Processing of compound-word characters in reading Chinese: An eye-movement-contingent display change study

This article was downloaded by: [Turku University]On: 30 March 2014, At: 09:52Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office:Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

The Quarterly Journal of ExperimentalPsychologyPublication details, including instructions for authors and subscriptioninformation:http://www.tandfonline.com/loi/pqje20

Processing of compound-word charactersin reading Chinese: An eye-movement-contingent display change studyLei Cui a b , Guoli Yan a b , Xuejun Bai a b , Jukka Hyönä c , Suiping Wang d &Simon P. Liversedge ea Academy of Psychology and Behavior , Tianjin Normal University ,Tianjin , P.R. Chinab Department of Education , Jining University , Qufu , P.R. Chinac Department of Psychology , University of Turku , Turku , Finlandd Department of Psychology , South China Normal University , Guangzhou ,P.R. Chinae School of Psychology , University of Southampton , Southampton , UKAccepted author version posted online: 21 Feb 2012.Published online: 24Apr 2012.

To cite this article: Lei Cui , Guoli Yan , Xuejun Bai , Jukka Hyönä , Suiping Wang & Simon P.Liversedge (2013) Processing of compound-word characters in reading Chinese: An eye-movement-contingent display change study, The Quarterly Journal of Experimental Psychology, 66:3, 527-547, DOI:10.1080/17470218.2012.667423

To link to this article: http://dx.doi.org/10.1080/17470218.2012.667423

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)contained in the publications on our platform. However, Taylor & Francis, our agents, and ourlicensors make no representations or warranties whatsoever as to the accuracy, completeness, orsuitability for any purpose of the Content. Any opinions and views expressed in this publicationare the opinions and views of the authors, and are not the views of or endorsed by Taylor &Francis. The accuracy of the Content should not be relied upon and should be independentlyverified with primary sources of information. Taylor and Francis shall not be liable for anylosses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilitieswhatsoever or howsoever caused arising directly or indirectly in connection with, in relation to orarising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantialor systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, ordistribution in any form to anyone is expressly forbidden. Terms & Conditions of access and usecan be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

Processing of compound-word characters in readingChinese: An eye-movement-contingent display

change study

Lei Cui1,2, Guoli Yan1, Xuejun Bai1, Jukka Hyönä3, Suiping Wang4, andSimon P. Liversedge5

1Academy of Psychology and Behavior, Tianjin Normal University, Tianjin, P.R. China2Department of Education, Jining University, Qufu, P.R. China3Department of Psychology, University of Turku, Turku, Finland4Department of Psychology, South China Normal University, Guangzhou, P.R. China5School of Psychology, University of Southampton, Southampton, UK

Readers’ eyemovementsweremonitored as they readChinese two-constituent compoundwords in sentencecontexts. The first compound-word constituent was either an infrequent character with a highly predictablesecond constituent or a frequent character with an unpredictable second constituent. The parafoveal previewof the second constituent was manipulated, with four preview conditions: identical to the correct form; asemantically related character to the second constituent; a semantically unrelated character to the second con-stituent; and a pseudocharacter. An invisible boundary was set between the two constituents; when the eyesmoved across the boundary, the previewed character was changed to its intended form. The main findingswere that preview effects occurred for the second constituent of the compoundword. Providing an incorrectpreview of the second constituent affected fixations on the first constituent, but only when the second con-stituentwas predictable from thefirst.The frequencyof the initial character of the compound constrained theidentity of the secondcharacter, and this in turnmodulated the extent towhich the semantic characteristics ofthe preview influenced processing of the second constituent and the compoundword as a whole. The resultsare considered in relation to current accounts of Chinese compound-word recognition and the constrainthypothesis of Hyönä, Bertram, and Pollatsek (2004). We conclude that word identification in Chinese isflexible, and parafoveal processing of upcoming characters is influenced both by the characteristics of thefixated character and by its relationship with the characters in the parafovea.

Keywords: Reading; Eye movements; Chinese; Compound words; Display change.

In Chinese script, a word can be composed of one toseveral characters. Since there are no explicitmarkers between words in written Chinese (thespace between characters and words is identical), it

is unclear how the constituent characters of a mul-tiple-character word are processed. Zhu (2005)reported the frequency of words with one, two,and more than two characters among 10 million

Correspondence should be addressed to Jukka Hyönä, Department of Psychology, University of Turku, FI-20014 Turku, Finland.

E-mail: [email protected]

# 2013 The Experimental Psychology Society 527

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013

Vol. 66, No. 3, 527–547, http://dx.doi.org/10.1080/17470218.2012.667423

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

word tokens. Within the corpus, 53.8% of wordswere composed of one character, 42.2% werecomposed of two characters, and only 4% werecomposed of more than two characters. Thus,approximately 93% of multiple-character words inChinese are composed of two characters, andalmost all of the two-character words are compoundwords. In order to understand word recognition inChinese reading, it is of importance to gain insightinto how two-character compound words are pro-cessed. The present study makes an effort in thatdirection.

A reasonable number of studies exist on readingmorphologically complex compound words inalphabetic scripts. Many of these studies haveemployed the registration of readers’ eye fixationpatterns as the measure of online word processing(for a review, see Pollatsek & Hyönä, 2006). Thecritical question addressed in these studies iswhether individual compound-word constituents(breast and bone in breastbone) are identified separ-ately one at a time or whether they are processedtogether as one recognition unit. Adopting thelogic introduced by Taft and Forster (1976), thefrequency of first and second constituents as wellas whole-word frequency has been manipulatedfor two-constituent compound words. Accordingto the logic, if constituent frequency exerts aneffect on compound-word processing, then indi-vidual constituents play a significant role therein.On the other hand, if only whole-word frequencyshows an effect, identification is taken to be holisticin nature.

Compound-word processing in alphabeticscripts

The previous eye movement studies showed thatboth first- and second-constituent frequency exertan effect on the fixation time spent on the com-pound word during reading. Studies on long (onaverage about 12–13 letters) compound words inFinnish (Hyönä, Bertram, & Pollatsek, 2004;Hyönä & Pollatsek, 1998; Pollatsek & Hyönä,2005; Pollatsek, Hyönä, & Bertram, 2000) estab-lished an early effect of first-constituent frequency,followed by a later effect of second-constituent and

whole-word frequency. These data suggest that theidentification of long compound words is initiatedby the access of the first constituent, followed bythe access of the second constituent and that ofthe whole word. On the other hand, Bertram andHyönä (2003) reported data suggesting thatshorter (7–9 letters) Finnish compound words areidentified holistically. They found that whole-word frequency but not first-constituent frequencyexerts a significant effect on fixation times on shortcompounds (however, see Juhasz, 2008).

The privileged role of the first constituent has notemerged as clearly in English studies. Andrews,Miller, and Rayner (2004) and Juhasz, Starr,Inhoff, and Placke (2003) only obtained a margin-ally significant first-constituent effect and amargin-ally significant second-constituent effect. Theweaker effects in English may be due to thesestudies employing somewhat shorter compoundwords than the Finnish studies. The compoundwords employed by Andrews et al. contained 6–11characters, and those studied by Juhasz et al. con-tained 9 characters. Bertram and Hyönä (2003)argue that visual acuity constraints modulate theextent to which compounds are processed seriallyor holistically. If the entire word falls within fovealvision, it may be identified holistically, whereas con-stituent-based processing is required when a part ofthe word is outside foveal reach.

A study of Kuperman, Schreuder, Bertram, andBaayen (2010) on the processing of Dutch com-pound words employed a regression methodinstead of an experimental design involving theorthogonal manipulation of variables. Theyobserved reliable constituent frequency effects (forboth left and right constituents) combined withearly whole-word frequency effects as well aseffects of the morphological family size of the leftand right constituents. As will become clear below,these effects are relevant to the present study.

A final study that is of interest comes fromJuhasz, Pollatsek, Hyönä, Drieghe, and Rayner(2009; see also Häikiö, Bertram, & Hyönä,2010). Juhasz et al. demonstrated that Englishreaders attend to the second constituent of a com-pound to a greater extent when it is a part ofspatially unified compound (basketball) than when

528 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

the two constituents are separated by a space (tennisball). The issue of spacing in relation to compoundprocessing is potentially important in Chinese sinceit is an unspaced language. What emerges from thisdiscussion is a quite consistent picture for theidentification of long compounds words: The leftconstituent is processed prior to the right constitu-ent, and the compound word becomes activated atthe same time with the right constituent or evenearlier (see Kuperman et al., 2010). We now turnour attention to how compound-word processingmight take place during Chinese reading.

Compound-word processing in Chinese

Evidence from the lexical decision paradigmsuggests that constituent morphemes play a signifi-cant role in Chinese compound-word recognition(e.g., Zhou, Marslen-Wilson, Taft, & Shu, 1999).Zhou et al. have argued that “the processing of theinitial morphemes activates the semantic represen-tations of whole words. This semantic activationserves as contextual constraint in interpretingthe ambiguous second character” (p. 561). Thus,according to Zhou et al., compound-word proces-sing in Chinese appears to bear similarity to suchprocessing in alphabetic scripts. On the otherhand, it may differ between Chinese and alphabeticscripts in that no whole-word orthographic–phono-logical representations are assumed to exist forChinese compound words (Zhou et al., 1999).Thus, we were keen to know more about whetherChinese compound words are processed in asimilar manner to compound words in alphabeticscripts.

Chinese is a logographic script. Equally spacedbox-like symbols, characters, represent the basicunits of meaning (morphemes); characters mayalso represent tonal characteristics relevant to pro-nunciation. The Chinese script is visually compactin that adjacent characters appear close to eachotherwith no extra spacing betweenwords. The per-ceptual span extends one character to the left andtwo or three characters to the right of fixation (atpresent Chinese is typically written from left toright, Inhoff & Liu, 1998). Inhoff and Wu (2005)demonstrated that characters within the perceptual

span can be combined to form different words.Studies show that reading of Chinese appears tobe both word based and character based. G. L.Yan, Tian, Bai, and Rayner (2006) found thatreading time was longer for low-frequency wordsthan for high-frequency words. Moreover, aneffect of first-character frequency in two-characterwords is obtained when the word is infrequent butnot when it is frequent. Bai, Yan, Liversedge,Zang, and Rayner (2008) found that the insertionof spaces between characters slows down reading,which is not the case when illegal spaces are addedat word boundaries. Tsai and McConkie (2003)provided evidence suggesting that words notextending more than two characters to the right ofthat currently fixated may serve as recognitionunits. Also, Yang, Staub, Li, Wang, and Rayner(2010) found, with respect to the process of incre-mental semantic interpretation of a two-characterword, that the word, not the character, was the func-tionally relevant processing unit.

Given these studies, an interesting questionconcerns how Chinese readers process characterpairs that form a compound word. This is particu-larly the case given the findings of Juhasz et al.(2009) and that Chinese is an unspaced language.We investigated this question using the eye-contin-gent display change boundary paradigm (Rayner,1975). In this technique, an invisible boundary ispositioned immediately prior to a target character.When the reader makes a saccade such that thepoint of fixation crosses the boundary, the characterimmediately to the right of the boundary may bechanged. In our experiment, the boundary waspositioned between the first and second charactersof a two-character compound word, and wemanipulated the relationship between the parafo-veal preview and the target character. We usedtwo sets of two-character Chinese compoundwords: In one set, the first character was infrequent,and in the other set, it was frequent. When infre-quent, the second character was predictable fromthe first; when frequent, the second character wasnot predictable from the first. We selected ourstimuli in this way because there were insufficientwords with low-frequency initial characters thatprovided low constraint over the second character

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 529

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

and insufficient words with high-frequency initialcharacters that provided high constraint over thesecond character. Thus, it was not possible toemploy a fully crossed design in respect of thetwo variables.

We hypothesized that if the identity of the initialcharacter of the compound target did constrain theidentity of the second character, then this mightaffect the time course of processing of the secondconstituent (see Hyönä et al., 2004; Kupermanet al., 2010). If this were the case, then we mightexpect that providing a preview that was a differentcharacter to a predictable second character wouldlead to greater disruption to processing than provid-ing a preview that was a different character to anunpredictable second character. To assess any suchinfluence, we examined the size of the previewbenefit effect. We were also very keen to investigatethe extent to which Chinese readers process thecharacters of the compound word sequentially, orin parallel. If readers did process the second charac-ter of the compound whilst still fixating the firstcharacter of the compound, then we should obtainparafoveal-on-foveal effects (Kennedy, 2000). Inorder to examine the type of information thatreaders obtained from the second character whilstfixating the first character, we compared four typesof second-character preview: (a) an identicalpreview (i.e., no display change), (b) a semanticallyrelated preview, (c) a semantically unrelatedpreview, and (d) a pseudocharacter preview. IfChinese readers acquire semantic informationabout the second character while fixating on thefirst character, and the meaning of the second char-acter is accessed in parallel with the first one, fixationtime on the initial character would be shorter whenthe preview was semantically related than when itwas unrelated to the second constituent (a semanticparafoveal-on-foveal effect). An analogous effectobserved during processing of the second characterwould index a semantic preview effect: Previewinga semantically related character would lead to abenefit over a preview of an unrelated characterwhen the second character was foveally processed.On the other hand, if character preview effects aredriven by the lexical status or the orthographic fam-iliarity of a character, pseudocharacter previews

should produce longer fixation times than real char-acter previews (i.e., readers may be able to differen-tiate parafoveally between a pseudocharacter and areal character) on the first character, or on thesecond character. We refer to such effects on thefirst character as lexical parafoveal-on-foveal effectsand those on the second character as lexicalpreview effects. Finally, if only the identical previewsdiffer from all other previews, parafoveal processingmay be considered to be orthographic in nature, asthe nonidentical previews share the feature ofbeing orthographically distinct from the identicalpreview.

Parafoveal text processing in reading

When reading alphabetic scripts, readers acquirevisual–orthographic information, such as wordlength and letter identity information (for reviews,see Hyönä, 2011; Rayner, 1998, 2009), for theword(s) to the right of the fixated word. However,evidence for parafoveal lexical–semantic processingis hard to find across two adjacent words separatedby a space (for a review, see Rayner, White,Kambe, Miller, & Liversedge, 2003; however, seeHohenstein, Laubrock, & Kliegl, 2010). On theother hand, White, Bertram, and Hyönä (2008)obtained a semantic preview effect in readingFinnish compound words (no spaces appearbetween constituents). Using the display changeparadigm, they found that replacing the second con-stituent with a semantically unrelated lexeme pro-duced longer regression path durations than whenthe second constituent was replaced with a semanti-cally related lexeme. Regression path time is alsosometimes referred to as “go-past reading time”and can be taken to reflect an effect with a slightlylater time course, particularly when no differencesare obtained on earlier measures such as first-fix-ation duration or gaze duration on a word.

A boundary study conducted in Chinese(M. Yan, Richter, Shu, & Kliegl, 2009) hasprovided evidence for semantic processing ofparafoveally presented pictographic and indicativecharacters. These characters represent meaning asrelatively analogical, picture-like representations.The effect was apparent both as a parafoveal

530 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

preview effect and as a parafoveal-on-foveal effect.On the other hand, Wang, Tong, Yang, andLeng (2009) obtained only a weak, nonsignificantparafoveal-on-foveal effect of contextual predict-ability in Chinese. Yang, Wang, Xu, and Rayner(2009) found a lexical parafoveal-on-foveal effectin one-character words when there was a previewwith another one-character word (Experiment 1).However, in their second experiment, the effectwas only obtained when character N – 1 was a com-ponent character of a two-character word (for a par-afoveal lexicality effect, see also Yen, Tsai, Tzeng,& Hung, 2008). More recently, Yang, Wang,Tong, and Rayner (in press) reported two relevantexperiments. In Experiment 1, they manipulatedthe semantic relatedness of the parafovealpreview, but found no evidence for a semantic par-afoveal-on-foveal effect, nor for a semantic previeweffect. In a second experiment, Yang et al. exam-ined joint effects of semantic relatedness andplausibility of the parafoveal previews. This exper-iment yielded a preview effect for plausibility andsome evidence for a semantic relatedness previeweffect (the related and plausible condition differedfrom the unrelated and plausible in first-fixationduration but not in gaze duration). In sum, itappears that parafoveal lexical–semantic effectsmay be somewhat easier to establish in Chinesethan in alphabetic scripts.

The present study is similar to that of Yang et al.(in press) in that we also manipulated the semanticrelatedness of parafoveal previews. However, itdiffers from it in that we manipulated local lexicalpredictability (the second character is constrainedto a varying degree by the first character of two-character compound words), whereas the predict-ability manipulation of Yang et al. was moreglobal in nature (i.e., created by the sentenceframe leading to the target word).

Method

ParticipantsSixty-eight undergraduates from Tianjin NormalUniversity (native Chinese speakers) with normalor corrected-to-normal vision participated in theexperiment. Four subjects were discarded (two

because their comprehension accuracy was below75%, and two because more than 25% of thedisplay changes occurred during a fixation—suchchanges occurred due to either the change beingtriggered by a microsaccade during a fixation veryclose to the boundary, or instances where the eyescrossed the boundary within the last few millise-conds of the saccade, leaving too little time tomake the change during the saccade). After theexperiment, participants were asked whether theyexperienced anything unusual during reading.Two subjects reported seeing something flicker onthe screen on only one or two trials. No participantwas able to report exactly what it was that they hadseen. None of these participants took part in therating studies reported below.

ApparatusEye movement data were collected using anEyelink2000 (SR Research, Canada). The datawere sampled at the rate of 1,000 Hz (0.01°spatial resolution, 0.25°–0.5° accuracy). Althoughviewing was binocular, the movement of the righteye was monitored by the eye-tracker. A 21-inchDell Trinitron monitor was used to display thestimuli. The resolution of the monitor was1,024× 768 pixels. The monitor’s refresh ratewas 150 Hz, and it took approximately 13 ms tocomplete a display change.

MaterialsA set of 64 low-frequency (high-constraint) first-constituent compounds was paired with a set of 64high-frequency (low-constraint) first-constituentcompounds (see Table 1 for their lexical properties).All compounds were two-noun, semantically trans-parent compoundwords.Other than their frequencycharacteristics, the compound words did notdiffer across frequency conditions. Each compoundpair contained the same second constituent, and themeaning of thewholewordwas related to each of theconstituents (and to this extent, they were semanti-cally transparent).

There was an invisible boundary between thefirst and second constituents. Prior to the eyescrossing this boundary, the second constituentwas (a) identical to the correct form, (b) a

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 531

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

semantically related character to the second con-stituent, (c) a semantically unrelated character tothe second constituent, or (d) a pseudocharacter.The first constituent and the second-character pre-views never produced a real compound, except inthe identical condition. The previews did notshare the same radical, nor did they share aradical with the first constituent. The previewswere matched on the number of strokes (F, 1)and character frequency, F(2, 189)= 1.31,p= .27 (see Table 2). During the saccade crossingthe invisible boundary, the preview was changedto its intended form. As vision is suppressedduring saccades, readers do not see the actualdisplay change taking place.

The pseudocharacters were picked from theTrue Font collection. They very closely resembledreal characters but were completely meaningless,as they comprised inappropriate radical combi-nations (though the number of radicals wasmatched with the target as closely as possible).Furthermore, the pseudocharacter previews did

not contain any of the radicals (semantic or pho-netic) of the target character.

Each compound with an infrequent first con-stituent was paired with one that had a frequentinitial constituent, and a sentence frame wascreated for each pair that was at least identical upthrough the word following the target word. Anexample sentence pair is given in Table 3.

The target sentences occupied a maximum oftwo text lines; the target word always appeared onthe first line but was never the initial or the finalword of the line. Thirty-two filler sentences werealso included; they contained no display changes.Eight lists of 96 sentences were constructed; fourparticipants were randomly allocated to each list.Each list contained an equal number of itemsfrom each preview condition. The items were coun-terbalanced using a Latin square design such thatthe participants saw each target compound onlyonce. After every three sentences, a comprehensionquestion was asked about one of the three sen-tences. The participants answered the questionsby pressing a Yes or No key on a gamepad.

The text was presented in simple Song font inblack on a white background. Each character wasabout 2.4× 2.4 cm in size; with a viewing distanceof 71 cm, each character subtended approximately1.93°. The characters were separated from eachother by a space equalling 0.3°. Given the largefont size, the second character often appeared inthe parafovea while fixating the first character,although this was not always the case. For conven-ience, we refer to the second character as the paraf-oveal character.

Rating of materialsSecond-constituent predictability. A list of first con-stituents were given on a sheet of paper, and 30

Table 2. Lexical–statistical properties of the four preview conditions for the target compound words

Lexical property Identical Related Unrelated Pseudo character

Number of strokes 7.33 (2.94) 7.88 (3.29) 7.44 (2.67) 7.61 (2.27)

Character frequency 451 (662) 331 (437) 326 (422) —

Note: Character frequency is measured as a character-per-million figure using the Modern Chinese Character Frequency Dictionary

(1992). Standard deviations in parentheses.

Table 1. Lexical–statistical properties for the compound words

with a low-frequency versus a high-frequency first constituent

Lexical

property

Low-frequency first

constituent

High-frequency first

constituent

Character frequency 6.41 (3.21) 661 (683)

Word frequency 0.31 (0.38) 0.23 (0.39)

Number of radicals 2.75 (0.74) 2.75 (0.78)

Number of strokes 11.23 (2.82) 10.69 (2.31)

Note: Character frequency is measured as a characters-per-

million figure using the Modern Chinese Character

Frequency Dictionary (1992), and word frequency is

measured as a words-per-million figure using the Modern

Chinese Common Word Frequency Dictionary (1990).

Standard deviations in parentheses.

532 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

students were asked to add to the first constituent asecond one that first came to mind, in order tomake a compound word. Those second constitu-ents were selected for the experiment when at

least 13 participants came up with the intendedcharacter for the low-frequency first constituents(M= 20.05, SD= 7.69); for the high-frequencyfirst constituents, participants suggested a highly

Table 3. An example sentence pair for the different preview conditions

The English translation of the identical condition is: “We all know that the sternum is an important part of the human body.”

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 533

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

variable set of second constituents (M= 0.5, SD=1.21), with no more than 3 participants making thesame suggestion (and this only occurred for oneitem).

Semantic relatedness between the second constituentsand their previews. Another 30 participants wereasked to rate on a 5-point scale the semantic relat-edness between the second constituent and itssemantically related and unrelated preview. Therelatedness was significantly greater, t(63)=21.56, p, .001, for the semantically related(M= 3.95, SD= 0.69) than for the semanticallyunrelated condition (M= 1.64, SD= 0.52).

Semantic relatedness between the first constituents andthe previews. We also investigated how semanticallyrelated the first constituents were to their previews.Another 48 participants used a 5-point scale to ratethe semantic relatedness of the first constituent(high or low frequency) and the semanticallyrelated and unrelated preview (64 word pairs).Fourteen filler character pairs with a similarformat and a strong semantic relation were alsoincluded. Four lists of 78 character pairs were con-structed. Each list contained an equal number ofitems from each condition. The items were coun-terbalanced using a Latin square design such thatthe participants saw each character pair only once.Participants were randomly allocated to each list.We found a significant main effect of preview char-acter, F(1, 63)= 42.11, MSE= 0.56, p, .001.Semantic relatedness was higher for the relatedthan for the unrelated preview. There was no sig-nificant main effect of first-constituent frequency,nor frequency by preview interaction, Fs, 1.Although semantic relatedness was higher for therelated preview than for the unrelated preview,this effect was similar for both high- and low-fre-quency first constituent (see Table 4).

Semantic relatedness between the whole compoundword and the previews. Forty-eight different partici-pants used a 5-point scale to rate how semanticallyrelated the compound word was to the semanticallyrelated and unrelated preview. Again, 14 similarfiller word pairs with a strong semantic relation

were also included. The design of the rating assess-ment was identical to that used previously. Therewas significant main effect of preview, F(1, 63)=90.61,MSE= 0.72, p, .001; semantic relatednesswas higher for the related preview than for the unre-lated preview. As before, there was no effect of first-constituent frequency, nor was there a frequency bypreview interaction (Fs, 1, see Table 4).

Plausibility rating. A final plausibility pretest wasconducted to guarantee that the target wordsfitted well in the sentences. Twenty-eight studentswere asked to rate the target sentences for theirplausibility, using a 5-point scale (1= very implau-sible, 5= very plausible). No difference, t, 1, wasobserved between the high- (M= 3.64, SD= 0.44)and low-frequency (M= 3.66, SD= 0.42) first-constituent condition.

Results and discussion

Trials in which the display change occurred duringa fixation were excluded, and so were trials wherethe display change was completed after a fixationhad already started on the second character.Fixations less than 60 ms or greater than 800 mswere also excluded. For the 64 participants includedin the analyses, 13.6% of the data was excluded,including track losses. The experimental designwas a 2 (frequency of first constituent: high vs.low)× 4 (type of second constituent preview: iden-tical, semantically related, semantically unrelated,pseudocharacter) within-participant design.

Table 4. Semantic relatedness between the first constituent or

compound word and semantically related and unrelated preview for

the low-frequency and high-frequency first-constituent conditions

Constituent/word Preview Low frequency High frequency

First constituent Related 2.29 2.40

Unrelated 1.76 1.72

Compound word Related 2.60 2.59

Unrelated 1.51 1.65

Note:A 5-point scale was used to rate relatedness: 1= unrelated;

5= related.

534 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

Prior to reporting effects of the variables that wemanipulated, we first provide some descriptive datathat indicate the frequency with which differentpatterns of inspection occurred for the targetword. We identified five categories of inspection(see Figure 1).

Our analyses showed that for approximately half ofthe trials, readers fixated each of the two characters ofthe compound sequentially, triggering the boundarychange immediately after fixating the first characterof the compound and before fixating the second.Thus, on the majority of occasions, participants readthe target word by fixating each character in turn.However, this sequential pattern of fixation acrosscharacters does not always occur to this extent (e.g.,Yang et al., 2009), and it is likely that it occurredhere because theChinese characters used in this exper-iment were quite large. Two other types of inspectionpattern accounted for the vast majority of the remain-ing trials andoccurred approximately equally often.Onabout a fifth of the trials, readers fixated the first char-acter of the compound and skipped the second, and ona further fifth of trials they launched from a characterbefore the target word, skipping the first characterand fixating the second. Readers then made a right-ward saccade to fixate upcoming text. In the followingsection, we present analyses of the data separately forthe first and second constituents, followed by eye fix-ation measures indexing the processing of the wholecompound word.

The eye movement measures for the first con-stituent index compound-word processing thattakes place before the display change (i.e., whenthe different second-character previews were stillpresent), while the eye fixation measures for thesecond constituent reflect processing that takesplace after the preview has changed to its intendedform. Thus, the data for the first constituent

potentially index parafoveal-on-foveal effects, whilethe data for the second constituent index somewhatlater processing.1 The measures for the whole com-pound word are composite measures reflecting bothearly and later effects. Note that for all measures wereport separate analyses for the high- and low-con-straint conditions. This was done to directly investi-gate whether the nature of the first character of thecompound target constrained the identity of thesecond and impacted on processing.

Eye fixation measures for the first constituentFixation durations on the first constituent (seeTable 5) reflect the foveal processing of the initialconstituent; moreover, they also reflect possibleparafoveal-on-foveal effects during compound-word processing.

First-fixation duration. For the first-fixation dur-ation, there were significant main effects of preview[F1(3, 183)= 3.49, MSE= 1,009, p, .05; F2(3,186)= 3.94, MSE= 981, p, .01] and first-con-stituent frequency [F1(1, 61)= 5.58,MSE= 1,373,p, .05; F2(1, 62)= 4.65, MSE= 923, p, .05]and a significant Frequency× Preview interaction[F1(3, 183)= 6.21, MSE= 1,087, p, .001; F2(3,186)= 8.09, MSE= 953, p, .001]. The previeweffect reflects the fact that identical previews wereassociated with shorter first fixations (215 ms) thanwere nonidentical previews (an average of 224 ms).The main effect of first-constituent frequency wasdue to high-frequency first constituents producingshorter first fixations (218 ms) than low-frequencyfirst constituents (226 ms).

For the low-frequency first-constituent com-pounds, there was a significant main effect ofpreview: F1(3, 189)= 5.52, MSE= 1,224,p, .01; F2(3, 189)= 9.97, MSE= 937, p, .001.

1 In addition to these analyses, we considered fixations on the first character of the target compound word contingent on whether

they were part of Inspection Pattern 1 or Inspection Pattern 2, as well as fixations on the second character of the target word contingent

on whether they were part of Inspection Pattern 1 or Inspection Pattern 3 (see Figure 1). We did this to determine whether the par-

ticular way in which the target word was processed across fixations influenced the nature of the effects that occurred. In both sets of

analyses, the effects were either very similar for fixations from both patterns of inspection, or were null effects. None of the analyses of

the different data sets produced patterns of effects that were suggestive of anything that the overall analyses did not also show. We also

examined fixations prior to the saccade launched onto the target region contingent on inspection pattern, and again effects were similar

(though the power of these analyses was reduced relative to the overall analyses). Given the similarity of effects across analyses, we

simply report detailed effects based on the overall data sets in the Results section.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 535

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

First fixation was shorter in the identical previewthan in the other preview conditions (ps, .01).As the nonidentical previews share the feature of

being orthographically distinct from the identicalpreview, this effect may be considered a parafovealorthographic preview effect. Given the fact that all

Figure 1. The five predominant patterns of fixation on the target word that occurred in the experiment. The sequential pattern, whereby each

character was inspected in turn from left to right, probably occurred more predominantly than is often the case because the characters in the present

experiment were comparatively large.

536 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

target items in the nonidentical conditions werenonwords, the effect may also be interpreted as aparafoveal lexical effect. However, as there was nodifference in the pseudocharacter and the real-char-acter preview conditions, this interpretation maynot be fully warranted. There were no reliableeffects for the high-frequency condition (Fs, 1).

Gaze duration. For gaze duration, the main effectsof preview [F1(3, 183)= 10.72, MSE= 1,683,p, .001; F2(3, 186)= 11.83, MSE= 1,829,p, .001] and first-constituent frequency [F1(1,61)= 25.76, MSE= 2,037, p, .001; F2(1, 62)=26.13, MSE= 2,151, p, .001] were modified bya Frequency× Preview interaction [F1(3, 183)=3.94, MSE= 2,197, p, .01; F2(3, 186)= 5.54,MSE= 1,949, p, .001]. Identical previews wereassociated with shorter gaze durations (236 ms)than nonidentical previews (259 ms). The maineffect of first-constituent frequency in turn reflectsthe fact that gaze duration was longer when thefirst constituent was infrequent thanwhen it was fre-quent (263 ms vs. 243 ms).

For the low-frequency first-constituent com-pounds, there was significant main effect of preview[F1(3, 189)= 9.84, MSE= 2,207, p, .001; F2(3,189)= 13.32, MSE= 2,173, p, .001]. Gaze dur-ation for the identical previews was shorter than

that for the other previews (ps, .001). As arguedabove, the effect indexes a parafoveal orthographic(or possibly lexical) effect. There were no reliableeffects for the high-frequency condition.

Probability of skipping. The probability of skippingover the first constituent showed no significanteffects (Fs, 2.98).

To sum up the results for the first character, weobtained reliable first-constituent frequency andpreview main effects both in first-fixation durationand in gaze duration. However, our primary focusin this experiment was to assess how the frequencyof the initial constituent modulated parafoveal pro-cessing of the second constituent of the compoundtarget. Thus, our main finding was that a reliablepreview effect (the nonidentical conditions produ-cing longer first fixation and gaze durations thanthe identical condition) was observed in the low-fre-quency first-constituent condition but not in thehigh-frequency first-constituent condition. Thispattern of results is assumed to result from differ-ences in lexical predictability: Given a low-fre-quency first character, the identity of the secondcharacter is more constrained than is the case witha high-frequency character. Consequently, whenthe previewed character did not accordwith the con-straint, a parafoveal-on-foveal effect ensued.2

Table 5. Eye fixation measures for the first compound-word constituent, as a function of first-character frequency and preview type

Measure

Identical Related Unrelated Pseudocharacter

M SD M SD M SD M SD

Low frequency

FFD 209 45 233 40 231 44 230 45

GD 234 51 270 60 277 59 272 66

Skip .23 .19 .24 .23 .22 .23 .24 .23

High frequency

FFD 220 39 209 39 222 42 221 41

GD 238 48 237 53 249 51 246 53

Skip .28 .23 .25 .23 .25 .24 .23 .23

Note: Eye fixation measures in ms. FFD= first-fixation duration; GD= gaze duration; Skip= probability of skipping.

2 Curiously, the effect emerged in the nonidentical preview conditions but not in the identical preview condition. One possible

explanation is that the lexical status of the entire compound word (a nonword in the nonidentical preview conditions) induced difficulty

on the processing of the first character exacerbated in the low-frequency condition. We are grateful to SarahWhite for this suggestion.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 537

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

Eye fixation measures for the second constituentThe eye fixation measures for the second constitu-ent index the processing of the second constituentafter the parafoveal preview has been changed toits intended form. Thus, these data reflect parafo-veal preview effects. The data are presented inTable 6.

First-fixation duration. The duration of first fix-ation revealed a significant main effect of preview[F1(3, 189)= 41.18, MSE= 1,916, p, .001;F2(3, 186)= 42.67, MSE= 1,814, p, .001] anda marginal main effect of first-constituent fre-quency [F1(1, 63)= 3.50, MSE= 1,527, p= .07;F2(1, 62)= 3.02, MSE= 2,028, p= .09], but noFrequency× Preview interaction (Fs, 1). Thepreview effect reflects the finding that the identicalpreview produced shorter first fixations (210 ms)than the related (254 ms), unrelated (257 ms),and pseudocharacter conditions (265 ms). The fre-quency effect reflects the tendency for the first fix-ation being shorter in the low-frequency (243 ms)than in the high-frequency (249 ms) condition,suggesting that the low-frequency first constituentsconstrained the second constituents more stronglythan the high-frequency first constituents (for asimilar finding in Finnish, see Hyönä et al.,2004). Despite the lack of a reliable interaction(here, and below), we conducted analyses to evalu-ate effects for the high- and low-frequency first-constituent compounds separately. This allowed

us to explore whether patterns of effects were con-sistent with our a priori expectations.

For the low-frequency first-constituentcompounds, there was a significant main effectof preview: F1(3, 189)= 23.61, MSE= 1,872,p, .001; F2(3, 189)= 26.21, MSE= 1,736,p, .001. First fixations on the second characterwere shorter for the identical previews than forthe other previews (ps, .001); moreover, semanti-cally related previews produced shorter first fix-ations than pseudocharacter previews (p, .05).This is an interesting finding that is suggestive ofthe possibility that the semantic characteristics ofthe parafoveal preview character can modulate theduration of the first fixation on the second characterwhen the first character constrains the identity ofthe second character. Such an effect constitutes asemantic parafoveal influence. Note, however,that this finding should be treated cautiously, asthe overall frequency by preview interaction wasnot reliable, and the semantically related previewcondition did not differ from the unrelatedpreview condition.

For the high-frequency first-constituent com-pound words, there was also a significant maineffect of preview: F1(3, 189)= 26.14, MSE=1,359, p, .001; F2(3, 186)= 20.56, MSE=1,685, p , .001. Identical previews producedshorter first fixations than the other previews(ps, .001), once again reflecting a parafovealorthographic effect.

Table 6. Eye fixation measures for the second compound-word constituent, as a function of first-character frequency and preview type

Measure

Identical Related Unrelated Pseudocharacter

M SD M SD M SD M SD

Low frequency

FFD 205 43 248 53 255 55 264 46

GD 222 51 277 54 302 70 304 60

Skip .37 .18 .25 .17 .20 .15 .11 .12

High frequency

FFD 214 37 260 47 258 47 265 52

GD 239 54 301 56 302 69 313 65

Skip .30 .19 .22 .18 .23 .17 .17 .15

Note: Eye fixation measures in ms. FFD= first-fixation duration. GD= gaze duration. Skip= probability of skipping.

538 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

Gaze duration. For gaze duration, there were signifi-cant main effects of preview [F1(3, 189)= 51.66,MSE= 3,147, p, .001; F2(3, 186)= 48.72,MSE= 3,375, p, .001] and first-constituent fre-quency [F1(1, 63)= 7.50, MSE= 2,789, p, .01;F2(1, 62)= 7.30, MSE= 4,046, p, .01], but noFrequency× Preview interaction. The previeweffect reflects the fact that identical previews pro-duced shorter (231 ms) gaze durations than related(289 ms), unrelated (302 ms), or pseudocharacter(309 ms) previews. The frequency effect occurreddue to shorter gaze durations after the low- (276ms) than after the high-frequency (289 ms) preced-ing character, reflecting a difference in predictabilityof the second constituent given the first. Again, weexplored the data for the high- and low-frequencyfirst-constituent compounds separately to assessour a priori expectations.

For the low-frequency first-constituent com-pounds, there was a significant main effectof preview: F1(3, 189)= 34.90, MSE= 2,700,p, .001; F2(3, 189)= 33.89, MSE= 2,833,p, .001. Gaze duration for identical previewswas shorter than that for the other previews(ps, .001), again, reflecting a parafovealorthographic effect. Interestingly, semanticallyrelated previews produced shorter gaze durationsthan semantically unrelated and pseudocharacterpreviews (p, .05). One might cautiously interpretthis effect as indicative of a parafoveal semanticinfluence, though note again that the overall inter-action was not reliable.

For the high-frequency first-constituent com-pounds, there was a significant main effect ofpreview: F1(3, 189)= 26.88, MSE= 2,666,p, .001; F2(3, 186)= 23.45, MSE= 3,160, p ,.001. Gaze duration for identical previews wasshorter than that for the other previews (ps, .001),indexing a parafoveal orthographic effect, thoughagain, this interpretation should be regarded withcaution.

Probability of skipping. The probability of skippingover the second constituent produced a significantmain effect of preview [F1(3, 189)= 40.10,MSE= .02, p, .001; F2(3, 189)= 23.99,MSE= .03, p, .001] and a Frequency× Preview

interaction [F1(3, 189)= 4.37, MSE= .02, p, .01;F2(3, 189)= 5.68, MSE= .02, p, .001], but nomain effect of first-constituent frequency, Fs, 1.The preview effect reflects the fact that the skippingprobability was highest for the identical previews(.34) and lowest for the pseudocharacter previews(.11), while the related and unrelated preview con-ditions fell between them (.24 and .22, respectively).

For the low-frequency first-constituentcompounds, there was a significant main effectof preview: F1(3, 189)= 34.86, MSE= .02,p, .001; F2(3, 189)= 26.52, MSE= .03,p, .001. The skipping rate for identical previewswas higher than that for the other previews(ps, .001); semantically related and unrelated pre-views were associated with a higher skipping ratethan pseudocharacter previews (ps, .001), pre-sumably reflecting a parafoveal effect of the lexicalstatus or orthographic familiarity of the preview.The semantically related previews were skippednumerically more frequently than semanticallyunrelated previews, but this effect was not reliable(p, .1).

For the high-frequency first-constituent com-pounds, there was also a significant main effect ofpreview: F1(3, 189)= 6.92, MSE= .02, p, .001;F2(3, 189)= 6.47, MSE= .02, p , .001. Theskipping rate for identical previews was higherthan that for the other previews (ps, .05); seman-tically related and unrelated previews were associ-ated with more skipping than pseudocharacterpreviews (ps, .05), reflecting a parafoveal effectof lexical status, or orthographic familiarity.

In summary, the eye fixation measures on thesecond character of the target are something of amixture. Therewas a frequency effect forfirstfixationand gaze duration on the second character. Thiseffect derived from the frequency characteristics ofthe first character; low-frequency first constituentsconstrained the second constituent more stronglythan the high-frequency first constituents, such thatprocessing time on the second character wasreduced when the first character constrained theidentity of the second. There were consistent androbust preview effects such that identity previewswere processed more quickly than the other con-ditions reflecting a parafoveal orthographic effect.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 539

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

This is a standard finding in parafoveal previewexperiments.

There were also effects of whether the previewwas a pseudocharacter compared to a real characterfor the first fixations and gaze duration on thesecond character. However, the pairwise compari-sons were not backed up by reliable interactionsin the omnibus analyses. We have interpreted thiseffect as a parafoveal lexical effect indicating thatthe lexical status or the orthographic familiarity ofthe parafoveal character influenced processing ofthe second character when it was directly fixated.Finally, for the fixation measures, while therewere numerical differences suggesting that first-constituent frequency may have modulated influ-ences of the semantic relationship between thepreview and second character of the target word,these effects were not statistically robust, and there-fore caution is required in forming interpretationsof them.

Turning to the skippingdata,we found, consistentwith thefixation timemeasures, that the second char-acter of the compound was skippedmore often whenit was a character thanwhen it was a pseudocharacter.Again, the lexical status or orthographic familiarity ofthe preview influenced the degree to which it wasskipped. This effect was stronger for the low-fre-quency than for the high-frequency condition.However, there was no reliable modulatory influenceof the semantic relationship between the preview andthefirst character of the compound target on skipping

of the second character. Thus, the analyses of thesecond constituent indicate standard identitypreview effects along with effects of the lexicalstatus or orthographic familiarity of the preview.

Eye fixation measures for the whole compound wordFor the whole compound word, we report the ana-lyses for the skipping and gaze duration data. Themeans and standard deviation are presented inTable 7, which also includes the total fixationtimes. However, as total fixation time is not veryinformative regarding preview effects, no statisticsare reported for that measure.

Probability of skipping. The probability of skippingover the whole compound word showed no signifi-cant effects (all Fs, 2.46).

Gaze duration. For the gaze duration measure, allfixations made on the compound word before a fix-ation elsewhere in the sentence were summed.Thus, unlike the gaze duration measures reportedabove, these measures were computed for thewhole compound word rather than the individualconstituent characters. There was no main effectof first-constituent frequency (Fs, 1), but therewas a significant main effect of preview [F1(3,189)= 75.39, MSE= 9,951, p, .001; F2(3,189)= 88.83, MSE= 8,364, p, .001] and aFrequency× Preview interaction that approachedsignificance [F1(3, 189)= 2.21, MSE= 7,572,

Table 7. Eye fixation measures for the whole compound word, as a function of first-character frequency and preview type

Measure

Identical Related Unrelated Pseudocharacter

M SD M SD M SD M SD

Low frequency

GD 330 75 452 135 500 146 530 152

Total 542 175 709 210 786 212 754 203

Skip .004 .024 .008 .038 .006 .028 .006 .027

High frequency

GD 363 95 454 152 488 159 511 151

Total 593 171 752 223 777 242 771 204

Skip .013 .039 .006 .036 .012 .044 .011 .037

Note: Eye fixation measures in ms. GD= gaze duration. Total= total fixation time. Skip= probability of skipping.

540 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

p= .09; F2(3, 189)= 2.07, MSE= 8,613, p=.10]. The preview effect was such that the identicalpreview produced clearly the shortest gaze duration(347 ms), and the related condition was associatedwith the second shortest gaze duration (453 ms),followed by the unrelated (494 ms) and the pseudo-character condition (521 ms).

For the low-frequency first-constituentcompounds, there was a significant main effectof preview: F1(3, 189)= 56.52, MSE= 8,766,p, .001; F2(3, 189)= 59.45, MSE= 8,328,p, .001. Gaze duration for the identical previewcondition was shorter than that for the otherpreview conditions (ps, .001); the semanticallyrelated preview produced shorter gaze durationsthan the semantically unrelated and the pseudo-character previews (ps, .01); the semanticallyunrelated previews produced shorter gaze durationsthan the pseudocharacter previews (p, .05). Thus,when the first character of the compound target waslow frequency, there was evidence that the parafo-veal preview had an influence in terms of both itslexical status and its semantic relationship to theconstituents and the whole compound word.Note that the gaze duration measure for thewhole word is composed of fixations that reflectprocessing of the compound target that occurredboth before and after the boundary change. Thatis to say, during initial fixations on the first charac-ter, the first character and the preview are visuallyavailable to the reader for visual processing. In con-trast, during fixations on the second character, thewhole compound target word (and therefore, notthe preview) is available to be processed. Thus,the measure reflects both parafoveal-on-foveal andstandard previews effects. It may be of someconcern that the visual (and, consequently, linguis-tic) information that is available to be processed bythe reader differs for these different types of fixationand that these different fixations are being summedwithin the same measure. However, it is importantto note that the preview character or pseudocharac-ter was never available for direct fixation during anyof these fixations. Thus, any influences of thepreview character that we might observe in thesemeasures do reflect effects that derive from parafo-veal processing. Once again, however, we note that

caution must be applied given that the overall inter-action did not attain significance.

For the high-frequency first-constituent com-pounds, there was also a significant main effect ofpreview: F1(3, 189)= 30.99, MSE= 8,757,p, .001; F2(3, 189)= 30.72, MSE= 8,649,p, .001. Gaze durations for the identical previewswere shorter than those for the other previews(ps, .001); moreover, the semantically relatedpreviews produced shorter gaze durations than thepseudocharacter previews (p, .01). It is unclearwhy semantically unrelated previews did notproduce a similar effect.

In sum, gaze duration on the whole compoundword demonstrated both a parafoveal lexical and asemantic preview effect for the low-frequency first-constituent condition. That is, when the secondcharacter was constrained by the first character,and the preview was related to the intended secondcharacter, gaze duration was shorter than in thecase where the previewed character was semanticallyunrelated to the intended character. Moreover, gazeduration was shorter when the previewed characterwas an existing character than when it was a pseudo-character, suggesting a parafoveal lexical effect.However, the results were less clear in the low-con-straint (i.e., high-frequency first-constituent) con-dition, where the pseudocharacter conditiondiffered reliably from the semantically related butnot from the unrelated condition.

GENERAL DISCUSSION

Parafoveal processing of the second character oftwo-character compound words was examined inreading Chinese. We obtained a number of inter-esting effects, and we deal with each of these inturn below.

First, we obtained a very reliable preview effectfor the eye fixation measures. Identity previewsalways produced shorter fixation times and moreskipping of the second character of the targetword than nonidentity previews. This is a standardpreview benefit effect (Rayner, Well, Pollatsek, &Bertera, 1982). It indicates that the boundary para-digm was working appropriately and that readers

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 541

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

obtained maximum preview benefit when the par-afoveal preview character was identical with thesecond character upon fixation.

A second aspect of the results that is noteworthyconcerns the frequency effects that occurred. Weobtained a reliable frequency effect on the firstcharacter of the target word. When it was high fre-quency, fixation times were shorter than when itwas low frequency. Again, this is an unsurprisingresult, and it is consistent with the findings ofG. L. Yan et al. (2006), who showed that the fre-quency of a character within a Chinese compoundword determines the ease with which it is processed(and particularly so for word initial characters).This is an important result in the present study,because the frequency of the initial constituentalso indexed the frequency with which it appearedwith other, different, second characters to form atwo-character compound word. Thus, the reliablefrequency effects that we obtained for the first char-acter indicate that the degree of constraint that itexerted over the identity of the second characterwas differential across conditions. In other words,this effect demonstrates that our manipulation ofconstraint was very effective.

Consistent with this suggestion was the comp-lementary finding that initial fixations and gazeduration on the second character of the targetword were shorter when the first character waslow frequency than when it was high frequency.The most obvious explanation of this effect isthat the frequency of the first character of thetarget did indeed differentially constrain the iden-tity of the second character such that when thefirst character could potentially combine withmany second characters to form words, then fix-ations on the second character were longer. In con-trast, when the first character could potentiallycombine with comparatively few other characters,then processing time on the second character wasreduced. Thus, this effect provides further evidencethat the manipulation of constraint over the iden-tity of the second character was effective andreflects a higher degree of holistic processing ofcompound words when the second character is con-strained by the first character due to its frequency.That is to say, this finding does support the

suggestion that when the first character of a com-pound word constrains the identity of the secondthrough being low frequency, then processing ofthe second character occurs, to some extent atleast, during the fixation on the first character.The processing does not necessarily have to resultin full identification of the second character, butinstead in a reduction in the candidate set generatedfor the second character when fixating the firstcharacter. This would lead to facilitation in proces-sing the second character once it was fixated. Thefinding is generally consistent with the morpho-logical family size effects observed by Kupermanet al. (2010) in Dutch. They found that with alarge morphological family for the left compound-word constituent, the effect of the right-constituentfrequency was greater than that for a small family.When the family is large, the right constituent isnot predictable from the left, and more processingeffort is needed to identify it.

In addition to the frequency effects, we obtainedan effect of the lexical status of the preview on pro-cessing of the target word in relation to skipping ofthe second character of the target. When thepreview was a pseudocharacter, the second con-stituent was skipped less often than when it was areal character (regardless of whether it was relatedor unrelated). This effect of the lexical status ofthe preview may have occurred for several reasons.It could be due to readers being more familiarwith the orthography of a real character than apseudocharacter. If so, then this indicates readers’sensitivity to the orthographic characteristics ofChinese characters when they are available in theparafovea. An alternative possibility is that readersmay be sensitive to the lexical status of the charactereven prior to its direct fixation. If this were the case,then it would suggest that readers were able todetermine that the character either was or was notrepresented in their mental lexicon prior to itbeing fixated (i.e., a lexical parafoveal-on-fovealeffect). It is also possible that prior to skippingthe second character it is not fully identified, butrather a significant reduction has emerged in thecandidate set for possible second characters. Thisin turn may lead to a realization that identificationis imminent, which may have triggered a saccade

542 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

skipping over the second character. The presentdata do not allow us to discriminate betweenthese three possible explanations.

The data on the probability of skipping over thesecond character also revealed a Preview Type×First-Constituent Frequency interaction. Thenature of the interaction is twofold. First, in theidentical preview condition, the high-constraintcondition (low-frequency first character) producesmore skipping than the low-constraint condition.This is a “standard” predictability effect:Predictable lexical items are skipped more oftenthan less predictable items (e.g., Rayner, Li,Juhasz, &Yan, 2005). Second, in the pseudocharac-ter condition, there is an opposite pattern,suggesting that readers are less willing to skip overa pseudocharacter in the high-constraint than inthe low-constraint condition. This may be takento suggest that readers are more knowledgeable ofthe orthographic features of the second characterin the high-constraint condition (we return to thisissue below).

Another important aspect of our results thatrequires discussion concerns our findings in relationto whether the semantic characteristics of thepreview character influenced processing on thetarget word. The effects that we obtained in relationto this question are far less clear than those discussedalready. It is certainly the case that there werenumerical differences suggestive of reduced proces-sing difficulty on the target word when the previewwas a character that was semantically related to thesecond character of the target word than when itwas a semantically unrelated character. We obtainedpatterns like this when the first constituent was lowcompared with high frequency for gaze duration onthe second character, gaze duration on the wholeword, and total fixation time on the whole word.Note, however, that we examined these effects forthe low- and high-frequency initial characters ofthe target words on a priori theoretical grounds,even though the more global interaction did notlicense this through significance. Thus, there islimited evidence that information about the seman-tic characteristics of a character were available whenit was present in the parafovea (i.e., adjacent to thefixated character).

What is also very clear from the findings,however, is that these effects were comparativelylate. They did not occur for fixations on the firstcharacter of the target word. They also did notoccur on the initial fixation on the second characterof the target word after the boundary change. Asnoted already, the earliest effects occurred on gazeduration for the second constituent and for theeven later measures of gaze duration and totaltime for the whole target word. Thus, eventhough these effects clearly derived from thenature of the second constituent whilst in the par-afovea, they appeared comparatively late duringprocessing of the target word. These results arecomplementary to those from a similar study ofWhite et al. (2008) conducted with Finnish com-pound words.

There are other points to make in relation to thiseffect. The rating studies that we conducted priorto the eye movement experiment indicated thatfor the semantically related condition, the previewcharacter was semantically related to the characterthat replaced it, but also to both the first characterof the compound and the compound word as awhole. Thus, to the extent that such effectsoccurred, it is not clear whether they arose due tothe relationship of the preview character with thefirst character, or the second character, or indeedthe whole compound word. However, what isclearly the case is that those effects that did occurcertainly derived from the preview character whenit was in the parafovea, and, therefore, theseeffects reflect a parafoveal semantic influence onprocessing of the target word. Note, however,that these effects held only when the first characterof the target word was low frequency. No similareffects occurred when the initial character washigh frequency. Thus, it appears that any parafovealpreview effects that do occur only do so under highconstraint conditions—that is, when the first char-acter of a two character compound indicates thelikely identity of the second character.

The finding that a parafoveal semantic effect wasonly obtained when the second character wasstrongly constrained by the first character is consist-ent with the constraint hypothesis of Hyönä et al.(2004) proposed for compound-word processing

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 543

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

in Finnish. This notion predicts that attentionshifts earlier to the second constituent when it ispredictable from the first, thus giving rise togreater parafoveal preview effects. The parafoveal-on-foveal effect and the semantic effects observedin the present study are consistent with the con-straint hypothesis. They are also generally consist-ent with the results of Kuperman et al. (2010)reported for the processing of Dutch compoundwords.

Before discussing the implications of the resultsfor understanding how compound words are pro-cessed in Chinese, one caveat may be warranted.In order to increase the probability that thesecond character was located in the reader’s paraf-ovea when fixating the first character, we used arelatively large font size (one character extendednearly two degrees of visual angle). We opted fora relatively large font in order to make the exper-iment more comparable to studies conducted inalphabetic scripts, where (at least in case of longcompound words), most of the letters comprisingthe second constituent lie in the parafovea.This was likely to increase the probability ofreading the compound words with two fixations,compared to previous studies using smaller fontsize.

The effects that we obtained have consequencesfor modelling of compound-word processing inChinese. According to Zhou et al.’s (1999) modelof Chinese compound-word processing, access tothe first compound-word character activates notonly its meaning but also the meaning of a com-pound word it is a part of. This type of semanticactivation in turn serves as contextual constraintin interpreting the second character. According tothe model, compound-word meaning may be acti-vated on the basis of the first character. Our pre-screen experiments provided evidence that therewas a strong semantic relationship between the par-afoveal preview and each of the target word con-stituent characters as well as the target word as awhole. Thus, the preview could have provided asemantic cue from the parafovea as to the identityof the second character and the target word as awhole. According to an alternative view, themeaning of both compound-word characters must

be accessed before access to the compound-wordmeaning is achieved.

A compromise between these two views may beoffered on the basis of the fact that meanings ofChinese characters are not fixed but shaped bythe meanings of the other characters theycombine with. Thus, it is conceivable that a paraf-oveal preview that is semantically related to theintended character, which in turn is highly con-strained by the first character, gives rise to semanticactivation that is sufficiently close to the intendedcompound-word meaning. This in turn maysignal to the recognition system that compound-word recognition is imminent, which then resultsin facilitation to processing of the target word.Inherent in this view is the idea that meaning acti-vation in Chinese compound-word processing isnot an all-or-nothing phenomenon. This in turnseems reasonable given that the meanings ofChinese compound words vary greatly in semantictransparency.

Given the evidence that 2–3-character wordsmay serve as recognition units (Tsai &McConkie, 2003), it is surprising to find no paraf-oveal-on-foveal effect in the low-constraint con-dition. This finding suggests that many two-character compound words are processed seriallyin Chinese: When visually attending to the firstcharacter, the second character is not simul-taneously processed. Even when the second charac-ter is highly constrained by the first character, it canbe argued that parallel processing of the two char-acters was found to be only orthographic innature. Given the nature of our constraint manipu-lation (the processing of the second character iscontingent on that of the first character), it maybe argued that what appears parallel processing infact reflects serial processing. According to thisaccount,2 the first character of a compound isalways processed first; if there are few possible com-pletions of the word, attention quickly turns to thesecond character to establish whether it fits with theexpectations provided by the first character. Theattention shift triggers a saccade, and only lowerlevel nonlexical processing is carried out of thesecond character prior to fixating it. This accountis generally consistent with the theorizing of

544 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

Zhou et al. (1999), who assume that Chinesereaders do not possess orthographic–phonologicalrepresentations for compound words.

The above results accord with those of Yanget al. (2009) by demonstrating that even in a visu-ally compact script, parafoveal-on-foveal effects arenot a standard finding. On the other hand, the lackof a semantic parafoveal-on-foveal effect is incon-sistent with the results of M. Yan et al. (2009),who did observe such an effect. It is noteworthy,however, that their effect was obtained with picto-graphical characters. Thus, as regards parallelsemantic processing of two adjacent characters,the rare pictographical characters might form aspecial case.

Studies on compound-word processing in alpha-betic scripts have not established reliable parafoveal-on-foveal effects (only suggestive numerical trends)of any kind (Hyönä et al., 2004; Juhasz et al., 2009;Pollatsek & Hyönä, 2005; White et al., 2008).However, a recent study of Häikiö et al. (2010)did observe such an effect for the processing ofletter identity information. Interestingly, the effectonly appeared for highly frequent compounds, forwhich the second constituent is predictable fromthe first. Moreover, Häikiö et al. reanalysed thedata of White et al. (2008), by computing the selec-tive regression path duration for thefirst constituent.This reading time measure includes both first-passfixations and any second-pass fixations that werepreceded by a regression from the first constituentback to a preceding word. These analyses revealeda reliable orthographic parafoveal-on-foveal effect.Note that the materials in the White et al. studyonly contained compoundwords whose second con-stituents were highly constrained by the first con-stituents. Thus, the results of these two studies arein agreement those of the present study, whichpoint to similarities in compound-word processingbetween Finnish and Chinese.

A final aspect of the data that deserves commentconcerns the word skipping that occurred in thisexperiment. In the present study, the probability ofskipping a character varied from 11% to 37%,depending on the type of preview. These estimatesdo not differ much from what was observed byChen, Lau, and Wong (1998), who obtained a

skipping rate of 42% in Chinese sentence reading,or from the study of M. Yan et al. (2009) whofound the one-character skipping rate to be approxi-mately 30%. However, our skipping rates are muchlower than those observed byYang et al. (2009), whofound the skipping rate for one-character words tobe as high as 50%. These differences may haveoccurred due to two reasons. First, the size of thecharacters in visual angle may affect the skippingrate. In support of this claim, consider threestudies. Yang et al. (2009) used characters thatwere about 1.1× 1.1 cm in size and subtendedapproximately 0.9 degrees of visual angle.However, in the present study, the visual angle ofeach character was much larger. Each characterwas about 2.4× 2.4 cm in size, and with a viewingdistance of 71 cm, each character subtendedapproximately 1.93°. Similarly, in the study ofM. Yan et al. (2009), the character size was 1.5° ofvisual angle, and for this size of character, the skip-ping rate wasmore comparable to that of the presentstudy. Note also that their skipping rates were muchlower than those of Yang et al. (2009). Second,differences in skipping rates may also have occurreddue to character frequency. The frequency of targetcharacters was much higher in the study of Yanget al. (2009) than in the present study.

CONCLUSIONS

The main findings from the current study are thatpreview effects occurred for the second constituentof a compound word in Chinese. Providing anincorrect preview of the second constituent affectedfixations on the first constituent, but only when thesecond constituent was predictable from the first;this preview effect was orthographic (or possiblylexical) in nature. We also found that the frequencyof the initial character of the compound con-strained the identity of the second character, andthis in turn modulated the extent to which lexicaland semantic characteristics of the preview influ-enced processing of the second constituent andthe compound word as a whole. Our primary con-clusion is that during Chinese reading, parafovealprocessing of upcoming characters occurs to

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 545

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

differing degrees contingent on the lexical charac-teristics of the fixated character (e.g., its frequency)and its relationship to the upcoming character orcharacters in the sentence. In future studies, itmay be worthwhile to study in more detail whatother textual (and reader) characteristics modulatethe nature of, or at least, the degree to which paraf-oveal characters are processed when they are in theparafovea, rather than continue arguing whether allword processing in reading is serial or parallel. Forexample, it remains to be seen whether effects suchas those we report are limited to compound words,or whether they extend similarly across two adja-cent characters that do not form a word. Asnoted in the Introduction there is evidence(Inhoff & Wu, 2005) that Chinese characterswithin the perceptual span are free to combinewith each other to form words. Thus, this kindmodulation may be more characteristic in Chinesethan in alphabetic scripts, where even a stronglexical constraint across two words cannotproduce a parafoveal-on-foveal effect (see theFinnish study of Vainio, Hyönä, & Pajunen, 2009).

Original manuscript received 26 November 2010

Accepted revision received 12 January 2012

First published online 26 April 2012

REFERENCES

Andrews, S., Miller, B., & Rayner, K. (2004). Eye move-ments and morphological segmentation of compoundwords: There is a mouse in mousetrap. European

Journal of Cognitive Psychology, 16, 285–311.Bai, X. J., Yan, G. L., Liversedge, S. P., Zang, C. L., &

Rayner, K. (2008). Reading spaced and unspacedChinese text: Evidence from eye movements.Journal of Experimental Psychology: Human Perception

and Performance, 34, 1277–1287.Bertram, R., & Hyönä, J. (2003). The length of a

complex word modifies the role of morphologicalstructure: Evidence from reading short and longFinnish compounds. Journal of Memory &

Language, 48, 615–634.Chen, H.-C., Lau, W. Y.L., & Wong, K.F.E. (1998,

December). Reading Chinese: Implications from charac-

ter reading times and eye movements, Paper presented

at the First International Workshop of WrittenLanguage Processing, Sydney, Australia.

Häikiö, T., Bertram, R., & Hyönä, J. (2010).Development of parafoveal processing within andacross words in reading: Evidence from the boundaryparadigm. Quarterly Journal of Experimental

Psychology, 63, 1982–1998.Hohenstein, S., Laubrock, J., & Kliegl, R. (2010).

Semantic preview benefit during eye movements inreading: A parafoveal fast-priming study. Journal ofExperimental Psychology: Learning, Memory &

Cognition, 36, 1150–1170.Hyönä, J. (2011). Foveal and parafoveal processing

during reading. In S. P. Liversedge, I. Gilchrist, &S. Everling (Eds.), Oxford handbook of eye movements

(pp. 819–838). Oxford, UK: Oxford University Press.Hyönä, J., Bertram, R., & Pollatsek, A. (2004). Are long

compoundwords identified serially via their constituents?Evidence from an eye-movement-contingent displaychange study.Memory & Cognition, 32, 523–532.

Hyönä, J., & Pollatsek, A. (1998). Reading Finnish com-pound words: Eye fixations are affected by componentmorphemes. Journal of Experimental Psychology:

Human Perception and Performance, 24, 1612–1627.Inhoff, A.W.,&Liu,W. (1998). The perceptual span and

oculomotor activity during the reading of Chinese sen-tences. Journal of Experimental Psychology: Human

Perception and Performance, 24, 20–34.Inhoff, A. W., & Wu, C. (2005). Eye movements and

the identification of spatially ambiguous wordsduring Chinese sentence reading. Memory &

Cognition, 33, 1345–1356.Juhasz, B. J. (2008). The processing of compound words

in English: Effects of word length on eye movementsduring reading. Language and Cognitive Processes, 23,1057–1088.

Juhasz, B. J., Pollatsek, A., Hyönä, J., Drieghe, D., &Rayner, K. (2009). Parafoveal processing within andbetween words. Quarterly Journal of Experimental

Psychology, 62, 1356–1376.Juhasz, B. J., Starr, M. S., Inhoff, A. W., & Placke, L.

(2003). The effects of morphology on the processingof compound words: Evidence from naming, lexicaldecisions and eye fixations. British Journal of

Psychology, 94, 223–244.Kennedy, A. (2000). Parafoveal processing in word rec-

ognition. Quarterly Journal of Experimental

Psychology, 53, 429–455.Kuperman, V., Schreuder, R., Bertram, R., & Baayen, R.

H. (2010). Reading polymorphemic Dutch com-pounds: Toward a multiple route model of lexical

546 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3)

CUI ET AL.

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4

processing. Journal of Experimental Psychology: Human

Perception & Performance, 35, 876–895.Modern Chinese Character Frequency Dictionary. (1992).

Beijing, China: Chinese Press.Modern Chinese Common Word Frequency Dictionary.

(1990). Beijing, China: Yuhang Press.Pollatsek, A., & Hyönä, J. (2005). The role of semantic

transparency in the processing of Finnish compoundwords. Language and Cognitive Processes, 20, 261–290.

Pollatsek, A., & Hyönä, J. (2006). Processing of mor-phologically complex words in context: What can belearned from eye movements. In S. Andrews (Ed.),From inkmarks to ideas: Current issues in lexical proces-

sing (pp. 275–298). Hove, UK: Psychology Press.Pollatsek, A., Hyönä, J., & Bertram, R. (2000). The role of

morphological constituents in reading Finnish com-pound words. Journal of Experimental Psychology:

Human Perception & Performance, 26, 820–833.Rayner, K. (1975). The perceptual span and peripheral

cues during reading. Cognitive Psychology, 7, 65–81.Rayner, K. (1998). Eye movements in reading and infor-

mation processing: 20 years of research. PsychologicalBulletin, 124(3), 372–422.

Rayner, K. (2009). The thirty-fifth Sir Frederick Bartlettlecture: Eye movements and attention during reading,scene perception, and visual search. Quarterly Journalof Experimental Psychology, 62, 1457–1506.

Rayner, K., Li, X., Juhasz, B. J., & Yan, G. L. (2005).The effect of word predictability on the eye move-ments of Chinese readers. Psychonomic Bulletin and

Review, 12, 1089–1093.Rayner, K., Well, A. D, Pollatsek, A., & Bertera, J. H.

(1982). The availability of useful information to theright of fixation in reading. Perception &

Psychophysics, 31, 537–550.Rayner, K., White, S. J., Kambe, G., Miller, B., &

Liversedge, S. P. (2003). On the processing ofmeaning from parafoveal vision during eye fixationsin reading. In J. Hyönä, R. Radach, & H. Deubel(Eds.), The mind’s eye: Cognitive and applied aspects

of eye movement research (pp. 213–234). Amsterdam,The Netherlands: North-Holland Elsevier SciencePublishers.

Taft,M.,&Forster,K. I. (1976).Lexical storage and retrie-val of polymorphemic and polysyllabic words. Journal ofVerbal Learning and Verbal Behavior, 15, 607–620.

Tsai, J. L., & McConkie, G. W. (2003). Where doChinese readers send their eyes? In J. Hyönä,R. Radach, & H. Deubel (Eds.), The mind’s eyes:

Cognitive and applied aspects of eye movements

(pp. 159–176). Amsterdam, The Netherlands:North-Holland Elsevier Science Publishers.

Vainio, S., Hyönä, J., & Pajunen, A. (2009).Lexical predictability exerts robust effects on fix-ation duration, but not non initial landing dur-ation during reading. Experimental Psychology,56, 66–74.

Wang, S., Tong, X., Yang, J.,&Leng, Y. (2009). Semanticcodes are obtained before word fixation inChinese sen-tence reading: Evidence from eye-movements. ActaPsychologica Sinica, 41, 220–232.

White, S. J., Bertram, R., & Hyönä, J. (2008). Semanticprocessing of previews within compound words.Journal of Experimental Psychology: Learning,

Memory & Cognition, 34, 988–993.Yan, G. L., Tian, H., Bai, X. J., & Rayner, K. (2006).

The effect of word and character frequency on theeye movements of Chinese readers. British Journal of

Psychology, 97, 259–268.Yan, M., Richter, E. M., Shu, H., & Kliegl, R. (2009).

Readers of Chinese extract semantic informationfrom parafoveal words. Psychonomic Bulletin &

Review, 16, 561–566.Yang, J., Staub, A., Li, N.,Wang, S., & Rayner, K. (2010,

May). Plausibility effects when reading one- and two-

character words in Chinese: Evidence from eye movements,Talk given at the 4th China International Conferenceon Eye Movements, Tianjin, China.

Yang, J., Wang, S., Tong, X., & Rayner, K. (in press).Semantic and plausibility effects on preview benefitduring eye fixations in Chinese reading. Reading

and Writing.Yang, J., Wang, S., Xu, Y., & Rayner, K. (2009). Do

Chinese readers obtain preview benefit from charactern+ 2? Evidence from eye movements. Journal of

Experimental Psychology: Human Perception and

Performance, 35, 1192–1204.Yen, M.-H., Tsai, J.-L., Tzeng, O. J. I., & Hung, D. L.

(2008). Eye movements and parafoveal word proces-sing in reading Chinese. Memory & Cognition, 36,1033–1045.

Zhou, X., Marslen-Wilson, W., Taft, M., & Shu, H.(1999). Morphology, orthography, and phonologyin reading Chinese. Language and Cognitive

Processes, 14, 525–565.Zhu, Q. M. (2005). Chinese information processing tech-

nology tuition, Beijing, P.R. China: QinghuaUniversity Press.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2013, 66 (3) 547

READING CHINESE COMPOUND WORDS

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

09:

52 3

0 M

arch

201

4