the effect of radical-based grouping in character learning ... · the effect of radical-based...

The Effect of Radical-Based Groupingin Character Learning in Chinese as aForeign LanguageYI XUUniversity of PittsburghDepartment of East Asian Languages and Literatures113C Old Engineering Hall3942 O’Hara StreetPittsburgh, PA 15260Email: [email protected]

LI–YUN CHANGUniversity of PittsburghDepartment of Psychology648 Learning Research and Development Center3939 O’Hara StreetPittsburgh, PA 15260Email: [email protected]

CHARLES A. PERFETTIUniversity of PittsburghDepartment of Psychology833 Learning Research and Development Center3939 O’Hara StreetPittsburgh, PA 15260Email: [email protected]

The logographic nature of the Chinese writing system creates a huge hurdle for Chinese as a foreignlanguage (CFL) learners. Existing literature (e.g., Shen, 2010; Taft & Chung, 1999) suggests that radicalknowledge facilitates character learning. In this project, we selected 48 compound characters in eightradical groups and examined how grouping characters based on their radicals affected the form, sound,and meaning representations of characters and radical knowledge development. We found that forbeginning learners, learning radical-sharing characters in groups consistently led to better recall andbetter radical generalization than learning in distribution. For intermediate level learners, the groupingfactor did not lead to significant differences, while participants in both conditionsmade improvement inradical perception and radical semantic awareness generalization.We concluded that there is a benefit topresenting learners with recurring radicals in compound characters in groups in character learning andin the autonomous generalization of radical knowledge. We also noted the differences betweenbeginning and intermediate learners in their character perception and learning, and put forwardimplications for CFL pedagogy.

Keywords: character; Chinese as a foreign language; orthography; proficiency; radical knowledge

PRODUCING CHARACTERS FROM MEMORYand retrieving the form, sound, and meaning ofcharacters are frequently identified as majorchallenges for students learning Chinese as aforeign language (CFL) (e.g., Shen, 2004). The

difficulty comes from the large number ofcharacters, the visual complexity of the graphe-mes, and the absence of systematic grapheme–phoneme correspondence. Earlier researchinvestigating effective teaching and learningmethods to develop second language (L2)reading skills in Chinese confirmed that radicalknowledge facilitates character recognition andproduction (e.g., Shen, 2005, 2010), but research-ers have also noted that character teaching in CFLis not systematically guided by the structural

The Modern Language Journal, 98, 3, (2014)DOI: 10.1111/j.1540-4781.2014.12122.x0026-7902/14/773–793 $1.50/0© 2014 The Modern Language Journal

principles of the characters, making it difficult forlearners to extract recurring radicals from com-pound characters (Taft & Chung, 1999). Wang,Liu, and Perfetti (2004) found that beginningCFL learners cannot infer themeaning of radicalsin unknown characters without probing, andhypothesized that the difficulty in extractingradicals’ functional regularities may be due tolimited exposure. The current research usescompound characters with recurring radicalsas learning materials and investigates whethergrouping characters based on their sharedradical facilitates orthographic knowledgedevelopment—such as learning the form, sound,and meaning of new characters and developmentin knowing radicals’ shapes, fixed positions, andsemantic functions.

REVIEW OF LITERATURE

Chinese characters correspond to morphemesand are typically monosyllabic, so Chinese hasbeen referred to as a morphosyllabic language(e.g., DeFrancis, 1989). The series of strokeswithin a character compose recurring subchar-acter components, including radicals and chunks.These terms are defined as follows:

1. Radicals (bushou): the smallest orthographicunits within a character that have semanticor phonetic functions. In this article, “radi-cal” refers to semantic radicals and “phonet-ic component” refers to phonetic radicals.

2. Chunks (bujian): the smallest visually inte-grated unit “separated by a visible diminutivespace from other units” in a character (Shen& Ke, 2007, p. 99). Unlike radicals, chunksare not consistently associated with a partic-ular function. For instance, (hun, ‘wed-ding’) consists of (nu, ‘female’) as aradical and (hun, ‘dusk’) as a phoneticcomponent; but it is also composed of threechunks: , coincidentally a radical; andand , which do not serve semantic orphonetic functions in this compound char-acter. Chunks and radicals can overlap whenthe radical is not further decomposable intosmaller chunks; moreover, certain graphicforms, such as , may be a radical in somecharacters (e.g., ) but a chunk in others(e.g., , an, ‘case’). “Chunk” also has ameaning not specific to Chinese characters:In the study of cognition, a chunk is “ameaningful unit of information built fromsmaller pieces of information” (Gobet &Lane, 2012, p. 541). For example, chunks of

information become functional in memory.This definition is used when we refer to“chunking theory.”

3. Subcharacter component: an orthographic unitinternal to a character. This is an umbrellaterm that can include chunks, radicals, orphonetic components, but does not includestrokes, the lowest level component in acharacter.

Approximately 97% of Chinese characters aresemantic–phonetic compounds (xıngsheng) with aradical and a phonetic component (DeFrancis,1989). Commonly-used characters are formedwith about 200 semantic radicals (e.g., Feldman&Siok, 1999a). Although the radical does notspecify the precise meaning of a compound, theinterpretation of the radical is generally consis-tent with the meaning of the whole character(Feldman & Siok, 1999b).1 The semantic value ofthe radical is therefore useful for recognizing thecharacter.

Learning theories and empirical evidenceindicate that radical knowledge supports charac-ter learning (Shen, 2010; Taft & Chung, 1999).Perceiving characters as organized subunits ofradicals and other subcharacter components(e.g., bujian) creates units of encoding that aremore efficient for processing than perceivingcharacters as a pile of interwoven strokes. Thisargument is in line with the cognitive process ofchunking that binds lower level elements (e.g.,strokes) into larger units (e.g., radicals), reducingmemory burden and increasing informationcapacity (Chase & Simon, 1973). In the applica-tion of chunking to language acquisition, Ellis(2003) contends that “[a] chunk that activatessome meaning representations makes (. . .) itselfmore salient in the input stream” (p. 78). Radicalsare examples of suchmeaning-bearing perceptualunits. When a reader develops the awareness thatcharacters are composed of subcharacter compo-nents, including some with consistent meaningassociations, memorizing a thousand charactersas independent units gives way to buildinginterrelated characters connected through amuch smaller number of recurring radicals andchunks.

Empirically, the facilitation of radical knowl-edge is attested among both L1 and L2 speakers.First, L1 speakers’ recognition of compoundcharacters involves radical processing (Feldman& Siok, 1999a; Taft, Zhu, & Peng, 1999; amongothers). Research also shows that radical knowl-edge is associated with Chinese L1 children’scharacter acquisition and reading performance

774 The Modern Language Journal 98 (2014)

(Shu&Anderson, 1997; Tong et al., 2009). In Taftand Chung (1999), the effect of radical instruc-tion was confirmed in an experiment with a groupof naıve learners who had no prior experiencewith Chinese characters. The authors reportedthat giving radical instruction before teachingcharacters and in the first presentation ofcharacters both yielded better recall than noradical instruction in character teaching. In theCFL context, Shen and Ke (2007) found thatbeginning and intermediate students’ radicalapplication knowledge (i.e., whether they canmake meaning inferences in unknown charactersbased on radicals) is moderately correlated withtheir memory retention of vocabulary knowledge.Jackson, Everson, and Ke (2003) and Shen (2005)also reported that learners learn new charactersusing an assortment of radical knowledge, includ-ing locating the radical in unfamiliar charactersand making meaning association.

At the same time, the number of radicals thatstudents encounter makes it difficult for studentsthemselves to generate systematic rules to decom-pose characters (e.g., Everson, 2011), especiallybecause students often do not have enoughopportunities to reencounter radicals in differentcharacters after initially learning them (Shen,2010). An item is better memorized when itrepeatedly occurs (Ellis, 2003). Beyond repetitionis the question of when to repeat—close togetheror spaced apart. In many learning situations,widely spaced repetition is effective (Cepedaet al., 2006). However, the effect of spacingdepends on the knowledge components involved(Koedinger, Corbett, & Perfetti, 2012), andnarrow spacing can be beneficial when what isto be learned—the whole character—changesfrom trial to trial. Repeating a radical with broadspacing limits the opportunity to make a connec-tion with previously learned characters thatcontain the same radical. Having radical-sharingcharacters occur together should help learnersmake the connection. The present study presentsradicals that repeatedly occur in fixed position indifferent compound characters, varying whetherthe characters with shared radicals are blocked ingroups or distributed over instruction. Closerepeated presentation of radicals could potential-ly bootstrap the learning of radicals and charac-ters. Thus the current study investigates whetherclassifying characters based on their sharedradicals and presenting characters to learnersin radical-based groups might raise learners’awareness of these recurring radicals, includingthe radical’s regular positions and semanticfunctions.

Teaching characters in radical or orthography-based groups has been used in L1 literacyeducation (Guo & Zhang, 1991; Zhang, 2012).For instance, Chang and Han (2004) reportedthat, when instructors supplemented the teachingof a key character in text (e.g., pao) with theteaching of orthographically similar characters( pao, pao, bao), L1 children developedbetter knowledge regarding the radical andphonetic components than children who learnedonly the key character . Radical-based groupingwas also employed in Chen et al. (2013), in whichgroups of radical-sharing characters were used aslearning material in the experimental group.However, because Chen et al.’s control group didnot learn the same set of characters, and explicitorthographic instruction was available only in theexperimental group, the beneficial effect associ-ated with their experimental group could be dueto the learning of more characters containingtarget radicals, coupled with instructors’ ortho-graphic instruction. Chen et al.’s research did notdirectly address the role of character grouping,but their study extended radical-based characterteaching to adult heritage learners, and their setsof radical recognition and semantic awarenesstasks set examples for the current study.

The current project aims to discover whetherclassifying radical-sharing characters in groups ina learning sequence supports character learningand helps learners extract the visual and func-tional regularities of radicals. This groupedcondition is compared to a distributed conditionin which radical-sharing characters are assignedto different learning sessions. The project asksthe following questions:

RQ1. DobeginningCFL learners establishbetterform, sound, andmeaning representationsof characters learned in the groupedcondition or the distributed condition?

RQ2. Do beginning learners in the groupedcondition or the distributed conditiondevelop better radical knowledge, includ-ing the ability to visually identify thegraphic form of a radical in unknowncharacters and in understanding under-standing the radical’s semantic function inassociation with its fixed position?

Existing research is scarce on radical knowl-edge development and character learning beyondthe beginning level (Everson, 2011), so anotherpopulation of interest is intermediate learners.We asked the same questions for that proficiencygroup.

Yi Xu et al. 775

RQ3. Do intermediate learners learn charac-ters’ form, sound, and meaning better inthe grouped condition or the distributedcondition?

RQ4. Do intermediate learners develop radicalknowledge better in the grouped condi-tion or the distributed condition?

For the more experienced group, it is possiblethat increased familiarity with the radical basedcomposition of compounds reduces the advan-tage of closely repeated exposure.

METHOD

Participants

Forty-eight nonheritage adult CFL learners in1st-year Chinese classes at a northeastern UnitedStates university constituted the beginning learnersin this research. These learners had learnedChinese in a 7-hour per week program for eightweeks and had acquired approximately 180characters prior to the experiment. As part of thefirst week of their regular curriculum, theseparticipants received brief instruction in 40 com-mon radicals introduced in their textbook, withbrief exposure to two character examples for each.Forty CFL learners in 2nd-year Chinese classes atthe same institution constituted the intermediatelearners. These learners had learned Chinese for36 weeks and had learned approximately 530 totalcharacters. Twenty-four 1st-year participants and202nd-year participants were assigned to the groupedcondition, while the remaining participants wereassigned to the distributed condition. Participants’overall course grades confirmed that proficiency inthe two conditions did not differ (first year: t(46)¼ .92, p¼ .36; second year: t(38)¼ .20,p¼ .84).

Materials

Target radicals in this study are among the 40radicals introduced in participants’ textbook:(‘money’), (‘fire’), (‘metal’), (‘wood’),

(‘female’), (‘sun’), (‘food’), and(‘heart’). Six characters for each radical wereselected based on the following criteria: (1) nocharacter was taught to either 1st- or 2nd-yearparticipants in their course curriculum; (2) aradical always appears in the same position forcharacters in its group; and (3) characters in eachradical group are matched by their frequencies inEnglish translations (Brysbaert & New, 2009),their mean number of chunks (M¼ 2.98) (Chen

et al., 2011), and their mean number of strokes(M¼ 10.11), with at least one character in eachradical groupwith low (6–8),medium(9–11), andhigh stroke numbers (12–14). In all cases, radicalscontribute to the meaning of the wholecharacter.2

The key manipulation was the assignment ofthose 48 characters into eight character sets. Inthe grouped condition, each set consisted of sixcharacters from the same radical group; forexample, (‘wedding’), (‘to marry’),(‘daughter-in-law’), (‘baby’), (‘lovely’),and (‘father’s sister’) all contain the‘female’ radical. In the distributed condition,characters sharing the same radical are distribut-ed so that each set always contains six characterswith different radicals. Appendix A indicates howcharacters were assigned in the conditions.

Reading texts were created based on eachcharacter set, so that learning sessions resembledtraditional CFL classroom practices in whichcharacter learning is integrated with word-basedvocabulary teaching and text comprehension.When the character does not constitute anindependent word (e.g., ‘father’s sister’), thesame compound word containing the character(e.g., ‘father’s sister’) appears in bothconditions. The same texts were used for 1st- and2nd-year participants. The average length of eachtext was 46 characters, with no significant lengthdifference between conditions, t(14)¼ 1.77,p¼ .10. The number of unfamiliar words pertext (including words containing target charac-ters) in both conditions also matched (1st-year:10.5 vs. 9.5 unknown words in the distributed andthe grouped conditions, t[14]¼ 1, p¼ .33; 2nd-year: 7.5 vs. 7.25 unknownwords in the distributedand grouped conditions, t[14]¼ .51, p¼ .62). SeeAppendix B for sample texts in the conditions.

Procedure

The experiment consisted of the following: apretest session, four learning sessions eachfollowed by an immediate test session in fourconsecutive days (with four immediate testsessions in total), a posttest session on the fifthday, and a delayed test session two weeks after thefirst learning session. Since there were eight setsof characters, participants learned two sets ofcharacters in each learning session. All sessionswere conducted in a language lab classroom onthe participants’ campus.

Pretests. In the pretests, participants wereasked to complete a pinyin and meaning


production test, that is, to write the pinyin andmeaning for any of the 48 characters that theyrecognized. Participants also completed a radicalrecognition (RR) and a semantic awareness (SA)test as generalization tasks. The radicals tested arethe aforementioned eight target radicals. Thesetasks are explained in detail in the AssessmentInstruments section.

Learning Sessions. In each session, participantslearned two sets of characters embedded in twotexts. Each class period consisted of two parts. Inthe first 20 minutes of the class period, partic-ipants’ regular instructors conducted class usingword-based vocabulary teaching and text compre-hension questions in both conditions. Oneinstructor taught all 1st-year sessions, while asecond instructor taught all 2nd-year sessions.The instructors did not provide etymological ororthographic explanations for the characters.

In the second part of class, participants learnedthe 12 target characters in that session onindividual computers, with approximately 30seconds per character; that is, participants inthe grouped condition learned 12 charactersfrom two radical groups, for example, six with the

radical and six with the radical, andparticipants in the distributed condition learned12 characters crossing all eight radical groups.The computer consecutively displayed characters’three lexical constituent information elements:form, sound (pinyin, accompanied by a femalenative speaker’s voice pronouncing the charac-ter), and meaning. All participants experiencedthe same variety of encoding methods with a

balanced order to establish a robust orthographicrepresentation of each character.3

Immediate Tests. Immediately after each learn-ing session, participants completed three tasks oncomputers: lexical decision, soundmatching, andmeaning matching. These tasks were enabled bythe E-Prime software. Participants also completeda written character production task, in which theywere asked to write the 12 characters learned inthat session based on a meaning prompt.

Posttests. On the fifth day, participants re-ceived a rehearsal for all learned characters. All48 characters were reviewed one at a time forsix seconds each: In the first three seconds, thecharacter form, pinyin, and meaning were pre-sented each for one second; in the next threeseconds, the character’s form, pinyin, and mean-ing remained on the screen. Participants experi-enced the characters in the same order as theylearned them in the learning sessions. After thisrehearsal, participants took a pinyin and meaningproduction task, in the same format as the pretest,as well as a form production task with all 48characters. They also completed a RR and a SAtask in the same format as in the pretests.

Delayed Tests. Delayed tests included a pinyinand meaning production task, a form productiontask, a RR task, and a SA task.

Assessment Instruments

The tasks in various testing points and theirpurposes are summarized in Table 1. Among

TABLE 1Assessment Measures and Testing Points

Measures

Testing Points

Knowledge/Skill AssessedPretestImmediate

test PosttestDelayedtest

Lexical decision U Visual–orthographic form representation(receptive)

Sound matching U Form–sound linkMeaning matching U Form–meaning linkPinyin production U U U Sound representation and form–sound linkMeaning production U U U Meaning representation and form–meaning

linkCharacter production U U U Visual–orthographic form representation

(productive)Radical recognition U U U Recognition of the radicals’ form in un-

familiar charactersSemantic awareness U U U Knowledge of the radicals’ form, semantic

function and fixed position; application ofthe knowledge in unfamiliar characters

Yi Xu et al. 777

these testing instruments, pinyin, meaning, andcharacter production tasks are common inprevious CFL character research (e.g., Shen,2004). We explain other tasks in detail.

Radical Recognition (RR) and Semantic Awareness(SA). The RR task evaluates how well partic-ipants visually recognize a radical’s graphic formwhen encountering unfamiliar characters. In thisparticular task, radical refers to general subchar-acter components and is not restricted tosituations where they serve semantic functions.We accommodate variations in the graphic form’sspatial position by including keys such as(‘case’), where the target form does not servesemantic functions. The SA task assesses whetherlearners can identify radicals’ shape,meaning andthe positional specificity (Taft et al., 1999) of itssemantic function in unknown characters. Inother words, cannot be a key for a SA taskitem on the target radical of , because doesnot occur in a character-internal position indicat-ing meaning. These two tasks were designedfollowing Chen et al. (2013). There are 16 itemseach (2 for each radical) in each version of thetest.

Strict procedures were taken in design toensure the construct validity of those tests. First,we selected options from the Chinese Orthogra-phy Database Explorer (CODE) (Chen et al.,2011), with characters from participants’ text-books excluded and low frequency charactersgiven priority. Keys for the SA task are charactersdefined as compound characters with

as semantic radi-cals from the database. Second, the distractorsgenerally have two more or fewer strokes than thekey to control for the visual complexity of theoptions (Liu & Han, 2000). Third, heedingsuggestions from King et al. (2004), we createddistractors that reflect errors in different naturesand at different sophistication levels. For the RRtask, we generally supply a Type A distractor thatdoes not contain any component orthographical-ly similar to the radical, and supply two Type Bdistractors that contain components orthograph-ically similar to the target radical, representing anincreasing level of sophistication for participants’perception. For instance, in item (1), is aType A distractor while and are Type Bdistractors because the shape of and resemble. In item (2), is a Type A distractor while

and are Type B distractors because andat the bottom resemble . Several other

examples follow; the numbers preceding the itemstem indicate keys.

The SA task follows the same principles toprovide one Type A distractor and two Type Bdistractors for most items. For instance, in item(6), is a Type A distractor and and areType B distractors. In other cases and whenpossible, we supply one option each for Type Aand Type B distractors, and supply a Type Cdistractor that contains the radical’s form in aposition unassociated with semantic functions. Initem (7), is a Type A distractor, is a Type Bdistractor (with resembling ), and is a TypeC distractor (with occurring in a non-meaningindicating position). Items (8–10) each containthese three types of distractors.

Additionally, we prevented bias among distrac-tors (Liu & Han, 2000) by avoiding selectingdistractors orthographically too close to the key(e.g., and ) and controlled the overalldifficulty in the three versions of tests bymatchingthe number of items with distractors at a highsophistication level.

After designing the tests following strict criteria,we consulted three experienced instructors whotaught in the program from which participantswere recruited. The instructors confirmed thatlanguage used in the items’ stem was appropriateand understandable to participants (Liu &Han, 2000), and that items were at approximatelythe same level of difficulty for participants.

EXAMPLE 2. Sample Items From SA Task

(6) Which word is related to “female” or “woman”?

(7) Which word is related to “sun” or “day”?

(8) Which word is related to “metal” or “gold”?

(9) Which word is related to “wood” or “timber”?

(10) Which word is related to “money” or “cash”?

EXAMPLE 1. Sample Items From RR Task

(1) Which word has the “ ”radical?

(2) Which word has the “ ” radical?





We tested the RR and SA tests’ reliability byasking a group of 45CFL students (with nooverlapwith the participant pool in this study) with onesemester’s learning to complete all the itemsincluded in the tests (48 items each for the RR andthe SA tasks). Cronbach’s alpha was .71 for the RRtask and .83 for the SA task, both higher than thegeneral standard of .70 (Nunnally, 1978), showinggood internal consistency.

Lexical Decision and Matching Tasks. The lexi-cal decision task assessed learners’ visual–orthographic representation of characters. Thestimuli consisted of 24 trials: 12 target characterslearned in that particular session, 6 familiarcharacters not composed of the 8 key radicals,and 6 novel characters containing the 8 keyradicals. For each stimuli type, half were realcharacters and half were noncharacters. Forfamiliar and target characters, noncharacterswere created by deleting or adding a stroke. Fornovel characters, noncharacters were created bymoving radicals to illegal positions. Participantswere asked to judge whether the stimuli was a realcharacter by pressing one of two keys, and torespond as quickly and as accurately as possible.Each stimulus was presented for 1000 milli-seconds (ms) followed by a blank of 3000msbefore the screen moved to the next trial.Participants’ accuracy rate and reaction time(RT) on the 12 target characters were analyzed.

The meaning and sound matching tasks assessparticipants’ form–meaning and form–soundmapping. In sound matching, the computershowed a character’s form for 1000ms, and thendisplayed the character’s pinyin, accompanied by afemale native speaker’s voice pronouncing thesound. Participants were instructed to judgewhether the pinyin and sound represented thecorrect pronunciation of the character. If noresponse was made within 3000ms, the screenproceeded to the next trial. Inmeaningmatching,a character was presented on the screen for1000ms, followed by an English translation.Participants were asked to judge whether thetranslation was the exact meaning of the charac-ter. If participants made no response within3000ms, the screen moved to the next trial.Both matching tasks had twelve pairs createdbased on that day’s learning session, with sixmatching pairs and six mismatching pairs. Accu-racy and RT for matching pairs were analyzed.

RESULTS AND DISCUSSION

For all tasks, mean accuracies and standarddeviations were calculated using the proportion

of correct responses. For lexical decision andmatching tasks, reaction time (RT) results arereported in milliseconds. Below, means are alwaysreported with the distributed conditions first,followed by the grouped condition. Results for1st-year participants are discussed first, followedby results for 2nd-year participants. An overalldiscussion for both proficiency groups is offeredafterwards.

Beginning Learners’ Form/Sound/MeaningRepresentation

We report results from the lexical decision andmatching tasks before reporting results in pinyin,meaning, and character production measures.Pretest results, reported later in the Pinyin,Meaning, and Character Production section, verifythat participants generally had no knowledge ofcharacters’ sound and meaning before learningand that there was no difference betweenconditions.

Lexical Decision and Matching. Accuracies andRT results are reported in Table 2. A one-wayANOVA with character grouping (distributed vs.grouped) as an independent variable was con-ducted on each measure. In lexical decision,there were no differences between conditions onaccuracy (.75 vs. .72), F< 1, or on RT (1291 vs.1195), F(1, 46)¼ 2.30, p¼ .14. At the same time,participants’ accuracy in both conditions wassignificantly higher than the .50 chance level(distributed: t[23]¼ 14.86, p< .01, Cohen’sd¼ 6.20; grouped: t[23]¼ 8.08, p< .01, Cohen’sd¼ 3.37), suggesting that participants developedrobust orthographic representation of the targetcharacters and were able to differentiate themfrom noncharacters.

For sound matching, the distributed conditionhad a significantly higher accuracy than thegrouped condition (.65 vs. .56), F(1, 46)¼ 4.26,p< .05, Cohen’s d¼ .60, while there was nodifference on RT (1225 vs. 1167), F< 1, suggest-ing that grouping characters based on radicalsresulted in inhibition of form–sound link activa-tion. Results in the meaning matching taskshowed a reverse pattern: There was no differencefor accuracy (.72 vs. .73), F< 1, but the groupedcondition had a shorter RT (1131 vs. 1010),F(1, 46)¼ 4.32, p¼ .04, Cohen’s d¼ .60, suggest-ing that grouping strengthened the form–meaning link.

In the grouped condition, the form–soundmatching inhibition is likely due to the ortho-graphic similarity in shared radicals among the

Yi Xu et al. 779

target characters, which always belonged to twoparticular radical groups, making the discrimina-tion task more challenging. On the other hand,the faster decision time in form–meaning match-ing in the grouped condition can be a result of thesemantic relatedness of those characters. That is,learning in a groupmade iteasier for participants to attend to the relatedmeaning of ‘female’ in those characters, and suchassociation cued participants’ recall of thecharacters’ exact meanings. Previous studiesdemonstrated that the visual attributes and thesemantic transparencies of shared radicals affectcharacter recognition in different ways (e.g.,Zhou & Marslen–Wilson, 1999). For instance,Feldman and Siok (1999a) showed that a primecharacter sharing the same radical as the targetcharacter can facilitate recognition, when themeaning of the radical is transparent in boththe prime and target, but the prime inhibitedrecognition when there was semantic inconsisten-cy between the prime and target. These resultsconfirm that radicals can be a psychologicallyrelevant unit of processing in the L2 reading ofChinese (Taft & Zhu, 1997; Wang, Perfetti, &Liu, 2003).

Pinyin, Meaning, and Character Production.Aside from lexical decision and matching tasks,all other testing instruments were applied in threetesting points. Those include pinyin/meaning/character production and two generalizationtests assessing radical knowledge. Table 3 showsdescriptive statistics for all these measures indifferent testing points.

A mixed ANOVA was conducted on thesemeasures with grouping as a between-subjectsfactor and testing points as a within-subjectsfactor. When the difference reached significance,

pairwise comparisons with Bonferroni adjust-ments are reported.

Results reveal that the grouped condition led toa posttest advantage in pinyin and meaning, butthe grouping advantage did not remain in thedelayed test. Specifically, in pinyin production, thegrouping� testing time interaction was signifi-cant (F[2,92]¼ 5.23, p< .01, hp

2¼ .10). A simplemain effect analysis of grouping did not reveal apretest difference, where participants all had a .00accuracy rate (F< 1), and did not reveal adifference between conditions in the delayedtest (.04 vs. .06), F< 1. Meanwhile, the groupedcondition showed a posttest advantage (.18 vs..30), F(1, 46)¼ 5.13, p¼ .03. In meaning pro-duction, the interaction was significant, F(2,92)¼ 6.50, p< .01, hp

2¼ .12. There were nodifferences in conditions in the pretest (.01 vs..01, F< 1) or the delayed test (.11 vs. .14),F(1, 46)¼ 1.29, p¼ .26, but the grouped condi-tion was more beneficial in the posttest (.27 vs..40), F(1, 46)¼ 6.31, p¼ .02.

The simplemain effect analysis of time revealedthat, under both conditions and in both pinyinand meaning production, participants’ posttestperformance was better than their delayed testperformance, which in turn was better than theirpretest performance (ps< .05). Figure 1 illus-trates 1st-year participant accuracy in thesemeasures over time.

These results indicate that the grouped condi-tion had a short-term advantage over the dis-tributed condition in sound and meaningrepresentations. The disappearance of the advan-tage associated with a particular learning condi-tion over an interval beyond 48 hours is consistentwith studies on CFL character learning (e.g.,Shen, 2004; Wang & Thomas, 1992; Xu et al.,2013). Results also indicate that sound and

TABLE 2Mean and StandardDeviations (SD) of Accuracy andReactionTime inLexical Decision andMatchingTasks for1st-Year Students (N¼ 48)

Distributed (n¼ 24) Grouped (n¼ 24)

Mean SD Mean SD

AccuracyLexical decision .75 .08 .72 .14Sound matching .65 .15 .56 .15Meaning matching .72 .15 .73 .14

Reaction timeLexical decision 1290.63 218.97 1194.83 218.46Sound matching 1224.78 283.97 1167.32 201.87Meaning matching 1130.76 235.05 1010.27 159.59

Note. Accuracies were calculated using the proportion of correct responses. Reaction time statistics are inmilliseconds.


meaning representation were subject to retentionlosses without repeated practice, regardless ofhow characters were sequenced. This suggests theimportance of repeated practice to maintain theadvantage induced by instructional interventions.

Character production accuracy was assessedin an immediate test, posttest, and delayed test.Results indicate that the grouped condition did

not offer an advantage in the immediate test, butled to more beneficial outcomes than thedistributed condition in both the posttest anddelayed test. Using a mixed ANOVA, the group-ing� testing time interaction was significant,F(2, 92)¼ 4.34, p¼ .02, hp

2¼ .08. The simplemain effect analysis of grouping did not reveala grouping difference in the immediate test

TABLE 3Mean Accuracies and Standard Deviations (SD) in Production and Generalization Tasks for 1st-Year Students(N¼ 48)


Mean SD Mean SD

PretestPinyin .00 .00 .00 .00Meaning .01 .04 .01 .02

Radical recognition .96 .05 .96 .04Semantic awareness .65 .20 .62 .28Immediate test

Character .20 .15 .21 .15Posttest

Character .15 .11 .24 .17Pinyin .18 .16 .30 .20Meaning .27 .19 .40 .17

Radical recognition .96 .05 .99 .03Semantic awareness .72 .26 .87 .16Delayed test

Character .06 .06 .12 .08Pinyin .04 .05 .06 .07Meaning .11 .12 .14 .07

Radical recognition .97 .04 .98 .03Semantic awareness .78 .24 .91 .12

Note. Accuracies were calculated using the proportion of correct responses.

FIGURE 11st-Year Participants’ Pinyin and Meaning Production Accuracies Over Time

Yi Xu et al. 781

(.20 vs. .21), F< 1, but there were significantgrouping differences in the posttest (.15 vs. .24),F(1, 46)¼ 5.48, p¼ .02, and in the delayed test(.06 vs. .12), F(1, 46)¼ 7.60, p< .01. The simplemain effect analysis of time under the distributedcondition reached significance, F(2, 46)¼ 25.28,p< .01, hp

2¼ .52. There were retention lossesfrom the immediate test to posttest (p¼ .04), andfrom posttest to delayed test (p< .01). The simplemain effect analysis of time under the groupedcondition was also significant, F(2, 46)¼ 24.02,p< .01, hp

2¼ .51. Character production accuracyremained from the immediate test to the posttest(p¼ .27), but there was retention loss fromthe posttest to delayed test (p< .01). Figure 2illustrates the pattern.

Discussion of Beginning Learners’ Form/Sound/Meaning Representation. The above results showthat presenting radical-sharing characters ingroups affected characters’ form, sound, andmeaning representations. We suggest that begin-ning learners, depending on their learningconditions, may encode characters in differentways, such as undifferentiated whole characters orwith attention to various areas of the visual space.Grouping characters with the same radical helpsthem by drawing their attention to the radical as asemantic unit across various characters. When thecharacters with shared radicals are spaced overinstruction, this advantage is lost as memory forthe radical fades. Spacing radical-sharing charac-ters narrowly (such as in the grouped condition)

may require equal or more effort in the learningstage, as shown by the inhibition effect in soundmatching in the grouped condition and the lackof difference between the two conditions incharacter production in the immediate test. Butin the posttest, the advantage of radical-basedgrouping was consistently evident in form, sound,and meaning representations, suggesting thatrelying on radicals as functional orthographicunits in learning is an effectivemethodofmemoryorganization.

The grouping advantage is related to whatCraik and Lockhart (1972) refer to as “deeperprocessing” (p. 679). The depth of processing isthe degree of semantic and cognitive analysisinvolved in learning; deeper processing leads toimprovement in memory. Specifically, learningthrough sheer rote (e.g., copying words) con-stitutes shallow processing, whereas a deeper levelof processing results from associating new infor-mation to one’s past experience, extractingmeaning through pattern recognition, or orga-nizing information into meaningful structures.Taft and Chung (1999) suggest that analyzingcharacters in terms of radicals constitutes deeperprocessing. When presenting radical-sharingcharacters in groups, recurring radicals are likelyto draw learners’ attention to the shape, position,and consistent semantic function of those char-acters. Learners have opportunities to realize thattarget characters can be decomposed into sharedcomponents and distinctive components. For instance,when (‘to marry’) and (‘father’s sister’) are

FIGURE 21st-Year Participants’ Character Production Accuracies Over Time


presented together, it is easily observable thatis the shared component, contributing to thegeneral concept of ‘female’ in these characters,while the two characters need to be differentiatedfrom one another by and , which, inaddition to , serve to make their form–soundand form–meaning links distinctive. In makingthese observations, participants conducted se-mantic and structural analysis of the charactersand made connections between the radical andthe whole character, as well as connections amongthe characters. These elaborations facilitatedacquisition.

Previous studies indicate that beginning learn-ers could not automatically use radical knowledgeto make meaning inferences in unknown charac-ters without probing (Wang et al., 2004). McGin-nis (1999) also reported that beginning learnersmore frequently used repeated writing and“making up a story” (p. 160) than relying onradicals to learn new characters. In other words,beginning learners might benefit from some typeof external aid that would encourage them toapply radical knowledge. Such external aidsmight be achieved through question probing(Wang et al., 2004) or instructor elaboration (e.g.,Shen, 2004). The current study suggests thatpresenting characters in radical-based groupscould be a supportive method that encouragesbeginning learners to autonomously use radicalknowledge in learning new characters.

Thus, to answer our RQ1 regarding beginninglearners’ character form, sound, and meaningrepresentation in the two conditions, the grouped

condition offered a short-term advantage in allthese three aspects of character learning. Thenext section examines results in Radical Recogni-tion and Semantic Awareness tasks.

Beginning Learners’ Radical Knowledge Application

RR and SA tasks measure how well participantscan generalize and apply radical knowledge(including form recognition and meaning associ-ation) to unfamiliar characters. Whereas descrip-tive results were shown in Table 3, Figure 3 belowpresents 1st-year participants’ response accuraciesin RR and SA tasks in the pretest, posttest, anddelayed test.

In 1st-year participants’ performance in RR,there was no evidence of improvement acrosstime and no difference between conditions. Thegrouping� testing time interaction was not sig-nificant, F(2, 92)¼ 1.60, p¼ .21. Mean accuraciesin the conditions were .96 vs. .96 in the pretest, .96vs. .99 in the posttest, and .97 vs. .98 in the delayedtest. Neither the main effect of grouping, F(1,46)¼ 1.80, p¼ .19, nor the main effect of testingtime, F(2, 92)¼ 2.45, p¼ .09, was significant.

The above suggests that beginning learners canvisually perceive the radical’s graphic symbol incompound characters with limited experience inChinese orthography. Participants’ high accuracyrate is consistent with studies reporting earlyacquisition of character’s orthographic structuresamong alphabetic readers as beginning CFLlearners (Shen & Ke, 2007; Wang et al., 2003).Shen and Ke (2007) argued that adult CFL

FIGURE 31st-Year Participants’ Response Accuracies in RR and SA Tasks Over Time

Yi Xu et al. 783

learners can utilize their cognitive maturity anddo not necessarily need to use radical knowledgeto accomplish a task asking for visual identifica-tion of a radical’s graphic form. In this study, thelack of significant results could be explained ifbeginning learners relied on simple visual skillsinstead of orthographic structural knowledge tocomplete the RR task; that is, learners with limitedexposure to the graphic components of charac-ters may merely try to identify packs of visualsymbols in this perception task, without necessar-ily understanding these symbols as orthographicunits. Since adults’ basic visual skill as a cognitivemechanism is unlikely to be affected by a fewhours of learning, treatment in the experimentdid not impact participants’ performance. Later,when analyzing 2nd-year participants’ patterns,we suggest that experienced learners handle theRR task differently.

1st-year participants’ performance in the SAtask showed that the grouped condition led tobetter outcomes than the distributed condition inthe posttest and delayed test. Interaction of thetwo variables (grouping� testing time) was signif-icant, F(2, 92)¼ 3.95, p¼ .02, hp

2¼ .08. Thesimple main effect analysis of grouping did notreveal a pretest difference between conditions(.65 vs. .62), F< 1; but the grouped condition wasmore advantageous in the posttest (.72 vs. .87),F(1, 46)¼ 5.78, p¼ .02, and the grouped condi-tion outperformed the distributed condition inthe delayed test (.78 vs. .91), F(1, 46)¼ 5.72,p¼ .02. In the distributed condition, the simplemain effect analysis of testing time reachedsignificance, F(2, 46)¼ 3.32, p< .05, hp

2¼ .13;the difference between the delayed test andpretest, but not any other paired comparisonbetween testing points was significant (ps> .05).In the grouped condition, the simple main effectanalysis of testing time was significant, F(2,46)¼ 21.16, p< .01, hp

2¼ .48, and both theposttest and delayed test had higher accuracythan the pretest (ps< .01), with no differencebetween the posttest and delayed test (p¼ .89).

In addition to showing the advantage of thegrouped condition in this measure, improvementin the grouped condition from pretest to posttestand delayed test suggests that radical-basedgrouping helped participants make associationsamong the radical’s form, position, and semanticfunction, and to apply such knowledge inunfamiliar characters. Compared to earlier stud-ies demonstrating the role of explicit orthograph-ic instruction in developing learners’ semanticawareness (e.g., Chen et al., 2013), these resultsindicate that radical-based grouping can help

beginning learners derive radicals’ structural andfunctional regularity through implicit learning.

Note that under the distributed condition,although gains were observed from pretest todelayed test, this effect might be attributed toparticipants’ learning beyond the experiment,because no improvement in SA from pretest toposttest was evident in the distributed condition,and participants had 14 additional hours oflearning in their regular curriculum over thetwo-week interval between the posttest anddelayed test. Thus, it is not clear whether learningin the distributed condition enhanced 1st-yearparticipants’ radical semantic awarenessdevelopment.

In sum, to answer our RQ2 regarding begin-ning learners’ radical knowledge development inthe two conditions, we found no differencebetween the two conditions in radical formrecognition, while the grouped condition hadan advantage in developing radical semanticawareness.

Intermediate Learners’ Form/Sound/MeaningRepresentation

Pretest results confirmed that 2nd-year partic-ipants in both conditions had little knowledge oftarget characters’ sound (.01 accuracy in pinyinproduction in both conditions) and meaning(.00 accuracy in meaning production in bothconditions). The same data analysis design for2nd-year participants indicated that charactergrouping did not affect 2nd-year participants asit did 1st-year participants. At the same time,the overall learning outcome showed a patterngenerally comparable to 1st-year participants inthe grouped condition, indicating significantimprovement from pretest to posttest in pinyinand meaning production, and retention loss inthese measures after the two-week interval.Meanwhile, form representation measuredthrough character production remained fromthe immediate test to posttest.

Lexical Decision and Matching. Participants’performance in lexical decision and matchingtasks are reported in Table 4. There were nodifferences between conditions in either accuracyor RT in these tasks: lexical decision accuracy (.79vs. .76), F(1, 38)¼ 1.50, p¼ .23; lexical decisionRT (1105 vs. 1187), F(1, 38)¼ 3.32, p¼ .08; soundmatching accuracy (.67 vs. .62), F(1, 38)¼ 1.47,p¼ .23; soundmatching RT (1143 vs. 1167), F< 1;meaning matching accuracy (.74 vs. .74),F< 1; and meaning matching RT (1081 vs.


1138), F< 1. In lexical decisions, participantresponse accuracies in both conditions weresignificantly higher than chance, confirmingthat participants established visual–orthographicrepresentations of these characters after learning(distributed: t[19]¼ 26.10, p< .01, Cohen’sd¼ 11.97; grouped: t[19]¼ 9.19, p< .01, Cohen’sd¼ 4.21).

Pinyin, Meaning, and Character Production. De-scriptive statistics in 2nd-year students’ pinyin,meaning, and character production in differenttesting points are reported in Table 5. Descriptive

results in the two generalization tasks on radicalknowledge application are also included in thistable.

In pinyin production, mean accuracies for thetwo conditions were .40 vs. .47 in the posttest and.18 vs. .20 in the delayed test. The grouping�testing time (pre, post, delayed) interaction wasnot significant, F(2, 76)¼ 1.01, p¼ .37. Groupingwas not a significant main factor, F< 1, but testingtime was a significant main factor, F(2, 76)¼158.52, p< .01, hp

2¼ .81. Pairwise comparisonswith Bonferroni adjustments indicated that theposttest had higher accuracy than the delayed test

TABLE 4Mean and StandardDeviations (SD) of Accuracy andReactionTime in Lexical Decision andMatchingTasks for2nd-Year Students (N¼ 40)


Mean SD Mean SD

Accuracy rateLexical decision .79 .05 .76 .12Sound matching .67 .15 .62 .11Meaning matching .74 .14 .74 .15

Reaction timeLexical decision 1105.20 174.02 1187.34 101.67Sound matching 1143.25 248.93 1167.08 271.96Meaning matching 1080.97 226.34 1138.39 176.25

Note. Accuracies were calculated using the proportion of correct responses. Reaction time statistics are inmilliseconds.

TABLE 5Mean Accuracies and Standard Deviations (SD) in Production Task and Generalization Task for 2nd-YearStudents (N¼ 40)


Mean SD Mean SD

PretestPinyin .01 .02 .01 .01Meaning .00 .01 .00 .01


Immediate testCharacter .39 .16 .38 .21

PosttestCharacter .39 .19 .43 .18Pinyin .40 .20 .47 .22Meaning .45 .17 .52 .16


Delayed testCharacter .22 .15 .25 .17Pinyin .18 .11 .20 .14Meaning .23 .13 .25 .13


Note. Accuracies were calculated using the proportion of correct responses.

Yi Xu et al. 785

and pretest, and the delayed test had higheraccuracy than the pretest (ps< .01).

In meaning production, mean accuracies forthe two conditions were .45 vs. .52 in the posttestand .23 vs. .25 in the delayed test. The interactionwas not significant, F(2, 76)¼ 1.53, p¼ .22.Grouping was not a significant main factor, F(1,38)< 1, but testing time was a significant mainfactor, F(2, 76)¼ 263.58, p< .01, hp

2¼ .87. Theposttest had higher accuracy than the delayed testand pretest, and the delayed test had higheraccuracy than the pretest (ps< .01).

In character production, means for the twoconditions at the three testing points were .39 vs..38 in the immediate test, .39 vs. .43 in the posttest,and .22 vs. .25 in the delayed test. The grouping� testing time (immediate, post, delayed) inter-action was not significant, F(2, 76)¼ 1.21, p¼ .30.Grouping as a main factor was not significant,F< 1, but testing time was a significant mainfactor, F(2, 76)¼ 58.59, p< .01, hp

2¼ .61. Therewas no difference between the immediate test andposttest (p¼ .42), while both had higher accuracythan the delayed test (ps< .01). Figures 4 and 5respectively illustrate 2nd-year participant accura-cy in pinyin/meaning productions and characterproduction over time, indicating little differencebetween conditions at all testing points.

Discussion of Intermediate Learners’ Form/Sound/Meaning Representation. The above provides ananswer to our RQ3 regarding the form, sound,

andmeaning representation among intermediatelearners: There was no advantage associated witha particular condition in any of these measures.The lack of a grouping effect for 2nd-yearparticipants could be due to two factors. First,learning materials in this study included radicalswith high combination frequencies: based on Fan(2010), those eight radicals form 6 to 31characters in textbooks that 2nd-year participantslearned from before the experiment. Becauseinstructors regularly decomposed characters andprovided explanation of radicals in lectures, 2nd-year participants had considerably more radicalknowledge than 1st-year participants, includingunderstanding those radicals’ semantic functionsand combinability. The second and more impor-tant factor is that learners with more than a year’sexperience in CFL learning develop radicalapplication knowledge and consciously applythis knowledge when learning new characters(Shen & Ke, 2007). Shen’s (2005) survey showedthat, as learners’ proficiency levels increase,learners more strongly perceive the importanceof orthographic knowledge-based strategies suchas relying on known radicals. Critical learning hasoccurred between the first and second year, as2nd-year students have come to perceive charac-ters as the composition of radicals and othersubcharacter components and utilize radicalknowledge in learning new material. This elimi-nates the advantage of close spacing of characterswith shared radicals, as 2nd-year participants

FIGURE 42nd-Year Participants’ Pinyin and Meaning Production Accuracies Over Time


more routinely apply radical knowledge whenencountering new characters.

Intermediate Learners’ Radical KnowledgeApplication

Next we examine intermediate learners’ resultsin radical form recognition and semantic aware-ness. For the RR task, means in conditions at eachtesting time were .95 vs. .97 in the pretest, .99 vs..99 in the posttest, and .99 vs. .99 in the delayedtest. There was no significant interaction, F(2,76)¼ 2.83, p¼ .07. Grouping was not a significantmain factor, F< 1, but testing time was asignificant main factor, F(2, 76)¼ 10.54, p< .01,hp

2¼ .22. There was no difference between theposttest and delayed test (p¼ .61), while they bothhad higher accuracy than the pretest (p¼ .01 andp< .01, respectively). For the SA task, means ineach condition were .74 vs. .75 in the pretest, .88vs. .89 in the posttest, and .92 vs. .93 in the delayedtest. Interaction of the grouping condition andtesting time was not significant, F< 1. Groupingwas not a significant main factor, F< 1, but testingtime was a significant main factor, F (2,76)¼ 33.27, p< .01, hp

2¼ .47. There was nodifference between the posttest and delayed test(p¼ .19), while they both had higher accuracythan the pretest (ps< .01). Figure 6 illustrates2nd-year participant response accuracy in the twotasks in each condition across time.

In sum, to answer our RQ4 regarding radicalknowledge application including radical form

recognition and semantic awareness, neither thegrouped condition nor the distributed conditionhad an advantage over the other. Meanwhile,learning in both conditions resulted in improve-ment in RR and SA tasks, and this improvementremained in delayed tests.

We discuss results in the semantic awarenesstask first. As we argued, 2nd-year participantsgenerally applied radical knowledge in learning,regardless of material sequencing. Thus, theirimprovement from pretest to posttest and todelayed test adhered to the same pattern observedamong 1st-year participants in the groupedcondition. Results indicate that, for experiencedlearners who receive explicit orthographic in-structions regularly in their curriculum, the abilityto infer a radical’s semantic functions in unfamil-iar characters can be further developed throughfrequent exposure to radical-sharing compounds.In an early study, Shen and Ke (2007) observeda plateau phenomenon in learners’ radicalapplication knowledge development during theirsecond year of study, in which improvementappeared to be less significant than in learners’first or third year of study. Shen and Keacknowledged that the plateau phenomenon isnot necessarily fixed in duration and can possiblybe prolonged or shortened depending on prac-tice. Our study indicates that learning a list ofradical-containing compound characters in highconcentration might be an effective way toovercome the potential plateau period for inter-mediate learners. In this experiment, participantsin both conditions experienced six compound

FIGURE 52nd-Year Participants’ Character Production Accuracies Over Time

Yi Xu et al. 787

characters for each radical over four days. Thisrecurrence rate is likely to be more frequent thanlearning in regular CFL classrooms. Accumulatedexperience with radicals’ structural and function-al regularity within a short period of time couldenable learners to make rapid improvement inapplying radical knowledge in unfamiliarcharacters.

Next, 2nd-year participants’ performance alsoimproved in Radical Recognition. This mightseem initially surprising, since no improvementwas observed among 1st-year participants in thismeasure from pretest to posttest. An explanationcan be provided if beginning and intermediatelearners handle the RR task differently. While wesuggested earlier that 1st-year participants reliedmostly on visual perception skills to complete thistask, existing research indicates that learnersfinishing one year of study, in comparison withbeginners, especially value using subcharactercomponents including radicals as a method tolearn (e.g., Ke, 1998; cf., McGinnis, 1999). Resultsin this study so far also suggest that experiencedlearners automatically apply orthographic knowl-edge to process and learn new characters. That is,2nd-year participants perceive characters not as asimple visual symbol, but as the combination ofradicals and chunks with organization principles.To complete this task, those participants werelikely to use strategies such as recognizing familiarradicals/chunks to decompose characters’ struc-tures in meaningful ways. Radical knowledge isnot necessarily needed to visually perceive aradical in a compound character, but it can

facilitate the recognition process (Shen &Ke, 2007). When learners relied on radicalknowledge to respond in this perceptual recogni-tion task, as was the case for 2nd-year participantsin this study, their performance in identifyingradical shape within a compound could improvebecause of increasing experience with the radi-cals’ form and combinational possibilities.

Overall Discussion of Learners in Different ProficiencyGroups

We argued that beginning learners (defined asstudents with no more than a few weeks ofexperience in Chinese orthography) might stilllack insight into a character’s internal structures,whereas intermediate learners may be able toautomatically perceive compound characters asthe logical composition of radicals and othersubcharacter components. This could explainwhy the grouping manipulation affected 1st-yearparticipants but not 2nd-year participants at asignificant level. Evidence for difference betweenbeginning and intermediate learners in characterperception and learning is reported in L1 readingacquisition: A meta-analysis by Yang et al. (2013)summarizing 64Chinese as anL1 studies reportedthat three visual skills (visual perception, thespeed of visual information processing, and purevisualmemory) have amedium correlation (.34 to.44) with reading acquisition in lower grades untilthe 2nd grade, but those visual skills have nosignificant correlation (.12 to .20) with reading

FIGURE 62nd-year Participants’ Response Accuracies in RR and SA Tasks Over Time


acquisition for 2nd to 6th graders. Note that mostL1 children develop an understanding of com-pound characters’ basic configuration in 2ndgrade (Shu & Anderson, 1999). Similarly, Lu(2002) showed that beginning and intermediateCFL learners had different levels of sensitivitytoward characters’ structural configurations. InChang et al. (2014b), we analyzed the correlationbetween participants’ visual skills and learningoutcomes in the current experiment: 1st-yearparticipants’ visual skill of pattern discriminationcorrelated with gains in character form andmeaning learning (r¼ .33, r¼ .34, respectively,ps< .05), but 2nd-year participants’ visual skillsdid not correlate with their gains (form: r¼ .15,meaning: r¼ .12, ps> .05). Such evidence indi-cates that learners who have not yet developed afull understanding of a Chinese character’sorthographic structures may rely on basic visualskills in character perception and learning,whereas those with metalinguistic awarenessregarding a character’s structures analyze sub-character components as orthographic units andorganize them in meaningful ways to learn.

CONCLUSION

For beginning learners, sequencing radical-sharing characters together led to a short-termadvantage in the form, sound, and meaningrepresentations of characters. Radical-basedgrouping can be an instructional method thatencourages learners to analyze characters’ struc-tures and generalize radicals’ structural andfunctional regularities, thus achieving effectivecharacter learning. Intermediate learners rou-tinely rely on radical knowledge to learn, so thatradical-based grouping offered no particularadvantage. Meanwhile, experiencing recurringradicals in multiple compound characters canfacilitate intermediate learners’ further develop-ment in radical knowledge application.

Radical-based grouping in CFL classrooms isunderrepresented in research. While we wouldnot argue for a radical-based approach insequencing new characters in primary CFLteaching materials, the current research indicatesthe benefit of frequently presenting recurringradicals to learners. Several implications forteaching are proposed. First, when teaching atarget character, instructors could refer to radical-sharing characters in previous and subsequentlessons. For instance, when teaching in(‘time’), it is desirable to review and previewcharacters such as (‘evening’ or ‘late’),(‘tomorrow’), and (‘yesterday’), which all

contain as a radical on the left side, withmeaning associations to ‘time’ or ‘day.’ Instruc-tion can also extend to characters such as(‘star’), (‘early’), (‘to be’), and (‘easy’),with the component in a position differentfrom . Whereas students are expected learnthese characters in the first semester followingtext sequences in their theme-based textbooks,instructors can ensure that those characters arepresented together in the classroom to achieve agrouping effect to support form, meaning mem-orization, and radical knowledge development.Second, this study points to the possibility ofachieving orthographic knowledge developmentwithin meaningful context. Reading passages canbe created to contain a range of radical-sharingcharacters, and such reading texts have beenapplied in L1 classrooms and as E-learningmaterials (Chen et al., 2013; Ki et al., 2003).With appropriate character selection, short pas-sages similar to the ones used in this study can beconstructed for CFL students as supplementary orextensive reading materials with the primary goalof developing orthographic awareness. Third,many textbook and workbooks provide exercisesin which learners must recycle learned charactersby grouping them based on shared radicals or bymaking radical-based meaning inferences inunfamiliar characters. This research providesevidence supporting the validity of these exer-cises. Finally, we found that intermediate learn-ers’ radical knowledge is subject to furtherdevelopment. That is, introducing multiple newcharacters containing familiar radicals continuesto be a meaningful instruction method forsecond-year CFL students, as it can expedite thefurther development of their radical applicationskills.

The beneficial effect of character groupingamong beginning learners is likely to depend onsome existing orthographic knowledge amonglearners. Also, since there is variability in how andto what extent radicals contribute to the precisemeaning of a character, generalization based onknown radicals cannot provide exactly accuratemeaning inferences in unknown characters. Inthese aspects, the implications of the study shouldbe interpreted with caution.

ACKNOWLEDGMENTS

This project was funded by the first author’s FY12Small Grant from the Office of Research, University ofPittsburgh. We thank the participating students, the two

Yi Xu et al. 789

language instructors, and the lab assistants for their helpin the study.

NOTES

1 In addition to semantic–phonetic compounds,characters may have internal structures that followother compounding patterns. For instance, associativecompounds (huıyı) are characters that consist of twosubcharacter components that both contribute to themeaning of the character. Six types of charactersfollowing different compounding principles are catego-rized in the etymological dictionary Shuowen Jıezı( ).

2 Most of the 48 characters are semantic–phoneticcompounds with the eight radicals as semantic radicalsbased on Shuowen Jıezı, with the exception of , which iscategorized as a semantic–phonetic compound withas the semantic radical and as an associativecompound (Gao, 2003). According to Xınhua Dictionary( ), both characters are associative com-pounds, and simultaneously acts as a phoneticcomponent for character . Since associative com-pounds are characters in which both radicals contributeto the meaning of the whole character (Gao, 2003), theeight radicals for all 48 characters can be consideredsemantic radicals.

3 All participants in both conditions experiencedreading, writing, character chunk presentation, andstroke counting as encoding methods in differentorders. Since this design applies to all participants, itdoes not affect our analysis of the grouping factor here.Part of the encoding method effect among 1st-yearparticipants is reported in Chang et al. (2014a).

REFERENCES

Brysbaert, M., & New, B. (2009). Moving beyond Kuceraand Francis: A critical evaluation of current wordfrequency norms and the introduction of a newand improved word frequency measure forAmerican English. Behavior Research Methods, 41,488–496.

Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., &Rohrer, D. (2006). Distributed practice in verbalrecall tasks: A review and quantitative synthesis.Psychological Bulletin, 132, 354–380.

Chang, L. Y., Xu, Y., Perfetti, C. A., Zhang, J., & Chen,H.–C. (2014a). Supporting orthographic learningat the beginning stage of learning to read Chineseas a second language. International Journal ofDisability, Development, and Education, 61, 289–306.

Chang, L. Y., Xu, Y., Tseng, A. M., & Perfetti, C. A.(2014b, March). Basic visual skills modulate Chinesecharacter reading acquisition in learning Chinese asforeign language: An in-vivo study. Paper presentedat the annualmeeting of theAmericanAssociationfor Applied Linguistics, Portland, OR.

Chang, S.–J., & Han, M.–J. (2004).

[The effects of different word recognition teach-ing methods for lower grade elementary schoolstudents]. Educational Review, 22, 71–88.

Chase, W. G., & Simon, H. A. (1973). Perception inchess. Cognitive Psychology, 4, 55–81.

Chen, H.–C., Chang, L.–Y., Chiou, Y.–S., Sung, Y.–T., &Chang, K.–E. (2011). Chinese orthography data-base and its application in teaching Chinesecharacters. Bulletin of Educational Psychology, 43,269–290.

Chen, H.–C., Hsu, C.–C., Chang, L.–Y., Lin, Y.–C.,Chang, K.–E., & Sung, Y.–T. (2013). Using aradical-derived character E-learning platform toincrease learner knowledge of Chinese characters.Language Learning and Technology, 17, 89–106.

Craik, F. I. M., & Lockhart, R. (1972). Levels ofprocessing: A framework for memory research.Journal of Verbal Learning and Verbal Behavior, 11,671–684.

DeFrancis, J. (1989). Visible speech: The diverse oneness ofwriting system. Honolulu, HI: University of Hawai’i.

Ellis, N. C. (2003). Constructions, chunking, andconnectionism: The emergence of second lan-guage structure. In C. J. Doughty & M. H. Long(Eds.), The handbook of second language acquisition(pp. 63–103). Oxford, UK: Blackwell.

Everson, M. E. (2011). Best practices in teachinglogographic and non-Roman writing systems toL2 learners.Annual Review of Applied Linguistics, 31,249–274.

Fan, H.–M. (2010). Developing orthographic awarenessamong beginning Chinese language learners: Investi-gating the influence of beginning level textbooks.(Unpublished doctoral dissertation). Universityof Iowa, Iowa City, IA.

Feldman, L. B., & Siok, W. W. T. (1999a). Semanticradicals in phonetic compounds: Implications forvisual character recognition in Chinese. In J.Wang, A. W. Inhoff, & H.–C. Chen (Eds.), ReadingChinese script: A cognitive analysis (pp. 19–35).Mahwah, NJ: Lawrence Erlbaum.

Feldman, L. B., & Siok, W. W. T. (1999b). Semanticradicals contribute to the visual identification ofChinese characters. Journal of Memory and Lan-guage, 40, 559–576.

Gao, S.–F. (2003). [ZhengZhong Chinese dictionary]. Taipei: ZhengZhong ShuJu.

Gobet, F., & Lane, P. C. R. (2012). Chunkingmechanisms and learning. In N. M. Seel (Ed.),Encyclopedia of the sciences of learning (pp. 541–544).New York: Springer.

Guo, L., Zhang, T. (Eds.). (1991).[Theories and

practices in teaching characters with the orthographicallyclassified approach]. Beijing: Educational SciencePublishing House.

Jackson, N. E., Everson, M. E., & Ke, C. (2003).Beginning readers’ awareness of the orthographic


structure of semantic–phonetic compounds: Les-sons from a study of learners of Chinese as aforeign language. In C. McBride–Chang & H.–C.Chen (Eds.), Reading development in Chinese children(pp. 142–153). Westport, CT: Praeger.

Ke, C. (1998). Effects of strategies on the learning ofChinese characters among foreign languagestudents. Journal of the Chinese Language TeachersAssociation, 33, 93–112.

Ki,W.W., Lam,H. C., Chung, A. L. S., Tse, S. K., Ko, P. Y.,Lau, E. C. C., & Lai, S. M. S. (2003). Structuralawareness, variation theory and ICT support. L1–Educational Studies in Language and Literature, 3,53–78.

King, K. V., Gardner, D. A., Zucker, S., & Jorgensen,M. A. (2004). The distractor rationale taxonomy:Enhancing multiple-choice items in reading andmathematics. Pearson Assessment Report. Accessed2 March 2014 at http://images.pearsonassess-ments.com/images/tmrs/tmrs_rg/Distractor_-Rationales.pdf?WT.mc_id¼TMRS_The_Distrac-tor_ Rationale_Taxonomy_Enhancing

Koedinger, K. R., Corbett, A. C., & Perfetti, C. (2012).The Knowledge–Learning–Instruction (KLI)framework: Bridging the science–practice chasmto enhance robust student learning. CognitiveScience, 36, 757–798.

Liu, R., & Han, B. (2000).[Language testing and its methods]. Beijing: ForeignLanguage Teaching and Research Press.

Lu, S. (2002).[A study on the orthographic awareness amongthe students processing alphabetic writing]. Lan-guage Teaching and Linguistic Studies, 3, 53–57.

McGinnis, S. (1999). Student’s goals and approaches. InM. Chu (Ed.), Mapping the course of the Chineselanguage field: Chinese Language Teachers Associationmonograph series (Vol. 3, pp. 151–168). Kalamazoo,MI: Chinese Language Teachers Association.

Nunnally, J. C. (1978). Psychometric theory (2nd ed.). NewYork: McGraw–Hill.

Shen, H. H. (2004). Level of cognitive processing:Effects on character learning among non-nativelearners of Chinese as a foreign language.Language and Education, 18, 167–182.

Shen, H. H. (2005). An investigation of Chinese-character learning strategies among non-nativespeakers of Chinese. System, 33, 49–68.

Shen, H. H. (2010). Analysis of radical knowledgedevelopment among beginning CFL learners. InM. E. Everson &H. H. Shen (Eds.), Research amonglearners of Chinese as a foreign language (pp. 45–65).Honolulu, HI: University of Hawai’i, NationalForeign Language Resource Center.

Shen, H. H., & Ke, C. (2007). Radical awareness andword acquisition among nonnative learners ofChinese. Modern Language Journal, 91, 97–111.

Shu, H., & Anderson, R. C. (1997). Role of radicalawareness in the character and word acquisition ofChinese children. Reading Research Quarterly, 32,78–89.

Shu, H., & Anderson, R. C. (1999). Learning to readChinese: The development of metalinguisticawareness. In J. Wang, A. W. Inhoff, & H.–C.Chen (Eds.), Reading Chinese script: A cognitiveanalysis (pp. 1–18). Mahwah, NJ: LawrenceErlbaum.

Taft, M., & Chung, K. (1999). Using radicals in teachingChinese characters to second language learners.Psychologia, 42, 234–251.

Taft, M., & Zhu, X. (1997). Submorphemic processingin reading Chinese. Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 23,761–775.

Taft, M., Zhu, X., & Peng, D. (1999). Positionalspecificity of radical in Chinese character recogni-tion. Journal of Memory and Language, 40, 498–519.

Tong, X., McBride–Chang, C., Shu, H., & Wong, A. M.–Y. (2009).Morphological awareness, orthographicknowledge, and spelling errors: Keys to under-standing early Chinese literacy acquisition. Scien-tific Studies of Reading, 13, 426–452.

Wang, A. Y., & Thomas, M. H. (1992). The effect ofimagery-based mnemonics on the long-termretention of Chinese characters. Language Learn-ing, 42, 359–376.

Wang, M., Liu, Y., & Perfetti, C. A. (2004). The implicitand explicit learning of orthographic structureand function of a new writing system. ScientificStudies of Reading, 8, 357–379.

Wang, M., Perfetti, C. A., & Liu, Y. (2003). Alphabeticreaders quickly acquire orthographic structure inlearning to read Chinese. Scientific Studies ofReading, 7, 183–208.

Xu, Y., Chang, L.–Y., Zhang, J., & Perfetti, C. A. (2013).Reading, writing, and animation in characterlearning in Chinese as a foreign language. ForeignLanguage Annals, 46, 423–444.

Yang, L.–Y., Guo, J.–P., Richman, L. C., Schmidt, F. L.,Gerken, K. C., & Ding, Y. (2013). Visual skills andChinese reading acquisition: A meta-analysis ofcorrelation evidence. Educational Psychology Review,25, 115–143.

Zhang, T. R. (2012).[Literacy-focused learning, mass reading, step-by-stepwriting]. Beijing: Central Compilation and Trans-lation Press.

Zhou, X., & Marslen–Wilson, W. (1999). Sublexicalprocessing in reading Chinese. In J. Wang, A. W.Inhoff, &H.–C. Chen (Eds.),Reading Chinese script:A cognitive analysis (pp. 37–63). Mahwah, NJ:Lawrence Erlbaum.

Yi Xu et al. 791

http://images.pearsonassessments.com/images/tmrs/tmrs_rg/Distractor_Rationales.pdf?WT.mc_id=TMRS_The_Distractor_Rationale_Taxonomy_Enhancing





APPENDIX AAssignment of the 48 Characters in the Two Conditions

Character sets in the grouped condition Character sets in the distributed condition

Set Code Character Set Code CharacterA1 ‘peddler’ G4 ‘full’A2 ‘accounts’ D1 ‘shack’

1 A3 ‘to be defeated’ 1 B5 ‘smoke’A4 ‘to stand a loss’ H1 ‘suddenly’A5 ‘thief’ C3 ‘diamond’A6 ‘to earn’ E1 ‘wedding’B1 ‘stove’ E4 ‘baby’B2 ‘to warm by fire’ F2 ‘spacious’

2 B3 ‘flame’ 2 B6 ‘candle’B4 ‘firecracker’ A3 ‘to be defeated’B5 ‘smoke’ H5 ‘to endure’B6 ‘candle’ C5 ‘key’C1 ‘silver’ D2 ‘pail’C2 ‘bank note’ C6 ‘to fish’

3 C3 ‘diamond’ 3 F3 ‘to sun-dry’C4 ‘to lock’ A2 ‘accounts’C5 ‘key’ H2 ‘to contract’C6 ‘to fish’ G6 ‘to forgive’D1 ‘shack’ E6 ‘father’s sister’D2 ‘pail’ C1 ‘silver’

4 D3 ‘ladder’ 4 D6 ‘approximately’D4 ‘board’ F1 ‘noon’D5 ‘material’ G5 ‘to lose’D6 ‘approximately’ A4 ‘to stand a loss’E1 ‘wedding’ B4 ‘firecracker’E2 ‘to marry’ F6 ‘dark’

5 E3 ‘daughter-in-law’ 5 C4 ‘to lock’E4 ‘baby’ D3 ‘ladder’E5 ‘lovely’ H3 ‘anxious’E6 ‘father’s sister’ A5 ‘thief’F1 ‘noon’ E2 ‘to marry’F2 ‘spacious’ A1 ‘peddler’

6 F3 ‘to sun-dry’ 6 C2 ‘bank note’F4 ‘to dry in the air’ G2 ‘stuffing’F5 ‘warm’ F4 ‘to dry in the air’F6 ‘dark’ B2 ‘to warm by fire’G1 ‘steamed bun’ E3 ‘daughter-in-law’G2 ‘stuffing’ G1 ‘steamed bun’

7 G3 ‘greedy’ 7 B1 ‘stove’G4 ‘full’ D5 ‘material’G5 ‘to lose’ F5 ‘warm’G6 ‘to forgive’ H6 ‘to worry’H1 ‘suddenly’ E5 ‘lovely’H2 ‘to contract’ G3 ‘greedy’

8 H3 ‘anxious’ 8 H4 ‘to avoid’H4 ‘to avoid’ A6 ‘to earn’H5 ‘to endure’ D4 ‘board’H6 ‘to worry’ B3 ‘flame’

Note. Code is used to illustrate how characters under the same radical group (i.e.,markedby a letter) are assigned to thesame set in the grouped condition and distributed across different sets in the distributed condition. Letters from A toH are associated with the following radical groups: (‘money’), (‘fire’), (‘metal’), (‘wood’), (‘female’),(‘sun’), (‘food’), and (‘heart’), in this sequence.


APPENDIX B

Sample Texts Containing Target Characters inthe Two Conditions

Target characters are in block font, while wordsthat they form are underlined. The two texts sharethe same word (‘father’s sister’).(1) Text in the grouped condition:

‘In 2021, Xiaoai had her wedding. She marriedXiaogao and became a daughter-in-law of the Gao

family. Xiaoai gave birth to a baby. The baby isdainty and lovely, and is now five years old. Shecalls Gao Xiaoyin ‘aunt (father’s sister)’.(2) Text in the distributed condition:

‘Xiaogao’s aunt (father’s sister) works in asilverware shop. She is busy every day, recordingtransactions from morning till noon. She neverloses her capital assets or suffers losses in business.She is a very smart business person.’

2014 ACTFL/NFMLTA/MLJDissertation Support Grants

THEAMERICANCOUNCILONTHETEACHINGof Foreign Languages (ACTFL), the NationalFederation of Modern Language Teachers Associ-ations (NFMLTA), and theModernLanguage Journal(MLJ) invite proposals for supporting graduatestudents in the fields of applied linguistics andlanguage education with small grants at thedissertation writing stage. The purpose of thegrants is to provide resources to graduate studentsat any stage of dissertation writing, e.g., datagathering, familiarization with relevant research,data analysis, and write-up of the findings.

The award amount is $2,500; up to two grantsper year may be awarded.

Applicants must

� have completed by the application deadlineall requirements for the Ph.D./Ed.D. exceptthe dissertation;

� have an approved dissertation proposal with,if applicable, an approved IRB;

� be enrolled in a doctoral program at aninstitution of higher education in the UnitedStates; they need not be citizens or perma-nent residents.

A complete application includes:

� Applicant contact information: Name, physi-cal address, e-mail address, name of appli-cant’s institution of higher education

� Brief statement signed by the dissertationadvisor/dissertation committee chair stat-ing that the applicant has: successfullydefended his/her dissertation proposal;the IRB process has been cleared (ifappropriate to the study); the applicanthas successfully completed all but thedissertation writing at the time of applica-tion for this grant.

� Proposed dissertation title and a descriptionof the proposed research. The descriptionshould include: purpose of the study;theoretical framework(s); research ques-tions; methodology; importance of the study;limitations (maximum two pages, Times NewRoman 12).

Complete applications must be submitted onthe ACTFL website www.actfl.org by September30, 2014. Look for the announcement for theDissertation Support Grants on the ACTFLHomepage, scroll to the bottom of the announce-ment, and submit where it says “here.”

Awardees will receive a check in the amount of$ 2,500. Recipients will be announced at theACTFL 2014 Convention, via diverse electronicand print media, on the web page of theModern Language Journal, and in the journal’sspring issue.

Yi Xu et al. 793

http://www.actfl.org/

the effect of radical-based grouping in character learning ... · the effect of radical-based...

Documents