Children use canonical sentence sch A crmslinguistic study of word

Cognition, 12 (1982) 229-265

Children use canonical sentencesch A crmslinguistic study of word order and infl L DAN

I. SLOBIN”

University of CMiomia at Berkeley THOMAS

G. BEVER

ColumMa University “We speak .. . by certain well-practiced schemes,-senftence-skeletons that require but the variation of a few words from utterance to utterance.” Leonard 13loomfield The Study of Lmgvage (New York: lienry Holt, 1914, p* 190)

Abstract We propose that children consftruct a canonical sentence schema as a p,eltminary organizing structure for language behavior. The canonical se?ttence embodies the typical features of complete clauses in the input language, and seSrves as a framework for th!e application of productive and pefcizptual strategies. ThIecanonical sentence schema offers a ftAnctiona1 explanation o.f word-order and inflectional strategies based on the chir’lci :slattempts to quickly master basic communication skills in his or her language. The present research explores sensitivity to the canonical”sentence form and to word-order and inflectional perceptual strategies for comprehending simple transitive sentences in monolingual children aged 2;0 to 4,:4!In four languages: English (ordered, uninflectional), Italian (we,akly ordered, weakly inflectional), Serbo-Croatian (weakly ordered, inflectional), Turkish (minimally ordered, inflectional). The results show that children fail to respond systematically to sequences that violate the canonical sentence -form of their particular language. They develop distinct word-order and inflectional strategieis appropriate to the regularitfes of their language. The early behavioral emergence *The research reported here is part of the Berkeley Crossling~istic Acquisition Project, carriedout with support from the WilliamT. Grant Foundation ar:ldthe NMxM Science Foundation to the In&itute of Human Learning (Dan L Slobiq ?rinindp&bveskgator) and from the National Institute of’ Mental Health to the Language-Behavior Research Laboratory, University of California, Berkeiley. Ayhan Aksu, Francesco Antinucci, Susan Ervin-Tripp, Judith R. Johnston, and Ljubica Radulovi6 collaborated with us in designing the investigation. We gratefully acknowledge the labors of our testers: Penny Boyes-Braem, Judith R, Johnstca, and Gail Loewenstein Holland in the United States; Rosanna Bosi and Wanda Gianelli in Italy; Ljubica Radulovid and Emilia Zaloui6 in Yugoslavia; and Alev -Ala& and Ayla Algar in Turkey. Our thanks also go to LaurieWagner,who drew the figures and prepared the tables. Order of listing of co-authors of this paper is based on the universal linguistic constrahlt that two u’ zances cannot be produced simultaneously. Reprint requests should be sent to D. Slobin, Psycho&.,jr Department, University of California, Berkeley, Cal. 94720, U.S.A. 001~b277/82/06022937/$09.50

0

Elsevier Sequoia/Printed in The Nartherlmds

approprhzte canonical sentences and prmessin,g strategies 8zI foundatiopt for linguistic constrairzlson the surjbce

of devices of word order, inflection, and prosody to tities in sentences and to orient sentences to the flow ular lanDgages rely more heavily on one or another device d pragmatic functions. It is the task of the st-12devices and the range of form-function For example, a child acqui.ring English must f a noun in relation to a role of that noun in ~1senouns and verbs are not a an utterance. Iss2zesof othier ways, requiring such as syntactic variphrase, and prosody. A child acquiring quite different conclusions, since sufsemantic roles of nOi.ins, while simple tions without altering the semantic and pragmatic functions of basic e course of everyday communication, word onder, are presented in complex interaction; furthermore, is guided by nonlinguistic factors of context and the English-speaking child -to have come to opposite order in their respective languages. Such ated by two types of factors: (1) attenctior, of ‘privileged’ features ce types as diagnostic, or language. In ascribing such assume an ability to identify canonical we lowest in terms of what Given has ). In %tering such canonptions about the backener. The neutralsentence type which transmits active affmative declarative. In regard to of ow study-such forms tend to have

Children use cam-mica!sentence schetnas

231

definite, human agents which function as discourse topics; finite active verbs; and item arrangements exhibiting the basic word order of the language- - for example, SVO in English and SOV in Turkish. To the extent th,at children receive and orient to such sentences, they are provided with linguistic models which most clearly reveal basic schemas of syntactic organization in the language. Canonical forms are not the most frequent, since input contains a high proportion of questions, imperatives, fragments, and focused forms. And even among declaratives, canonical forms are not always in the majority --for example, although SVO sentences predominate in English, in Turkish only 48% of simple sentences are in canonical SOV order in our sample of adult speech to preschool children. What is important is that they require the least processing of implicit discourse presuppositions in addition to the basic semantic content of the sentence. We do not have data on children’s processing of canonical sentence forms in natural discourse, but our cxperimental resuits suggest that children extract scahemas of canonical sentences and use such schemas to guide comprehension. In devclaping a schema for the canomcal sentence type, the child rn*rybe predisposed to orient to particular linguistic features. For example, earlier work on, English development focused on sensitivity to word-order regular_ ities, especially the agent-first and SVO order of English. The simplest canonical sentence schema for an English-speaking clhild to ;urive at would be one based on word order. And, indeed, in an extensive series of studies of scanterrce comprehension in English (beginning with Fraser, Bellugi, and Brown in 1963; summarized recently in Bridges, 1980) it has been repeatedly demonstrated that children extract the basic SVO order of that language. Clear evidence comes from a common developmental phase, sometime in the fourth year, in which the noun-verb-noun sequence of passive sentencee, is interpreted as SVO rather than the appropriate OVS (e.g., Bever, 197@, b; Maratsos, 1974). On the basis of such work, Bever (197Oa, 6) proposed ti basic inductive capacity to form perceptual strategies which assign syntactic functions according to sequential arrangements of words. Applied to English, such a capacity woulcl result in a language-specific processing strategy: ““Any Noun_ Y.1 v Grb-Noun (NVN) sequence within a potential iniemal unit in the surface structure corresponds, to ‘actor_actrTon-object”’ (Bever, 197k p. 298). However, an acquisitional model based on a predisposition for canonical word-order strategies encounters two basic difficulties: (1) HOW does the child determine the functions of other devices, besides word order, in languages like English? (2) How does the child acquire languages which use variable word order? The fast question can be generalized to the problem of discovering means

o:.$cal strategies. In order for English-speaking children d passive sentences, they must eventuahy note that the contravened by a participial form of the verb and a &wing noun. Other linguistic devices, however, such as not contravene the SVO schema. Orientation to ne cannot lead the child to discover the range of funcr constellation of word order, inflection, the neutral form. Such facts can only be discovwhat is variable in comparisons of caat is to say, the child must record what main constant in sentences across the range of presupn our research we seek to clarify the role played by of canonical sentence forms, asking whether children predisposed to attend to some features (e.g., word further, attempting to reveal developmental patterns hich can contravene strategies based on canonical languages. Our conclusion is that a schema formed cal sentence form is the point of departure for the related schemas for sentence types which are noncond problem noted above-that of crosshnguistic schema LEuseful to the child learning Engiish, where ent. But what of languages with greater freedom of inductive process proposed above when suit in an early strategy to interpret any V (provided that strategies are induced only on the f wordqrder strategies are basic, all lanly be approached as fmed word-order languages. Indeed, ch a suggestion on the basis of an early phase in one of Russian (Gvozdev, 1949), in which fmed word order exhibits a range of word orders. Along sts in the late sixties and early seventies, it seemed to c syntactic device had a special primaus kinds of inflectional and variable word-order rkish and Serbo-Croatian, studied here), also orient to case inflections in organizing Slobin, 1973, 1982). The primacy assigned to acquisition principle was unduly influenced by the to identify characteristics of simple sentences which sentence processing strategies in different

Children use canonkal sen tense sclwms

types of languages. Specifically, we compare the roles of word order ant. inflection in the comprehension of transitive sentenctss with animate agents and patients. Accordingly, we selected four languages which vary in regard to the use of word order and inflections to signal agent-patient relations. Two of the languages, Slerbo-Croatian and Turkish, are inflectional and therefore allow all six word orders of subject, verb, and object (SVO, OVS, SOV, OSV, VSO, VOS). These languages differ in that the inflections in Turkish are always regular and explicit, while in Serbo-Croatian there are several declensions, as well as forms that do not differentiate phonetically between subject and object. The other two languages, English and Italian, are non-M’lectional, and therefore do not have formal marking systems that allow variation in word order. However, Italian shows considerably more flexible -word order in usage than English, relying on prosodic and contextual1 cues to disambiguate re-ordered sentences. In the process of our work we come to the conclusion that children in the third and fourth year of life arrive ,at a basic understanding of the major means used in their language for signa!iing syntactic relations --at least in the transitive sentence types we have investigated---along with some grasp that deviations from canonical patterns can bc: interpreted in consistent ways. We also conclude that some types of sentence interpretatirall strategies are more accessible to the very young child Ithan others---but that is getting ahead of our story. To begin with, we examine statistics of convcrsational data in the four languages, and then move on to our experimental findings. 1. Speech patterns Reliance on word order should be shaped by the roles played by word order in the particular language. Accordingly, we present a lrery general survey of speech production patterns of children and adults in the four languages under study. Little system-tic statistical information is avaliable about the relative frequency of subject-first and object-first constructions in Englishspeaking children and their caregivers. We assume that subject-first forms predominate. One reason is that object-first constructions are relatively rare for adults and involve complex constructions. In addition, the role of word order in guiding comprehension in English (and other languages) must be considered in the light of the structure of early adult-child discourse. The sentences exchanged between p*reschool children and their caretakers hardly allow for ambiguity as regards the roles of agent an6 patient in transitive clauses, suggesting that children’s sensitivity to word order must be based more on propensities for syntactic analysis than on discourse pressures for se:yrBntic interpretation of utterances irl context. For exampld, in an extenm

disccWse, me discovered that almost every uttery nonreversible, in that imate. That is, word order is not one of the nouns xamples of reverstwo animates), at least one of the participants was e inflections on person pronouns &ores would not have e children’s sensitivity to word ary to present them with contmllerd stiment described below. y for background information. sentences in adult-child inter1i:shis marginally infle&ional amongthe different theories of trategies. Italian, as noted above, is more flexible. o children., collected between along with a small sample of adult speech to these wn in Table 1. It is evident that even though Italian alfrom SVO order, this is still the most frequent in both speech. (Here we consider only object nouns, since object e the verb. Given the pattern of results reported below, we nt sentence schemas may be developed for sentences .) Furthermore, ignorult utterances follow

a&&child

re diwtiminative

data on the basis of word-order children and their careir native country as a from this samplle, 0 order of subject, verb, . As in Italian (and Enare very similar: the NVN pattern the subject. At first this of a developmental priority of ‘word na.l and allows all possib&e word idid position, even more oftten in a language that does ication clarity, chilcIren tend to impose

r, b only imperfectly

inflectional, which leaves

CrhiIdrenuse canonical sentence schemas

Table 1.

Percentages ofoccurrence o$ orders speech’ Word order

svo ovs sc9v QSV VSQ vos

of subject, verb, and object in Italian

Claudia (ls-2;9)

na Francesco (1;9--3;9)

74 7 1 1 12 5

Adult input PP. 82 5 2 0 I1 0

70 9 1 1 10 13

%gures from Bates (L376, Tables J-9, pp 188-190). centage.

Table 2.

235

rounded to nearest whole pep

Percentages of occurrence of word-order types in Serbo-Croatianand Ttdrkish speech -_ -SerboCroatiar@ -_ Children Adults

Turkishh _L__ Chiklren Adults

NVN (percentage of NVNs that are SVQ)

73 (99)

57 (97)

37 (46)

38 (66)

NNV (percentage of NNVs that are SSV)

12 (83)

24 (67)

53 (87)

56 (86)

VNN (percentage of VNNs that are VSO)

14 (79)

19 (84)

10 ( II. OO!

Total Nr = Subject

94

87

73

6 (100)

78

“Based on a sample of 48 children between the ages of 2$ and 4;8. bBased on a sample of 14 children between the ages of 2;3 an&38.

an interpretation ambiguous. There are three major noun deelensions: masculine, feminine, and neuter. In some: instances the subject and object case are differentiated phonetically, but in others the inflectional suffixes are identical. This means that some sentences in Servo-Croatian do not indicate by phonetic infIec,tion whieh noun is the subject and which the object. Speakers must make use of some convention to deal with those cases, and the convention is to follow SW order in inflectionally ambiguous sentences. Since at least some of our data were uniquely phoneticaly inflected, it is striking that the tendency for the first noun to be the subject is SO strong, even when it is often not required. This suggests that if required at all, an order schccia is overgeneralized in speech.

such

bema is never required by’a grammatical sentence in a iI1 it be used anyway? Turkish approaches being such noun is uniformly marked by the addition ofa unique n is a? indefmite object noun, which is uninflectedl. *, is marginally relevant to the language-learning which it wo4d “leadto ambiguity would be a and an indefInite subject and object (e.g., ‘a boy mple of 500 adult utterances to a child of 3;2 we they would bc vew frequent in ummrary data for adult and child conversations in Turkin the same way as in the case of Serbothe relative frequencies of NNV (intermediate), and VNN are nearly identical for and, in one sentence order, NVN, the children show place the subject before the object than do the adults: 213 of the NVN constructions, f the wrresponding utterances by children. es with greater flexibility of eless, even in Turkish we did find overall the usually he object. Flawever given the h statistics come from conversations, it is not possible to urs factors which may contribute to this tendency. Subnitively salient human actors ceptual ‘naturalness’ such as that proposed by 1977; &good and Tanz, 19771). On the other e objects because sub_iects are most likely to be on discourse function such as that proposed by her, these two tendencies interact in that agents tend our speech data can only be sis. This we have not done. Rather, having ility across our four languages, we g0 perimental task. ch b-e control the speech input and observe the different positions. The goal is to distinguish d relies on word order to identify agent and -if such there be. The following uence comprehension in children in four different lanquestion directly. ‘The task reqtiired children to act out with nolcontextual sgqmt, thus forcing them to differentiation. Such tasks have

children ustfcanonical serrtemx s&ems

237

been criticized for ignoring children’s real-life capacities to make use alf many types of information in understanding speech (e.g., Bti~~es, 1980, Chapman, 1978). Clearly, such tasks do not establish the ~.,rmal age at which particular sentences types can be understood in natural settings, noi do they presume to model natural speech comprc!hensinn in its everyday richness. Rather, by putting a child in a stringent task, one hopes to reveal the lin,guistic capacity when pushed. to its limit. Some researchers have questioned whether it is indeed linguistic capacity which we are tapping in such settings (e.g., Karmiloff-Smith, 1979). However, in thz present study we have an additional independent variable of crucial significance: the differences between English, Italian, Serbo-Croatian, and Turkish. All of our child.ren have been put in the same communicatively unnatural setting. To the extent that their performance varies systematically on the basis of definable characteristics of the particular language that they are acquiring, we can be satisfied that the experimental task is not merely a device to elicit ad hoc strategies from confused children. With this anticipation of the questions of some methodologically concerned readers, we turn to the crosslinguistic sentence comprehension experiment and its results. 2. Method of experiment 2.1. Subjects Subjects were all monolingual children of professional parents, residing in Berkeley, Rome ~ ,Dubrovnik, Istanbul, or Ankara. For each language, ‘rhe design provided for th.e testing of 48 ‘children (three boys and three girls in each of eight age groups: 2;0, 2;4, 2;8, 3;0, 3;4, 3;8, 4;0, 4;4). Due to various vicissitudes of field research, the actual numbers of children tested were: English, 48; Italian, 44; Serbo-Croatian, 38; and Turkish, 37. 2.2. Task The child was presented with a pair of toy animals or dolls (identical in \all field Tettings) and was asked to demonstrate an action of one object upaaa the other. (See Bever, 197Oa, b, for the study on which our technique was modelled.) Rach instruction contained itwo animate nouns and a reversible verb (a verb which could take either n.oun as agent or patient, e.g., ‘The squirrel scratches the dog”). The verb-noun combinations are given in Table Al in. the Appendix. There were 18 veils and 18 forms of the test, allowing for permutations sf ~::ard order and case inflections, as described below. Each child received three different forms of the test, administered 3n differ-

a ten-day period: the six children of each age group thus rms of the test. menter ascertained th? child’s ability to name the intransitive verb instructions (e.g., “Here’s a story about him. How about, ‘The camel is hat that would be like?“). In the main task, three times, in a medium flat intonation at normal the the verb and equal stress on both ish and Italian, as dlescribed below). were given in the present tense (e.g. the tences e gnzff"ur ilgatto); the Serbs-Croatian sentences mperfective tense (e.g., gtene grebe maze ‘dog the Turkish sentences were given in the optative mood cat scratch-let’). Response was scored in terms . If the child’s response, after three presentawas probed in two ways: (1) the experimenter ibte events, asking the child to choose the appropribe questioned directly (e.g., “Who is touchere are based only on clearly interpretab’le re-

two nouns and one verb occurred equally often e nouns were chosen so as to be either neuter or n, allowing for ambiguous and unambiguous ‘utterminative and accusative cases are not distinguished for neuter nouns (zero ending -e in both cases), while the ive -a versus accusative 94). fully ambiguous utterances with two neuter nouns,, martig of subject only, using feminine nominative ked forms, and (3) utterances with clear marking of object cusative and neuter unmarked forms. Consistent ree use of a word-order strategy reas consistent responses to utterances with on either word-order rules (resulting in errors t second) or inflectional rules There were 15 possible seneuter -e, S : feminine nomiWC,V =verb): NVN, SVN, NVO, NVS, , VNO, VNS, VON. Each of

Chikiren use canonical sentence schernas

239

the unmarked types occurred twice (NVN, NNV, VNN), resulting in 1$ sen:tence types for each test. Each type was represented by a different const tion af verb and nouns. Turkish has a uniformly regular and obligatory definite direct obj suffix’, thus allowing for six unarLrbiguous and grammatical orders: SVO, OVS, SOY, OSV, VSO, VOS. In order to probe for the ap@cation of a word-order strategy in the absence of inflectional cues, some sentences were presented with two uninflected nouns, corresponding to the Serbo-Croatian NVN, NNV, and VNN sentences. It should be noted that these sentences are ungrammatical in Turkish (lacking the obligatory case suffix) but grammatical in Scrbo-Croatian (since both nouns are neuter). The piaradigms in English and Italian were simpler than the other languages, sine-b VNN > NNV; while English has only NVN. This is reflected in the fact that word-order strategies appear only in NVN forms in English, whereas in Itlian, both the stress and order strategies appear Imost strongly in NVN, then VNN, then NNV (see Table 7). Thus children not only identify the canonical sentence pattern in forming schemas for speech processing, but they also come to categorize sentence patterns in terms of degree of deviair basic tion from the basic schema. Italian-speaking children include in extent, transitive sentence schema the possibility that VPJN and, to a les NNV orders are susceptible to interpretation, and that nominal stress may be significant in guiding interpretation. Our findings have suggestive parallels with a recent study of English and Italian sentence processing by adults. Bates ef al. (1982), in similar tasks, found greater reliance on word order in English speakers, as cornpar Italians, and greater reliance on stress information by Italians. Adult speakers showed some tendency to apply word-order strategies to those non-

ofmun

choke in EngI&h and Italian on the basis of word order

English

Italian -

58 57 54

Choice of first noun

Choice of stressed noun

79 46 45

48 47 52

ave limited currency in colloquial English, interprett-dislocation) and NNV as 63SV (left-dislocatian), ertained the possibility of a greater variety of word stress cues to come to consistent interpretations of nones ef al. propose that their study “suggests that Italian an English? Thrs difference in language patident at our preschool level, but it is clear that sentence evelopmental history beyond the early levels nghsh seem far more similar from our perspecof the adult study. What must be added is an ignaling pragmatically motivated variants are inczor-

generalization about the Turkish data is that childTen are itive to inflection, regardless of word order. Every single le correctly acted out the inflected forms with an average is sensitiviF~ de= increase with MLU for some senk$4 in the Appendix, but it is significantly of 67% even in the earliest two MLU groups. frst (SW, SW, VSO) and object-first (QSV, of average percentage correct. Correct response an inflectional strategy, since the second noun ch a strategy appflrently deve!ops in Turkish marked differerxe bletween the types of NNV, VNN) in their sensitivity to in-

Children use canonical sentence schemas

Figure 3.

249

Percentage correct in Turkish by order type and MLU (Subject first = SO V, S VO, WO; Object first = OS’V, 0 VS, VOS) 100 90 80 70 _ z:

60

s

40

0) g z 0) u 5 0

20

50 30 IO 12345 MLU

678 Group

There is, however, considerable difference in the sensitivity of the three ungrammatical (uninflected) sequences to word order. Uninflected NNV forms show a significant tendency for interpretation of the first noun as agent; uninflected NVN shows a strong trend in this direction; while uninflected VNN shows no word order effect. This order mirrors the order of frequency of occurrence of the three order types in Turkish, as shown in Table 2, above. The ordering of the three sequence types is also weakly revealed in patterns of younger children’s choice of first noun as agent in inflected sentences. We calculated the tendency to .pick first noun in inflected forms by averaging the percentage of first noun choices for each version of a sequence type (e.g., NOV + IONV). If there were no effect of word order, independent of inflections; this sum would be 50%; if there is a tendency to take the first noun as agent, this sum would be greater than 50%. For the fiist four MEU groups the average of this sum for the NNV sequences is 53%, for the NVN sequences S2%, and for the VNN sequences 49%. In brief, there is a very weak ordering of word order strategy in the inflected fomis, which has the sa:me relative strength as in the uninflected forms. Again, while the effect in the inflected forms is only a trend, the fact that it ~2” be detected at all,, and that it shows the same relative ordering as in the uniltlflected forms, is importarrt in the light of considerations of canonical sentena:e schemas which we raise in more detail below. 4.5. Serbo-Croatian We caIculated means of the tendency to pick the first nrxrn as the actor, separating inflected and uninflected forms, and subject-marked versus object-

I,

bin and T. G. Bever

r-noun choice in Serbo-Ckoatianby order type and ML U IS/N VSN; N”Q= NW, NQV, VON; N/N = NVN, NNV, VNN; V, VNS,;O/N = OVN, ON V, VQN)

thin the inflected constructions. Figure 4 displays the ntage correct according to increasing MLU group. Note groups of sentences according to inflection: inflected S), inflected object only (N/O, O/N), and neither noun or object (N/N). The figure highlights the different opposed to object inflectional marking, displaying the o object-marked forms develops sooner than to subjecthe correct performance is to interpret nd noun as subject. s might be so by considering the effect of word order e developmental pattern of the tendency to take the agent in inflected and uninflected forms is displayed in MLU group there is a strong tendency egy, even in the inflected.foms. The increase in this in the uninflected forms is paralleled by an increase in in inflected forms in which the first noun is the agent. e the first noun as agent increases at the same at is the correct response; this rate is constant cted noun in the sentence or not, and ction is in the form of a subject marking on on the second noun. e infIection does affect perfoDrmance on sentences in ti to interpret ,the first noun as the object. In itid object inflection are not affected by the deegy. On the contrary, as the word-order

Children use canonical sentence schernm

r

251

strategy emerges, correct performance on object-initial sentences also increases, while in subject-final sentences it decreases at first, and is worse than performance on olbject-initial sentences until the last MLU group. Si.milar findings have been reported for German (Mills, 1977) and Hebrew (Frankel and Arbel, 1.979; Frankel et al., 1980). In sentence interpretation experiments with children in those languages, it was found that the tendency to pick first noun as agent was more readily blocked by an Lritlal object inflection (masculine laccusative afiicle in German and direct objctct particle in Hebrew) than by subject marking on the second noun (masculine nominative article in German and subject-verb gender agreement in Hebrew) where the first noun was neutral in regard to subject or object marking. We can offer the following observation as a way of summurizing this resul&, The word-orde;r strategy and inflectional sensitivity lead tc the same result except ir1 sen.tences without inflections, and sentences with objectinitial1 or subject-final constructions. The children’s performance on constructions with no inflections shows that there is a word-order strategy available in Serbo-Croatian. However, their performance on subject-final versu!s object-initial inflected sentences suggests that the location of the inflection can influence whether the word-order strategy applies or not. The initial object inflection apparently blocks the application of the word-order strategy much more istrongly than the final subject inflection does (as in German and Hebrew also). We have here the kernel of develoament of strategies for contravening processing based on the canonical seritence schema. Early strategies of this sort seem bound to on-line processing, as evidenced by their sen.sitivil’-yto a sentence-initial local cue on an object noun. More mature strategies will require deferral of interpretation until an entire clause has been received. All of this is not to say that subject has no effect at the preschool level. Firs& even the subject-final inflected forms are performed correctly by the fourth MLU group significantly more often thlan not. Furthermore, the presence of initial subject inflection consistently is reflected in a gre:ater tendency to take the first noun as agent, in comparison with the corresponding uninflected N...N forms? even in the earliest MLU groups. The standard word order (_NVN)of Serbo-Croatian shows all of the effects we have discussed to a greater degree than the non-standard word orders (NW, VNN). (See ‘Table A5 in the Appendix) First, the effect of word order on uninflected forms is stronger in NVN sequences than in the others. Second, performance on object-initial sentences is better in NVN sequences~ Finally, performance ion initial subject inflected sentence forms is wome ir, the third MLU group in NVN sequences. Recall that this is the overall pattern-that is, whatever pattern is typical of the sentences overall, is even

standard cquence orders than in the non-standard orders. a general sensitivity to the canonical word order of the rmance on our task.

can be compared in terms of the effects of wcxd-order prehension strategies. The languages are directly comregard to the object-inflected sentences, since the object is in Turkish sentences and all the uninflected forms are unsh. In the object-inflected forms, the Turkish sentences ended better than the corresponding Serbo-Croatian first three MLU groups. This difference is surpristian object-inflected sentences are as uniquely responding Turkish forms. There are several facts which menon First + unlike Turkish, Serbo-Croatian presents each with numerous excepthe youngest Turkish chilsystem of their language, ing phonetic form for each oatian children must not only learn several possible phowhich nouns go with each . Consequentliy, the Serbosame level of performance s does not occur until the n correctly act out all S), as shown in Table AS in thle Appendix. omprehension in Serbos is the fact that many sentences have no uniq_ue inflecy interpr&ed as having the agent when the verb is semantically reversible. Serboonding strategy, as demonstrated by their ine third MLU group. Thus, unlike Turksystem, but must also ically recognizable; in addition they have to ded. Turkish children to dzvelop a similar word-order strategy even though But this strategy in Turkish this does occur strongly in

Children use canonical sentence schcmas

25.3

,§. The role of canonical sentence schemas Many aspects of our data can be interpreted on the view that the child is building up a representation of a ‘typical’ or ‘canonical’ sentence in his or her language. We have used the term ‘schema’ at a number of points in the preceding presentation. What we have in mind is a summary representation of the characteristics of a linguistic entity-in this case, a simple active affirmative declarative sentence involving a transitive verb and the related subject and object nouns. (For a theoretical discussion of ‘schemas’ in another area of language development, see Bybee and Slobin, 1982, on English past-tense forms. For related theoretical discussion, see Bever, 197Oa, 19’706, 1981; Bever and Langendoen, 197 1.) Schemas are formed on the basis of linguistic experience and are used for recognizing utterances as interpretable (e.g., accessible to comprehension strategies in our experiment). We propose, further, that sentence schemas have their origin in orientation to canonical or presuppositionally most neutral sentence forms in adult speech. The data of our study fill in only part of the picture. (Slobin, 1981, discusses relations between canonical sentence forms and prototypical event perception, but we need to know much more about the contextual settings of such form-event pairings as well as the application of linguistic schemas outside of the laboratory.) Our findings in this study point to language-specific schema development, drawing on features of word order, inflection, and prosody occurring in the canonical sentence r’~rms of each of the four languages. The English and Italian children, overall, responded consistently only to NVN sentences, and the Turkish children only responded consistently to forms with an object inflection. The failure of a strong pattern to emerge in response to the non-standlard forms (NNV and VNN in English and Italian, uninflected forms in Turkish) suggests that the child does not apply systematic analysis to a string wlhich does not correspond to the canonical sentence schema. Further evidence of sensitivity to canonical forms is shown in those cases where consistent responses are made to deviant sequences. A tendency to show any consistent response is related to the frequency of occurrence of the order type of the deviant construction in the language. Turkish children are consistent most frequently in response to NNV strings, corresponding to the dominant SOV order of their language, and least frequently to verbinitial strings, which are the most infrequent in normai Turkish discourse. In Italian the relative frequency of application of stress and order strategies corresponds to the relative frequencies of the order types in the language from NVN to VNN to NNV.

strategies in response to grammatto the relative typicality of the form. Roth rategies in Serbo-Croatian (emerge first in rehemas we are aracteristics of ‘prototypes’, in that they define not ction of less-related forms in terms ine if verb-final forms play a similar role in Turkish ng prototypical schemay since the earliest MLU group. However, the imitate grammatical sentence:s in to explore the possibility that processing grammatical . The 18 sentences of one form of were offered as stimuli for imitation to 30 of the 3;9. The findin. are summarized s were imitateld correctly (again of a!1 possible orders). Re-orderings were rare, they reflected a sensitki:y to the frequency of $1 verb-final strings were (2) NW strings we)-: r&rdered kss frequently N s: kgs were re-ordered rally into NNV order, but also into NVN. Younger uently (from 46% at 3;0 to 11% by 3;8), and atte:mpts to move less frequent orders (NVN er, suggestmg a greater early role for canonical sharpIy after 3;4. Thus, in some circumstances, tendency to assknilate noncanonical to canonsentence schema can a30 be used to clarify f sentence comprehension. Children’s early in&e sentences suggests that these forms are which case the strategy would be to

emas include not only characteristics. At children begin to

ChJdm use canonical sentence schemas

Table 8.

255

Percentages of Thkish irentence imitations by word-order type. (N = 30; Mean age = 45 months) Stimutus Sentence

svo ovs

Response Type ovs

sov

osv

76 2

73

14 4

2 15

9 7

98 6

81

2 13

20 8

2 2

_

sov osv vso vos

Other

-svo

2 3

vso

67 2

vos

2 69

7 18

--

The French findings of Sinclair and Bronckart (1972) further support the suggestion that the child% notion of canonical sentence form includes both word-order and inflectional information (nominal inflections, as ;hown in Turkish, anId verbal inflections, as shown in English and French). Sinclair and Bronckart, in a task similar to ours, did not find consistent acting out of NVN sequences until later than age five. Their stimuli were all ungramlnatical in French, consisting of two nouns and a verb in the infmitive. We suggest that their younger children, applying a canonical sentence schema which requires an inflect-d verb in such structures, did not corjlsider such strings to be interpretable sentences of French. In Japanese, similar to lour Turkish findings, Hakuta (1976) found random response to NNV ?;tri.ngs without case inflections. The interpretation of our results in terms of the early development of a prototype-like canonical sentence schema also offer!%an interesting interpretation of the behavior of our Serbo-Croatian subjects. We noted that the children respond non-systematically to object-first sentences, even at ages when they interpret N”SrNor object-final NV0 sentences systematically. Apparently the childrea construct an initial canolnical sentence schemla which includes both the! SVO order convention and the separate information that the object should be explicitly marked. This canonical schema is redundant for certain sentence types (e.g., SVO, NVO...), but it is independentl~l~ motivated by other frequent forms (e.g., NVN). On our view the child is sensitive to the fact that such sentences are most frequent in what s/he hears,. and (in the first MLU group) constructs an SVO canonical schema as an intitial hypothesis about all such sentences in the language. At a later stage, when the order strategy emerges as a separate phenomenon, an initial object-

er strategy while a final subject-marked noun that this asymmetry between object- and subject-markers ebrew. This raises the possibility that object an subject inflections generally. There is an er, that explains the specific phenomena in tences are processed. An initial noun that is terpreted as subject by applicatior sequent noun is unambiguously inflected as r has to recall the first noun and the intervening masemantic relation between them. However, if the y marked as an object, then the order strategy is the intermediate m&assignment of semantic ay play a role in guiding rosslinguistic results, and those in the literature, support the child is not only acquiring a schema of the ‘typical’ sentence but is also us&g that schema as a guide for the applicastrategies. The fact that the notion of the ‘canonical’ s so early in language development indicates that it may the possible variations in sentence forms cessed and learned during the later stages of acquisition. m acquisition-as an interactive counterfunctions-as to why languages have the extent that the child may be sensical consistencies in language, the acquisition of canonical tate construction of related linguistic schemas. For example, child might find it easy to acquire nd prenominal modifiers al verb-fiial order in lan-

result of this investigation is that children seem prepared tiund and word-order languages. Contrary to earlier exthe alleged naturalness of fixed word order, the acquisiimpaired by the fact that word order is not a cue r is there evidence from Turkish for a natural t-f&, or SVO schema. At most, our Turkish data weakly

CWdren use canonical sentence schenwrs

257

confirm a more general hypothesis that word order may play some role in comprehension and imitation even if it is not strongly modelled in the iinput la,nguage. Serbo-Croatian children, however, show a strong dependence on wordorder comprehension strategies, even in inflected sentence forms where such strategies would be unnecessary. Apparently the fact that the word-(Brder strategy is required for the comprehension of some forms that are not ret-n ognizably inflected stimulates the child to overgeneralize the use of the strategy to all forms. Ultimately the child learns when to attend to the inflections as contravening tAe word-order strategy. It is striking that in all four languages the strongest evidence for the word-order strategy occurs in MLU groups 3-4. The average age for these groups is around 3;6, which is the usual age at which overgeneralizations of word-oroer strategies have been previously reported for English. It is not clear whether this seeming universality is due to linguistic development or to some general maturational factors. Chapman and Kohn (1978) have found a general increase in systematic responding to comprcnension tests at this age, including word-order and nonlinguistic task-specific strategies. Bever (1970~) has made similar arguments in regard to other aspects of cognitive development at this age. (Also see Mehler and Bever, 1967.) A linguistic hypothesis might be that it is only at this stage that the child can process utterances with three major phrases, the minimum number needed to express the relation between a predicate, an agent and a patient. This is consistent with the fact that the MLU that corresponds to our MLU groups 3-4 is about five words for English and Italian and about three words for Turkish and Serbo-Croatian. English and Italian SVO sentences require two additional words for the determiners, as compared with Turkish and Serbo-Croatian which lack determiners. A more general explanation might lie in the development of immediate memory, which would account both for the MLU development and the emergence of the order strategy. We hope to have shown that children are attuned to canonical sentence forms, and that, early on, they develop schemas embodying the most typical features of such forms; and, furtrier, that canonical sentence schemas play a central role in processes of perception and interpretation of utterances.

NOUNA

NOUN B

NOUN C

the cat mai% kedi

the chick la gauiua pile civciv

the bird l’ucceito ptica

the tismb l%gnell.ino we kuzu

the horse il cavallo idrijebe at

kug the camel il cammello deva deve

the df il vitellino dana

the goat la czpa koale keqi

the giraffe la gisafa Zrafa ziirafa

the boy ii bambino djete aetan

the dog il cane &ne kapek

the fox la volpe lija kurt

the Pig ilmaialii prase domuz

the duck h papera pa33 iirdek

the goose Y’occhetta guska kaz

the baby

the deer il cer~i&to

the cow h mucca krava rnek

il gatto

tele

if PUP0

lane geya

thegoat lac_pra

kozle k* the baby mm

the boy il bambino djete

the girl la bambina djevoj%a krz

the horse il cavallo Zdrijebe at

the zebra la zebra zebra zebra

the lamb l’agnellino jaJlie kuzu

the camel il cammello deva deve

CFiMlen use ctmonicd sentence sche~nas

Table Al.

(continued) V’RB

NOUNA

NOUN B

NOUN C

touches tocca dira eMesin

the calf il vitellino tele dana

the chick la gallina pile civciv

the bird I’uccello ptica

hits colpisca tu’Ee tekmelesin

the cat il gat to ma& kedi

the deer il cerbiatto lane geyik

ku9 the llama it lama lama lama

picks up dEe kaldtrsm

the pig il maialino prase domuz

the dog il cane Ystene kopek

the fox la volpe lija kutt

bumps urta lupa dugursun

the boy il bambino djete o@an

the calf il vitellino tele dana

the donkey l’asirnello mazga e#ek

shakes. scuota trefe

the pig il maialino

the sheep la pecora ovca

sallasm

domuz

the goat la capra kozle kcci

kisses bacia ljubi iipsiin

the duck la papera pa&e iirdck

the baby il PUPo bebe bebck

the cow la mucca krava inck

slaps schiaffeggia pljuska tokatlasrn

the l,amb l’agnellino janje kuzu

the horse il cavallo ‘idrijebe at

the zebra la zebra zebra zxbra

scratches graff”la grebe ka3rsrn

the dog il cane Ztene kiipek

the cat il gatto ma& kedi

the squirrel io SC6iat td6

bites morde grize isrrsfn

the c>ick la gallim pib civciv

the deer il cerbiatto lane geyfi

the llama il lama lama lama

S6lkViii

priW2

kCQWl

Vervetic

sincap

aEach stimulus sentence contained the ve& and the noun from column A, pair either the rmoun from column B or column C. Columns A and B are neuter SerboCroatian; Column C feminine. Tire Mguages are listed in the order: Wian, SerboCroSatian, Turkish.

259

2.. perctievtqpof choi&s of fifft now as agmt in English __.~~~

Sentence type a

N

N

1_NvN m

NNV

YNV

N_bJV VNN

50

VljN

VN&I

72 47

58 55

58 56

49

28 48

46 33

38 45

56 53

81 90

65 88

84 81

32 40

Sl 53

32 3!)

40 43

48 28

31 47

780 6 92

80 86

89 88

44 46

26 42

4’1 51

40 61

43 58

54 67

6 5

83 9iI

90 100

51 56

45 49

4$ 35

42 56

53 33

62 24

4 6

SO

6 6

92

noun re&ed stress

of

choties of first noun as agent in Italkm

P Vhl

N_VN NVhJ

Sentenoe type” _-NNV N_NV N_NV VNN

V_NN VN_N

-6 6

64 61

6 6 6

95 90

5

93 90 79

5 4

$1

81 89 95 94 87 87 93 91

60 56 83 64 75 90 90 69

so

61 49 56 47 43 43 29 52

57 57

49 57 57 30 56

43 2:9 50 55 4#9 30 61 26

43 50 53 70 71 56 20 58

46 71 67 56 65 67 43 67

66 37 37 43 44 57 51 40

noun received dress

tage of choices ofjbt

noun as agent in Turkish Sentence type4

5 5 5

54 60 58

NV0

O’JN

NNV

NOV

ONV

VNN

VNO

VON

79

231 27 28

17 74

78 Y33

30 12

84 81

62

72

30

35 65 51

29 24 24

81 73

90

(contibwed on fxing page)

Odldren use canonicalsentence sehemas

Table A4

261

(cwtfinued) a-

MLU Group

4 5 6 7 8

N

5 5 5 4 3

QN = unmarked

Sentence

type= “1

-v NVN

NVQ

NPJO

NNV

NW

ONV

VNN

VNO

VON

53 67 65 53 52

86 83 83 100 94

23 10 16 13 0

42 73 70 82 67

90 90 77 8s 67

23 13 20 23 26

St 53 50 60 44

7 83 80 83 83

77 10 87 82 6

(subject

case), 0 = noun with definite direct object inflection.

nouns

Table. AS. Percentageofchoicesof frst noun as agent in Serbs-CPorrtian MLU Group

N

type -

1 2 3 4 5 (1 7 8 MLU Group

Sentence NVN

SVN

NOV

NVS

OVN

NNV

SNV

NW

NW

53 80 75 73 80 77 58 72

53 47 78 7 617 680 100 78

70 73 61 93 60 73 84 89

73 80 61 27 20 26 42 56

47 47 33 13 40 20 33 11

50 57 67 67 55 47 67 67

53 60 sil x? 60 80 70 66

53 SO 84 73 80 53 50 67

40 27 61 40 47 47 SO 33

N

5 5 6 5 5 5 4 3

Sentence type QNV

VNN

VSN

VNO

VNS

VUN

33 47 50 27 27 27 27 33

68 63 67 57 72 57 58 72

47 60 78 80 80 87 67 89

47 60 67 87 73 67 50 89

53 67 44 47 47 33 42 22

47

47 33 7 13 20 17 22

aN f neuter noun (unmarked for subject-object),--S = feminine nominative noun 0 = feminine accusative not r, ,ob,ject).

ject),

(sub-

and Brian MacWhinney have performed analyses of variiance the patterns of results presented in Section 4. We this independent analysis, and present a partial sumdependent variable in all analyses t movn~ as subject: choice of first noun was scored 8s 1, and no response or uninterpretable response was scored mber of subjects across language equalized. Subjects were grouped into four age levels. r of subjects for whom complete data were available in was IQ; accordingly, subjects were randomly eliminaked from of 10 subjects at each of 4 age summarize only the results of these 16 of vartiance (examining Word Order, Case/Stress, and /Stress at 4 age levels), along with a brief summary of languages. (Full details are available on request.) The analyses are fully consistent with our more fine-grained of8 MLU groups, as presented in Section 4, above. of the effects reached significance in the youngest of the ough the word-order main effect was marginal, p < age level, word order had reached significance, and ffect of stress never reached sigof the ages, nor were there interactions between order and only cue used by English children at any age level is word on NVN items and close to on the other two word-order types). ngest age level was also margin, with choice of the 1, both word-order and stress third and fourth gksti group, the oun on NVN, with random ignificance ait any level was ady sipflificant among the youngest e through the fourth age level. However, in choose the fmt noun more often than the case at every age level, with a sigthe fourth level. The word-order

Children use canonical sentence schemas

263

pattern reflected a bias Zoward first noun in NNV as compared with the other two orders. There were no significant interactions of word order and case/stress in ar.,y of the languages. Age effects within languages can be swmmarized very briefly as follows: English: an age X word order X stress ANOVA (age as a between-subjects variable, order and stress as repeated measures) yielded a main effect of age, a main effect of word order, and an age X wordl order interaction. None of the other effects reached significance. In short, English children rely entirely on word order, and the difference between NVN and the other two wordorder types increases with age. Italian: there were significant main effecu; of word order and stress, with none of the other effects reaching significance. Serbo-Croatian: the only significant effects were a main effect of case inflection and an age X case interaction. That is, processing of case inflections incKeases reliably across the four age levels, while word order does not reach significance and does not interact with age. Turkish: there was a main effect of word order, along with a very much larger main effect of case inflection.

References Aksu, A. A. and Slobin, D. II.(in press) Acquisition of Turkish. In Slobin, 1). I. (ed.), T/14 Clrc~sslirt~~ris~ic Study of Language Acquisition. Hihsdale, NJ, Lawrence Erlbaum Associates. Bates, E. (1976) Language and Context: Thl. Acquisition of P?agmatic~. New York, Academic Press. Bates, E., McNew, S., MacWhinney, B., Deves.ovi, A. and Smith, SF (1982) Functional constraints on sentence processing: A cross-linguistic study. Cog., II, 245-300. Bever, T. G. (197Oa) The cognitive basis fof Linguistic structures. Pp. 279-352 in 1. R. Itaym (cd..), Cognition and the Development of Language. New York, Wiley. Bever, ‘I’. G. (1970b) The integrated study of language. Pp. 158-206 in J. Morton (ed.), BMugi~al and Social Perspectives in Lunguage. London, Logos Press. Bever, T. G. (1981) Regression m the service of development. In T. G. Bever (ed.), Regression in Child Development. Hillsdale, NJ, Lawrence Erlbaum Associates. Bever, T. G. and Lapgendoen, D. T. (1971) A dynamic model of langualge evolution. LisTg. Iraq.. 2, 433-463. Bridges, A. (1980) SVO comprehension >trategies reconsidered: the evidence of’ individual patterns of response. J. child Lung., 7,89- 104. Brown, R. (1973) A First Language: The Early Stages. Cambridge, MA., Harvard University Press. Bybee, J. and Slobin, D. I. (in press) Rules and schemas in the development and use of the English past tense. img., 58. Chapm;an, R. S. (1978) Comprehension strategies in children: A, discussion of Bransrlbrd and Nitsch’s paper. Pp. 308-327 in J. F. Kavanagh and W. Strange (e:ds.), Speech and Language in Laboratory, School, and clinis. Cambridge, MA,, MIT Press. Chapman, R. S. and Kohn, L. L. (1978) Comprehension strategies in two and three year okis: Animate agents or probable events? J. Speech Hem. Res., 18, 35% 37 1.

264

D. I. Slobin and T. G. Bever

de Villiers, J. G. and de Villiers, P. A. (19730) A cross-sectional study of the development of grammatical morphemes in child speech. J. psychoZing. Res., 2,267-278. de Villiers, J. G. and de Villiers, P, A. (L973B) Development of the USCof word order in comprehension. J. psycholing. Res,, 2,33 I-341. Ekmckci, 0. F. (1979) Acquisitionof firkish: A LongitudinalStudy on the Early LanguageDevelop ment of a ‘ItrrkishChild. Unpublished doctoral dissertation, University of Texas at Austin. Frankel, D. G., Amir, M., Frenkel, E. and Arbel, T. (1980) A developmental study of the role of word order in comprehending Heblew. J. exper. child Psychof., 29,23-35. l:rankel, D. G. and Arbel, T. (1979) ProbabilisticAssignmentsof Sentence Relationson the Basisof Differentially-weightedInterpretive Cues. Unpublished paper, Department of Psychology, Hebrew University, Jerusalem. Fraser, C., Bellugi, U. and Brown, R. (‘1963) Control of grammar in imitation, comprehension, and production. J. verb. Leuw. vert,. Behm., 2, 121-l 35. Givbn, T. (1979) On UnderstandingGpamwiar. New York, Academic Press. Gvozdev, A. N. (1949) Formirovaniye u Rebenka Gramwuzticheskogo Stroya Russkogo Yazyka. Moscow, Izd-vo Akademii Pcda%ogichcskikh Nauk RSFSR. Hakuta, K. (1977) Word order and particles in the acquisition of Japanese. Papersand Reports on Child Language Development (Department of Linguistics, Stanford University), No. 13, 1 lo117. Karmiloff-Snlith, A. (1979) A Functional Approach to Child Language: A Study of Determiners and Reference. Cambridge, Cambridge University Press. Macirulay, R. K. S. (1978) The myth of female superiority in language. J. child Lang., 5, 353-374. Maratsos, M. (1974) Children who get worse at understanding the passive: A replication of Bcver. J. psychsling. Res., 3,65-74. Mehler, J. and Bever, T. G. (1967) Cognitive capacities of very young children. Sci., 158, 141-142. Mills, A. E. (1977) First and second language acquisition in German: A parallel study. Ludwigsburg Studies in Languageand Linguistics,No. 2. Osgood, C. E. and Bock, J. Ic. (1977) Salience and sentencing: Some production principles. In S. Rosenberg (ed.)o Sentence Production: Development in Research and Theory. Hillsdale, NJ., Jawrence Erlbaum Associates. Osgoou, 6. E. and Tanz, C. (1977) Will the ret1 direct object in bitnnsitive sentences please stand up? Pp. 537-590 in A. Juilland (ed.), Linguistic Stud&s offered to Joseph Greenberg on the 00 casionof his Sixtieth Birthday. Saratoga, CA., Amma Libri. Radulovit, L. (1975) Acquisition of Language: Studies of Dubrovnik Children. Unpublished doctoral dissertation, University of California, Berkeley. Sinclair, H. and Bronckart, J. P. (l972) S.V.0, A linguistic universal? A study in developmental psycholinguistics. J exper, child Psychol., 14,329-348. Slobin, D. I. (1966) The acquisition of Russian as a native language. Pp. 129-148 in F. Smith and G. A. Miller (eds.), The Genesisof Language: A Psycholinguistic Appmach. Cambridge, MA., MIT Press. Slobin, D. I. (1973) Cognitive -3rerequisites for the acquisition of grammar. Pp. 175-208 in C. A. Ferguson ;and D. I. Sloblrl (eds.), Studies of Child Language Development. New York, Holt, Rinehar: 8i Winston. Slobin, D. I. (1981) The origins af grammatical encoding of events. Pp. 185-200 in W Deutsch (ed.), The Child’s Construction of Language, London, Academic Press. Slobin, D. I. (1982) Universal and particular in the acquisition of language. In L. R. Gleitman and E. Wanner (eds.), Language Acquisition: State of the Art. Cambridge. Pp. 128-170, in Cambridge University Press.

CM&en use canonical sentence schema3

26 5

Les etlfarrts construisent un schema dc phrase canoniquc commc structure prdliminaire pour organiser la! ~~~~~~t~~~~t la ier. La phrase canonique in&t les traits t,Jpiques des clauses dans la langue maternefle ct sert dr taamc aux stratdgics de production ct de perception. Le schCma de phrase canonique fournit unc explication fonctionnelle des strat$ies selon l’ordrc dcs mats ou selon les flexions qui s’appuicnt sur fcs essais de I%nfant pour acyudrir la maitrise rapidc des techniques de communication fondamentalcs. On &tudic, dan- la rceherchc prdscntk, la sensibilit6 aux phrases de forme canoniquc et aux strat6gics fond&s sur I’ordrc dcs mots ou sur les fkxions dans la compkhcnsion dc phrases transitives simples. Les sujets sont dcs entints monolingucs dc 2;O A 4;4 ans. L’budc pottc sur 4 langucs: I‘Anglsis (languc A sontraintcs d’srdre non f%chi), I’ltalicn (contraintes d’ordre faible, pcu de flcxions); le Serbo-Croats (contraintes d’ordrc faiblc ct flexions); le Turque (contraintes d’ordrc minimalc ct flcxions).

Lcs rdsultats indiqucnt quc fcs enfants nc kpondcnt pas systdmatiquem‘nt aux s6qucnces qui violent la formc canoniquc de leur ianguc. 11sdkvcloppent des stratdgics distinctes et appropriks aux r6gularitds de lcur languc (ordre des mots ou flcxions). L’kmcrgencc d’un comportcment prkocc pour Its phrases canoniqucs pertinentcs de la langue et I’dmcrgencc de strategies de traiternent suggkent quc des bases comportcmcntalcs cxistcnt pour Its contraintcs linguistiques sur la formc de surface dcs phrases.