Jap Double Object and Causative Sents

8/3/2019 Jap Double Object and Causative Sents

1/27

Journal of Psycholinguistic Research, Vol. 27, No. 4, 1998

Processing and Learning of Japanese Double-Object Active and Causative Sentences: AnError-Feedback ParadigmYoshinori Sasaki1,2

Do native speakers always outperform second-language (L2) learners in terms of speechprocessing accuracy? Surprisingly, the answer to this seemingly obvious question is noaccording to the study reported here. Indeed, native speakers sometimes make moreerrors than learners in interpreting their own first-language (L1 ) speech. In this com-petition experiment of the double-object active and transitive causative sentence pro-cessing strategies, six native Japanese speakers and nine English-speaking learners ofJapanese participated. The participants were required to identify the agents of the mainlexical verb ("doers") of a series of Japanese sentences, each consisting of one verband three noun phrases, in which word order and case-marking cues either competedor were consistent with each other. In the first (pretest) and last (posttest) parts of thestudy, participants received no feedback about the accuracy of their responses, whereasin the middle part they received immediate feedback. Th e stimulus sentences were suchthat a listener could determine the semantic role of noun phrases (actor, causer, orrecipient) only by taking into consideration both the case markers and the verbs voice(active vs. causative).Learners of Japanese as a second language (JFLs) demonstrated an evident wordorder bias. Native Japanese speakers also made surprisingly numerous errors, by im-I would like to express my gratitude for comments on earlier versions of this article bythe reviewer of Journal of Psycholinguistic Research and its editor R. Rieber, and byTamar Kaplan, Chihiro Kinoshita-Thomson, Brian MacWhinney, Ak ira Miyake, andMarcus Taft. I would also like to express my gratitud fort voluntary participation of theexperiment subjects, wh o were also my students. Daniel Lepetit shared his phoneticsexpertise with me in the process of material preparation. I also acknowledge that I hadthe opportunity to develop this research plan while I spent the summer of 1993 at theUniversity of Illinois, under the sponsorship of Robert Hart.1 University of New South Wales, Sydney, NSW 2052, Australia2 Address all correspondence to Yoshinori Sasaki, University of New South Wales,School of Asian Business & Language Studies, Sydney, NSW 2052, Australia.

4530090-6905/98/0700-0453$ 15.00/0 1998 Plenum Publishing Corporation


2/27

454 Sasakiposing the canonical case-marker sequence in reconstructing noncanonical sentences.Indeed some of the native Japanese revealed an even stronger word order bias than thelearners, and they committed more errors than learners in interpreting noncanonicalword order sentences. The results are explained in terms of the working memory con-straint. Directions of further research are discussed.

INTRODUCTIONDo native speakers always outperform second-language (L2) learners interms of speech processing accuracy? Surprisingly, the answer to this seem-ingly obvious question is no according to the study to be reported here:Native speakers may commit more errors than learners in a sentence inter-pretation task, when subtle condition parameters are set in a certain way.This study applied the competition model (MacWhinney & Bates,1989) to analyze how native English speakers learn the form-function map-ping of Japanese double-object and causative sentences. According to thismodel, language processing is essentially the competition and/or conver-gence of multiple cues for a limited number of channels. It claims thatsemantic, grammatical,3 and phonological cues are processed concurrently.Earlier competition studies have shown that native speakers of different lan-guages interpret sentences by paying their primary attention to different as-pects (cues) of given speech, and this first-language (LI) comprehensionstrategy often affects their L2 processing in a complex manner. When sev-eral cues are in competition regarding the interpretation of speech (a conflictsentence), the relative strength of the cues determines the interpretation alistener/reader eventually gives. For example, A fence damage s a donkeyprovides a competition between lexical-semantic and word order cues.Japanese Word Order and Case-Marking SystemsSimple Transitive Sentences

In a Japanese sentence, a noun is marked for thematic roles (e.g., sub-ject, object) by the subsequent case marking postpositions: O usually marksits preceding noun for object, and ga typically is the subject marker.Sasaki (1994b) presented a series of Japanese w ord strings to English-speaking learners and native speakers of Japanese, in which three types ofcues (word order, case markers, noun an imacy) were consistent or divergentregarding the sentence interpretations they each suggested. The results in-3 The terms grammar and morphosyntax are used interchangeably in this article.


3/27

Japanese Causative 455dicated that native Japanese speakers characteristically relied heavily oncase-marking cues in interpreting Japanese simple transitive sentences, andnative English learners of Japanese assimilated this case-marker sensitivityfairly promptly: Even beginning learners relied on case markers to a muchgreater extent than on Japanese canonical subject-object-verb (SOV) wordorder schema.Double-Object and Causative Sentences

However, it may be premature to conclude that the approximation ofnative-lik e case-marker-referenced strategies is always a smooth process: AsFig. 1 illustrates, Sasaki's (1994b) set of stimulus materials maintained theone-to-one correspondence between the case-marker form and the semanticrole function (a noun followed by ga is always agent/experiencer; a nounfollowed by o is always patient). This easy-to-process feature should facil-itate acquisition, as Slobin (1979) predicted that: Universal 3: The closera grammatical system adheres to one-to-one mapping between semantic el-ements and surface elements, the earlier it will be acquired (p. 109)On the other hand, there are instances where such simple relations nolonger hold. Japanese double-object active and transitive causative sentencescollectively provide such an instance (see Fig. 2). The following active andcausative sentences, which are phonologically identical except for the finalverb (active kaku vs. causative kakaseru), each expresses a totally differentpropositional meaning:

(4-1) Naomi ga Ken ni tegami o kaku.Naomi SBJ Ken I-OBJ letter D-OBJ to write(Naomi writes a letter to Ken.)

(5-1) Naomi ga Ken ni tegami o kakaseru.Naomi SBJ Ken I-OBJ letter D-OBJ to make write(Naomi makes/lets Ken write a letter.)

Whereas Naomi marked by subject marker ga is the writer of a letter(=doer) in the first, double-object active sentence (4-1), the equivalentphrase in the second, causative sentence (5-1) denotes a causer, and Ken,marked by ni, is the one who was made to write a letter (causee = doer).In other words, it is possible to assign correct semantic roles to these nounphrases only by taking into consideration both their case markers and theverb form (active vs. causative).Th e processing is further complicated when the noun phrases are scram-bled: Whereas (4-1) and (5-1) each provide basic word orders of Japanesedouble-object active and causative constructions (i.e., N-ga N-HZ N-o V)


4/27

456 Sasaki

Fig. 1. Form-function mapping of Japanese simple transitive sentences.

Fig. 2. Form-function mapping of Japanese double-object and causative sentences.

which are presented in elementary Japanese textbooks (e.g., Nagara, 1990),it is possible to order the noun phrases in different orders without changingthe propositional meaning.4 For examples, the following five sentences con-vey the same semantic information as (4-1), each with different pragmaticemphases:

(4-2) Naomi ga tegami o Ken ni kaku.(4-3) Ken ni Naomi ga tegami o kaku.(4-4) Ken ni tegami o Naomi ga kaku.4 A sentence element is more likely to occupy the sentence-initial position when it istopicalized.


5/27

Japanese Causative 457(4-5) tegami o Naomi ga Ken ni kaku.(4-6) tegami o Ken ni Naomi ga kaku.

Comparable word-order variations are possible in causative sentences.Therefore, learners of Japanese as a foreign language (JFL) have toobserve at least the following three principles for properly processing thesetypes of constructions:

1. Semantic functions are not directly mapped onto word orders.2. Semantic roles (e.g., agent, patient) of noun phrases are assigned by thecase-marking particles that follow them.3. Th e semantic roles assigned by case markers depend on the verb form(voice).As was initially stated, a listener/reader has to pay attention to differentparts of a sentence in order to properly carry out this strategy. The strategyis more complex and diffused than the simple case-marker-referenced strat-egy adapted by Sasakis (1994b) JFL learners. In a complexity of form-

function mapping relation like this, JFL learners' adaptation of optimal pro-cessing strategies may be delayed.

Cue Validity Model and Strategy ChangeWhat does the optimization of processing stretegies entail? McDonald's(1989) cue validity model claims that development of comprehension ca-pability essentially is the adjustment process of sensitivity to types of cuecalled (cue strengths). Approximation of nativelike cue strengths allows forproper processing of linguistic input an d thus provides significant momen-tum for language acquisition. In McDonald's formulation, a change of thecue strength occurs when learners realize that their sentence interpretations

are incorrect. In that event, the strength of the cues that suggest a correctinterpretation will increase. If the interpretation is correct, no change in cuestrengths will take place.Accordingly McDonald (1987) reported that native Dutch speakers witha longer history of studying English were more dependent on native-English-like word-order-referenced strategies. Sasaki's (1994b) analysis of JFLlearners revealed a similar correlation between proficiency and case-markerdependency.Heilenman and McDonald (1993) applied this cue validity model tooffer an interesting pedagogical proposal: L2 learners should be exposed tononcanonical sentences, to facilitate their cue strength optimization. Non-conflict (canonical) sentences alone will not bring them very close to thenativelike cue strengths.


6/27

458 SasakiIn the following study, the pedagogical feasibility of this proposal wasassessed by simulating it w ithin a computer-controlled learning environm ent:Participants interpreted a series of noncanonical as well as canonical sen-

tences, followed by feedback to their responses. According to the cuevalidity model, learners should make increasingly fewer interpretation errorsin the course of the feedback learning. In other words, the present studyassessed feedback's short-term effect on the subsequent sentence interpre-tation strategy.This was a rare attempt to incorporate an explicit training componentin a competition experiment: So far, developmental comparisons under thecompetition model framework have been almost exclusively cross-sectionalbetween speakers with different proficiency levels and/or calendar ages,rather than longitudinal and/or interventional (c.f. McDonald & Mac-Whinney, 1991).Research Questions1. Do JFL learners count on w ord order or case marker cues more frequentlyin interpreting causative sentences?2. Do JFL learners count on word order or case marker cues more frequentlyin interpreting active sentences, when those sentences are mixed withcausative sentences?3. Does error feedback lead learners to employ strategies which yield gram-matically correct responses?

In addition to these three research questions regarding JFL learners,native Japanese speakers' overall superiority over learners was anticipated.Namely:4. Will native speakers respond more accurately w ithin a shorter time periodthan JFL learners in all types of sentences?

METHODParticipants

Participants in the study comprised two groups: The Native Japanesespeaker group was made up of six graduate and undergraduate Japanesestudents (Jl through J6) of a large state university in New England of theU.S. They were teaching assistants or voluntary tutors of the institution'sJapanese program. The English-speaking JFL learner group consisted of nine


7/27

Japanese Causative 459American students (El through E9) attending a fourth-semester intermedi-ate-level5 Japanese course at the same institution, which course the presentresearcher was in charge of. Participation of the subjects wa s voluntary.

Timing and LocationThe study was conducted in the spring of 1994. Learners participatedin the study a few weeks after they learned the causative construction in

class. The experiment w as conducted individually in the present researcher'soffice.

Instructional BackgroundsThe learners had been exposed to approximately 280 hours of formalinstruction of Japanese over four semesters before the experiment, or wereconsidered to possess equally high language proficiency.The third- and fourth-semester Japanese courses were influenced by theproficiency-oriented approach (Omaggio-Hadley, 1993) with abundantoral/aural as well as reading activities. These two particular courses at theinstitution were meticulously crafted and intricately coordinated. Whereastheir sequence of introducing grammatical structures and vocabulary itemsfollowed Nagara (1990) to maintain continuity from lower-level courses,classroom activities and assignments in these two courses centered aroundthe two instructors' original course packages (Sasaki, 1994a; Taira, 1994),to better meet the institution's needs and students' future goals.No noncanonical causative sentences were presented to studentsthroughout the course. Use of noncanonical active sentences, if any, wasinfrequent in class.

ProcedureApparatus

The experiment programmed with Cedrus Corporation's SuperLabewas run on a Macintosh LC Hie . The participants' choice responses andresponse time latency of 1-ms accuracy were automatically recorded.5 In the present article, learners' proficiency levels are defined as follows: elementary =less than 150 hours' formal instruction, or comparable fluency; intermediate = between150 and 300 hours' formal instruction, or comparable fluency; advanced = more than300 hours' formal instruction, or comparable fluency. According to these criteria, theintermediate group in Sasaki (1994b) is termed as advanced in the present article.


8/27

460 SasakiSessions

The study consisted of four sessions (pretest; feedback I; feedback II;posttest), which in total took about 25 min. After the experiment, the re-searcher interviewed each participant to solicit introspective protocol. Inaddition, five practice sentences were presented before the pretest. The se-quence of the study was as follows:

Instruction practice (5 sentences) pretest (24 sentences) > in-struction > feedback I (24 sentences) feedback II (24 sentences)instruction posttest (24 sentences) > retrospection.Feedback I and II were consecutively run without an intermission.Instructions to the Participants

Instructions in English were visually presented on the computer mon-itor, and participants read them at their ow n pace by pressing the space barto renew the screen. They had a chance to ask the experimenter questionsbefore pretest, feedback I, and posttest each started.Stimulus Sentences

In each session, participants listened to a series of 24 digitized Japanesesentences recorded in a male voice. Each sentence consisted of two animatenouns followed by a case-marking particle (ga or ni), an inanimate nounfollowed by a direct object marker (o), and the sentence-final verb. Lexicalitems are listed in Appendix A. Each phrase (noun + case marker; verb)was separately recorded and subsequently assembled to make a sentence, toavoid potential phonological biases.The 24 sentences comprised the six sentences of the four types shownin Table I. All sentences were morphosyntactically well formed and seman-tically interpretable, although they did not depict very likely events. Hereare some examples:

Table I. Types of Stimulus Sentences

Case-marker orderCanonicalNoncanonical

VoiceActive

A-ga A-ni I-o VactiveA-ni A-ga I-o Vactive

CausativeA-ga A-ni I-o VcausativeA-ni A-ga I-o Vcausative

A : animate noun; I: inanimate noun; V: verb; ga: Japanese subject marker; ni: Japanese indirect-object marker; o: Japanese direct-object marker.


9/27

Japanese Causative 461(6) Gorira ga raion ni tegami o kaku. (A gorilla writes a letter to alion.)(7) Inu ni pengin ga denwa o saseru. (A penguin makes a dog give acall.)

(See Appendix B for the complete sentence list.)The same set of 24 sentences was reused in the subsequent three ses-sions (feedback I; feedback II; posttest). In total, each participant listenedto 96 (= 24 X 4) sentences, in addition to the five practice items.Task.

Following each sentence, a set of pictures of the six animals and birdswas presented on the monitor (see Fig. 3) Learners were requested to reportwho they thought was the doer 6 (i.e., the agent of the lexical verb,namely, the one who conducted the action of writing or phoning) in eachsentence by pressing the corresponding key among the 10 numeric keys onthe keyboard. For example, if a learner who considered lion as the doer wassubsequently presented the screen image of Fig. 3, then s/he should press6 The term doer was used to refer to the subject of the lexical verb throughout the classespreceding the experiment. Also, it was ascertained in the retrospection session after theexperiment that the participants had interpreted the instruction as such.

Fig. 3. Visual stimuli. Dog, gorilla, lion, panda, pelican, and penguin from the lower lefteach corresponded to 1, 2, 3, 4, 5, and 6 among the 10 numeric ke ys grouped on theright-hand end of the keyboard. (Positions of these animals and birds were constantlyshifted between sentences.)


10/27

462 SasakiKey 3, which was in the corresponding low er right position on the keyboard.If s/he considered panda (upper left position in Fig. 3) as the doer, then s/heshould press Key 4. The positions of the animals/birds were constantlychanged between sentences. There was no specific time limit, although learn-ers were encouraged to respond as promptly as possible.After responding to each sentence, a learner pressed the space bar tosolicit the next sentence. In the second (feedback I) and third (feedback II)sessions, a feedback message was presented after each response: When thelearner provided a correct answer, a message Thats right! was presentedon the screen accompanied by a sound effect. When the response was in-correct, a message Thats wrong! appeared together with a differentsound effect. After one of these feedback m essages, the same set of pictureswas displayed again, and a circle appeared on the correct choice. In theinitial (pretest) and last (posttest) sessions, no feedback was provided.Analyses

The results were analyzed in terms of two measures: namely:1. Response time latency. Both correct and incorrect responses were takeninto account when analyzing the latency data.7 The latency data wasconverted into its natural logarithm before statistically analyses.2. Choice response. Th e choice data were analyzed in the following threeterms: (1) case marker (ga vs. ni) marking the chosen noun, (2) thesentential position of the chosen noun (first vs. second), and (3) correct-ness of the choice (correct vs. incorrect).

The five practice sentences were not included in the statistical analyses.Unless otherwise stated, choice data were excluded from frequency countswhen the pressed key did not correspond to one of the two animals/birds inthe stimulus sentence. Such errors are likely to have stemmed from lexical,phonological, sensory, and/or motor factors, which were beyond the scopeof the present study.A 2 X 2 X 2 X 4 analysis of variance (ANOVA) was run, with thetwo L1 groups (native Japanese vs. native English learners of JFL) as abetween-subject independent variables, two voices (ac tive vs. causative), twocase marker orders (. . .ga.. .ni vs. . . .ni . . .ga) and four sessions (pretest,feedback I, feedback II, posttest) as within-subject independent variables,and the correctness of the response as the dependent variable. There were7 Subsequent statistical analyses of latency data would have been impossible if incorrectresponses were excluded, since some participants never gave correct responses in somecells.


11/27

Japanese Causative 463six replications in each cell. The correctness of each response was coded asfollows: When the key corresponding to the correct age nt animal/bird waspressed, it was coded as +1. When the other animal/bird in the sentencewas chosen, it was coded as 1. Responses which did not fall into one ofthese were coded as zero.8A nother com parable 2 x 2 x 2 x 4 A N OV A w as run w ith th e re-sponse-time latency as the dependent variable.

RESULTSANOVA of Response Accuracy

Participants mostly chose their answers from one of the two ani-mals/birds presented in each stimulus word string: A third animal/bird waschosen less than 4% of the time (85 instances out of the total 2,160 trials.)Session [F (l , 13) = 9.00, p < .001] had a significant effect: Partici-pants' performance generally improved along the time course throughou t theexperiment (Figs. 2 and 3). Case-marker order also had a significant effect[F(l, 13) = 30.86, p < .00001]: Th e noncanonical . . .ni. . .ga sentencesinduced m ore errors than the canonical . . .ga. . .ni ones. Moreover, therewas a significant interaction between case-marker order and session [F(3,39) = 6.46, p < .05]: The accuracy rate in response to noncanonical sen-tences improved substantially through the feedback learning, whereas ca-nonical sentences constantly received high accuracy rates across the foursessions (Table II).Case-marker order also produced a significant interaction with L 1 [ F ( l ,13) = 6.82, p < .05]: Native speakers' performance deteriorated in responseto the noncanonical order more dramatically than learne rs' (Table III). There8 This coding convention was adapted because the repeated-measures AN OV A routineof SAS cannot be executed if the data contain missing values.

Table II. Accuracy Rates (in Percentages) by Case-marker Order and Session (AllParticipants)

PretestFeedback IFeedback IIPosttestTotal

. . .ga. ..ni86.087.490.290.288.4

... ni ... go45.870.577.478.768.1

Total66.179.083.984.578.4


12/27

464 Sasakiwas a higher interaction as well between LI, voice, and case marker [F(l,13) = 7.64, p < .05]: Native Japanese speakers made more errors in re-sponse to noncanonical active rather than noncanonical causative sentences,whereas JFL learners m ade fe wer errors in regard to active sentences acrossthe board.One of the Japanese speakers (J3) reflected that, in the middle of afeedback session after repeatedly receiving negative feedback, she had be-come unsure what a doer meant, whereas her initial understanding was con-sistent with the intent of the study. To confirm that the native speakers'particular difficulty in processing noncanonical active sentences was in effectbefore feedback was presented, participants' error frequencies in the pretestsession were separately counted. Table IV clearly shows the same tendencyin this particular session.On the other hand, voice was not a significant effect [ F ( 1 , 13) = 2.08,n.s.]. Neither did its interaction with L1 reach the significant level [F(l, 13)= 0.788, n.s.].ANOVA of Response-Time Latency

Case-marker order proved to be a significant effect [F(l, 13) = 19.23,p < .001]: Th e noncanonical . . .n i. . .ga sentences (2.81 s) generally tooklonger than the canonical . . .ga. . .ni sentences (2.44 s) to respond to.

Table IV. Accuracy Rates (in

Case-marker orderJFL learnersNative Japanese speakersTotal

Percentages) by Voice,Pretest3Case-Marker Order and LI

. . . ga . . . ni . . . ni . . . gaActive

84.9100.091.0

Causative68.897.180.7

Subtotal77.298.686.0

Active52.036.445.8

Causative Subtotal46.942.945.2

49.539.745.5

in the

Total63.569.866.1

L1 = first-language; JFL learners = learners of Japanese as a foreign language.

Table III. Accuracy Rates (in Percentages) by Voice, Case-Marker Order and Ll. . . ga ni ni . . . ga

Case-marker order Active Causative Subtotal Active Causative Subtotal TotalJFL learners 87.9 83 .2 85.5 77.7 68.8 73.4 79.5Native Japanese speakers 97.2 87.9 92.6 56.7 64.4 60.5 76.8Total 91.8 85.1 88.5 69.1 67.0 68.1 78.4L1 = first-language; JFL learners = learners of Japanese as a foreign language.


13/27

Japanese Causative 465

Voice X Session X LI was also significant [F(3, 39) = 5.69,p


14/27

466 SasakiOn the other hand, Table VII (incorrect choices) reveals latency's gen-eral increase toward the end of the experiment. On average, incorrect choices(3,388 ms) took longer than correct answers (2,354 ms) to be made.No further inferential statistics (e.g., ANOVA) were applied to the re-

sponse latency data tabulated by choice (in)accuracy, because the skeweddistribution of accurate and inaccurate choices across cells (see AppendixD) resulted in a shortage of the degree of freedom.Responses of JFL Learners in the Pretest Session

None of the nine English speakers presented an e rror-free performancein the pretest session: All of them made four or more errors in the 24 trials.Moreover, eight of them made at least one error in response to active sen-tences; only one (E9) provided an error-free performance to active sentences.Overall, a strong word order bias was generally evident in the Englishspeakers responses: When their responses to canonical and noncanonicalactive sentences in the pretest were consolidated, they chose the first noun67% of the time. They chose the second noun 61.6% of the time in responseto causative sentences. Moreover, in response to active ni-ga sentences, theyincorrectly chose the first ni noun 48% of the time. Namely, the word-ordereffect [SO(O)V schema] stood on a par with the case-marker effect (seeTable VIII for results from conflict sentences9).Responses of Native Japanese Speakers in the Pretest Session

Surprisingly numerous errors were committed by the native Japanesespeakers (Table IV). In total, they made errors more than 30% of the time.9 Note that collective tabulation can obscure certain individuals' response patterns. For

example, one JFL learner (E3) chose the first noun with ni in active sentences becauseof the case marker ni. Such idiosyncrasies are masked in Table VIII.

Table VII. Response-Time Latencies (in Milliseconds) by L1 and Voice: IncorrectAnswers: Unweighed Geometric Means (Third-Animal Choices Are Excluded)

JFL learners

PretestFeedback IFeedback IIPosttestTotal

Active2,7754,1654,8798,1213,956

Causative4,2872,7284,3075,2543,998

Subtotal3,5433,2764,4906,4243,980

Native Japanese speakersActive2,2552,3152,8032,7122,464

Causative Subtotal2,4503,0623,3014,2732,603

3,2502,6293,0603,4333,114

Total3,0502,9533,6934,6683,388

LI = first language; JFL learners = learners of Japanese as a foreign language.


15/27

Japanese Causative 467Table VIII. Agent Choice by Cues in Noncanonical Sentences (Values in Percentages)(Word-Order vs. Case-Marker Conflict Sentences, While Other Cues Were Neutralized)

StudySasaki(1994b)

This study[pretest]

Sentence typeSimple transitiveactive (N-o/0 N-0/ga V)

Double-object active(N-ni N-gaN-o Vactive)

Transitive causative(N-ni N-ga N- o Vcausative)

Subjects

Bgn. learnersAdv. learnersNative Japanese

Intmd. learnersNative Japanese

Intmd. learnersNative Japanese

Choicesconsistentwith wordorder cue

(wrong)1st accusative

noun*43.637.510.3

1st nounwith ni

48.063.6

2nd nounwith ga

53.157.1

Choicesconsistentwith casemarker cue(correct)

2nd nominativenoun56.462.589.7

2nd nounwith ga

52.036.4

1st nounwith ni

46.942.9

Bgn. = beginning; Adv. = advanced; Int md. = intermediate.* In half of the cases, the first noun was explicity marked for accusative by o. In other cases, the

first noun wa s unmarked, and instead, the second noun wa s marked for nominative by ga .c In half of the cases, the second noun was explicitly marked for nominative by ga . In other cases,

the second noun was unmarked, and instead, the first noun wa s marked for accusative by o.

Most of those errors took place in response to noncanonical . . . n i . . .gasentences, where they chose the first noun in active and the second noun incausative sentences each. When the six native Japanese speakers' responsesto canonical and noncanonical active sentences in the pretest were consoli-dated, they chose the first noun 82.6% of the time; they chose the secondnoun 77.1% of the time in response to causative sentences. In response tononcanonical active sentences, they (erroneously) chose the first noun 63.6%of the time, which is far beyond 48% by JFL learners. The comparable first-noun choice ratio in noncanonical sentences by Sasakis (1994b) native Jap-anese participants was 10.3% (Table VIII).Effects of Feedback on Performance Change

Figure 4 shows the longitudinal change of accuracy rates of the JFLlearner group in the four stages of the experiment (i.e., pretest, feedback I,


16/27

468 Sasaki

Fig. 4. Changes of accuracy rate by sentence type for learners of Japanese as a foreignlanguage.

feedback II, and posttest). The accuracy rates improved over time in re-sponse to all four sentence types. Overall, the effect of error feedback onstrategy change was confirmed.Th e effects of error feedback on Japanese speakers' strategy changeappeared less impressive (Fig. 5), though this group difference (Group XSession) did not reach a statistically significant level [F(3, 39) = 1.03, n.s.].All six native Japanese speakers made at least one error in the posttestsession, whereas three learners (E4, E5, and E9) presented an error-freeperformance there.

DISCUSSIONUnexpectedly, in the pretest there was only one JFL learner (E9) whoprovided an error-free performance in interpreting active sentences, despitethe fact that the learners should have been familiar with subject marker-gafor the preceding three and a half semesters. Their substantial reliance onword order provided a sharp contrast with the performance of beginning and


17/27


Fig. 5. Native Japanese speakers' Changes of accuracy rate by sentence type.

intermediate learners in Sasaki's (1994b) study in response to simple activeJapanese sentences (noun-noun-verb, or NNV).A retabulation of his original data (Sasaki, 1992, 1994b) revealed thatnative English-speaking beginning and intermediate learners each chose thefirst noun 58.6% and 63.6% of the time, respectively, in response to case-biased animacy-neutral active sentences. These figures were markedly ormarginally lower than 67%, the first-noun choice ratio of English speakersin response to comparable active sentences in the pretest of the present study.Moreover, the first-noun choice ratio of 48% in response to noncanonicalactive sentences (N-ni N-ga N-o V) in the present study's pretest exceededthe counterpart figures of 43.6% (beginning learners) and 37.5% (advancedlearners) in Sasaki (1992). Though inferential statistics are not applicable totest the significance of the difference between these two studies since eachused a different set of linguistic materials, the perceived pattern revealed astronger word order effect in the present study, as Fig. 6 scehmatically il-lustrates the data in Table VIII.This word-order bias was even more prominent in native speakers'performance: As opposed to the initial anticipation, they made staggeringly


18/27

470 Sasaki

Fig. 6. A schematic illustration of word order versus case-marker biases in Japaneseconflict sentences.

numerous errors due to positional biases. Indeed, the native speaker groupwas outperformed by the English-speaking JFL learner group in terms oftheir accuracy in interpreting noncanonical sentences (Tables III and IV).In total, they chose the first noun of active sentences in the pretest82.6% of the time, which w as substantially higher than the 55.2% calculatedfrom the data by Sasaki's (1994b) native Japanese-speaking subjects, inresponse to counterpart, case-biased animacy-neutral active sentences. Hisnative Japanese speakers were subject to a dominance of case-marking cuesover word-order cues when interpreting Japanese simple transitive sentencesin the active voice.This is a striking phenomenon, if the Japanese participants of the pres-ent study (U.S. residents) are representative of the native Japanese-speakingpopulation. Indeed the results from the present study provide a virtual mirrorimage to those of Sasaki (1994b). They also stands in contrast to precedingstudies which reported a positive correlation between L2 proficiency andsensitivity to grammatical cues which are valid in the target language (Kil-born, 1987; McDonald, 1987). In the face of this new finding, Sasaki's(1994b) results have to be reinterpreted within a broader context.One of the native Japanese speakers in the present study who receivedmany error correction messages (Jl) spontaneously started repeating the sen-tences to herself, before she chose the doer noun of each sentence. Thepresent researcher sitting behind her observed that in those repetitions she


19/27

Japanese Causative 471systematically inverted the case marker orders of noncanonical sentences tomake them into canonical ones (e.g., she heard Panda NI raion GA denwao suru and reconstructed Panda GA raion NI denwa o suru) while an in-version in the opposite direction (from canonical to noncanonical) was notobserved. Misrecollections of noun orders and verb voices were not ob-served, either.The present researcher had similar experiences of incorrectly recallingcase marker sequences when he ran the experiment with himself, despite thefact that those case markers were clearly audible.Essentially, many of the errors made by the native Japanese speakersseemed due to their imposing the canonical . . .ga. . .ni schema as the tem-plate when trying to retain/recall . . .ni. . .ga sentences: They responded toerroneously recollected case-marker sequences, thus yielding the positionalbias. When they chose the doer noun, the original case-marker sequence wasoften inaccessible to them.

Are Sentences Processed Differently in Different Contexts?Sasaki (1992, 1994b) and the present study provide marked differencesin terms of comparable participants word-order dependency in active sen-tences. This suggests that exposures to causative sentences affected the

learners' interpretation of subsequent active sentences. This hypothesis couldbe more rigorously tested by having learners process the same set of sen-tences under two different conditions (one voice at a time vs. active-and-causative mixture).This possible environment dependency of sentence processing strategycoincides with a common experience by language learners: Less proficientlearners often find it difficult to smoothly access a pertinent morphosyntacticschema to interpret a linguistic structure embedded in a text/speech, whichform they are supposed to be familiar with through a grammar class.Miyake (personal communication, 1992; 1994 Miyake & Friedman (inpress) presented an interesting perspective concerning this phenomenon: Heexplained the dominance of noun animacy over word-order cues reportedby Sasaki (1991), by hypothesizing that local-cue-referenced strategies areeasier to adapt because of their resource-thrifty nature: A local-cue-refer-enced strategy (e.g., noun animacy, case marking) does not occupy as muchworking memory capacity as a diffused one (e.g., word order, noun-verbagreement), once identifying the subject noun is the sole task. This hypoth-esis effectively explains the significant role of case markers in Sasaki(1994b) as well.The case-marker-cues were not as influential in the present study, pre-sumably because they did not serve as local strategies in the given condition:


20/27

472 SasakiIn order to reach a correct sentence interp retation, case cues must be retainedin memory until the sentence-final verb is presented. Thus the relative ad-vantage of the case-marker cues over word-order cues in terms of memoryefficiency diminished.

Granting that the case-marker-referenced strategies in the presentstudy's design were not as memory-efficient as in Sasaki's (1994b), therestill remains a question why the word-order cues so overwhelmingly dom-inated case marker cues in many participants' minds, among them those ofthe native Japanese speakers. This issue will be discussed in the followingsection.

Is Word Order a Prepotent Cue?Obviously, the strong word-order bias native Japanese speakers re-vealed must have developed through their massive exposure to the Japaneselanguage speech over their lifetimes, which corpus contains numerous ca-

nonical S-ga O-ni O-o V sentences.10 However, the linguistic input (amountof exposure) alone, at best, provides no more than a partial explanation fortheir heavy reliance on word order over case markers, because the ubiquitousJapanese case markers have an even higher cue validity than word order.Therefore, it would be prudent to suspect that word order's dominance ispartly grounded in the human cognitive system.This proposal might sound counterintuitive to some readers, who areaware that the rigid word-order-based system typical of English is typolog-ically exceptional (MacWhinney & Bates, 1981). Sasaki's (1994b) resultsalso suggest that case markers rather than word order is an easier cue tolearn to employ.

Cue Assignability and Vulnerabilities of StrategiesAt the ,same time, it is well known that Creole languages are highlydependent on word orders in assigning thematic roles, in the absence of richcase-marking devices. As was hinted in the above-cited retrospection by

10 As a matter of fact, it is possible that JFL learners' relative sensitivity to case-markerdislocations over that of native Japanese speakers was due to their lack of past exposureto noncanonical sentences: Encounters with . . . ni . . . ga sequences during the exper-iment might have provided violations of the familiar Japanese sentence schema whichwere memorable novel episodes to them, as memory research suggests. Th e nonca-nonical sequence was not nearly as irregular to the mind of native speakers (though itis unusual), and thus easily slipped their mind during the challenging processing task.


21/27

Japanese Causative 473some learners, one-syllable case-marking particles are difficult to retain inthe course of a demanding on-line speech processing task, unless their se-mantic functions are immediately determinable; content word-order infor-mation is more salient and easier to retain.

In terms of cue assignability, or memory load, namely, the amountof material that must be held in memory before a meaning assignment canbe made (Bates & MacWhinney, 1989, p. 58), morphological and case-marking cues under some circumstances can be more costly than word-ordercues. Indeed, this notion of processing resource constraint plays an increas-ingly important role in the recent competition model literature. The follow-ing characterization of morphological cues by Bates and Wulfeck (1989)applies to Japanese case-marker cues as well:...the assignability d imen sion shou ld have its biggest impact on closed-classmo rph olog y....A nd given the referential/predicative functions played by con-tent words, most of the communicative work an item must carry out canbe completed quickly as soon as the word is identified. By contrast, mostclosed-class morphem es must be bound to a governing word or phrase. . . .Thismeans that closed-class m orphemes make heavier demand of short-term mem-ory resourceswhich is problematic, given the fact that these items are lowin imagery and other factors that make an item easy to keep in mind. (p. 366)

The robustness of content word-order information under suboptimalsituations is evidenced by neurolinguistic studies as well: Bates, Friederici,and Wulfeck (1987) reported that brain damage impairs patients' sensitivityto morphological cues more strongly than their sensitivity to word-ordercues. Also, MacWhinney, Osman-Sagi, and Slobin (1991) reported Turkishaphasic subjects reduced sensitivity to case-marking cues, as opposed tothe resilience of word-order cues by English-speaking patients.

Effects of Feedback on Processing Strategy ChangeThe significant effect of Session on JFL learners accuracy rates suggeststhat error feedback provides a positive effect on performance: In a futurestudy, it would be desirable to use a control group which receives no errorfeedback throughout the experiment, to tease apart the effects of replicationversus feedback.On the other hand, the native Japanese speakers appeared as relativelyslow in their decrease of error rates throughout the error feedback learning.Figure 5, in contrast to Fig. 4, reveals that the strategy shifts of nativeJapanese speakers w ere not as sweeping as those of native English speakers.This can be explained in terms of their higher level of absolute cue strengthvalues. According to the cue validity model, the change of the cue strengths


22/27

474 Sasakiis always incremental: Weakening of a certain cues effect is explained interms of the increase of competing cues' strengths. After their life-time ex -posure to the language, the native Japanese speakers' cue strengths hadreached a very high level. The 48 feedback messages throughout this ex-periment would have made no more than a tiny bump on that monolith. Theabsolute values of JFL learners cue strengths were much lower, and thusthe short session was able to change their performance considerably.In this regard, those numerous native speaker errors may not be assurprising as they would seem: It is well known that native speakers areoften poor proofreaders of typographical errors. Their top-down processingis so efficient and influential that they tend to see a text as it should read,rather than as it does read (in terms of pragmatics, grammar, orthography,etc.). Top-down processing can be even more influential when the text ispresented orally, which does not allow for backtracking. The heavier pro-cessing load imposed in the present study would also have contributed tothe greater role of an automatized routine.Presumably this is why native Japanese speakers showed poorer per-formance in response to no ncanonical active rather than causative sentences(Table III): The instantiation of the active sentence -ga -ni schema wasparticularly highly automatized because of the native speakers' numerousencounters with them, to the extent that the fragile memory traces of case-marker positions were easily pushed away. Their generally faster responsesto active rather than causative sentences (Appendix D) are also consistentwith the hypothesis that the active sentence schema is stronger.11 Indeed theresponses to noncanonical causative sentences were slowed down after thenumerous error messages, whereas responses to noncanonical actives re-mained speedy: Native speakers continued to make position-biased errorsswiftly, which typifies an automatic behavior.It would be reasonable to hypothesize that a word-order-referencedrather than a case-marker-referenced strategy can be more thoroughly au-tomatized, because the form er is basically a sequential process (e.g., Thefirst noun is . ..; the second noun is . . .) which involves only minimalconditional branching. On the other hand, the latter strategy requires inces-sant detection of particular features of incoming speech segments, followedby each accordingly different processing routines (e.g., If a noun is fol-lowed by ga , then . . .; if it is followed by o, t hen . . . ) .If this automatization is a valid account for the native Japanese speak-ers' numerous interpretation errors, those errors would paradoxically reflecttheir native-level proficiency, which is characteristic of an optimal allocation11 Native Japanese speakers' longer latencies to canonical active (1,909 ms) than to caus-ative (1,785 ms) in the pretest provided an exception, which is difficult to interpret.


23/27


of mental resources. It delivers efficient speech comprehension under normalconditions. Under the unusual condition of the current study (i.e., an unu-sually high ratio of noncanonical sentences without extragrammatical clues),however, it induced many errors.Why Can Native Japanese Speakers Perform Well in Real Life?

It should be explained here that Japanese noncanonical sentences donot always present as much comprehension difficulty in real-life commu-nication: It has not been reported that native Japanese speakers often mis-interpret such sentences presented in normal conversation. One possibleexplanation would involve the form-function mapping schema: In a normalcorpus, the majority of sentences are active, and only occasionally are othervoices used. Therefore, listeners may initially assume all sentences to beactive: In most cases, it is safe to assign an agent/experiencer role to a ganoun. This assumption of a one-to-one form-function-mapping relation re-duces the processing load, making it easier for listeners to retain case-markerpositions. Only when they later realize that the sentence has a nonactivevoice, may they invoke an off-line strategy to reassign semantic roles. Thishypothesis triggers a suspicion that the unusually high ratio of causativesentences in the present study might have forced the participants to abandonthe heuristics they would have otherwise employed. In that particular event,they had to delay role assignments until the very end of the sentence.Also, the poverty of extragrammatical clues in the present study mighthave made the task harder: In normal communication, such additional in-formation is typically available. Presumably native speakers take advantageof some additional clues which are readily available in real-life speech toaccomplish their seemingly flawless performance. It is likely that such cluesinvolve meaning information (pragmatics and/or semantics), and possiblyphonology (intonation, stress, pause) as well. To test this, the current re-search paradigm should be extended to a discourse level, where such prag-matic/semantic clues are readily available.Pedagogical Applications and Implications

The present study has provided mixed results regarding one of its orig-inal aims of exploring the pedagogical application of the competition ex-periment paradigm: On the one hand, many learners correctly interpretedvirtually all sentences toward the end of the experiment, which took lessthan 30 min. The obvious effect of feedback learning on strategy changeencourages us to envision an implementation of the present study's paradigmin foreign language courseware, perhaps in the natural context of a com-


24/27

476 Sasaki

municative reading/listening activity: namely, presenting a mixture of ca-nonical and noncanonical sentences to learners, followed by feedback totheir interpretations. This can be efficiently carried out within a computer-controlled learning environment, as was prototyped in the present study.On the other hand, however, native Japanese speakers' surprisinglypoor performance casts serious doubt upon the validity of this particularcompetition experiment design as a measurement tool of Japanese languageproficiency. Before an all-out instructional application of the current researchparadigm is attempted, therefore, it is essential to elucidate what underliesnative speakers' seemingly flawless performance in the real-life communi-cation. That insight will help reconstruct an authentic speech situation withina classroom/laboratory, so that the ecological validity of a psycholinguistic

experiment as a simulation of human communication is enhanced. It is likelythat pragmatic/semantic clues play a crucial role in people's minds, whichthe current study did not systematically manipulate.

CONCLUDING REMARKSThe present study provides several insights into the nature of humanspeech processing: The major findings are as follows:

1. There were considerable word order biases in JFL learners responses toactive and causative sentences.2. Native Japanese speakers were subject to even stronger word-order bi-ases, to the extent that case-marking cues were overwhelmed.3. JFL learners made fewer errors after the error feedback learning.4. Native speakers' decrements of errors throughout the feedback trainingwere less impressive.

On the other hand, the generalization of these findings is still limitedbecause of some remaining questions, including the external validity of re-sults, the possible significant effects of discoursal pragmatic and/or semanticcues, listeners' psychological set on com prehension strategies, and the long-term effects of error feedback learning. A replication study with a largernumber of subjects including a control group will be necessary to test thereplicability of the tentative conclusions presented above.


25/27


APPENDIX ALexical Items in the StringsAnimate nouns

Case markers

Object + verb

gorira (gorilla) inu(dog)raion (lion) panda (panda)perikcm (pelican) pengin (penguin)ga (subject marker)ni (indirect object marker)denwa o suru/saseru (to give/make give a phone call)tegami o kaku/kakuseru (to write/make write a letter)

APPENDIX BStimulus sentences. (The following set of sentences was used in the four sessions

once each, in the order of blocks. Within each block, presentation order was randomized.)These sentences are all grammatical (i.e., morphosytactically well formed) andsemantically interpretable.

Block 1perikan-ga

raion-gainu-ni

gorira-niperikan-ga

raion-gainu-ni

gorira-ni

inu-nipanda-ni

penguin-garaion-ga

inu-nipanda-ni

penguin-garaion-ga

denwa-otegami-odenwa-otegami-odenwa-otegami-odenwa-otegami-o

sumkakusumkaku

sasemkakaseru

sasemkakaseru

Block 2imt-ga

raion-gaperikan-nipenguin-niinu-ga

raion-gaperikan-nipenguin-ni

gorira-nipanda-ni

inu-garaion-gagorira-nipanda-ni

inu-garaion-ga


surukakusumkakusaseru

kakaserusaseru

kakaseruBlock 3

penguin-garaion-ga

perikan-nigorira-ni

penguin-garaion-ga

perikan-nigorira-ni

gorira-nipanda-ni

inu-garaion-ga

gorira-nipanda-ni

inu-garaion-ga


surukakusurukaku

saserukakaseru

saserukakaseru


26/27

478 Sasaki

APPENDIX CSummary of Choice Results by Voice, Case-Marker(Accuracy Rate), all in Percentages order, Session, and L1

VoiceActive

Case-marker orderEnglishspeakers(n = 9)

Japanese

Total

PretestFeedback IFeedback IIPosttestSubtotalPretestFeedback IFeedback IIPosttestSubtotal

.. ga . . . ni84.9%87.0%90.0%89.8%87.9%

100.0%94.4%97.2%97.2%97.2%91.8%

Causative. . . ni . . . ga Subtotal .

52.0%79.2%92.5%86.3%77.7%36.4%55.6%66.7%66.7%56.7%69.1%

68.7%83.0%91.3%88.0%85.5%69.6%75.0%81.9%81.9%77.2%80.4%

. . ga . . .ni68.8%88.0%86.5%88.5%83.0%97.1%80.0%88.6%86.1%88.0%69.1%

. . . m . . . ga Subtotal46.9%69.4%75.6%83.7%68.8%42.9%75.8%67.6%72.7%64.4%67.0%

57.7%78.8%81.4%95.6%76.1%70.0%77.9%78.3%79.7%76.4%76.3%

Total63.5%68.1%68.1%87.1%79.5%69.8%76.4%80.1%80.9%76.8%78.4%

REFERENCESBates, E., Friederici, A. D., & Wulfeck, B. (1987). Comprehension in aphasia: A cross-linguistic study. Brain and Language, 32, 1967.Bates, E., & MacWhinney, B. (1981). Second-language acquisition from a functionalistperspective: Pragmatic, semantic, and perceptual strategies. In H. Winitz (Ed.), An-nals of the New York Academy of Sciences of conference on native & foreign lan-guage acquisition (pp. 190-214). New York: New York Academy of Sciences.Bates, E., & MacWhinney, B. (1989). Functionalism and the competition model. In B.MacWhinney & E. A. Bates (Eds.), The cross-linguistic study of sentence processing(pp. 3-73). New York: Cambridge University Press.Bates, E., & Wulfeck, B. (1989). Crosslinguistic studies of aphasia, in B. MacWhinney& E. A. Bates (Eds.), The cross-linguistic study of sentence processing, (pp. 328-371) New York: Cambridge University Press.Heilenman, K. L., & McDonald, J. L. (1993). Processing strategies in L2 learners ofFrench: The role of transfer. Language Learning, 43, 507-554.Kilborn, K. (1987). Sentence processing in a second language: Seeking a performancedefinition of fluency. Unpublished doctoral dissertation, University of California, SanDiego.Kilborn, K. (1991). Selective impairment of grammatical morphology due to inducedstress in normal listeners: Implications for aphasia. Brain and Language, 41, 275-288.MacWhinney, B., & Bates, E. (Eds.). (1989). Th e crosslinguistic study of sentence pro-cessing. New York: Cambridge University Press.MacWhinney, B., Osman-Sagi, J., & Slobin, D. I. (1991). Sentence comprehension inaphasia in two clear case-marking languages. Brain and Language, 41, 234-249.


27/27

Japanese Causative 479McDonald, J. L. (1987). Sentence interpretation in bilingual speakers of English and

Dutch. Applied Psycholinguistics, 8, 379-413.McDonald, J. L. (1989). The acquisition of cue-category mappings. In B. MacWhinney

& E. Bates (Eds.), The crosslinguistic study of sentence processing, (pp. 375-396)New York: Cambridge University Press.

McDonald, J. L. & MacWhinney, B.. (1991). Levels of learning: A comparison of con-cept formation and language acquisition. Journal of Memory and Language, 39,407-430.Miyake A. (1994). Toward a unified theory of capacity constraints: The role of working

memory in complex cognition. Cognitive Studies: Bulletin of the Japanese CognitiveScience Society, 1, 43-62.

Miyake, A. & Friedman, N. (in press). Individual differences in second language profi-ciency: Working memory as Language aptitude. In, Healey, A. & Bourne, L.(Eds.), Foreign language learning: Psycholinguistic studies on training and reten-tion. Mahwah, NJ: Erlbaum.

Nagara, S. (Ed.). (1990). Japanese for everyone. Tokyo, Japan: Gakken.Omaggio-Hadley, A. (1993). Teaching language in context (2nd ed.). Boston, MA:Heinle & Heinle.Sasaki, Y. (1991). English and Japanese interlanguage comprehension strategies. AppliedPsycholinguistics, 12, 4773.Sasaki, Y. (1992). Paths of processing strategy transfers in learning Japanese, an d Eng-lish as foreign languages: A competition model approach. Unpublished doctoral

dissertation, University of Illinois at Urbana-Champaign.Sasaki, Y. (1994a). A study guide an d exercise book for Japanese for everyone (Les-

son 14~26). Unpublished course packet, University of Massachusetts.Sasaki, Y. (1994b). Paths of processing strategy transfers in learning Japanese andEng-lish as foreign languages: A competition model approach. Studies in Second Lan-

guage Acquisition, 16 , 329350.Slobin, D. I. (1979). Psycholinguistics (2nd ed.). London, England: Scott, Foresman.Taira, N. (1994). Intermediate Japanese role-play kits. Unpublished course packet, Uni-

versity of Massachusetts.

Jap Double Object and Causative Sents

Documents