Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Memory & Cognition1998,26 (I), 75-87
Orthography and phonology in readingJapanese kanji words: Evidence from thesemantic decision task with homophones
NAOKOSAKUMATokyo Metropolitan Institute ofGerontology, Tokyo, Japan
SUMIKO SASANUMAInternational University ofHealth and Welfare, Ohtawara, Japan
ITARUF. TATSUMITokyo Metropolitan Institute ofGerontology, Tokyo, Japan
and
SHINOBUMASAKIThe ATR Human Information Processing Research Laboratories, Kyoto, Japan
Correspondences between spelling and sound for Japanese kanji are complex and deep. The meaning of kanji words has generally been assumed to be accessed directly from orthography withoutphonological mediation. Experiment 1,however, replicated the findings of VanOrden (1987) that subjects made more false-positive errors on homophone foils than they did on nonhomophone controls ina semantic decision task, although they did so only when the foils were orthographically similar to thecorrect exemplars, which indicates both orthographic and phonological activations of meaning. Experiment 2 showed the same results when subjects were not required to pronounce the target wordsafter semantic decisions, which indicates automatic phonological activation of kanji words. In Experiment 3, under pattern-masking conditions, this homophony effect was reduced but remained on errors,and the orthographic-similarity effect remained strong on both homophone and nonhomophone foils.These results suggest that both orthography and phonology play an important role in the comprehension of kanji words.
Orthographies differ in the degree ofcomplexity in therelationship between their print and their sound. In shallow orthographies, such as Serbo-Croatian and Japanesekana, the correspondences between print and sound aresimple and regular, at least at the segmental level. In deeporthographies, such as Hebrew and Japanese kanji, incontrast, these relationships are highly complex. Alphabetic English lies somewhere between these extremes(see, e.g., Besner & Hilderbrandt, 1987; Frost, Katz, &
This research was started as a study parallel to that of Wydell, Patterson, and Humphreys (1993) and was carried out independently oftheir study. A preliminary version of this article was reported in a paperat the meeting of the Japanese Psychological Association, Kyoto, 1992.This research was supported in part by the NTT Basic Laboratory research grant to S.S. We are grateful to Karalyn E. Patterson andTaeko N. Wydell for valuable discussion and advice for constructingearlier versions of this paper. We also wish to thank Geoffrey R. Loftusand two anonymous reviewers for helpful comments on the manuscript.Correspondence concerning this article should be addressed toN. Sakuma, Department of Language and Cognition, Tokyo Metropolitan Institute of Gerontology, 35-2 Sakae-cho, Itabashi-ku, Tokyo 173,Japan (e-mail: [email protected]).
-Accepted by previous editor, Geoffrey R. Loftus
75
Bentin, 1987; Patterson, 1990; Sasanuma, 1986, 1994; Seidenberg, 1985). An important question is whether variations in these correspondences between print and soundinfluence word processing in reading. In the present study,we consider the role of phonology in visual word recognition across different orthographies.
The role of phonology in visual word recognition hasbeen repeatedly discussed in reading research, primarilyon the basis of results obtained from experimental studies done with English words (see, e.g., Jared & Seidenberg, 1991; Van Orden, 1987). Early research on visualword recognition suggested that phonological representation is a primary source of access to the meaning ofwritten words (see, e.g., Rubenstein, Lewis, & Rubenstein,1971). This view has been termed phonologically mediated access. An alternative view, usually termed direct access, argued that the phonological code is not necessaryin skilled word recognition and that an orthographic representation activates its meaning directly (see, e.g.,Baron, 1973). A number ofcurrent models ofword recognition include both phonologically mediated access and direct access (see, e.g., Allport, 1977; M. Coltheart, Davelaar,Jonasson, & Besner, 1977; Monsell, Patterson, Graham,Hughes, & Milroy, 1992; Morton & Patterson, 1980; Sei-
Copyright 1998 Psychonomic Society, Inc.
76 SAKUMA, SASANUMA, TATSUMI, AND MASAKI
denberg & McClelland, 1989). The question of exactlyhow phonology is activated in reading words aloud hasbeen hotly debated (see, e.g., M. Coltheart, Curtis, Atkins,& Haller, 1993; Plaut, McClelland, Seidenberg, & Patterson, 1996; Van Orden, Pennington, & Stone, 1990),but the basic idea that both processes (or routes) to meaning-phonologically mediated access and direct access-operate to activate the meaning of written wordshas been widely accepted.
Van Orden and his colleagues (Van Orden, 1987; VanOrden, Johnston, & Hale, 1988; Van Orden et aI., 1990),however, argued against this view of two parallel formsof access to meaning and proposed the view that computed phonology based on print-and-sound correspondences exerts the major influence in the activation ofmeaning for written English words. Van Orden (1987)offered new evidence for the use of phonology in meaning activation that was based on a semantic-decision taskwith homophones. The subjects were presented with acategory name (e.g., a flower) followed by a target word(e.g., tulip, rows, or robs) and were asked to make a semantic decision to the target. VanOrden argued that ifthesound ofrows, which is a homophone ofrose (a true category exemplar), yielded activation of the meaning ofrose, then subjects should make more false-positive errorsto homophone foils such as rows than to nonhomophonecontrols such as robs. He examined this homophony effectfurther by manipulating orthographic similarity betweencategory exemplars and target foils. In his Experiment I,the subjects made more false-positive errors to homophone foils than to nonhomophone controls when the foilswere orthographically similar to category exemplars. Inhis Experiment 2, the homophony effect remained strongeven under pattern-masking conditions, although the orthographic-similarity effect disappeared. In his Experiment 3, he discovered an effect exerted by the frequencyof category exemplars (rose) but no systematic effect ofthefrequency of homophone target foils (rows) themselves.
These results were interpreted by Van Orden (1987)as being most consistent with the view that the primaryroute from orthography to meaning is mediated by phonology. Van Orden explained these findings by proposing a model based on the verification hypothesis ofvisual word recognition (see, e.g., Becker, 1976, 1980;Becker & Killion, 1977; Paap, Newsome, McDonald, &Schvaneveldt, 1982; Rubenstein et al., 1971). A visuallypresented word activates its phonological representation,which, in turn, activates a candidate set oflexical entries(meanings). The orthographic representation of the mostactive lexical entry (meaning) is then retrieved and compared with the orthographic representation of the stimulus target. This process is termed verification, which isessentially a spelling check. If a match occurs, the lexical entry (meaning) is selected; otherwise, the verification is performed on the next most active candidate untila match is found. In this model, a homophone target foil(rows) will activate its correct homophonic exemplar (rose)as a candidate; this availability of the correct exemplar in
the verification process causes more false-positive errorson homophone foils than on nonhomophone controls,particularly when the foil is orthographically similar to thecorrect exemplar.
Furthermore, Van Orden et al. (1988) obtained the positive findings for phonological mediation in their experiments using nonword homophone foils. Van Orden et al.(1990) developed a subsymbolic account based on twohypotheses-covariant learning and self-consistencyfor the mappings from orthographic codes to phonological codes and to other linguistic codes (see also Van Orden & Goldinger, 1994).
However, orthographies differ in the manner in whichthey represent phonology. Specifically, in Japanese kanji,individual characters are thought to represent words ormorphemes rather than phonological units, and the relationship between orthography and phonology is complex.It is open to question whether the phonological mediation hypothesis in word recognition could apply generally to writing systems other than English. In this paper,we investigate the role of phonology in Japanese kanjiword recognition. Before we address the theoretical issues involved in attempting Van Orden's (1987) experiments, a brief description of some features of Japanesekanji orthography is in order.
Japanese Kanji Orthography andPhonological Processes
There are three different nonalphabetic orthographieslogographic kanji and two types of syllabic kana (hiragana and katakana}-in written Japanese. Roughly speaking, lexical morphemes, such as nouns and the roots ofverbs and adjectives, are written in kanji (and, rarely, inhiragana), whereas grammatical morphemes and function words are written in hiragana, and loan noun wordsare written in katakana. Words in kanji, therefore, are themost popular forms for representing meaning in writtenJapanese. The number of kanji characters is quite large;one needs to know as many as 3,000 kanji characters toread newspapers and ordinary texts. Furthermore, manyof these characters are complex, as well as distinct fromone another, in visual configuration (see examples in Tables I and 2).
A single kanji character can often be a word, but themajority oflexical items are made up oftwo or more kanjicharacters (Morton & Sasanuma, 1984; Morton, Sasanuma, Patterson, & Sakuma, 1992).1 A single kanji character usually has two or more pronunciations (see, e.g.,kanji A, B, and C in Table 1),2 which can be categorized
Table 1Examples of Single Kanji Words and Two-Kanji-Character
Words Containing Those as Component Characters
Kanji A Kanji B Kanji C Word A Word B
Kanji word ~ll". ~ll. _faTranslation finger circle vehicle ring wheelPronunciation (KUN) Iyubil Iwal Ikurumal Iyubi-wal
(ON) Ishii Irinl Isyal Isya-rinl
ORTHOGRAPHY AND PHONOLOGY IN JAPANESE KANJI 77
into two types-KUN-readings and ON-readings-but amultikanji-character word has only one legitimate pronunciation (see, e.g., words A and B in Table 1). The KUNreading is usually used when a single kanji character occurs in isolation as a word and is also used for a small setof multikanji-character words. The ON-reading, on theother hand, is used for most multikanji-character wordsand, rarely, for a single-character word. Although there isa strong tendency to pair the same reading type (ON-ONor KUN-KUN, as in words A and B in Table I) rather thanto use the mixed-reading type (ON-KUN or KUN-ON)for the pronunciation of a two-kanji-character word, apparentely no rules exist for determining which reading type(ON or KUN) should be used (see, e.g., Morton & Sasanuma, 1984; Sasanuma, 1980, 1986). In addition, thereare many sets of homophonic single kanji characters aswell as homophonic multikanji-character words in Japanese (see Table 2).
The Role of Phonology inKanji Word Recognition
A number ofexperimental and clinical studies involving kanji word recognition have suggested that the meaning of kanji words can be directly accessed from orthographic representation (see, e.g., Goryo, 1987; Kimura,1984; Saito, 1981; Sasanuma, 1986). On the other hand,there have been few studies on the phonological processesofkanji word recognition, and they have provided no clearevidence either for determining whether the phonologyofkanji words could be activated directly by orthographywithout semantic mediation or for determining whetherphonology could contribute to the meaning activation ofkanji words (see note 3).
Recently, a neuropsychological study of Japanese patients with dementia of the Alzheimer's type suggestedthat the lexical phonology ofkanji words can be activateddirectly by orthography without semantic mediation(Sasanuma, Sakuma, & Kitano, 1992). The patients in thisstudy showed a near-normal ability to read kanji wordsaloud until they reached a very advanced stage ofthe disease process, but there was progressive deterioration ofcomprehension.
More recently, Wydell, Patterson, and Humphreys(1993) reported a study on Japanese kanji word recognition, using procedures similar to those used in Van Orden's (1987) English experiments, which suggested thatphonology may contribute to the meaning activation ofkanji words. The results oftheir no-masking experiment,
which were similar to those of Van Orden's English experiment' were that homophony affected semantic decisions for kanji words and that the effects were strongestwhen the homophone target words were visually similarto the correct exemplar ofthe category name. The resultsof their kanji masking experiment, however, differedfrom those of Van Orden's English masking experimentin that the effect ofvisual similarity was significant evenunder pattern-masking conditions, whereas there was noeffect oforthographic similarity in Van Orden's maskingexperiment. Moreover, the homophony effect in the kanjimasking experiment was marginal, whereas there was asignificant effect ofhomophony in Van Orden's maskingexperiment. Wydell et al. interpreted their results as positive evidence for an early activation ofphonology, as wellas oforthography, in accessing the meaning ofkanji words.
Unfortunately, however, there was a methodologicalproblem in Wydell et al.'s (1993) experiments. In somestimulus pairs, they allowed the category name and thetarget word to share identical kanji characters. Furthermore, they did not control the number ofthese occurrencesacross the experimental conditions; ofthe total of 16pairsof category names and targets, seven pairs of visuallysimilar homophones, three pairs of visually similar controls, and no pairs of visually dissimilar homophonesshared identical kanji characters. There is considerableevidence for orthographic priming effects in cases ofbriefly presented pairs of letter strings. Evett and Humphreys (1981), for example, found that identification ofthe target words under masking conditions was facilitatedmore when primes and targets contained a number ofcommon letters than when their letters differed. If oneassumes that orthographic identification of a visuallysimilar homophone foil might have had a greater likelihood of being facilitated by priming from the categoryname than did the orthographic identification of the otherfoils in the masking experiment ofWydell et al. (1993),then the activation of phonology should be facilitated inturn. Thus, it appears necessary to replicate the study usingstimulus sets in which no pairs ofcategory name and target contain an identical kanji character.
In the present study, we addressed the question of therole of phonology in reading comprehension of kanjiwords, using procedures similar to those used by Van Orden (1987) and Wydell et al. (1993). Specifically, we examined whether results similar to those found in Wydellet al.'s study would be obtained under a different set ofexperimental conditions (e.g., different stimulus sets, a
Table 2Examples of Homophonic Kanji Words [One Set of Single Kanji Words (/ki/)
and Another Set of Two-Kanji-Character Words (/ka-tei/)I
Single kanji word * ~ Ie JIllPronunciation /ki/ /ki/ /ki/ /ki/Translation tree spirit account period
Two-kanji-character word f&:iE ii:fiPronunciation /ka-tei I /ka-tei/Translation assumption home
.-/ki/opportunity
lA~
/ka-tei/process
78 SAKUMA, SASANUMA, TATSUMI, AND MASAKI
different subject group, and a more reliable apparatus).Inasmuch as no study other than that by Wydell et al. hasprovided evidence that the meaning ofkanji words can beactivated by the use of phonology, particularly undermasking conditions, it would be useful to replicate theirstudy using a broader range of kanji characters as stimuli with another group of subjects who are native speakers of Japanese.
EXPERIMENT 1
In this experiment, we examined the basic homophonyand orthographic similarity effects on kanji word recognition using the same procedures as Van Orden's (1987)study.
MethodSubjects. Twenty-four subjects (ages 21-35 years, 27.7 years
mean age) who were employees ofthe Tokyo Metropolitan Instituteof Gerontology participated. All were native speakers of Japaneseand had normal or corrected-to-normal vision.
Stimuli. The experiment involved 233 pairs of definitions (seenote 4) and target words, 50 for practice trials and 183 for the experimental trials. Experimental target words consisted of60 key targets and 123 filler targets. The key targets were 30 homophone foilsand 30 nonhomophone controls. Half of the homophone foils andhalf of the nonhomophone controls were orthographically similar,and the remaining half was orthographically dissimilar (see Appendix A).
The experimental key targets, two-kanji-character nouns, two tofour syllables in length, were selected on the basis ofthe followingprocedure. First, a preliminary study was made to select familiarkanji words for the stimulus pairs. One hundred students in a specialschool of nursing were shown a list of homophone candidate itemswritten in syllabic kana and were asked to transcribe them into asmany kanji forms as possible. For each item, the two most frequently transcribed kanji forms were selected as candidate homophone pairs, with the constraint that (I) the two kanji forms ofeachpair had the same accent pattern and that (2) over 50% of the subjects transcribed into that form. The 30 homophone pairs were chosen from these candidate pairs in such a way that they fell into twosets that were different in terms of orthographic similarity. One setwas composed of 15 orthographically similar pairs, with the constraint that each homophone pair shared an identical character (oridentical parts of a character) in the same position, either the firstor the second position in the two-kanji-character words. The otherset was composed of 15 orthographically dissimilar pairs, with theconstraints that the members of each homophone pair shared nocharacters and no component parts of a character and that the or-
thographic configurations of the whole words in each pair were asdifferent from each other as possible. Next, nonhomophone controlwords were selected to match the 30 homophone foils in orthographic similarity. Examples of these words and definitions areshown in Table 3.
A definition, written in kanji and lor kana, which ranged from 2to 10 characters in length, was created for each ofthe 30 exemplars.The main restriction observed in creating each definition was thatit not contain any kanji characters that were used for its correct exemplar, the corresponding homophone foil, or the control foil. Thisrestriction was necessary to avoid orthographic priming. The definitions were used twice--once for matched homophone foils andonce for nonhomophone controls.
In addition, 120 words and 90 definitions were chosen for fillertrials. Ofthe 120 filler words, 90 were exemplars oftheir categories(yes-fillers) and 30 were not exemplars (no-fillers). Of the 90 definitions, 30 appeared twice for both yes-fillers and no-fillers and 60appeared once for yes-fillers. Thus, the entire list (key trials plusfiller trials) had an equal number of yes and no trials. In addition,three new pairs of a target word and a definition were chosen forpractice trials presented at the start of the session.
Because the same 30 definitions were used twice, both for thematched homophone foils and for the control foils, the 183 pairs ofdefinitions and target words were compiled into two lists. In eachlist, halfofthe homophone foils appeared in the first half of the listand the matched control foils appeared in the last half of the list.The order ofpresentation within each list was pseudo-random, withthe two constraints that no more than two key trials appeared consecutively and that no more than five yes or five no trials appearedconsecutively.
An additional 25 definitions and 50 target words were chosen fora practice list. Half of the target words were exemplars of their definitions and the other half were not.
Apparatus. Stimuli were presented using three tachistoscopicshutters mounted on Kodak slide projectors with a rear projectionscreen. The shutters and projectors were interfaced with a computer(NEC PC-9800) that controlled the stimulus presentation order aswell as the exposure duration. The stimuli, phototyped in Japanesetextbook font, were presented in black against a white backgroundthrough a small 3.5 X 9 em window on the screen. The size ofeachcharacter was 6 X 6 mm on the screen. The longest stimulus (i.e.,a IO-character string) was approximately 0.6 X 6 em on the screen,producing a typical horizontal viewing angle ofabout 6° at the typical viewing distance of 57 cm.
Procedure. All subjects were tested individually. They were randomly assigned to one of the two experimental lists. The subjectsfirst saw the 50 practice trials and then the 183 experimental trials(the first 3 filler trials being for practice). The subjects saw each target word only once.
Each trial began with the presentation ofa warning signal by theexperimenter, followed immediately by a definition for 1,500 msec,which was displayed directly above a fixation point. The definition
Table 3Examples of Definitions, Correct Exemplars, Homophone Foils,
and Nonhomophone Control Foils
Definition Exemplar Homophone Control
OrthographicallySimilarExample iI~t.t ctn31W-Q z. t *. ~. ••Translation Burning of a building fire house work mealPronunciation Ikajil Ikajil Isyokuji/
OrthographicallyDissimilarExample 1Ill*H' -Q.A. i[!fJ' ;". _:ITTranslation A reporter journalist train electric lightPronunciation Ikisyal Ikisyal IdentouJ
ORTHOGRAPHY AND PHONOLOGY IN JAPANESE KANJI 79
Figure 1. The interaction between orthographic similarity andhomophony in mean percentage errors and in mean correct response times (RTs) in milliseconds from Experiment 1. The errorbars represent the 95% confidence intervals (Loftus & Masson,1994).
was then replaced by the target word for 500 msec, which was displayed directly below the fixation point. The subjects were instructedto respond to the presentation of the target word as quickly and asaccurately as possible by pressing either the yes key, if they thoughtthat the target word was an exemplar of the given category, or theno key, if they did not think so, and then to name the target word.The computer recorded response times (RTs) and response keys (yesor no). The experimenter recorded incorrect pronunciations of thetarget words.
DiscussionThe results ofthe error analysis ofthis experiment were
similar to those of Van Orden's (1987) English experiment, showing the impact ofboth orthographic similarityand homophony on semantic judgments. The homophonyeffect in our kanji experiment was only significant whenthe target foils were orthographically similar to their correct exemplars. The subjects were able to correctly rejectthe homophone foils when the foils were orthographicallydissimilar. Correct RT data also showed the strong effectof orthographic similarity on both homophone and nonhomophone foils and a reliable effect of homophony onorthographically similar foils. Although Van Orden didnot measure RTs in his Experiments I and 2, our subjectswere approximately 100 msec slower at rejecting the target foils when they were orthographically similar to thecorrect exemplar than when they were orthographicallydissimilar. RTs for homophone foils were slightly but significantly (approximately 30 msec) longer than RTs fornonhomophone controls. These findings suggest thatphonology contributes to the activation ofthe meaning ofkanji words. In addition, however, the greater effect oforthographic similarity, irrespective of homophony, onRTs suggests that orthographic processing still plays aprominent role in the semantic decisions with regard tokanji words.
These results can be interpreted as being consistent withboth the phonological-mediation view and the parallelaccess view. In the phonological-mediation view of VanOrden (1987), the meaning of written words is primarilyactivated via phonology and the orthographic-verificationprocess subsequently follows to uniquely identify a target word. In this model, a homophonic foil might activate
and a main effect of homophony [Fs(l,23) = 37.67,MSe = 0.59,p < .0001, and Fj(1,56) = 8.78, MSe = 4.02,p < .005]. The interaction of these two factors was alsosignificant [Fs(1,23) = 23.93,MSe = 0.63,p<.0001,andFj(1,56) = 5.99, MSe = 4.02,p < .02]. A significant effectofhomophony on errors was found only when the homophone foils were orthographically similar to the correctexemplars.
Correct RT data also showed the effects of orthographicsimilarity and homophony. There was a main effect oforthographic similarity [Fs(1,23) = 93.33, MSe = 3,019.0,p < .0001, and F j (l ,56) = 47.39, MSe = 4,196.1, p <.0001] and amain effect of homophony [Fs(1,23) = 10.32,MSe = 2,758.4, p < .004, and Fi(I,56) = 5.74, MSe =4,196.1, p < .02]. The interaction of these two factorswas not significant[Fs(I,23) = 0.96, MSe = 2,339.I,p >.3, and Fj(I,56) = 0.85, MSe = 4,196.I,p >.3]. CorrectRTs for homophone foils were longer than those for nonhomophone foils when they were orthographically similar [Fs(l,23) = 9.97, MSe = 23,320.1,p < .004, by subjects, and F, (1,56) = 5.50, MSe = 23,074.1, p < .02] forthe simple effect contrast, but not when they were orthographically dissimilar [Fs(1,23) = 3.15, MSe = 7,375.5,p = .089, and Fj(1,56) = 30.46,MSe = 4,563.3,p>.3].
950
t:=:J Error (Non-Homo.)-G-RT (Non-Homo.)
_ Error (Homo.)
_RT(Homo.)
40
~ 30900 en0
E-II) -... 850 ~0...... 20 a:w- 800 ~cQ) ...~
...0
Q) 10 750 oa..
0 700Similar Dissimilar
ResultsIn this and subsequent experiments, each subject's
RTs for the 180 experimental trials were normalized byexcluding RTs that were beyond 3 SD from his or hermean. The percentages ofoutliers were small-less than5% across the three experiments (2.36%, 1.67%, and4.24% for Experiments 1,2, and 3, respectively). We examined error rates (false-yes responses) and RTs (forcorrect no responses) for homophone foils and nonhomop hone controls in each condition of orthographicsimilarity. Mean error rates and mean RTs for both subjects (within-subjects factors) and items (between-subjectsfactors) were submitted to two-way analyses of variance(ANOVAs). The independent variables were orthographicsimilarity (similar or dissimilar) and homophony (homophone or nonhomophone control) (see note 5). In thisand all following experiments, subject means are reportedin the text and figures. The main results ofExperiment Iare presented in Figure I.
Error data clearly showed the effects of orthographicsimilarity and homophony. The subjects made 15.0% errors on orthographically similar homophones, but madefew errors on others (3.3%, 2.8%, and 1.7%). There wasamain effect ofsimilarity [Fs(1,23) = 36.39,MSe = 0.72,p < .0001, and FJI,56) = 10.37, MSe = 4.02,p < .002]
80 SAKUMA, SASANUMA, TATSUMI, AND MASAKI
its correct homophonic exemplar as a candidate; this tendsto increase the error rate on homophone target foils relative to nonhomophone controls. Moreover, it should bemore likely that a homophone foil would be misclassified as the correct exemplar if the foil was orthographically similar to the correct exemplar.To these assumptionsmade by Van Orden, we should probably add, in order toexplain the stronger effect of orthographic similarity onthe semantic decisions ofkanji words, the assumption thatthe orthographic-verification process for kanji wordsmight take longer for orthographically similar foils thanfor orthographically dissimilar foils, irrespective of phonological similarity.
On the other hand, in the parallel-access view, it is hypothesized that the meaning ofwritten words is activatedboth by orthography and by phonology. Once the subjects are presented with a definition, possible semanticcandidates may be activated, and these candidates maythemselves activate corresponding orthographic and corresponding phonological representations (see Jared &Seidenberg, 1991). Consequently, when the orthographically similar foil is presented, a partial match may takeplace in orthographic representations between the candidate words and the target foil, because the candidatewords may contain a kanji character identical to one of thetwo characters in the orthographically similar foil. Thisconflicting information-a partial match and the mismatchof the whole string-should slow down category judgments (Wydell et aI., 1993). In addition, a partial matchmay take place in semantic representations, becausewords that contain an identical kanji character are assumed to be related in the semantic network (see, e.g., Hirose, 1992; Sasanuma, 1986). Therefore, in order to reject the orthographically similar foil, the subjects willhave to carefully compare and check the meaning as wellas the orthography ofthe target foil with those of the corresponding exemplar. Furthermore, it should be moredifficult to reject a foil if it is homophonic to the exemplar, because the subjects may use the meaning activatedby the phonology of the target foil as a source for making a semantic decision. In sharp contrast, it should bequite easy for the subjects to reject orthographically dissimilar foils. Since the dissimilar foil, by definition, contains no kanji characters identical to either of the twokanji characters of the exemplar, the subjects can easilydetect that nothing overlaps between the orthographicrepresentation of the target foil and that of the exemplarand, therefore, can easily notice the difference in meaning between the two before the target foil is fully processed to activate the exact meaning.
The question remains, however, ofwhether the phonological activation of the meaning of kanji words mighthave been produced by a strategy under the control of thesubject. In Experiment 1, the subjects had to name the target words immediately after the semantic decision. Thismight have caused them to attempt to generate phonology and to use it strategically to activate meaning. Thereis a possibility, therefore, that different results may be
obtained under an experimental condition in which thesubjects are not required to name the target word aftermaking their semantic decision. Experiment 2 exploredthis possibility.
EXPERIMENT 2
In Experiment 2, we examined whether it is possibleto replicate the results of Experiment 1 under the condition that the subjects were not required to name the target words immediately after semantic decision. If thephonological-mediation view of Van Orden (1987) applies to the activation of kanji word meaning, then theuse of phonology should not be strategically controlledby the task demand of naming; thus, results similar tothose of Experiment 1 should be obtained in Experiment 2. Wydell et al. (1993) also adopted this no-namingprocedure. We compared, therefore, the results ofExperiment 2 with Wydell et al.'s results, as well as with the results of Experiment 1.
MethodSubjects. Fifteen new subjects (ages 19-35 years, 27.9 years
mean age) from the same source group as in Experiment I participated in this experiment. All were native speakers of Japanese andhad normal or corrected-to-normal vision.
Apparatus. Visual and auditory stimuli were presented using anAV tachistoscope (lwatsu Isel: IS-70IA) interfaced with a computer (NEC PC-980 IVX). The AVtachistoscope displayed a visualpattern on a CRT monitor (21-in.), using a raster scan method witha P-31 rapid-decay phosphor. The decay time to 10% luminancelevel was 0.1 msec after the display offset. The AV tachistoscopecan present a stimulus with an accuracy of I msec. The 32 X 32 dotmatrix characters were presented in green against a gray background.A lO-character string was approximately 0.8 X 8 em on the display,producing a typical horizontal viewing angle of about 3° at a typical viewing distance of 160 em. The AVtachistoscope was connectedto one start key and two response keys (yes and no) and measuredresponse latencies with l-msec accuracy.
Stimuli and Procedure. All stimuli were the same as those usedin Experiment I. The procedure was identical to that of Experiment I, except for the presentation ofa warning signal and the naming procedure. In Experiment 2, each trial began with a 1,000-msecwarning beep by the AV tachistoscope instead of a warning signalby the experimenter. The subjects were instructed to make yes or noresponses without naming the target word.
ResultsThe main results ofExperiment 2 are presented in Fig
ure 2. Error data clearly showed the effects of orthographic similarity and homophony. The subjects made12.9%errors on orthographically similar homophones, butmade few errors on others (3.7%, 1.7%, and 0.4%). Therewas a main effect oforthographic similarity [Fs( 1,15) =16.30, MSe = 1.17, P < .001, and F j(l,56) = 15.27,MSe = 1.41,p < .0003] and a main effect of homophony[Fs(1,l5) = 9.16, MSe = 1.07,p < .009, and Fj(l,56) =7.97,MSe = 1.41,p<.007]. The interaction of these twofactors was also significant [Fs(1,15) = 13.85, MSe =0.41,p<.002,andFj(l,56) = 4.71,MSe = 1.41,p<.03].A significant effect of homophony on errors was found
ORTHOGRAPHY AND PHONOLOGY IN JAPANESE KANJI 81
Figure 2. The interaction between orthographic similarity andhomophony in mean percentage errors and in mean correct response times (RTs) in milliseconds from Experiment 2. The errorbars represent the 95% confidence intervals (Loftus & Masson,1994).
not required to name the target words. The subjects mademore false-positive errors on homophone foils than theydid on nonhomophone controls only when the foils wereorthographically similar to their correct exemplars. In addition, the subjects could make decisions more easily tothe orthographically dissimilar foils than to the similarfoils even when they were nonhomophone foils. These results suggest that phonological activation is not strategically controlled but occurs automatically in kanji wordrecognition and that phonology, as well as orthography,contributes to the activation ofthe meaning ofkanji words.
In RTs for correct no responses, a large main effect oforthographic similarity was obtained; responses to orthographically similar foils involved longer latencies thanthose to orthographically dissimilar foils did, which replicates the results both ofExperiment I and ofthe experiment ofWydell et al. (1993). On the other hand, no effectof homophony was found in RTs, which does not replicate either the results ofExperiment I or those ofthe experiment of Wydell et al.
There are at least two possible explanations for the inconsistent effects ofhomophony in contrast to the stableeffects oforthographic similarity on correct RTs for kanjiwords. One possibility, consistent with Van Orden's (1987)interpretation, is that, although the meaning of writtenwords is primarily activated via phonology, orthographicverification takes longer for orthographically similar foilsthan it does for orthographically dissimilar foils, irrespective ofphonological similarity. The second possibilityis that the meaning of kanji words is primarily activateddirectly by orthography; the phonology may be availableonly in the late stage of processing for meaning. As discussed in the context ofExperiment I, when orthographically similar foils are presented, the subjects might haveto carefully compare and check the incorrect target foilswith the corresponding correct exemplars. It is likely,therefore, that the checking process costs more time fororthographically similar foils than it does for dissimilarfoils. A significant homophony effect was found only onthe orthographically similar foils. Ifthe time taken to checkthe representations for orthographically similar foils withthose for candidate words is long enough to bring aboutthe activation ofphonology for kanji words, then the opportunity for a phonological contribution to activate meaning should increase for these words. In Experiment 3, weexplored these alternative possibilities using tachistoscopic presentation with a pattern-masking procedure,as used by Van Orden.
950
t=::IError (Non-Homo.)-D-AT (Non-Homo.)
_ Error(Homo.)
-AT (Homo.)
40
~ 30900 en
I/) .s... 850 ~0......20 ex:::w- 800 gc:
Q) ...0 ...... 0Q) 10 750 oQ..
0 700Similar Dissimilar
only when the homophone foils were orthographicallysimilar to the correct exemplars. Additionally, there wasa significant effect of orthographic similarity betweenthe two nonhomophone foils in the subject analysis[F.(l,15) = 4.91, MSe = 2.0,p < .05] for the simple effect contrast, but not in the item analysis [F; (1,56) = 1.51,MSe = 2.I,p > .2].
Correct RT data showed a strong effect oforthographicsimilarity on both homophone foils and nonhomophonecontrol foils [Fs(1,15) = 51.10,MSe = 1,524.7,p< .0001,and FiC1,56) = 24.63, MSe = 3,287.0, p < .0001], as inExperiment 1. However, there was no main effect ofhomophony by either subjects or items [Fs(1,15) = 3.89,MSe = 1,273.8, P > .07, and FiCI,56) = 1.40, MSe =3,287.0,p> .2]. The interaction ofthese two factors reacheda significant level in the subject analysis [Fs(1,15) =5.93, MSe = 317.6, p < .03], but not in the item analysis[FiCI,56) = 0.80, MSe = 3,287.0,p > .4]. A significanteffect of homophony on orthographically similar foilswas found in the subject analysis [Fs(l,15) = 20.37,MSe = 6,469.5,p < .001] for the simple effect contrast,but not in the item analysis [Fj (1 ,56) = 2.16, MSe =7,084.0,p> .14, by items]. There was no homophony effect for orthographically dissimilar foils in either analysis (F < 1.2, p > .3).
DiscussionIn Experiment 2, we reexamined the effects of ortho
graphic similarity and homophony on semantic decisions with regard to kanji words using a modified procedure with no demand of naming the target words aftersemantic decisions. The results ofthe error analysis werelargely similar to those of Experiment I, where the subjects were required to name the target words, as well asto those of Wydell et al. (1993), where the subjects were
EXPERIMENT 3
Van Orden (1987) argued that if"pattern-masking conditions provided a situation in which word identificationwas best served by its most rapidly available sources ofactivation" (p. 186), then the outcome ofhis masking experiment (i.e., an effect of homophony and no effect oforthographic similarity) demonstrated that phonologicalactivation is an earlier source of constraint than is ortho-
82 SAKUMA, SASANUMA, TATSUMI, AND MASAKI
graphic activation. Van Orden interpreted his maskingresults as evidence that orthography is relatively vulnerable to the effects of pattern masking; he argued thatthese findings were in conflict with the prediction ofthe parallel-access view, particularly with the delayedphonology hypothesis, which assumes that phonologicalcodes are late sources ofconstraint in lexical coding relative to direct access from orthographic codes (see, e.g.,Allport, 1977; Seidenberg, Waters, Barnes, & Tanenhaus, 1984). Van Orden, instead, proposed a verificationmodel to explain these findings. Several other studiesusing English words have reported evidence for a rapidactivation of phonology under masking conditions (see,e.g., Humphreys, Evett, & Taylor, 1982; Lesch & Pollatsek, 1993; Perfetti, Bell, & Delaney, 1988; Underwood& Thwaites, 1982).
On the other hand, there have been no positive findingsthat directly support the early activation ofphonology forJapanese kanji words. Rather, several studies have demonstrated delayed activation of phonology for kanji words.Wang (1988), for example, compared the processing timeof visual, phonological, and semantic targets of twokanji-character compounds, using Neisser's visual-searchtask. In Wang's study, the Japanese subjects responded tovisual targets more quickly than to phonological or semantic targets, but there was no difference between thelatter two. Wang argued that phonological and semanticprocessings of kanji words finish at the same time. If thephonology of kanji words is available in a later stage ofword-recognition processes, any effect ofphonology thatwas observed under no-masking conditions should disappear under pattern-masking conditions. In Experiment 3, we examined whether the homophony effectsand/or the orthographic similarity effects that were observed in Experiments 1 and 2 could be eliminated bymasking.
MethodSubjects. Thirty-three new subjects (ages 21-35 years, 27.8
years mean age) from the same source group participated in this experiment. All were native speakers of Japanese and had normal orcorrected-to-normal vision.
Apparatus and Stimuli. The apparatus was the same as thatused in Experiment 2. The stimuli were the same as those in Experiments I and 2. To construct a pattern mask, we used three verylow-frequency kanji characters. These characters are not used ineveryday life; most skilled readers even are unlikely to name themcorrectly. The pattern mask was constructed and presented, usingan AVtachistoscope, by overlapping one character in normal imageand a two-character string in inverted image in the center of the character position that was used for presenting a target word. No subject was able to name the mask pattern, although everyone knew itconsisted of features ofkanji characters. Note that Van Orden (1987),following the masking assumptions of Johnston and McClelland(1980), used a pattern mask that was composed of letter featuresand was constructed by overlapping alphabetic and nonalphabeticcharacters. Wydell et al. (1993), however, presented a pattern maskthat consisted of crosses, not of kanji-character features.
Procedure. The procedure was the same as that in Experiment 2,with four exceptions: (I) all target words were followed by the maskpattern; (2) SOA between the target word and the pattern mask was
determined separately for each subject; (3) the subjects were permitted to respond to the target words using a third response mode,namely "I don't see anything," instead of pressing either the yes orthe no key; and (4) the subjects were asked to name the target wordafter making a response, as in Experiment I and in Van Orden(1987), in order to check the hit rate oftarget identification for eachsubject after the experiment.
In Van Orden's (1987) masking experiment, the critical SOA wasset to be so short that the subjects could not report any nonexemplar target words. Although we attempted to match Van Orden'sprocedure as closely as possible, a pilot study showed that our subjects could no longer make any response when the critical SOA wasset in that manner. We thus used the following modified procedurefor setting individual SOAs in the practice trials.
Practice began with a 150-msec SOA. After several warm-up trials, the SOA was decreased by 10-msec steps until it became sobrief that the subjects could not identify three successive targetwords. Next, 10 successive trials were used for estimation of thecritical SOA that would produce a 70% to 80% performance levelof identification of the targets. The SOA was adjusted by +10 or- 10 msec during the next 10 trials if the performance level wasoutside this criterion range. The critical SOAs for the subjects rangedbetween 40 and 110 msec; the average was 70 msec. The criticalSOA was fixed throughout the experimental trials for each givensubject.
Each trial began with the presentation of a warning beep for1,000 msec, followed by a definition for 1,500 msec. The definitionwas then replaced by the target word, followed immediately by apattern mask. The pattern mask remained on the display until thesubject responded. The subjects were informed of the sequence ofevents and were told that they should try to respond to the presentation of the target word by pressing either the yes or the no key andthen naming the target word. The subjects were also instructed that,in case they were unable to make any decisions for a given targetword, they should not press the no key, but say instead, "I don't seeanything." This procedure was used to avoid a possible bias to theno-key response. A computer recorded both RTs and key responses.The experimenter recorded incorrect pronunciations of the targetwords and "don't see" responses.
ResultsData from 5 subjects who had a hit rate oftarget iden
tification for the 180 experimental trials greater than80% were excluded from the analysis. Furthermore, datafrom 6 subjects who had a rate of "don't see" responsesto 60 experimental key trials greater than 10% were alsoexcluded. Thus, we analyzed the data from 22 subjects inthe same manner as in Experiments 1 and 2. The meanhit rate of target identification for 180 experimental trials was 64.1%, and the mean percentage of "don't see"responses in 60 experimental key trials was 2.4%. Themain results of Experiment 3 are presented in Figure 3.
The error data showed a strong effect of orthographicsimilarity on both homophone foils and nonhomophonecontrol foils and a reliable effect of homophony on orthographically similar foils. The subjects made 43.6%errors on orthographically similar homophones and 34.2%errors on orthographically similar controls, but made relatively few errors on orthographically dissimilar conditions (14.2% and 10.6%). There was a main effect oforthographic similarity [Fs(l ,21) = 165.20, MSe = 2.11,p< .0001, andFj(I,56) = 55.78, MSe = 9.15,p< .0001]and a main effect of homophony in the subject analysis
ORTHOGRAPHY AND PHONOLOGY IN JAPANESE KANJI 83
Figure 3. The differences in mean percentage errors and inmean correct response times (RTs) in milliseconds between orthographic similar and dissimilar foils in Experiment 3. Theerror bars represent the 95% confidence intervals (Loftus &Masson, 1994).
[Fs(l,21) = 9.54, MSe = 2.20, P < .006] and marginallyin the item analysis [F i(l ,56) = 3.37, MS e = 9.15,p =.072]. The interaction of these two factors was marginally significant in the subject analysis [Fs( I ,21) = 3.72,MSe = 1.l0,p = .067] but not in the item analysis [fi(l,56)= 0.66, MSe = 9.15,p > .4]. The subjects made more errors to homophone foils than to nonhomophone foilswhen they were orthographically similar [Fs(l ,21) =19.81, MSe = 21.84, P < .0002, and F j(I,56) = 3.50,MSe = 32.03,p = .067] for the simple effect contrast, butnot when they were orthographically dissimilar [Fs(l,21)= 2.97, MSe = 3.27, P > .10, and Fj(l,56) < 1, MSe =4.8,p> .47].
Although overall RTs in Experiment 3 were slowerthan those in Experiments I and 2, the RT data showed aclear effect of orthographic similarity on both homophone foils and nonhomophone control foils, but no effect of homophony even on orthographically similarfoils. There was a main effect oforthographic similarity[Fs(l,21) = 20.30, MSe = 28,206.3, p < .0002, andFj(l,56) = 17.90, MSe = 32,290.2,p < .0001], but nomain effect of homophony by either subjects or items[F.(I,21) = 2.63,MSe = 32,914.I,p>.12,andfi(l,56) =1.05, MSe = 32,290.2, P > .3]. The interaction of thesetwo factors was not significant either by subjects or byitems [Fs(l,21) = 0.002,p> .9, andFj(I,56) = 0.46,p >.5]. The difference between homophone and nonhornophone foils was not significant either in orthographicallysimilar foils [~(l,21) = 2.46, MSe = 45,120.0, P > .1,and FJ I ,56) < 0.1, MSe = 1,904.0, P > .8] or in orthographically dissimilar foils [Fs(l ,21) = 2.26, MSe =41,604.7, P > .1, and Fi (l ,56) = 1.45, MSe = 46,886.5,p> .2] for the simple effect contrast.
- Error (Homo.)_RT(Homo.)
70
600'~ 50e?e 40
UJ'E 30
~Q) 20c..
10
aSimilar
c::::::::::J Error (Non-Homo.)
--G-RT (Non-Homo.)
1500
1400 enE
1300 ;;I-
1200 ~
1100 ~o
1000
--'---' 900Dissimilar
DiscussionThe results of the present kanji masking experiment
showed that the effect of orthographic similarity remained strong both on false-positive errors and on correct RTs, whereas the effect ofhomophony was relativelyreduced but remained significant on errors for orthographically similar foils. The false-positive error rate of43.6% for orthographically similar homophone foils wassignificantly different from the corresponding error rateof 34.2% for the orthographically similar nonhomophones, as well as from that of 14.2% for orthographicallydissimilar homophone foils. The mean false-positive errorrate for homophone foils was 28.9%, and the corresponding error rate for nonhomophonic controls was 22.4%.No reliable effect of homophony was found in the correct RT data.
In contrast, in the results of the English masking experiment by Van Orden (1987), the orthographic-similarityeffect that was observed under no-masking conditionsdisappeared, but the homophony effect remained strong.The false-positive error rate of40% for similarly spelledhomophone foils was not significantly different from thecorresponding error rate of46% for less similarly spelledhomophone foils. The mean false-positive error rate forhomophone foils was 43%, and the corresponding errorrate for nonhomophonic spelling controls was 17.5%.
These contrasting findings for the English maskingexperiment and our Japanese kanji masking experimentsuggest clear-cut differences between the two orthographies in the early processes ofvisual word recognition. Inthe English experiment, phonological activation of written words occurs soon enough to exert an influence on semantic decisions under the brief-exposure conditions ofpattern masking. In our Japanese experiment, in contrast,partial phonological activation of written words may occur, but not fully or quickly enough to affect semanticdecisions under masking conditions. Rather, orthographyexerts the major influence on the activation of meaningfor kanji words under masking conditions.
The findings of our masking experiment were essentially similar to those ofWydell et a1.'s (1993) kanji masking experiment, although there were also some differences.They found a strong effect oforthographic similarity, irrespective ofhomophony, on errors but not on correct RTs.The absence ofan effect of orthographic similarity on correct RTs in Wydell et al.s experiment might have arisenbecause they used an inadequate stimulus set ofcategorynames and target words; some of their category nameswere written in an unusual sentence-style and some ofthem did not represent an accurate meaning for their correct exemplars. They also found a marginal effect of homophony on errors when the target words were orthographically similar foils: The VS.HOM (Visually SimilarHomophone) "foils were significantly more error pronethan any other foil type (p < .0 I) except for the VS.NONHOM foils (the critical difference, which had to be equalto or greater than .248 to be reliable at the .05 level was
84 SAKUMA, SASANUMA, TATSUMI, AND MASAKI
.247)" (Wydell et aI., 1993, p. 501). By contrast, the effectof homophony on errors under masking conditions wassignificant in our experiment in those cases in which thedefinitions and the targets of the stimulus pairs had noidentical kanji characters between them.
These effects were interpreted by Wydell et aI. (1993)as support for the contention that phonological representation as well as orthographic representation can be derived sufficiently quickly to affect judgments under masking conditions. We endorse their interpretation that thereis an early orthographic activation of meaning in kanjiword recognition. Wehesitate, however,to accept their interpretation that there is an early phonological activationof meaning, because a significant effect of homophonyunder the masking condition was found only on errorsand only on the orthographically similar foils in both ourkanji study and Wydell et aI.'sstudy.Correct RTs for orthographically similar foils were significantly longer thanthose for orthographically dissimilar foils; this longer timemight foster phonological contributions to the activationof meaning. Thus, there still remains the question ofwhether the activation of phonology is rapid enough toactivate the meaning of kanji words prior to the meaningactivation from orthography. Undoubtedly, further research needs to be done in order to clarify the time-coursefor the processes by which the phonology ofkanji wordsis computed from the orthography. We tentatively conclude that orthographic activation is an early constraintin lexical coding relative to phonological activation inkanji word recognition.
GENERAL DISCUSSION
The present study addressed the question of whetherthe role of phonology in visual word recognition andcomprehension differs across different orthographies. Toinvestigate this question, we carried out three experiments for familiar kanji words with Japanese subjects,using paradigms that were equivalent to Van Orden's(1987) semantic decision task for English words.
In Experiments 1 and 2, the subjects were more likelyto make false-positive errors on homophone foils thanon nonhomophone foils only when the target foils wereorthographically similar to the corresponding correct exemplars. They made few errors when the target foils wereorthographically dissimilar. The task demand of namingdid not affect the results; homophone effects were observed whether subjects were or were not required to namethe target words. The results of these experiments indicate that phonology as well as orthography contributes tothe activation of the meaning of kanji words, which replicates the findings of Van Orden's (1987) English studyand those ofWydell et aI.'s (1993) Japanese kanji study.
The results of Experiment 3, however, were contraryto those of Van Orden's (1987) English masking experiment. In the English study, the effect oforthographic similarity observed under no-masking conditions disap-
peared under masking conditions, but the effect of homophony remained strong. In our kanji masking experiment, on the other hand, the effect of orthographic similarity remained strong on both errors and correct RTs,whereas the effect of homophony that was observedunder no-masking conditions was relatively reduced butsignificant on errors to orthographically similar foilsunder masking conditions. The results ofWydell et aI.'s(1993) masking experiment for kanji words were essentially similar to those of our kanji masking experiment.These results ofkanji masking experiments indicate that,although the phonology of kanji words is automaticallyactivated and this activation may at least partly contribute to the meaning activation, it is orthography that isthe primary source of the activation of the meaning ofkanji words. These findings from kanji masking experiments are consistent with the parallel-access view but inconsistent with the phonological-mediation view.
One major difference between Van Orden's (1987) English study and the two kanji studies was in the effect oforthographic similarity under the masking conditionsno effect in English, as opposed to strong effects in kanji.How do we explain this difference? As has already beenpointed out by Wydell et aI. (1993), many homophones(e.g., rows vs. rose) in English that are not spelled alikehave substantial spelling overlap in general (although avery small number ofexceptional examples, such as eightand ate, exist), while orthographically dissimilar homophones in kanji have nothing in common orthographically(Wydell et aI., 1993, pp. 501--502). In an alphabetic orthographic system like English, it is difficult to avoid aspelling overlap between the two members of any homophonic pairs of words. In kanji orthography, on the otherhand, there are numerous homophonic words whose orthographic patterns are completely distinct from eachother. The differences between similarly spelled homophones and less similarly spelled homophones in Van Orden's English study can be thought of as being within therange oforthographically similar homophones in our kanjistudy. This difference in the degree oforthographic similarity might have been responsible for the different resultsfor the effects of orthographic similarity under maskingconditions. Suppose that orthographically "dissimilar"foils, rather than "less-similar" foils, in English (e.g.,eight/ate rather than rose/rows) were used for the targetfoils of their category names. One would then expect toobtain a significant effect of orthographic similarity on errors, even in the masking experiment, which, in turn,would provide evidence to support the view that the meaning of English words is activated directly by orthography.Thus, it can be concluded that, although phonological mediation may occur in the activation ofmeaning for writtenEnglish words, it is open to question whether the occurrence of phonological mediation is an obligatory processfor the activationof meaningforwritten Englishwords.Thisview of ours appears to be supported by recent experimental findings by V. Coltheart, Avons, Masterson, and
ORTHOGRAPHY AND PHONOLOGY IN JAPANESE KANJI 85
Laxon (1991), by V. Coltheart, Patterson,and Leahy(1994),and by Jared and Seidenberg (1991), among others.
In conclusion, the results of the present experimentswith kanji words support the parallel-access view thatboth orthography and phonology contribute to the activation ofthe meaning ofwritten words. This general conclusion is consistent with Wydell et al.'s (1993) kanji study.The contrasting findings for English and Japanese kanjiwords under masking conditions would also suggest thatthe relative time-courses ofthe parallel routes to meaning,one from orthography and the other from phonology, aredifferent for the two languages. These findings, taken together, suggest that, although the basic processes ofreading have common features across different orthographies,the detailed nature ofthese processes, especially the timecourse of the processes, differs across different orthographies (see, e.g., Besner & Hilderbrandt, 1987; Patterson, 1990; Perfetti & Zhang, 1991; Sasanuma, 1986,1994; Seidenberg, 1985; Tabossi & Laghi, 1992). Muchfuture work focusing on these differences, as well as onthe similarities, across different orthographies is requiredfor the further understanding of universal and languagespecific features of reading processes.
REFERENCES
ALLPORT, D. A. (1977). On knowing the meaning of words we are unable to report: The effects of visual masking. In S. Dornic (Ed.), Attention and performance IV (pp. 505-533). New York: AcademicPress.
BARON, J. (1973). Phonemic stage not necessary for reading. QuarterlyJournal ofExperimental Psychology, 25, 241-246.
BECKER, C. A (1976). Allocation of attention during visual word recognition. Journal of Experimental Psychology: Human Perception &Performance, 2, 556-566.
BECKER, C. A. (1980). Semantic context effects in visual word recognition:An analysis ofsemantic strategies. Memory & Cognition, 8, 493-512.
BECKER, C. A., & KILLION, T. H. (1977). Interaction of visual and cognitive effects in word recognition. Journal ofExperimental Psychology: Human Perception & Performance, 3, 389-401.
BESNER, D., & HILDERBRANDT, N. (1987). Orthographic and phonological codes in the oral reading of Japanese Kana. Journal of Experimental Psychology: Learning, Memory, & Cognition, 13, 335-343.
COLTHEART, M., CURTIS, B., ATKINS, P., & HALLER, M. (1993). Models of reading aloud: Dual-route and parallel-distributed-processingapproaches. Psychological Review, 100, 589-608.
COLTHEART, M., DAVELAAR, E., JONASSON, J. T., & BESNER, D. (1977).Access to the internal lexicon. In S. Dornic (Ed.), Attention and performance VI (pp. 535-555). London: Academic Press.
COLTHEART, v.. AVONS, S. E., MASTERSON, J., & LAXON, V. J. (1991).The role of assembled phonology in reading comprehension. Memory & Cognition, 19, 387-400.
COLTHEART, v.. PATTERSON, K. E., & LEAHY, J. (1994). When a ROWSis a ROSE: Phonological effects in written word comprehension.Quarterly Journal ofExperimental Psychology, 47 A, 917-955.
EVETT, L. J., & HUMPHREYS, G. W. (1981). The use of abstract graphemic information in lexical access. Quarterly Journal ofExperimental Psychology, 33A, 325-350.
FROST, R., KATZ, L., & BENTlN, S. (1987). Strategies for visual wordrecognition and orthographical depth: A multilingual comparison.Journal ofExperimental Psychology: Human Perception & Performance, 13,104-115.
GORYO, K. (1987). Yomuto iu koto [Psychology ofreading]. Tokyo: TokyoUniversity Press.
HIROSE, H. (1992). Jukugo no ninchi-katei ni kansuru kenkyu [An inves-
ligation of the recognition process for jukugo by use of priming paradigms]. Japanese Journal ofPsychology, 63, 303-309. (in Japanese)
HUMPHREYS, G. W., EVETT, L. J., & TAYLOR, D. E. (1982). Automaticphonological priming in visual word recognition. Memory & Cognition, 10, 576-590.
JARED, D., & SEIDENBERG, M. S. (1991). Does word identification proceed from spelling to sound to meaning? Journal of ExperimentalPsychology: General, 120, 358-394.
JOHNSTON, J. C., & MCCLELLAND, J. L. (1980). Experimental tests ofahierarchical model of word identification. Journal of VerbalLearning & Verbal Behavior, 19, 503-524.
KAIHO, H., & NOMURA, Y. (1983). Kanjijyoho syori no sinrigaku [Psychology of kanji information processing]. Tokyo: Kyoiku Syuppan.
KIMURA, Y. (1984). Concurrent vocal interference: Its effects on kanaand kanji. Quarterly Journal ofExperimental Psychology, 36A,117-131.
LESCH, M. E, & POLLATSEK, A. (1993). Automatic access of semanticinformation by phonological codes in visual word recognition. Journal ofExperimental Psychology: Learning, Memory, & Cognition,19,285-294.
LOFTUS, G. R., & MASSON, M. E. J. (1994). Using confidence intervalsin within-subject designs. Psychonomic Bulletin & Review, 1,476-490.
MONSELL, S., PATTERSON, K. E., GRAHAM, A, HUGHES, C. H., & MILROY, R. (1992). Lexical and sublexical translation of spelling tosound: Strategic anticipation of lexical status. Journal ofExperimental Psychology: Learning, Memory, & Cognition, 18,452-467.
MORTON, J., & PATTERSON, K. (1980). A new attempt at an interpretation, or, an attempt at a new interpretation. In M. Coitheart, K. Patterson, & 1. C. Marshall (Eds.), Deep dyslexia (pp. 91-118). London:Routledge & Kegan Paul.
MORTON, J., & SASANUMA, S. (1984). Lexical access in Japanese. InL. Henderson (Ed.), Orthographies and reading: Perspectives fromcognitive psychology, neuropsychology, and linguistics (pp. 25-42).London: Erlbaum.
MORTON, J., SASANUMA, S., PATTERSON, K. E., & SAKUMA, N. (1992).The organization of the lexicon in Japanese: Single and compoundkanji. British Journal ofPsychology, 83, 517-531.
PAAP, K. R., NEWSOME, S. L., McDoNALD, J. E., & SCHVANEVELDT, R. W.(1982). An activation-verification model for letter and word recognition: The word-superiority effect. Psychological Review, 89, 573-594.
PATTERSON, K. E. (1990). Basic processes of reading: Do they differ inJapanese and English? Japanese Journal ofNeuropsychology, 6, 4-14.
PERFETTI, C. A., BELL, L. c., & DELANEY, S. M. (1988). Automatic(prelexical) phonetic activation in silent word reading: Evidence frombackward masking. Journal ofMemory & Language, 27, 59-70.
PERFETTI, C. A., & ZHANG, S. (1991). Phonological processes in reading Chinese characters. Journal ofExperimental Psychology: Learning, Memory, & Cognition, 17, 633-643.
PLAUT, D. C., MCCLELLAND, J. L., SEIDENBERG, M. S., & PATTERSON, K. (1996). Understanding normal and impaired word reading:Computational principles in quasi-regular domains. PsychologicalReview, 103,56-115.
RUBENSTEIN, H., LEWIS, S. S., & RUBENSTEIN, M. A (1971). Evidencefor phonemic recording in visual word recognition. Journal ofVerbalLearning & VerbalBehavior, 10,645-657.
SAITO, H. (1981 ). Kanji to kana no yomi ni okeru keitaiteki- fugouka oyobioninteki fugouka no kento. [Use of graphemic and phonemic encoding in reading kanji and kana]. Japanese Journal ofPsychology, 52,266-273. (in Japanese)
SASANUMA, S. (1980). Acquired dyslexia in Japanese: Clinical featuresand underlying mechanisms. In M. Coitheart, K. Patterson, & 1. C.Marshall (Eds.), Deep dyslexia (pp. 48-90). London: Routledge &Kegan Paul.
SASANUMA, S. (1986). Universal and language-specific symptomatology and treatment of aphasia. Folia Phoniatrics, 38,121-175.
SASANUMA, S. (1994). Universal and language-specific features of reading impairment. In P. Bertelson, P. Eelen, & G. d'Ydewalle (Eds.), International perspectives on psychological science (Vol. I, pp. 105-125).Hove: Erlbaum.
SASANUMA, S., SAKUMA, N., & KITANO, K. (1992). Reading kanji with-
86 SAKUMA, SASANUMA, TATSUMI, AND MASAKI
out semantics: Evidence from a longitudinal study of dementia. Cognitive Neuropsychology, 9, 465-486.
SEIDENBERG, M. S. (1985). The time course of phonological code activation in two writing systems. Cognition, 19,1-30.
SEIDENBERG, M., & MCCLELLAND, J. L. (1989). A distributed, developmental model of word recognition and naming. Psychological Review, 96, 523-568.
SEIDENBERG, M. S., WATERS, G. S., BARNES, M. A., & TANENHAUS,M. K (1984). When does irregular spelling or pronunciation influence word recognition? Journal ofVerbalLearning & Verbal Behavior, 23, 383-404.
TABOSSI, P.,& LAGHI, L. (1992). Semantic priming in the pronunciationofwords in two writing systems: Italian and English. Memory & Cognition, 20, 303-313.
UNDERWOOD, G., & THWAITES, S. (1982). Automatic phonological coding ofunattended printed words. Memory & Cognition, 10,434-442.
VAN ORDEN, G. C. (1987). A ROWS is a ROSE: Spelling, sound, andreading. Memory & Cognition, 15, 181-198.
VAN ORDEN, G. C, & GOLDINGER, S. D. (1994). Interdependence offormand function in cognitive systems explains perception ofprinted words.Journal ofExperimental Psychology: Human Perception & Performance, 20, 1269-1291.
VAN ORDEN, G. C, JOHNSTON, J. C, & HALE, B. L. (1988). Word identification in reading proceeds from spelling to sound to meaning. Journal ofExperimental Psychology: Learning, Memory, & Cognition,14,371-386.
VAN ORDEN, G. C., PENNINGTON, B. E, & STONE, G. O. (1990). Wordidentification in reading and the promise of subsymbolic psycholinguistics. Psychological Review, 97, 488-522.
WANG, J. (1988). Kanji no onin-syori to imi-syori ha douji ni kanryosuru-ka [Do phonological and semantic processings ofkanji finish atthe same time?] Japanese Journal ofPsychology, 59, 252-255. (inJapanese)
WYDELL, T. N., PATTERSON, K. E., & HUMPHREYS, G. W. (1993).Phonologically mediated access to meaning for kanji: Is a rows stilla rose in Japanese kanji? Journal ofExperimental Psychology: Learning, Memory, & Cognition, 19,491-514.
YOKOSAWA, K, & UMEDA, M. (1988). Processes in human kanji word
recognition. Paper presented at the 1988 IEEE International Conference on Systems, Man, & Cybernetics, Beijing and Shenyang, China.
NOTES
I. According to Yokosawa and Umeda (1988), approximately 70% ofthe 51,962 words in one Japanese dictionary are two-kanji-characterwords and the average word length is 2.4 characters.
2. There are a small number of kanji characters which have only onetype of reading---either a KUN-reading or an ON-reading. Of the totalof 1,945 kanji characters of contemporary standard usage for Japanesewritten language-ealled "Joyo kanji" in Japanese-c-Zfitl (12.9%) kanjicharacters have only one type of reading in the "Joyo kanji" table.
3. Wydell, Patterson, and Humphreys (1993) state that the semantically mediated procedure is believed to be the only way to retrieve thephonology of kanji words. This commentary seems to be insufficient,because a nonsemantic direct route for recoding from orthography tophonology (sometimes a bypassing route) has been assumed in somemodels for kanji word recognition (see, e.g., Goryo, 1987; Kaiho & Nomura, 1983; Sasanuma, 1986). Unfortunately, however, the existence ofthe nonsemantic direct route for kanji words has not been demonstratedby clear experimental evidence thus far.
4. We used the term definition instead of category name in this paper,because our category names expressed very narrow, specific categories->for example a chance for the exemplar opportunity, the homophone foilmachine, and the control foil function (see also Appendixes A and B).
5. We used definitions twice for a given subject in the present experiments---once for the homophone foils and once for the matched nonhomophone controls, although the order of presentations of the foilswas controlled over the different halves of the lists. To examine whetherthis repetition ofdefinitions affected the performances, we analyzed thefalse-positive errors, comparing the first and second halves of the listsusing a three-way analysis of variance in each experiment. The only significant effect was the interaction of repetition and orthographic similarity in Experiment I : More errors were made on orthographically similar foils in the first halfthan were made in the second half of the lists[Fs(l,46) = 4.49, MS. = 65.43,p < .05]. There was no other significanteffect of repetition over Experiments 1,2, and 3.
APPENDIX ACorrect Exemplars and Target Foils Used in Experiments 1-3
Definition Exemplar Homophone Control Definition Exemplar Homophone Control
Orthographically Similar Foils Orthographically Dissimilar Foilsti~O)i£i! lIIii ~iii JE\:ii ~j\~~.99,A, ~affl A~ lti:
.Jt.l:cHad.l9.::t m!~. m!1II m!~ ~.O)ltA. 'Z'~ \ <i1& ilt£ ~Jl ill.il~t.l:C;li~tt9.::t ~. ~. •• llllM99,A, ittf ~. .:trJi'R* IIIH,' .tl.' iIItl.' .~t.l:tl.' ~A '!Tl-t ~1IiE
tt>~PL- IlI'Ifl IVlJi1 IVIIt ••1Jlg0).~~ •• .iii ~Jl:
T1'/A 1Iof< II~ liRE iRlA99.::t .~ 811 of<tI~.U'~991.1:i'L ~T ~~ ~f'F L-! t: f) I'll J8111l of<~
.tl'~.O)t.:d.lo)~. llfillJJ IiEIlJJ mtllJJ ,;J:;I)'C9.::t It'!T 1!f!/l 303ft!lt.ltAt~P9':: t ~;Jllfi ~;J" M~ 'Jft.: It .:!- 'II. iWA'It -::> ;1)'15 ~j\ ~lIJl ~i&: U't;l)~ ':> 11~ iEiI JlD.W~t.:~\ f1ti.l f1tff f1tUl til L-t;J:;l)'9':: t :1!Ht fUil II~
9 <'nHE1J0)M'153o 7'7::t 7'7:~ 7'7:j\ ftli';J!t:~d.ltd!HiI :iE1iIIi f1t1' ~tl.'
~U.HJ15W9':: t ~M ~. ~;t: ~tJ:? 7'7:_ ti~ '!Till
!i!J,~'?! ~:tlJ! ~* ~iI r.iil l ) '21." ~* if.f±.~ tt.:t.l:~\.:: t •• .~ .JI .!F~t-::>-r.~9.::t ~{t In ~~
ORTHOGRAPHY AND PHONOLOGY IN JAPANESE KANJI 87
APPENDIXBCorrect Exemplars and Target Foils Used in Experiments 1-3 Translated into English
Definition Exemplar Homophone Control
Orthographically Similar Foilsgardening entertainmentsraising of a curtain open a garden to the publicfire house workconcern admirationindirectness jointopportunity machineatom genesisillumination proofcontrast objectshort temper short termlow temperature low-pitched soundgenius natural calamityfiring departureidea forwardinginoccupation colorless
Orthographically Dissimilar Foils
doctor willprocess homejournalist traingoodwill actionbacteria recently
To grow plantsTo open a play or dramaBurning of a buildingAn interestNot in a straight wayA chanceThe constituent parts of a matterLights used to decorate a stageTo compare things with othersImpatientnessColdOne who has great capacityShooting a gunA plan that comes up incidentallyTo hold no job
One who treats a diseaseA course of going forwardA reporterKindheartednessA microorganism which is
organized with one cellTo agree with an opinionA customaryGoing forwardHeightPersonalityTo figure to oneselfThe formal price for saleThe look of the skyPackingTo get old
approvalhabitprogressstaturecharacterimaginationfixed priceweatherwrappingaging
acidityweekfaithprudenceaccuracycreationdeclinechange of schoolsbroadcastingcorridor
military artsdevelopmentmealcenter of the Metropolisfunction (math.)functionoriginalexplanationcountermeasureshortwavedullnessweathersaledevelopmentimpossible
criminal investigationelectionelectric lightdoctor's officeeffect
place of meetingtalkshousewifedictionaryblood vesselhallconsciencebehaviortransformationchorus
(Manuscript received May 20, 1996;revision accepted for publication December 17, 1996.)