal (e.g., ), vertical (e.g., ), slanted (e.g., ), pointed(e.g., ), and crooked (e.g., ).


HAN ET AL.

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

have caused an abnormally rapid decay of infor-mation or failure in the refresh mechanismduring the buffering process.

Finding 2 and the logographeme frequencyeffect (in predicting the error rates) in Finding 3indicate that logographemes, as opposed tostrokes or radicals, are the functional units rep-resented in this impaired component, and itsresistance to impairment is sensitive to frequency.For this reason, we named this buffering com-ponent in writing Chinese characters “logogra-pheme output buffer” (LOB). We furtherobserved that there was stroke-feature similaritybetween the target logographeme and the responselogographeme, indicating that LOB in writingChinese characters encodes graphic information(shape and/or stroke features).

Below we discuss the theoretical implications ofour findings for two issues: the universality of abuffering component in the writing process andthe language-specific parameters of such acomponent.

The universal and language-specific aspectsof the output buffer

In alphabetic languages, convincing empirical evi-dence for the existence of a graphemic outputbuffer comes from the strong association betweenoral spelling and written spelling performance incertain classes of patients. Nevertheless, the lackof oral spelling means in Chinese does not meanthat a buffering structure is not necessary forwriting Chinese characters. If we consider thetheoretical motivation for the proposal of a buffer-ing process in writing, the idea of it being universalin all languages becomes natural. In general, whenthe “unit” of information that is output from onerepresentation is larger than what the subsequentrepresentation can take as input for furtherprocessing, it is reasonable to assume the existenceof a buffer to hold the to-be-processed unitstemporarily. Our analysis demonstrates thatlogographemes are the basic units in writingChinese characters, in a sense comparable to thestatus of letters in alphabetic writing systems (seea similar position in Law, 2004), and that there

exists a universal output buffering component inwriting both in Chinese and in alphabeticlanguages.

Of course there are fundamental differencesbetween the units being represented in theoutput buffer for alphabetic languages andChinese. One critical difference is that the “gra-phemes” in alphabetic languages representsamodal graphemes without name, shape, or font.Shape and stroke information are implementedat level(s) beyond the graphemic output buffer—that is, the allographic representation andgraphic motor pattern. Indeed, Rapp andCaramazza (1997) reported that patients whohad selective graphemic buffer deficit showed notarget–response similarity in visual-spatial orstroke features in their substitution errors, whilethey showed similarity with regard to abstractproperties (e.g., consonant/vowel status or syllableorganization). Target–response similarity in termsof visual-spatial or stroke features was present inthe substitution errors made by patients who suf-fered from “graphic motor pattern” deficit.Furthermore, graphemes have been argued to cor-respond to phonemes (e.g., see Tainturier & Rapp,2004, for a discussion), and therefore the buffercontains multidimensional phonological-relatedfeatures along with the grapheme identities andtheir order, such as the consonant–vowel statusand possibly syllabic organization information.By contrast, logographemes in Chinese charactersare purely “orthographic” units and do not serve asthe connection between individual sounds andorthography. In fact, only 49% can be pronounced(CCCSGCSIP, 1998), and their pronunciationsdo not correspond to the pronunciations of thecharacters they appear in.

Thus, it is interesting why logographemes act asthe basic units in writing characters, as opposed toother units, such as strokes. We argue there arecomputational advantages to having logogra-phemes as the functional units. Consider the char-acter (luck, /fu2/), for example. The idea ofhaving strokes as the functional units seems unli-kely. First, strokes of a character are generallytoo many to be kept in the working-memorysystem. The normal capacity of working memory



Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

is from four to nine chunks (Cowan, 2001; Miller,1956), whereas has 13 strokes (mean number ofstrokes in Chinese characters is 12.85, StandardsPress of China, 1994). Second, the strokes consti-tuting a character are highly ambiguous in shapeand position. For example, has four ofvarious sizes in different positions, easily leadingto confusions in writing. By contrast, the logogra-phemes in a character can be easily kept in workingmemory. is segmented into four logographemeswith spatial specification ( , , and ; meannumber of logographemes in a character is 3.64,Standards Press of China, 1994), which is lowerthan the maximum memory load. Also, the ambi-guity among strokes disappears when strokes of acharacter are embedded in logographemes.Another possible candidate for functional unitsin the buffer are radicals, a position motivated bythe cases that made writing errors on radicallevels (e.g., Law, 1994, 2004; Law & Caramazza,1995; Law et al., 2005). However, it is unclearwhat can be gained by having radicals as functionalunits because semantic radicals most correspondto logographemes, and phonetic radicalsusually correspond to existing characters.Their patients’ writing errors can therefore beexplained by either logographeme errors orcharacter errors.

The differences of the intrinsic characteristicsof logographemes and alphabetic graphemesshould lead to structural differences betweenLOB for Chinese and the graphemic buffer inalphabetic languages. Present findings fromW.L.Z. might indicate that the logographemesin the buffer are represented by some kind oforthographic feature. W.L.Z.’s logographemesubstitution errors resulted in significantly higherproportion of logographemes having similarvisual/motoric properties (structure, strokerelation, first stroke shape) to the target logogra-pheme than what was expected by chance. Thiscould reasonably be due to the fact that, althoughthe brain damage led W.L.Z.’s buffer system to beunable to efficiently access and/or select identityinformation of the target logographeme, thestroke shape information of the logographemeswas preserved. Thus, the system tried to look for

a substitute with similar visual/motoric infor-mation to that of the target logographeme.

To conclude, by studying the delayed-copyingperformance of a Chinese dysgraphic patient, wehave evidence that an output buffering componentis universal in logographic and alphabeticlanguages, and that the structure within thebuffer is shaped by language-specific parameters.The output buffer structure in Chinese representsthe identity and visual/motoric properties oflogographemes.

Manuscript received 10 May 2006Revised manuscript received 29 March 2007Revised manuscript accepted 30 March 2007

First published online 24 May 2007

REFERENCES

Annoni, J. M., Lemay, M. A., de Mattos Pimenta,M. A., & Lecours, A. R. (1998). The contributionof attentional mechanisms to an irregularity effectat the graphemic buffer level. Brain and Language,63, 64–78.

Badecker, W., Hillis. A., & Caramazza, A. (1990).Lexical morphology and its role in the writingprocess: Evidence from a case of acquired dysgraphia.Cognition, 35, 205–243.

Bi, Y., Han, Z., Weekes, B., & Shu, H. (in press). Theinteraction between semantic and sublexical systemsin reading: Evidence from Chinese. Neuropsychologia.

Bub, D. N., & Kertesz, A. (1982). Lexical or ortho-graphic agraphia. Brain, 104, 21–49.

Buchwald, A., & Rapp, B. (2006). Consonants andvowels in orthographic representations. CognitiveNeuropsychology, 21, 308–337.

Caramazza, A., & Miceli, G. (1990). The structure ofgraphemic representations. Cognition, 37, 243–297.

Caramazza, A., Miceli, G., Villa, G., & Romani, C.(1987). The role of the graphemic buffer in spelling:Evidence from a case of acquired dysgraphia.Cognition, 26, 59–85.

Cipolotti, L., Bird, C. M., Glasspool, D. W., &Shallice, T. (2004). The impact of deep dysgraphiaon graphemic buffer disorders. Neurocase, 10,405–419.

Coltheart, M., Rastle, K., Perry, C., Langdon, R., &Ziegler, J. (2001). DRC: A dual route cascaded


HAN ET AL.

Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

model of visual word recognition and reading aloud.Psychological Review, 108, 204–256.

Cotelli, M., Abutalebi, J., Zorzi, M., & Cappa, S. F.(2003). Vowels in the buffer: A case of acquired dys-graphia with selective vowel substitutions. CognitiveNeuropsychology, 20, 99–114.

Cowan, N. (2001). The magical number 4 in short-termmemory: A reconsideration of mental storagecapacity. Behavioral and Brain Sciences, 24, 87–185.

Ellis, A. W. (1982). Spelling and writing (and readingand speaking). In A. W. Ellis (Ed.), Normality andpathology in cognitive functions (pp. 113–146).London: Academic Press.

Ellis, A. W. (1988). Normal writing processes and per-ipheral acquired dysgraphias. Language and CognitiveProcesses, 3, 99–127.

Fu, Y. H. (1991). Hanzi de bujian [Logographemes ofChinese characters]. Yuwenjianshe, 12, 3–6.

Gao, S., Wang, Y., Shi, S., Liu, J., Lin, G., Rao, B.,et al. (1993). Aphasia. Beijing, China: The UnionPress of Beijing Medical University–the BeijingUnion Medical College.

Graham, N. L., Patterson, K., & Hodges, J. R. (1997).Progressive dysgraphia: Cooccurrence of central andperipheral impairments. Cognitive Neuropsychology,14, 975–1005.

Hillis, A., & Caramazza, A. (1989). The graphemicbuffer and attentional mechanisms. Brain andLanguage, 36, 208–235.

Institute of Language Teaching and Research (ILTR).(1986). Frequency Dictionary of Modern Chinese(Xiandai Hanyu Pinlv Cidian). Beijing, China:Beijing Language Institute Press.

Kertesz, A. (1982). Western Aphasia Battery. SanAntonio, TX: Psychological Corporation.

Kokubo, K., Suzuki, K., Yamadori, A., & Satou, K.(2001). Pure Kana agraphia as a manifestation ofgraphemic buffer impairment. Cortex, 37, 187–195.

Law, S. P. (1994). The structure of orthographicrepresentations of Chinese characters: From theperspective of the cognitive neuropsychologicalapproach. Bulletin of Institute of History andPhilology, 65, 81–130.

Law, S. P. (2004). Writing errors of a Cantonese dys-graphic patient and their theoretical implications.Neurocase, 10, 132–140.

Law, S. P., & Caramazza, A. (1995). Cognitive pro-cesses in writing Chinese characters: Basic issuesand some preliminary data. In B. de Gelder &J. Morais (Eds.), Speech and reading: A comparative

approach (pp. 143–190). Hove, UK: PsychologyPress.

Law, S. P., & Leung, M. T. (2000). Structural represen-tations of characters in Chinese writing: Evidencefrom a case of acquired dysgraphia. Psychologia, 43,67–83.

Law, S. P., & Or, B. (2001). A case study of acquireddyslexia and dysgraphia in Cantonese: Evidence fornonsemantic pathways for reading and writing inChinese. Cognitive Neuropsychology, 18, 729–748.

Law, S. P., Yeung, O., Wong, W., & Chiu, K. M. Y.(2005). Processing of semantic radicals in writingChinese characters: Data from a Chinese dysgraphicpatient. Cognitive Neuropsychology, 22, 885–903.

Margolin, D. I. (1984). The neuropsychology of writingand spelling: Semantic phonological, motor and per-ceptual processes. Quarterly Journal of ExperimentalPsychology, 36A, 459–489.

McCloskey, M., Badecker, W., Goodman-Schulman,R. A., & Ajiminosa, D. (1994). The structure ofgraphemic representations in spelling: Evidencefrom a case of acquired dysgraphia. CognitiveNeuropsychology, 11, 341–392.

Miller, G. A. (1956). The magical number seven, plus orminus two: Some limits on our capacity for proces-sing information. Psychological Review, 63, 81–97.

Orliaguet, J. P., & Boë, L. J. (1993). The role of linguis-tics in the speed of handwriting movements: Effectsof spelling uncertainty. Acta Psychologica, 82,103–113.

Rapp, B., & Caramazza, A. (1997). From graphemes toabstract letter shapes: Levels of representation inwritten spelling. Journal of Experimental Psychology:Human Perception and Performance, 23, 1130–1152.

Rapp, B., & Kong, D. (2002). Revealing the componentfunctions of the graphemic buffer. Brain andLanguage, 83, 112–114.

Reich, S., Chou, T. L., & Patterson, K. (2003).Acquired dysgraphia in Chinese: Further evidenceon the links between phonology and orthography.Aphasiology, 17, 585–604.

Sage, K., & Ellis, A. W. (2004). Lexical influencesin graphemic buffer disorder. CognitiveNeuropsychology, 21, 381–400.

Schiller, N. O., Greenhall, J. A., Shelton, J. R., &Caramazza, A. (2001). Serial order effect in spellingerrors: Evidence from two dysgraphic patients.Neurocase, 7, 1–14.

Shu, H. (2003). Chinese writing system and learning toread. International Journal of Psychology, 38, 274–285.



Dow

nloa

ded

By:

[Bei

jing

Nor

mal

Uni

vers

ity] A

t: 07

:49

3 A

ugus

t 200

7

State Language Commission. (1998). The Chinese char-acter component standard of GB13000.1 character set forinformation processing. Beijing, China: Language &Culture Press.

State Technology Supervision Bureau. (1994).Information technology — UCS: Universal multiple-octet coded character set (Part 1 Architecture and basicmultilingual plane). Beijing, China: Standards Pressof China.

Su, P. C. (1994). Xiandai Hanzixue gangyao [An intro-duction to the study of contemporary Chinese char-acters]. Beijing, China: Beijing University Press.

Sun, M. S. (1998). The Chinese written language corpus.Beijing, China: Tsinghua University. RetrievedSeptember 3, 1999, from http://www.lits.tsinghua.edu.cn/ainlp/source1.htm

Tainturier, M. J., & Caramazza, A. (1996). The status ofdouble letters in graphemic representations. Journal ofMemory and Language, 35, 53–75.

Tainturier, M. J., & Rapp, B. (2004). Complex gra-phemes as functional spelling units: Evidence fromacquired dysgraphia. Neurocase, 10, 122–131.

Wang,W. S. Y. (1973). The Chinese language. ScientificAmerican, 228, 50–60.

Ward, J., & Romani, C. (2000). Consonant–vowelencoding and orthosyllables in a case of acquired dys-graphia. Cognitive Neuropsychology, 17, 641–663.

Weekes, B. S., Yin, W., Su, I. F., & Chen, M. J. (2006).The cognitive neuropsychology of reading andwriting in Chinese. Language and Linguistics, 7,595–617.

Zesiger, P., Martory, M. D., & Mayer, E. (1997).Writing without graphic motor patterns: A case ofdysgraphia for letters and digits sparing shorthandwriting. Cognitive Neuropsychology, 14, 743–763.

Zesiger, P., Orliaguet, J. P., Boë, L. J., &Moundoud, P.(1994). The role of syllabic structure in handwritingand typing production. In C. Faure, P. Keuss,G. Luretter, & A. Vinter (Eds.), Advances in hand-writing: A multidisciplinary approach. Paris: Europia.

Zhang, P. (1984). Hanzi bujian fenxi de yuanjiu helilun [The method and theory of analyzinglogographemes in Chinese characters]. Yuwenjanjiu,1, 37–43.

Zhang, J., & Sheng, H. (1999). Study on the influenceof the relationship of the wholes and their parts inthe perceptual separation of Chinese characters.Acta Psychologica Sinica, 31, 369–376.


HAN ET AL.

Documents

Cognitive Neuropsychologybrain.bnu.edu.cn/home/yanchaobi/recent papers/Han et al 2007CN.pdf · writing to dictation, written naming), or the output modalities (e.g., written spelling,