Haskins Laboratories Status Report on Speech Research1989, SR-99 / 100, 118-136

Competence and Performance in Child Language*

Stephen Craint and Janet Dean Fodortt

1. INTRODUCTIONThis paper presents the results of our recent

experimental investigations of a central issue inlinguistic theory: which properties of humanlanguage are innately determined? There are twomain sources of information to be tapped to findthe answer t9 this question. First, universalproperties ofhuman languages are plausibly (evenif not necessarily) taken to be innatelydetermined. In addition, properties that emerge inchildren's language in the absence of decisiveevidence in their linguistic input are reasonablyheld to be innate. Clearly, it would be mostsatisfactory if these two diagnostics for what isinnate agreed with each other. In some cases theydo. For example, there is a universal principlefavoring transformational movement of phrasesrather than oflexical categories e.g., topicalizationof noun phrases but not of nouns. To the best ofour knowledge children abide by this principle;they may hear sentences such as Candy, you can'thave now~ but they don't infer that nouns can betopicalized. If they did, they would say things like*Vegetables, I won't eat the. But this is not anerror characteristic of children. Instead, from themoment they produce topicalized constructions atall, they apparently produce correct NP­topicalized forms such as The vegetables, I won'teat.

In recent years, this happy convergence ofresults from research on universals and research

This research was supported in part by NSF GrantBNS 84.18537, and by a Program Project Grant to HaskinsLaboratories from the National Institute of Child Health andHuman Development (HD·01994). The studies reported in thispaper were conducted in collaboration with several friends andcolleagues: Henry Hamburger, Paul Gorrell, Howard LasnikCecile McKee, Keiko Murasugi, Mineharu Nakayama, Jay~Sarma and Rosalind Thornton. We thank them for theirpermission to gather this work together here.

118

on ~cquisition has been challenged byexpenmental studies reporting various syntacticfailures on the part of children. The children inthese experiments are apparently violatingputatively universal phrase structure principles orconstraints on transformations. Failure todemonstrate early knowledge of syntacticprinciples is reported by Jakubowicz (1984), Lust(1981), Matthei (1981, 1982), Phinney (1981),Roeper (1986), Solan and Roeper (1978)Tavakolian (1978, 1981) and Wexler and Chie~(1985). Some explanation is clearly called for if asyntactic principle is respected in all adultlanguages but is not respected in the language ofchildren.

Assuming that the experimental data accuratelyreflect children's linguistic competence, there areseveral possible responses to the unaccom­modating data. The most extreme would be to giveup the innateness claim for the principle inquestion. One might look for further linguisticdata which show that it isn't universaL Or onemight abandon the hypothesis that all universalprin~iples are innate. For instance, Matthei (1981)obtamed results that he interpreted as evidencethat universal constraints on children'sinterpretation of reciprocals are learned notinnate. However, this approach is plausible o~ly ifone ~an offer so~e other explanation (e.g.,functIonal explanatIon) for why the constraintsshould be universal. But this is not always easy;as Ch~msky (1986) has emphasized, manypropertIes of natural language are arbitrary andhave no practical motivation.

A different response to the apparent failure ofchildren to respect constraints believed to be~nnate.is to argue that the constraints are as yetmapphcable to their sentences. The claim is thatas soon as a child's linguistic analyses havereached the level of sophistication at which a

The Perception of Phonetic Gestures 117

Stoll, G. (1984). Pitch of vowels: Experimental and theoreticalinvestigation of its dependence on vowel quality. SpeechCommunication, 3, 137-150.

Studdert-Kennedy, M. (1986). Two cheers for direct realism.Journal ofPhonetics; 14, 99-104.

Studdert-Kennedy, M., &: Shankweiler, D. (1970). Hemisphericspecialization for speech perception. Journal of the AcousticalSociety ofAmerica, 48, 579-594.

VanDerVeer, N. (1979). Ecological acoustics: Human perception ofentlironmental sounds. Unpublished doctoral dissertation,Cornell University.

Warren, W. &: Verbrugge, R. (1984). Auditory perception ofbreaking and bouncing events: A case study in ecologicalacoustics. Journal of Experimental Psychology: Human Perceptionand Performance, 10, 704-712.

Whalen, D. (1984). Subcategorical mismatches slow phoneticjudgments. Perception and Psychophysics, 35, 49-64.

Whalen, D. &: Liberman, A. (1987). Speech perception takesprecedence over nonspeech perception. Science, 237,169-171.

FOOTNOTES"In I. G. Mattingly &: M. Studdert-Kennedy (Eds.), Modularityand the motor theory of speech perception. Hillsdale, NJ: LawrenceErlbaum Associates, in press.

tAlso Dartmouth College, Hanover, New Hampshire.ttAlso University of Connecticut, Storrs. Now at the University

of California at Riverside, Department of Psychology.1There is a small qualification to the claim that listeners cannot

tell what contributions visible and audible information eachhave to their perceptual experience in the McGurk effect.Massaro (1987) has shown that effects of the video display canbe reduced but not eliminated by instructing subjects to look at,but to ignore, the display.

2Liberman et al. identify a cipher as a system in which eachunique unit of the message maps onto a unique symbol. Incontrast, in a code, the correspondence between message unitand symbol is not 1:1.

3Liberman et al. propose to replace the more conventional viewof the features of a phoneme (for example, that of Jakobson,Fant &: Halle, 1951) with one of features as "implicitinstructions to separate and independent parts of the motormachinery" (p. 446).

4With one apparent slip on page 2: "The objects of speechperception are the intended phonetic gestures of the speaker,represented in the brain as invariant motor commands...."

Sane can certainly challenge the idea that listeners recover thevery gestures that occurred to produce a speech signal.Obviously there are no gestures at all responsible for mostsynthetic speech or for "sine-wave speech" (e.g., Remez, Rubin,

Pisoni, &: Carrell, 1981) and quite different behaviors underlie aparrot's or mynah bird's mimicking of speech. The claim thatwe argue is incontrovertible is that listeners recover gesturesfrom speech-like signals, even those generated in some otherway. (We direct realists [Fowler, 1986a,b] would also argue that"misperceptions" (hearing phonetic gestures where there arenone) can only occur in limited varieties of ways-the mostnotable being signals produced by certain mirage-producinghuman artifacts, such as speech synthesizers or mirage­producing birds. Another, however, possibly, includes signalsproduced to mimic those of normal speakers by speakers withpathologies of the vocal tract that prevent normal realization ofgestures.)

6There are two almost orthogonal perspectives from whichperception can be studied. On the one hand, investigators canfocus on processes inside the perceiver that take place from thetime that a sense organ is stimulated until a percept is achievedor a response is made to the input. On the other hand, they canlook outside the perceiver and ask what, in the environment,the organism under study perceives, what information instimulation to the sense organs allows perception of the thingsperceived, and finally, whether the organisms in fact use thepostulated information. Here we focus on this latterperspective, most closely associated with the work of JamesGibson (e.g., 1966; 1979; Reed &:Jones, 1982).

7It is easy to find examples in which perception is heteromorphicwith respect to the proximal stimulation and homomorphicwith respect to distal events-looming, for example. We canalso think of some examples in which perception appearshomomorphic with respect to proximal stimulation, but in theexamples we have corne up with, they are homomorphic withrespect to the distal event as well (perception of a line dfawnby a pencil, for example), and so there is no way to decidewhether perception is of the proximal stimulation or of thedistal event. We challenge the motor theorists to provide anexample in which perception is homomorphic with structure inproximal stimulation that is not also homomorphic with distalevent structure. These would prOVide convincing cases ofproximal stimulation perception.

8Bregman (1987) considers duplex perception to disconfirm his"rule of disjoint allocation" in acoustic scene analysis bylisteners. According to the rule, each acoustic fragment isassigned in perception to one and only one environmentalsource. It seems, however, that duplex perception does notdisconfirm the rule.

9Using a more sensitive, AXB, test, however, we have found thatlisteners can match the metal door chirp, rather than a woodendoor chirp, to the metal door slam at performa~ce levelsconsiderably better than chance.

Competence and Performance in Child Language 119

universal constraint becomes relevant, then thatconstraint will be respected. For example, Otsu(1981) has argued that children who give theappearance of violating a universal constraint onextraction may not yet have mastered thestructure to which the constraint applies; theymay have only some simpler approximation to theconstruction, lacking the crucial property thatengages the constraint. We will discuss theevidence for this below.

A different approach also accepts therecalcitrant data as valid, but rejects the inferencethat the data are inconsistent with the innatenesshypothesis. It is pointed out that it is possible fora linguistic principle to be innately encoded in thehuman brain and yet not accessible to thelanguage faculty of children at early stages oflanguage acquisition. The principle in questionmight be biologically timed to become effective ata certain maturational stage. Like aspects ofbodydevelopment (e.g., the secondary sexcharacteristics), linguistic principles might liedormant for many years. One recent proposalinvoking linguistic maturation, by Borer andWexler (1987), contends that a syntactic principleunderlying verbal passives undergoesmaturational development. They maintain thatbefore a critical stage of maturation is reached,children are unable to produce or comprehendpassive sentences (full verbal passives with by­phrases).

It may eventually turn out that the innatenesshypothesis must be augmented by maturationassumptions in certain cases. But suchassumptions introduce new degrees of freedominto the theory, so its empirical claims areweakened. Unless some motivated predictions canbe made about exactly when latent knowledgewill become effective, a maturational approach iscompatible with a much wider range of data thanthe simplest and strongest version of theinnateness hypothesis, viz. that children haveaccess to the same set of universal principles atall stages of language development. This morerestricted position is the one to be adopted until orunless there is clear evidence to the contrary, e.g.,clear evidence of a period or a stage at which allchildren violate a certain constraint, in allconstructions to which it is applicable, simple aswell as complex, and in all languages. So far, nosuch case has been demonstrated.!

Our research has taken a different approach.We argue that the experimental data do notunequivocally demonstrate a lack of linguistic

knowledge. We do not (leny that children dosometimes misinterpret sentences. But the properinterpretation of such failures is complicated bythe existence of a variety of potentiallyconfounding factors. Normal sentencecomprehension involves lexical, syntactic,semantic, pragmatic and inferential abilities, andthe failure of anyone of these may be responsiblefor poor performance on an experimental task. Itis crucial, therefore, to develop empirical methodswhich will distinguish between these variousfactors, so that we can determine exactly where achild's deficiencies lie. Until this has been done,one cannot infer from children's imperfectperformance that they are ignorant of thegrammar of their target language.

In fact it can be argued in many cases that it isnon-syntactic demands of the task which are thecause of children's errors. We propose that taskperformance is weak at first and improves withage in large part because of maturation of non­linguistic capacities such as short term memory orcomputational ability, which are essential in theefficient practical application of linguisticknowledge. This does not deny that many aspectsof language must be learned and that there is atime when a child has not yet learned them. Butour interpretation of the data does make itplausible that young children know more of theadult grammar than has previously beendemonstrated, and also, most significantly, thattheir early grammars do abide by universalprinciples.

In support of this non-linguistic maturationhypothesis, we have reexamined andsupplemented a number of earlier experimentalfindings with demonstrations that nonsyntacticfactors were responsible for many of the children'serrors. The errors disappear or are greatlyreduced when these confounding factors aresuitably controlled for. In this paper, we report ona series of experimental studies along these linesconcerned with three kinds of nonsyntactic factorsin language performance: parsing, plans, andpresuppositions. We will argue that these otherfactors are crucially involved in the experimentaltasks by which children demonstrate theirknowledge, and that they impose significantdemands on children. If we underestimate thedemands of any of these other components of thetotal task we thereby underestimate the extent ofthe child's knowledge of syntax. As a result, thecurrent estimate of what children know aboutlanguage is misleading.

120 Crain and Fodor

2. CHILDREN'S ERRORS INCOMPREHENSION

In this section we attempt to identify and isolateseveral components of language-related skills, inorder to gain a better understanding of each, andto clarify the relationship between the innatenesshypothesis and early linguistic knowledge. Verylittle work has been done on this topic. Themajority oflanguage development studies seem totake it for granted that the experimentalparadigms provide a direct tap into the child'slinguistic competence. An important exception is astudy by Goodluck and Tavakolian (1982), inwhich improved performance on a relative clausecomprehension task resulted from simplificationof other aspects of the syntax and semantics of thestimulus sentences (Le., the use of intransitiverather than transitive relative clauses, andrelative clauses with one animate and oneinanimate noun phrase rather than twoanimates). The success of these manipulations isexactly in accord with our general hypothesisabout the relation between competence andperformance. As other demands on the child'sperformance are reduced, greater competence isrevealed.

Our experiments focus on three factors involved. in many child language experiments which mayinterfere with estimation of the extent ofchildren's linguistic knowledge in tasks which aredesigned to measure sentence comprehension.These factors-parsing, presupposition andplans-are of interest in their own right, but havereceived very little attention in previous researchon syntax acquisition. In this section we willreview our. recent work on these topics. In thefollowing section we will tum to an alternativeresearch strategy for assessing children'sknowledge, the technique of elicited production.

Parsing. Sentence parsing is a complex taskwhich is known to be governed (in adults) byvarious decision strategies that favor onestructural analysis over another where both arecompatible with the input word sequence. Evenadults make parsing errors, and it would hardlybe surprising, given the limited memory andattention spans of children, to discover that theydo too. These parsing preferences must somehowbe neutralized or factored out of an experimentaltask whose objective is the assessment ofchildren's knowledge of syntactic rules andconstraints.

Plans. The formation of an action plan is animportant aspect of any comprehension task

involving the manipulation .of toys or otherobjects. If the plan for manipulating objectsappropriately in response to a test sentence isnecessarily complex to formulate or to execute, itsdifficulty for a child subject may mask his correctcomprehension of the sentence. Thus we need to

. develop a better understanding of the nature andrelative complexity of such plans, and also todevise experimental paradigms in which theirimpact on performance is minimized.

Presuppositions. A variety of pragmaticconsiderations must also be taken into account,such as the contextual fixing of deictic reference,obedience to cooperative principles ofconversation, and so forth. In particular, ourresearch suggests that test sentences whosepragmatic presuppositions are unsatisfied in theexperimental situation are also unlikely to provideresults allowing an accurate assessment of achild's knowledge of syntactic principles. It isnecessary to establish· which kinds ofpresuppositions children are sensitive to, and toensure that these are satisfied in experimentaltasks.

2.1. Parsing

2.1.1. Subjacency. One universal constraintwhich should be innate is Subjacency. Subjacencyprohibits extraction of cohstituents from variousconstructions, including relative clauses.However, in an experimental study by Otsu(1981), many children responded as if theyallowed extraction from relative clauses inanswering questions about the content of pictures.For example, children saw a picture of a girl usinga crayon to draw a monkey who was drinking milkthrough a straw. They were then asked to respondto question (1).

(1) What is Mary drawing a picture of amonkey that is drinking milk with?

Otsu found that many children responded to (1) ina way that appeared to violate Subjacency. In thiscase, the answer that is in apparent violation ofSubjacency is "a straw." This is because "a straw"is appropriate only if the what has been movedfrom a position in the monkey drinking milkclause as shown in (2a), rather than from theMary drawing picture clause as shown in (2b).

(2) a) *What is Mary drawing a picture of amonkey [that is drinking milk with _]?

b) What is Mary drawing a picture of amonkey [that is drinking milk] with_?

Competence and Perfonnance in Child Language 121

But the monkey drinking milk clause is a relativeclause, and Subjacency prohibits the what frommoving out of it. Thus the only acceptablestructure is (2b), and the only acceptable answeris "a crayon." If these data are interpreted solelyin terms of children's grammatical knowledge,then the conclusion would then have to be thatknowledge of Subjacency sets in quite late in atleast some children.

As we noted earlier, Otsu suggested that theinnateness of Subjacency could be salvaged byshowing that the children who appeared to violateSubjacency had not yet mastered the phrasestructure of relative clauses (of sufficientcomplexity to contain an extractable noun phrase).When he cOftducted an independent test ofknowledge of relative clause structure, he found,as predicted, a correlation between phrasestructure and Subjacency application in thechildren's performance. However, the children'sperformance was still surprisingly poor: 25% ofthe children who were deemed to have masteredrelative clauses gave responses involvingungrammatical Subjacency violating extractions

. from relative clauses.We have argued (Crain & Fodor, 1984) for an

alternative analysis of Otsu's data, which makes itpossible to credit children with knowledge of bothphrase structure principles and constraints ontransformations from an early age. We claim thatchildren's parsing routines can influence theirperformance on the kind of sentences used in theSubjacency test; in particular, that there arestrong parsing pressures encouraging subjects tocompute the ungrammatical analysis of suchsentences. Until a child develops sufficientcapacity to override these parsing pressures, theymay mask his syntactic competence, making himlook as if he were ignorant of Subjacency.

A powerful general tendency in sentence parsingby adults is to attach an incoming phrase low inthe phrase marker if possible. This has beencalled Right Association; see Kimball (1973),Frazier and Fodor (1978). In sentence (3), forexample, the preferred analysis has with NPmodifying drinking milk rather than modifyingdrawing a picture, even though in this case bothanalyses are grammatically well-formed becausethere has been no WH-movement.

(3) Mary is drawing a picture of a monkey thatis drinking milk with NP.

To see how strong this parsing pressure is, notehow difficult it is to get the sensible interpretationof (3) when a crayon is substituted for NP. This

Right Association preference is still present if theNP in (3) is extracted, as in (1). The word with in(1) still coheres strongly with the relative clause,rather than with the main clause. The result isthat the analysis of (2) that is most immediatelyapparent is the ungrammatical (2a) in which whathas been extracted from the relative clause. Sincethis 'garden path' analysis is apparent to mostadults, it is hardly surprising if some of Otsu'schild subjects were also tempted by it andresponded to (1) in the picture verification task bysaying "a straw" rather than "a crayon."

We conducted several experiments designed toestablish the plausibility of this claim that therelatively poor performance of children onsentences like (1) is due to parsing pressuresrather than to ignorance of universal constraints.In the first experiment, we tested children andadults on complement-clause questions as in (4).Subjacency does not prohibit extraction fromcomplement clauses, so if there were no RightAssociation effects this sentence should be fullyambiguous, with both interpretations equallyavailable.

(4) What is Bozo watching the dog jumpthrough?

That is, given a picture in which Bozo the clownis looking through a keyhole at a dog jumpingthrough a hoop, it would be correct to say eitherthe "the keyhole" or "a hoop." Intuitively, though,the interpretations are highly skewed for adults,with a strong preference for the Right Associationinterpretation ("hoop") in which the prepositionattaches within the lower clause. Our experimentshowed that the same is true for children. Wetested twenty 3- to 5-year-olds (mean age 4;6) onthese sentences using a picture verification taskjust like Otsu's, and 90% of their responses werein accord with the Right Associationinterpretation.2

Thus children and adults alike are stronglyswayed by Right Association. This is an importantresult. To the best of our knowledge the questionof whether children's parsing strategies resemblethose of adults has not previously beeninvestigated. But children certainly should showthe same preferences as adults, if the humansentence parsing mechanism is innatelystructured. And the parsing mechanism certainlyshould be innately structured, because it would bepointless to be born knowing a lot of facts aboutlanguage if one weren't also born knowing how touse those facts for speaking and understanding. Itis satisfying, then, to have shown that children

122 Crain and Fodor

exhibit Right Association. And the fact that theydo offers a plausible explanation for why so manyof them failed Otsu's Subjacency test-they werelistening to their parsers rather than to theirgrammars.

Our other experiments in support of thisconclusion were designed to show that even peoplewhose knowledge of Subjacency is not in doubt­Le., adults-are also tempted to violateSubjacency when it is in competition with RightAssociation. We ran Otsu's Subjacency test onadults just as he did with the children. The adultsgave Subjacency-violating low attachmentresponses to 21% of these questions. This was notquite as high a rate as for the children, but as wehave noted, adults surely have a greater capacitythan children do for checking an illicit analysisand shifting to a less preferred but well-formedanalysis before they commit themselves to aresponse. In an attempt to equalize adult self­monitoring capacities with those of children, were-ran the Subjacency experiment with anadditional distracting task (= listening for adesignated phoneme in the stimulus sentence).Under these conditions the adults gaveSubjacency violating responses to 29% of therelative clause constructions, a slightly higherrate than the 25% for Otsu's child subjects.

Escalating still further, we changed thesentences so that the grammatically well-formedanalysis was semantically or pragmaticallyanomalous, as in (5).

(5) What color hat is Barbara drawing a pictureof an artist with?

Under these circumstances, where the semanticsclearly favored the Subjacency-violating analysis,75% of adults' responses violated Subjacency. Thismakes it very clear that linguistic competencemay not always be revealed by linguisticperformance.

Finally, we ran another study, in which weasked adults to classify sentences as ambiguous orunambiguous. The sentences were spoken in turnwith only a few seconds between them, and therewere 72 of them, so the task was fairlydemanding. The materials included complementquestions like (4) and relative clause questionslike (1), as well as ambiguous and unambiguouscontrol sentences of many varieties. The resultsshowed a 62% ambiguity detection rate for theambiguous control sentences, with a 16% 'falsealarm' rate for the unambiguous controlsentences. Thus the subjects were able to copewith the task tolerably well, though not perfectly.

What was interesting was that the ambiguity ofthe complement questions was detected only 48%of the time, in line with our claim that RightAssociation obscures the alternative reading withthe prepositional phrase in the main clause. Andmost interesting of all was that 80% of the relativeclause questions were judged to be ambiguous,even though Subjacency prohibits one analysis

. and renders them unambiguous. Our explanationfor this extraordinary result is that the subjectsfirst computed the Right Association analysisfavored by their parsing routines, then recognizedthat this was unacceptable because of Subjacency,and so rejected it in favor of the analysis with theprepositional phrase in the main clause. Weassume that it was this rapid shift from oneanalysis to the other that gave our subjects such' astrong impression that these sentences wereambiguous. Note that if this misanalysis-with­revision occurs 80% of the time for adults, only aslight handicap in children's ability to revisewould be sufficient to account for their errors.

To sum up: we still have no positive proof thatSubjacency is innate, but at least now there is noevidence against it. Our experiments make itplausible that children as young as can be testedare like adults both with respect to theirknowledge of this universal constraint and withrespect to their parsing routines-they are justnot very good yet at coping with conflicts betweenthe two.

2.1.2. Backward pronominalization. Afundamental constraint on natural language is thestructure dependence of linguistic rules. Theinnateness hypothesis implies that children'searliest grammars should also exhibit structuredependence-even if their linguistic experiencehappens to be equally compatible with structure­independent hypotheses. However, it has beenproposed that children initially hypothesize astructure-independent constraint on anaphora,prohibiting all cases of backwardpronominalization (Solan, 1983). 3 Backwardpronominalization consists of coreference betweena noun phrase and a preceding pronoun, asindicated by the indices in (6).

(6) That hei kissed the lion made the duckihappy.

We will argue that children do in fact permitbackward pronominalization, subject to structure­dependent constraints. We contend that theappearance of a general restriction againstbackward pronominalization is due to a parsingpreference for the alternative 'extra-sentential'

Competence and Performance in Child Language 123

reading of the pronoun in certain comprehensiontasks. The results of a new comprehensionmethodology show that children as young as 2;10admit the same range of interpretations forpronouns as adults do.

Two sources of evidence have been cited asevidence that children up to 5 or 6 years uniformlyreject backward pronominalization. First, childrenwho are asked to repeat back a sentence such as(7) often respond by converting it into a forwardpronominalization construction, as in (8) (Lust,1981).

(7) Because she was tired, Mommy was sleeping.

(8) Because Mommy was tired, she was sleeping.

The fact that these children took the trouble toexchange the pronoun and its antecedent certainlyindicates that they disfavor backward pronom­inalization in their own productions. But it doesnot show that the backward pronominalizationinterpretation is not compatible with the child'sgrammar, as suggested by Solan (1983). To thecontrary, the conversion of (7) to (8) shows thatchildren do accept backward pronominalization incomprehension; for they would think of (8) as anacceptable variant of (7) only if they wereinterpreting the pronoun in (7) as coreferentialwith the subsequent lexical noun phrase (Lasnik& Crain, 1985).

Second, it has been found that when the acting­out situation for a sentence like (6) includes apotential referent for he other than the duck (e.g.,a farmer), this unmentioned object is usuallyfavored by the children as the referent of thepronoun (Solan, 1983; Tavakolian, 1978). Incontrast to the prevailing view, we would attributethis to a parsing preference for the extra­sentential interpretation of the pronoun; it doesnot have to be taken as evidence that childrenhave a grammatical prohibition against backwardanaphora. Our suggestion, then, is that children'sknowledge might be comparable to, that ofadults, even if their performance differs.

It is particularly important to keep thisdistinction in mind for potentially ambiguousconstructions such as these. When a sentence hasmore than one possible interpretation, theinterpretation that children select can tell uswhich interpretation they prefer; it cannot showthat others are unavailable to them. After all,adults also exhibit biases in connection withambiguous constructions, but this does not lead usto accuse them of ignorance of alternativeinterpretations. To establish how much children

actually do know, we should look for the factorsthat might be biassing their interpretations, andalso for ways of minimizing this bias so thatinterpretations which are less preferred butnevertheless acceptable to them have a chance ofshowing through.

The most likely general source of bias againstbackward pronominalization is the fact thatinterpretation of the pronoun would have to bedelayed until the antecedent is encountered laterin the sentence. This retention of uninterpreteditems may strain a child's limited workingmemory. There is some evidence for thisspeculation. Hamburger and Crain (1984) havenoted that children show a tendency to interpretadjectives immediately, without waiting for theremainder of the noun phrase, even in caseswhere this leads them to give incorrect responses.And Clark (1971) has observed errors attributableto children's tendency to act out a clauseimmediately without waiting for other clauses inthe sentence. The only way to interpret thepronoun immediately in a sentence like (6) is toassign it an extra-sentential referent, as childrentypically do.

If this proposal is correct, it should be thatchildren will accept backward pronominalizationin an experimental task that presses subjects toaccess every interpretation they can assign to asentence. Crain and McKee (1985) used atrue/false paradigm in which subjects judge thetruth value of sentences against situations actedout by the experimenter. The sentences were as in(9), where either a coreferential reading or anextra-sentential reading of the pronoun ispossible.

(9) When he went into the barn, the fox stolethe food.

On each trial, a child heard a sentence followinga staged event acted out by one of twoexperimenters, using toy figures and props. Thesecond experimenter manipulated a puppet,Kermit the Frog. Following each event, Kermitsaid what he thought had happened on that trial.The child's task was to indicate whether or not thesentence uttered by Kermit accurately describedwhat had happened. Children were asked to feedKermit a cookie ifhe said the right thing, that is,if what he said was what really happened. In thisway, 'true' responses were encouraged in theexperimental situation. But sometimes Kermitwould say the wrong thing, if he wasn't payingclose attention. When this happened, the childwas asked to make Kermit eat a rag. (In pilot

124 Crain and Fodor

work without the rag ploy, we had found thatchildren were reluctant to say that Kermit hadsaid something wrong.)

To test for the availability of bothinterpretations of an ambiguous sentence like (9),children judged it twice during the course of theexperiment, once following a situation in which afox stole some chickens from inside a barn (for thebackward pronominalization interpretation), andonce following a situation in which a man stolesome chickens while a fox was in a barn (for theextra-sentential interpretation).

Children accepted the backward anaphorareading for all the ambiguous sentences 73 of thetime. The extra-sentential reading was accepted81 of the time, but the difference was notsignificant. Much the same results were obtainedeven for the 7 youngest children, whose ages werefrom 2;10 to 3;4. Only two of the 62 subjectsconsistently rejected the backward anaphorareading. Thus most children find the backwardanaphora reading acceptable, although it mightnot be preferred if they were forced to choosebetween inte.rpretations, as in previouscomprehension studies.

We should note that a variety of controlsentences were also tested to rule out other, lessinteresting, explanations of the children'sperformance. For example, the children rejectedsentence (10) following a situation in whichStrawberry Shortcake did eat an ice cream, butnot while she w~s outside playing. This showsthat they were not simply ignoring thesubordinated clauses of sentences in decidingwhether to accept or reject them.

(10) When she was outside playing,Strawberry Shortcake ate an ice cream.

Sentences like (11) were also tested in order toestablish that subjects were not merely givingpositive responses to all sentences, regardless oftheir grammatical properties.

(11) He stole the food when the fox went intothe barn.

The difference between (11) and the acceptablebackward pronominalization in (9) is that in (11)the pronoun is in the higher clause and c­commands the fox, while in (9) the pronoun is inthe subordinate clause and does not c-commandthe fox. (A node A in a phrase marker is said to c­command a node B if there is a route from A to Bwhich goes up to the first branching node above A,and then down to B. Note that c-command is astructure-dependent relation.) There is auniversal constraint that prohibits a pronoun from

c-commanding its antecedent. And indeed thechildren did reject (11) 87% of the time. Note thatthis positive result shows that the children haveearly knowledge not only of the absence of linearsequence conditions on pronominalization, butalso of the existence of structural conditions suchas c-command. (See also Lust, 1981, and'Goodluck, 1986.)

2.1.3. Subject/Auxiliary Inversion. Anotherstudy (Crain & Nakayama, 1987) also exploredthe tie between children's errors in acquisitiontasks and sentence processing problems. Thisstudy was designed to test whether children givestructure-dependent or structure-independentresponses when they are required to transformsentences by performing Subject/Auxiliaryinversion. As Chomsky (1971) pointed out,transformational rules are universally sensitive tothe structural configurations in the sentences towhich they apply, not just to the linear sequenceof words.

The procedure in this study was simply for theexperimenter to preface dec1aratives like (12) withthe carrier phrase "Ask Jabba if ...," as in (13).

(12) The man who is running is bald.

(13) Ask Jabba if the man who is running isbald.

The child then had to pose the appropriateyes/no questions to J abba the Hutt, a figure from"Star Wars" who was being manipulated by one ofthe experimenters. Following each question,Jabba was shown a picture and would respond"yes" or "no."

The sentences all contained a relative clausemodifying the subject noun phrase. The correctstructure-dependent transformation moves thefirst verb of the main clause to the front of thesentence, past the whole subject noun phrase, asin (14). An incorrect, structure-independenttransformation would be as in (15), where thelinearly first verb in the word string (whichhappens to be the verb of the relative clause) hasbeen fronted.

(14) Is the man who is running bald?

(15) *Is the man who running is bald?

For simple sentences with only one clause suchas (16), which are more frequent in a young child'sinput, both versions of the transformation rulegive the correct result.

(16) Ask Jabba if the man is bald.

(17) Is the man bald?

Competence and Perfonnance in Child Language 125

It is only on the more complex sentences that theform ofthe child's rule is revealed.

The outcome was as predicted by the innatenesshypothesis: children never produced an incorrectsentence like (15). Thus, a structure-independentstrategy was not adopted in spite of its simplicityand in spite of the fact that it produces the correctquestion forms in many instances. The findings ofthis study thus lend further support to the viewthat the initial state of the human languagefaculty contains structure-dependence as aninherent property.

The children did make some errors in thisexperiment, and we observed that most of themwere in sentences with a long subject noun phraseand a short main verb phrase, as in (18).

(18) Is the boy who is holding the plate crying?

By contrast, there were significantly fewererrors in sentences like (19), which has a shortersubject noun phrase and a longer verb phrase.

(19) Is the boy who is unhappy watchingMickey Mouse?

This kind of contrast is familiar in parsingstudies with adults. In particular, Frazier andFodor (1978) showed that a sequence consisting ofa long constituent followed by a short constituentis especially troublesome for the (adult) parsingroutines;4 a short constituent before a long one ismuch easier to parse. The distribution of thechildren's errors in the Subject-AuxiliaryInversion task may therefore be indicative not ofinadequate knowledge of the inversion rule, but ofan adult-like processing sensitivity to interactionsbetween structure and constituent length.

A follow-up study to test this possibility wasconducted by Nakayama (1987). Nakayamasystematically varied both the length and thesyntactic structure of the sentences to betransformed by the children. The children madesignificantly fewer Subject-Auxiliary Inversionerrors in response to embedded questions withshort relative clauses (containing intransitiveverbs) as compared to those with long relativeclauses (containing transitive verbs). With lengthheld constant, the children had more difficultywith relative clauses that had object gaps, as in(20), than with relative clauses that had subjectgaps, as in (21) (although this effect was not quitesignificant).

(20) The ball the girl kicked is rolling.

(21) The boy who was slapped is crying.

The ease of subject gap constructions, ascompared to object gap constructions, has beenfound in a number of other studies in languagedevelopment, in language impaired populations,and in experiments on adult sentence processing(where the question of syntactic competence is notin doubt). It seems reasonable to interpret theseresults as confirming that children's error rates inlanguage tests are highly sensitive to thecomplexity of the sentence parsing that isrequired.

2.2. PresuppositionSyntactic parsing is not the only factor that has

been found to mask knowledge of syntacticprinciples. Test sentences whose pragmaticpresuppositions are unsatisfied in theexperimental situation have been found to resultin inaccurate assessments of children's structuralknowledge. In this section we consider twoexperiments that point to the relevance ofpresuppositional content in sentenceunderstanding.

The structures we discuss here are relativeclauses and temporal adverbial clauses. A word ofclarification is needed before we proceed. Up tillnow we have restricted the scope of the innatehypothesis to universal constraints (likeSubjacency, and structure-dependence), whichcould not in principle be learned from normallinguistic experience (Le., without extensivecorrective feedback). But now we want to extendthe innateness hypothesis to a broader class oflinguistic knowledge, knowledge of universal typesof sentence construction. We cannot plausiblyclaim that every aspect of these constructions isinnate. Rather, every construction will have someaspects that are determined by innate principles,and other aspects that must be learned. And thebalance between these two elements varies fromconstruction to construction. So it is perfectlyacceptable on theoretical grounds that someconstructions should be acquired later thanothers. However, the innateness hypothesis is notcompatible with just any order of acquisition. Itpredicts early acquisition of constructions thatChomsky calls 'core' language, i.e., theconstructions that have strong assistance frominnate principles with just a few parameters to beset by learners on the basis of experience. It wouldbe surprising to discover that knowledge of theseconstructions was significantly delayed once therelevant lexical items had been learned. In theabsence of a plausible explanation, this would putthe innateness hypothesis at risk.

126 Crain and Fodor

We noted in section 1 a range of possibleexplanations of apparently delayed knowledge oflinguistic facts. In the present case they wouldinclude the following:

• the construction does not, after all, belong tothe core but is 'peripheral' and hence shouldbe acquired late;

• children don't hear this construction until quitelate in the course oflanguage development andso could not be expected to know it exists;

• the core principles in question undergomaturation and so are not accessible at earlystages of acquisition;

• the experimental data are faulty and childrendo indeed have knowledge of this construction.

We will argue for this last alternative. And just asin the previously described studies of innateconstraints, we will lay the blame for themisleading experimental data on the fact thattraditional experimental paradigms do not makesufficient allowance for the limited memory andcomputational capacities of young children. Onceagain, our story is that non-linguistic immaturitycan create the illusion of linguistic immaturity.

2.2.1. Relative Clauses. Children typicallymake more errors in understanding sentencescontaining relative clauses (as in 22) thansentences containing conjoined clauses (as in 23),when comprehension is assessed by a figuremanipulation (act-out) task.

(22) The dog pushed the sheep that jumpedover the fence.

(23) The dog pushed the sheep and jumpedover the fence.

The usual finding that (22) is more difficult forchildren than (23) up to age 6 years or so has beeninterpreted as an instance oflate emergence of therules for subordinate syntax in languagedevelopment (e.g., Tavakolian, 1981). However,though coordination may be innately favored oversubordination, it is also true that subordination isubiquitous in natural language; relative clauseconstructions are very close to the 'core.' Soignorance of relative clauses until age 6 wouldstretch the innateness hypothesis.

Fortunately this is not how things stand.Hamburger and Crain (1982) showed that thesource of children's performance errors on thistask is not a lack of syntactic knowledge. Byconstructing pragmatic contexts in which thepresuppositions of restrictive relative clauses weresatisfied, they were able to demonstrate mastery

of relative clause structure by children as youngas 3 years. There are two presuppositions in (22):(i) that there are at least two sheep in the context,and (ii) that one (but only one) of the sheepjumped over a fence prior to the utterance. Thereason why previous studies failed to demonstrateearly knowledge of relative clause constructions,we believe, is that they did not pay scrupulousattention to these pragmatic presuppositions. Forexample, subjects were required to act out themeaning of a sentence such as (22) in contexts inwhich only one sheep was present. The poorperformance by young children in theseexperiments was attributed to their ignorance ofthe linguistic properties of relative clauseconstructions. But suppose that a child did knowthe linguistic properties, but that he also wasaware of the associated presuppositions. Such achild might very well be unable to relate hiscorrect understanding of the sentence to theinappropriate circumstances provided by theexperiment. Adult subjects may be able to 'seethrough' the unnaturalness of an experimentaltask to the intentions of the experimenter, but it isnot realistic to expect this ofyoung children.

Following this line of reasoning, Hamburger andCrain (1982) made the apparently minor change ofadding two more sheep to the acting out situationfor sentence (22), and obtained a much higherpercentage of correct responses. The mostfrequent remaining 'error' was failure to act outthe event described by the relative clause, butsince felicitous usage presupposes that this eventhas already occurred, this is not really an errorbut is precisely the kind of response that iscompatible with perfect- comprehension of thesentence. This interpretation of the data issupported by the fact that there was a positivecorrelation between incidence of this responsetype and age.5

We have conducted another series of studies onrelative clauses, trying several other techniquesfor assessing grammatical competence. In onestudy, we employed a picture verificationparadigm to see if children could distinguishrelative clauses from conjoined clauses, despitethe claim of Tavakolian (1981) that theysystematically impose a conjoined clause analysison relatives. In this study, seventeen 3- and 4­year-oIds responded to relative clauseconstructions like (24).

(24) The cat is holding hands with a man whois holding hands with a woman.

(25) The cat is holding hands with a man andis holding hands with a woman.

Competence and Performance in Child Language 127

This sentence was associated with a pair ofpictures, one that was appropriate to it and onethat was appropriate to the superficially similarconjoined sentence (25). 8eventy percent of the 3­year-oIds' responses and 94 of the 4-year-olds'responses matched sentences with the appropriatepicture rather than with the one depicting theconjoined clause interpretation.

A second technique we tried used a 'silliness'judgment task (see Hsu, 1981) to establishwhether children can differentiate relative clausesfrom conjoined clauses. Ninety-one percent of theresponses of the twelve 3- and 4-year-olds testedcategorized as 'silly' sentences such as (26),although sentences such as (27) were accepted assensible 87% of the time.

(26) The horse ate the hay that jumped overthe fence.

(27) The man watched the horse that jumpedover the fence.

Notice that sentence (26) would not be anomalousif the that -clause were misinterpreted as an and ­clause, or if it were interpreted as extraposed fromthe subject NP; in both cases, the horse would bethe understood subject of the relative clause. Theresults therefore indicate that most childreninterpret the that-clause in this sentencecorrectly, i.e., as a subordinate clause modifyingthe hay. Informal testing of adults suggests thatthe only respect in which children and adultsdiffer on the interpretation of relative clauses isthat the adults are somewhat more likely toaccept the extraposed relative analysis as well,though even for adults this analysis is much lesspreferred.

A third experiment, on the phrase structure ofrelative clause constructions, indicates thatchildren, like adults, treat a noun phrase and itsmodifying relative clause as a single constituent,inasmuch as they can construe it as theantecedent for a pronoun such as one.

In a picture verification study, fifteen 3- to 5­year-olds responded to the instructions in (28).

(28) The mother frog is looking at an airplanethat has a woman in it. The baby frog islooking at one too. Point to it.

Ninety-three percent of the time the subjectschose the picture in which the baby frog waslooking at an airplane with a woman in it, inpreference to the picture in which the baby frogwas looking at an airplane without a woman in it.That is, the relative clause was included in thenoun phrase assigned as antecedent to thepronoun.

In short: the weight of evidence now indicatesthat children grasp the structure and meaning ofrelative clause constructions quite early in thecourse of language acquisition, as would beexpected in view of the central position of theseconstructions in natural language..

2.2.2. Temporal Terms. Another line ofresearch has yielded support for the claim thatpresupposition failure is implicated in children'spoor linguistic performance. These studiesemployed sentences containing temporal clauseswith before and after, as in (29).

(29) Push the red car to me before/after youpush the blue car.

Clark (1971) and Amidon and Carey (1972) haveclaimed that most normal, 3- to 5-year-olds do notunderstand these sentences appropriately. 8inceAmidon and Carey established that the childrenwere familiar with concepts of temporal sequence(e.g., as expressed by words like first and last), theimplication is that the structure of these adverbialclauses is beyond the scope of the child's grammarat this age.

However, the acting-out tasks employed in thesestudies were once again unnatural ones whichignored the presuppositional content of the testsentences. Felicitous usage of sentence (29)demands that the pushing of the blue car hasalready been contextually established by thehearer as an intended, or at least probable, futureevent; but this was not established in theseexperimental tasks. It is very likely, then, thatthese studies underestimated children's ability tocomprehend temporal subordinate clauses. Forexample, Amidon and Carey reported that fiveand six year old children who were not given anyfeedback frequently failed to act out the actiondescribed in the subordinate clause. Johnson(1975) found that four and five year old childrencorrectly acted out commands such as those in (30)only 51% of the time; again, the predominanterror was failure to act out the action described inthe subordinate clause.

(30) a Push the car before you push thetruck. (81 before 82)

b. Mter you push the motorcycle, pushthe bus. (After 81, 82)

c. Before you push the airplane, push thecar. (Before 82, 81)

d Push the truck after you push thehelicopter. (82 after 81)

128 Crain and Fodor

Table 1.

Note that the relevant factor is whethersubordinate clause information was provided inadvance. An analysis of variance confirmed thatthe mere amount of information provided makesno significant difference. The FI group performedbetter than the F group by only 6 percentagepoints, which does not approach statisticalsignificance. And the I group performed just alittle worse (non-significantly again) than the NCgroup.

Although our study was not specificallydesigned to assess age differences, we performed apost hoc breakdown of correct responses by twoage groups: under 4;4, and 4;4 and over. The oldergroup, as one would expect, performed somewhat

If he chose the bus, for example, a typicalcommand would be (31).

(31) Push the car before you push the bus.

On the other hand, a subject in the I group whohad chosen the bus would be given the command(32).

(32) Push the bus before you push the car.

Fifty-six children participated in the study,ranging in age from 3;4 to 5;10 (mean 4;5). Eachchild was assigned to one of the four groups,which were of equal size and approximatelymatched for age. The 'game' equipment consistedof 6 toy vehicles arranged in a row on a tablebetween the child and the experimenter. Thestimulus set consisted of 12 commands spoken bythe experimenter which the child was to act out.There were three sentences of each of the fourtypes illustrated in (30) above. We were careful tobalance order of choice with order of action andassignment to clause type.

The results showed a significant differencebetween the F and FI groups on one hand, and theI and NC groups on the other. Table 1 shows thepercentages of correct responses, where a correctresponse consisted of performing both actions inthe sequence specified by the sentence.

Main Clause Information

55%

77%

mean

66%65%

+

FI 80% F 74%

I 51% NC 59%

+

mean

Subordinate

Clause Information

Crain (1982) satisfied the presupposition of thesubordinate clause by having the subordinateclause act correspond to an intended action by thesubject, and observed a striking increase inchildren's performance. To satisfy thepresupposition, children were asked, before eachcommand, to choose a toy to push on the nexttrial. The child's intention to push a particular toywas incorporated into the command that wasgiven on that trial. For instance, sentence (30d)could be used felicitously for a child who hadexpressed his intent to push the helicopter.Correct responses (i.e., responses in which boththe main clause and subordinate clause actionwere performed, and in the correct order) wereproduced 82% of the time. Crain's interpretationof these results was that the children's improvedperformance was due to the satisfaction of thepresupposition of the subordinate clause.

However, we now note that the results of thatstudy are open to another interpretation. It maybe that improved performance was not duespecifically to the contextual appropriateness ofthe sentence, but to the fact that the child's taskwas simplified because he was provided with moreadvance information concerning what his taskwould be. In the act-out or 'do-what-I-say'paradigm applied to temporal terms, the childmust discern two aspects of the command: (i)which two toys to move, and (ii) in which order tomove them. If the child has established his intentto move a particular toy, his task involving (i) issimplified. Thus, improved performance may bedue to the satisfaction of presuppositions or it maybe due to the additional information the childpossesses.

Another study was conducted to disentanglethese two factors (Gorrell, Crain, & Fodor, 1989).In this study, there were four groups of subjects.One group, the Felicity Group (F), was givencommands containing before and after with priorinformation about the subordinate clause action,just as in the previous experiment. A secondgroup, the Information Group (I) received priorinformation about the main clause action; notethat this does not satisfy the presupposition of thesentence. There was also a third group, the NoContext Group (NC), who received no advanceinformation at all, and a fourth group, the Felicityplus Information Group (FI), who receivedinformation over and above what would satisfy thefelicity conditions since they chose both actions inadvance. Consider, for example, a subject in the Fgroup. He would be asked to choose a toy to push.

Competence and Performance in Child Language 129

an unnatural experimental situation, butchildren, with their more limited cognitive andsocial skills, apparently do not have this ability.Consequently, they are highly sensitive topragmatic infelicities. And therefore theirlinguistic knowledge can be accurately appraisedonly by tests which include controls to insure thatthey are not penalized by their knowledge ofpragmatic principles.

2.3. PlansAnother possible source of poor performance by

children is in formulating the action plans whichare needed in order to obey an imperative, or actout the content of a declarative sentence whichthey have successfully processed and understood.As we use the term, a plan is a mentalrepresentation used to guide action. A plan maybe simple in structure, consisting of just a list ofactions to be performed in sequence; or it may beinternally complex, with loops and branches andother such structures now familiar in computerprograms.

Formulating a plan is a skill that makesdemands on memory and computationalresources. In certain experimental tasks, thesedemands may outweigh those of the purelylinguistic processing aspects of the task. So whenchildren perform poorly, it is important toconsider the possibility that formulating, storingor executing the relevant action plan is the sourceof the problem, rather than imperfect knowledgeof the linguistic rules or an inability to apply themin parsing the sentence at hand.

2.3.1. Prenominal Modifiers. The first studyon plans that we conducted was in response to theclaim by Matthei (1982) and Roeper (1972) that 4­to 6-year-olds have difficulty in interpretingphrases such as (33) containing both an ordinaland a descriptive adjective.

(33) the second striped ball

?onfronted with an array such as (34), manychIldren selected item (ii), i.e., the ball which issecond in the array and also is striped, ratherthan item (iv) which is the second of the stripedballs (counting from the left as the children weretrained to do).

(34) Array for "the second striped ball"

better than the younger group. What is perhapsmost interesting is that the younger group appearto be even more sensitive than the older group tothe proper contextual embedding ofutterances.6

A breakdown of the types of errors that occurredreveals that the predominant errors are (i) actingout the main clause only, and (ii) reversing thecorrect order of the actions. As noted above forrelative clause constructions, acting out the mainclause only is a quite reasonable response giventhat the context failed to satisfy the subordinateclause presupposition. And in fact most of theseerrors were found in the I and NC groups.7

Reversals were the most frequent error type in thestudy though they constituted only 19% of allresponses. These errors may reflect a genuine lackof comprehension of either the temporal terms orthe relevant syntactic structure. However no childin either the F or FI groups produced a consistentresponse pattern which would indicate that thiswas the case, so it seems more likely that theseerrors were due primarily just to occasionalinattention.

The main conclusion we draw from these resultsis that children, from a very young age, are indeedsensitive to the proper contextual embedding oflanguage. Their performance is facilitated bysatisfying the presuppositions of temporalsubordinate clauses, and information which doesnot satisfy the presuppositions does not result infacilitation.

A secondary conclusion is that children doconstruct the appropriate syntactic structure forsentences with embedded clauses. If the childrenin our study had failed to distinguish main fromsubordinate clauses (e.g., by assigning a 'flat'conjunction-type structure to the experimenter'scommands), we would not expect to find thedifference between the F and I groups weobserved. Nor is it plausible to suggest that thechildren relied upon a structure-independentformula of 'old information precedes newinfor~ation.' For example, for the F group, thenew mformation was always in the main clause. Ifchildren were assuming that old informationwould be first, we would have expected relativelypoor performance from the F group on sentencesin which the main clause preceded thesubordinate clause. In fact, no such effect wasobserved.

In sum: once again, the linguistic knowledge ofyoung children, when freed of interferinginfluences, appears to be quite advanced. Adultshave the ability to set aside contextual factors in

~~

(i) (ii)

~~

(iii) (iv) (v)

130 Crain and Fodor

Hamburger and Crain found that the childrenconsistently responded to the second instructionby pointing to (v) rather than (iv), showing thatthey took the proform one to corefer withexpressions like striped ball. Thus it appears thatthey do know the structure (35).

Finally, we note two experiments by Hamburgerand Crain (1987). The purpose of theseexperiments was to provide empirical support forour claim that response planning is an importantfactor in psycholinguistic tasks, independently ofsyntax and semantics.

The first experiment attempts to show thatchildren's ability to comprehend a phrase isinversely related to the complexity of theassociated plan. For this purpose we comparephrases that are arguably equal to each other insyntactic complexity but differ in plan complexity.Examples are shown in (39), in increasing order ofcomplexity of plans.

(39) 1. John's biggest bookii. second green book

Ill. second biggest book

change in method is to withhold the display whilethe sentence is being uttered, so that formationand execution of the plan are less likely tointerfere with each other. A dramaticimprovement in performance on a phrase like (33)also results from first asking the child to identifythe first striped ball, which forces him to plan andexecute part ofthe procedure he will later need for(33). Facilitating the procedural aspects of thetask thus makes it possible for the child to revealhis mastery of the syntax and semantics of suchexpressions.

Hamburger and Crain also found quite directevidence that children do not assign the 'flatstructure' analysis. The standard assumption inlinguistics is that proforms corefer with asyntactic constituent. In the correct structure (35),the words striped and ball form a completeconstituent, but in the incorrect structure (36)they do not. Thus the children should permit theproform one to corefer with striped ball only ifthey have the correct hierarchical structure. Tofind out whether they permit this coreference,they were tested on the instructions in (37), withthe array in (38).

(37) Point to the first striped ball; point to thesecond one.

The empirical finding, then, appears to be thatchildren assign an interpretation that is not thesame as an adult would assign to expressions ofthis kind. This difference is attributed by Mattheito children's failure to adopt the hierarchicalphrase structure internal to a noun phrase thatcharacterizes the adult grammar. This structureis shown in (35). Instead, Matthei argues thatchildren adopt a 'flat structure' for phrases of thiskind, with both the ordinal and the descriptiveadjective modifying the noun directly as in (36).

(35) (36)

JIo'P NP

", / ,

DET N' DET N'

/,

" I'ADI N' ADI ADI N

/ ,I I IADI N

the second slriped ball

I Ilite second striped ball

Any divergence between children's and adults'grammars poses a problem from the standpoint oflanguage acquisition theory; namely, explaininghow the child ultimately converges on the adultgrammar without correction or other 'negative'feedback. Fortunately, there is no need to assumean error in the children's grammar in this case, forthere is an alternative component of the languageprocessor in which the errors might have arisen.In a series of experiments, Hamburger and Crain(1984) show that most children do assign the adultphrase structure and do understand the phrase.correctly as referring to the second of the stripedballs. The difficulty that children experiencearises when they attempt to derive from thisinterpretation a procedure for actually identifyingthe relevant item in the array. An analysis of thelogical structure of the necessary procedure showsit to be quite complex, significantly more so thanthe procedure for "count the striped balls," thekind of phrase Matthei used in a pretest in anattempt to show that children were able to copewith the nonsyntactic demands of the task.

This procedural account of the children's errorsis supported by the sharp improvement inperformance that results from three changes inmethod. One change is the inclusion of a pretasksession in which the children handle and counthomogeneous subsets of the items which aresubsequently used in an array. This experience isassumed to prime some of the procedural planningrequired in the main experimental task. A second

(38)

(i)

,-.-,_.(ii) (iii)

~~

(iv) (v)

Competence and Performance in Child Language 131

Planning complexity can also be deconfoundedfrom the complexity of semantic constituentprocessing. Note that semantic considerationslead to the prediction that (39i), would be hardest,because its word meanings have to combine in away contrary to the surface sequence of words(= the biggest ofJohn's books).

The pattern of children's responses supports thepredictions of our procedural account, and doesnot conform to the account based on syntactic orsemantic constituency. Children respondedcorrectly to phrases like (39i) 88% of the time.They gave correct answers only 39% of the timefor examples like (39ii), and they were only 17%correct for phrase like (39iii). Thus, thisexperiment provides clear evidence that plans, notlinguistic structures (syntactic or semantic) candetermine processing success and failure foryoung children.

The second experiment addresses the cognitivedifficulty of planning by prefacing the test with asequence of exercises designed to alleviate theplanning difficulty. This activity does not provideany extra exposure to the phrases tested;nevertheless, we anticipate a reduction in errorson these phrases. Consider a phrase such as "thesecond tallest building." This plan requires theinterpreter to identify its referent. The child mustintegrate sequential pairwise comparisons ofrelative size. In the pre-test activity the childwould be shown a display of several objects of onetype (say boxes), but of different sizes, and askedto hand the experimenter the biggest one. Then,once this object was removed from the array, theexperimenter asked the child to perform the taskagain, saying, "Now, find the biggest box in thisgroup." In this way the child would identify thesecond biggest box without ever hearing thephrase "the second biggest box" uttered.Children's comprehension of the phrase wastested before and after the preparatory task. Theygave significantly more correct responses (46%)following the preparatory task than before it (8%in this experiment). This result suggests thattheir difficulty with phrases of this sort stemsfrom the complexity of the response plans.

3. SENTENCE PRODUCTIONTo acquire a language is to learn a mapping

between potential utterances and associatedpotential meanings. Successful mastery shouldreveal itself in both comprehension andproduction. In the previous section we wereconcerned with studies of children'scomprehension, in which their knowledge is tested

by presenting utterances and observing theinterpretations that they assign. We now turn totests of children's competence which proceed inthe other direction: the input to the child is asituation, which has been designed to suggest aunique sentence meaning, and the behavior weobserve is the utterance by which the childdescribes that situation.

It would have been reasonable to expect that thesorts of nonsyntactic problems that presentobstacles for children in comprehension tasksmight prove to be as hard or even harder for themto overcome in production tasks. But we have notfound this to be the case. The results of recentelicited production studies are dramatically betterthan those of comprehension studies directed tothe same linguistic constructions. For example,Richards (1976) elicited appropriate uses of thedeictic verbs come and go from children age 4;0 ­7;7, while Clark and Garnica (1974) reported thateven 8-year-olds didn't consistently distinguishbetween come and go in a comprehension task.

The disparity between production andcomprehension studies is particularly strikingbecause it is the reverse of what one would expect.To find production superior to comprehension inchildren's language is as surprising as it would beto find production superior to comprehension inadult second-language learning, or to find recallsuperior to recognition in any psychologicaldomain. It is plausible to argue, therefore, thatthe superiority of production is only apparent, andis due to differences in the sensitivities ofproduction tests and comprehension tests. And thelogic of the situation suggests that it is thecomprehension tests that are deficient. After all,success is hard to argue with. With suitablecontrols, successful production by children is astrong indicator of underlying linguisticcompetence, as long as their productions are asappropriate and closely attuned to the context asadult utterances are. Because there are so manyways to combine words incorrectly, consistentlycorrect combinations in the appropriate contextsare not likely to come about by accident. On theother hand, failure on any kind of psychologicaltask cannot be secure evidence of lack of therelevant knowledge, since the knowledge may bepresent but imperfectly exploited.

As we saw in the previous section, com­prehension studies seem to be particularlysusceptible to problems of parsing, planning andso forth which impede the full exploitation oflinguistic knowledge. Production tasks appear tobe less hampered by these extra-grammatical

132 Crain and Fodor

factors. This is probably because productionavoids non-verbal response planning, which wehave seen is a major source of difficulty in act-outcomprehension tasks. It is worth noting also that

. in constructing contexts to elicit particularutterance types, we have no choice but to attend tothe satisfaction of the presuppositions that areassociated with the syntactic structures inquestion, because otherwise the subjects won'tutter anything like the construction that is beingtargeted. In elicited production it is delicatemanipulations of the communicative situationthat give one control over the subject's utterances.

3.1. Relative Clauses. In section 2 wepresented evidence ofyoung children's competencewith relative clauses. Further confirmation wasobtained by Hamburger and Crain (1982), usingan elicited production methodology. Pragmaticcontexts were constructed in, which thepresuppositions of restrictive relatives weresatisfied. It was discovered that children as youngas three reliably produce relative clauses in thesecontexts.

A context that is uniquely felicitous for arelative clause is one which requires the speakerto identify to an observer which of two objects toperform some action on. In our experiment, theobserver is blindfolded during identification of atoy, so the child cannot identify it to the observermerely by pointing to it or saying this / that one.Also, the differentiating property of the relevanttoy is not one that can be encoded merely with anoun (e.g., the guard) or a prenominal adjective(e.g., the big guard) or a prepositional phrase (e.g.,the guard with the gun), but involves a morecomplex state or action (e.g., the guard that isshooting Darth Vader). Young children reliablyproduce meaningful utterances with relativeclauses when these felicity conditions are met. Forexample:

(40) Jabba, please come over to point to the onethat's asleep. (3;5)

Point to the one that's standing up. (3;9)Point to the guy who's going to get killed.(3;9)

Point to the kangaroo that's eating thestrawberry ice cream. (3;11)

Note that the possibility of imitation is excludedbecause the experimenter takes care not to useany relative clause constructions in the elicitation

. situation. This technique has now been extendedto younger children (as young as 2;8), and to theelicitation of a wider array of relative clause

constructions, including relatives with object gaps(e.g., the guard that Princess Leia is standing on).

3.2. Passives. Borer and Wexler (1987) haveargued that A-chains, which are involved in thederivation of verbal passive constructions, are notavailable to children in the first few years.8 Borerand Wexler maintain that knowledge of A-chainsis innate, but becomes accessible only after thelanguage faculty undergoes maturational change.We were not convinced, however, that thismaturation hypothesis is necessitated by the facts.Rather, the facts seem to be consistent with A­chains being innate and accessible from theoutset.

One source of data cited in support of thematuration hypothesis is the absence of fullpassives in the spontaneous speech of youngchildren. But this of course is not incontrovertibleevidence that children's grammars are incapableof generating passives. Full passives are rarelyobserved in adults' spontaneous speech either, orin adult speech to children. But their paucity isnot interpreted in this case as revealing a lack ofgrammatical knowledge. Instead, it is understoodas due to the fact that the passive is a markedform which it is appropriate to use only in certaindiscourse contexts; in most contexts the active isacceptable and more natural, or a reduced passivewithout a by-phrase is sufficient. That is, theabsence of full verbal passives in adult speech· isassumed to be a consequence of the fact that it'sonly in rare situations that the full passive isuniquely felicitous. But the same logic thatexplains why adults produce so few full passivesmay apply equally to children. Perhaps they toohave knowledge of this construction, but do notuse it except where the communicative situation isappropriate.

We have tested this possibility in an experimentwith thirty-two 3- and 4-year-old children. (Crain,Thornton, & Murasugi, 1987). One experimenterasked the child to pose questions to anotherexperimenter. The pragmatic context wascarefully controlled so that questions containing afull verbal passive would be fully appropriate. Thefollowing protocol illustrates the elicitationtechnique:

Adult See, the Incredible Hulk is hitting one of thesoldiers Look over here. Darth Vader goes overand hits a soldier. So Darth Vader is also hittingone of the soldiers. Ask Kciko which one.

Child to Keiko: Which soldier is getting hit by DarthVader?

Competence and Performance in Child Language 133

Note that the child knows what the correct answeris to his question, and that he cannot expect toelicit this answer from his interlocutor (Keiko)unless he includes the by-phrase. In fact, exactly50% of responses were passives with full by-phrases). Of course, active constructions are alsofelicitous in this context (e.g., Which soldier isDarth Vader hitting?), even though the contextualcontrast with another agent (the Incredible Hulk)may tend to favor the passive stylistically. Andindeed 31% of responses were active questionswith object gaps. The other 19% of responsesincluded mostly sentences that were grammaticalbut not as specific as the context demanded (e.g.,passive lacking by-phrases).

Using this technique, we were able to elicit fullverbal passives from all but three of the thirty-twochildren tested so far, including ones as young as3;4. Some examples are shown in (41).

(41) She got knocked down by the Smurfie. (3;4)Which girl is pushing, getting pushed by acar? (3;8)He got picked up from her. (3;11)It's getting ate up from Luke Skywalker.(4;0)Which giraffe gets huggen by Grover? (4;9)

Note that these utterances contain a variety ofmorphological and other errors, but they allnevertheless exhibit the essential passivestructure (underlying subject in pre-verbalposition; agent in post-verbal prepositionalphrase). 9 It might be argued that the children'spassives elicited in this experiment do not involvetrue A-chains. However, since they are just likeadult passives (disregarding morphologicalerrors), the burden of proof falls on anyone whoholds that adult passives involve A-chains andchildren's passives do not. No criterion has beenproposed, as far as we know, which distinguishesadult's and children's passives in this respect. Forexample, it is true that the children almost alwaysuse a form of get in place of the passive auxiliarybe, but get is acceptable in adult passives also.Wet is more regular and phonologically moreprominent than forms of be, and this may be whyit is more salient for children.)

Children's considerable success in producingpassive sentences appropriate to thecircumstances (i.e., their correct pairing ofsentence forms and meanings) constitutescompelling evidence of their grammaticalcompetence with this construction. Comparison ofthese results with the results of testing the samechildren with two comprehension paradigms (act-

out and picture-verification) confirms that, likespontaneous production data, these measuresunderestimate children's linguistic knowledge.

The finding that young children evince masteryof the passive obviates the need to appeal tomaturation to account for its absence in earlychild language. Maturation cannot of course beabsolutely excluded; but a maturation account ismotivated only where a construction is acquiredsurprisingly late-where this means later thanwould be expected on the basis of processingcomplexity, pragmatic usefulness in children'sdiscourse, and so forth. (Also, as noted in section1, some important cross-language and cross­construction correlations need to be established toconfirm a maturational approach; see Borer &Wexler, 1987, on comparison of English passiveswith passive and causative constructions inHebrew.) The elicited production results suggestthat the age at which passive is acquired inEnglish falls well within a time span that iscompatible with these other factors, and somaturation does not need to be invoked.

3.3. Wanna contraction. Another phenomenonthat can be shown by elicitation to appear quiteearly in acquisition is wanna contraction inEnglish. The facts are shown in (42) and (43).

(42)a. Who do you want to help?

b. Who do you wanna help?

(43)a. Who do you want to help you?

b. Who do you wanna help you?

Every adult is (implicitly) aware thatcontraction is admissible in (42b) but not (43b).However, on the usual assumption that childrendo not have access to 'negative data' (Le., are notinformed of which sentences are ungrammatical)it is difficult to see how this knowledge about theungrammaticality of sentences like (43b) could beacquired from experience (at any age). So this isyet another candidate for innate linguisticknowledge. (What is known innately would bethat a trace between two words prevents themfrom contracting together. The relevant differencebetween (42b) and (43b) is that in (43b) the who isthe subject of the subordinate clause and has beenmoved from a position between the want and theto. The trace of this noun phrase that is leftbehind blocks the contraction. In (42b), bycontrast, the trace is in object position after help,and therefore is not in the way of the contraction.)

Crain and Thornton (in press) used the elicitedproduction technique to encourage children to askquestions that would reveal violations like (43b) if

134 Crain and Fodor

these were compatible with their grammars. Thetarget productions were evoked by having childrenpose questions to a rat who was too timid to talkto grown-ups. The details of the procedure areillustrated in the following scenarios:

Protocol for Object Extraction

Experimenter: The rat looks hungry. I bet he wants toeat something. Ask him what.

Child: What do you wanna eat?

Protocol for Subject Extraction

Experimenter: One of these guys gets to take a walk,one gets to take a nap, and one gets to eat acookie. So one gets to eat a cookie, right? AskRatty who he wants.

Child: Who do you want to eat the cookie?

Using this technique, questions involving bothsubject and object extraction were elicited from 21children, who ranged in age from 2;10 to 5;5, withan average age of 4;3. The preliminary findings ofthe experiment are clearly in accord with theexpectations of the innateness hypothesis,although we must verify our own subjectiveassessment of these data using a panel ofjudges. lO In producing object extraction questions(which permit contraction in the adult grammar),children gave contracted forms 59% of the timeand uncontracted forms 18% of the time. (Therewere 23% of other responses not of the targetform, such as What can you eat to see in the dark?)By contrast, children's production of subjectextraction questions (where contraction is illicit)contained contracted forms only 4 of the time anduncontracted forms 67% of the time (with 29% ofother responses).

The systematic control of this subtle contrastcould perhaps have been shown on the basis ofspontaneous production data, but the crucialsituations (particularly those that call for subjectextraction questions) probably occur quite rarelyin children's experience, just as they do in the caseof the full passive. So it is not easy to gather datain sufficient quantity for statistical analysis. Bycontrast, the elicitation technique is obviously anefficient way of generating data, and thusfacilitates testing for early acquisition of a varietyof constructions relevant to the innatenesshypothesis.

4. CONCLUSIONIn this paper we have reviewed a great many

empirical studies. The thread that ties themtogether is the idea that, when performance

problems are minimized in testing situations,children show early knowledge of a wide range ofbasic constructions. As early as 1965, Bellugisuggested that children's errors on Wh-questionswere due, not to a lack of knowledge of the tworelevant transformations (Wh-movement andSubjectJAuxiliary Inversion), but to a not yet fullydeveloped capacity to apply both rules in the samesentence derivation. Our work extends thisgeneral idea to a broader set of linguisticphenomena. Our particular emphasis has beenconstructions which linguistic theory predictsshould require little or no learning because theyinvolve principles which are universal and henceinnate. Our findings suggest that the innatenesshypothesis for language is still secure even in itssimplest form (in which different innate principlesare not timed to mature at differentdevelopmental stages). Maturation ofnonlinguistic abilities appears to be sufficient toaccount for the time course of linguisticdevelopment.

REFERENCESAmidon, A., & Carey, P. (1972). Why five-year-olds cannot

understand before and after. Journal of Verbal Learning and VerbalBehavior, 11,417-423.

I3ellugi, U. (1965). The development of interrogative structures inchildren's speech. In K. Riegel (Ed.), The development oflanguagefunctions. University of Michigan Language DevelopmentProgram, Ann Arbor, Report No.8.

Borer, H., & Wexler, K. (1987), The maturation of syntax. In T.Roeper & E. Williams (Eds.), Parameter setting. Dordrecht,Holland: D. Reidel Publishing Company.

Chomsky, N. (1971). Problems ofknowledge and freedom. New York:Pantheon Books.

Chomsky, N. (1986). Knowledge of language: Its nature, origin, anduse. New York: Praeger.

Oark, E. V. (1971). On the acquisition of the meaning of before andafter. Journal of Verbal Learning and Verbal Behavior, 10,266-275.

Oark, E. V., & Garnica, O. K. (1974). Is he corning or going? Onthe acquisition of deictic verbs. Journal of Verbal Learning andVerbal Behavior, 15, 559-572.

Crain, S. (1982). Temporal terms: Mastery by age five. Papers andReports on Child Language Development, 21, 33-38.

Crain, S., & Fodor, J. D. (1984). On the innateness of Subjacency.Proceedings of the Eastern States Conference on Linguistics (Vol. 1).Columbus: Ohio State University.

Crain, S., & McKee, C. (1985) Acquisition of structural restrictionson anaphora. Proceedings of the North Eastern Linguistic Society,16. Amherst: University of Massachusetts.

Crain, S., & Nakayama, M. (1987), Structure-dependence ingrammar formation. Language, 63,522-543.

Crain, S., Thornton, R., & Murasugi, K. (1987). Capturing theevasive passive. Paper presented at the 12th Annual BostonUniversity Conference on Language Development.

Crain, S., & Thornton, R. (in press). Recharting the course oflanguage acquisition: Studies in elicited production. In N.Krasnegor, D. Rumbaugh, R. Schiefelbusch, & M. Studdert­Kennedy (Eds.), Biobehavioral foundations of languagedevelopment. Hillsdale, NL: Lawrence Erlbaum Associates.

Competence and Performance in Child Language 135

Frazier, L., &; Fodor, J. D. (1978). The sausage machine: A newtwo-stage parsing model. Cognition, 6, 291-325.

Goodluck, H. (1986). Children's interpretation of pronouns andnull NPs: Structure and strategy. In P. Fletcher &; M. Garman(Eds.), umguage acquisition: Studies in first langUllge deoelopment(2nd ed.). Cambridge, MA: Cambridge University Press.

Goodluck, H., &; Tavakolian, S. (1982). Competence andprocessing in children's grammar of relative clauses. Cognition,8,389-416.

Gorrell, P., Crain, S., &; Fodor, J. D. (1989). Contextual informationand temporal terms. Journal of Child Language, 16,623-632.

Hamburger, H., &; Crain, S: (1982). Relative acqUisition. In S.Kuczaj (Ed.), Language deoelopment (Volume II, pp. 245-274).Hillsdale, NJ: Lawrence Erlbaum Associates.

Hamburger, H., &; Crain, S. (1984). Acquisition of cognitivecompiling. Cognition, 17, 85-136.

Hamburger, H., &; Crain, S. (1987). Plans and semantics in humanprocessing of language. Cognititle Science, 11, 101-136.

Hsu, J. R. (1981). The deoelopment of structural principles related tocomplement subject interpretation. Doctoral dissertation, The CityUniversity of New York.

Jakubowicz, C. (1984). On markedness and binding principles.Proceedings of the Northeastern Linguistic Society, Amherst,Massachusetts.

Johnson, H. (1975). The meaning of before and after for preschoolchildren. Journal of Experimental Child Psychology, 19.

Kimball, J. (1973). Seven principles of surface structure parsing innatural language. Cognition, 2, 15-47.

Lasnik, H., &; Crain, S. (1985). On the acquisition of pronominalreference. Lingua, 65,135-154.

Lust, B. (1981). Constraint on anaphora in child language: Aprediction for a universal. In S. Tavakolian (Ed.), Languageacquisition'and linguistic tht!Dry (pp. 74-96). Cambridge, MA: MITPress.

Matthei, E. M. (1981). Children's interpretations of sentencescontaining reciprocals. In S. Tavakolian (Ed.), Languageacquisition and linguistic theory (pp. 97-115). Cambridge, MA:MIT Press.

Matthei, E. M. (1982). The acquisition of prenominal modifiersequences. Cognition, 11, 301-332.

Nakayama, M. (1987). Performance factors in subject-auxinversion by children. Journal ofChild Language, 14, 113-125.

Otsu, Y. (1981). Uniuersal grammar and syntactic deuelopment inchildren: Toward a theory of syntactic deoelopment. Unpublisheddoctoral dissertation, M.l.T.

Phinney, M. (1981) The acquisition of embedded sentences andthe NIC. Proceedings of the North Eastern Linguistic Society, 11.Amherst: University of Massachusetts

Richards, M. M. (1976). Come and go reconsidered: Children's useof deictic verbs in contrived situations. Journal of VerbalLearning and Verbal Behauior, 15, 655-665.

Roeper, T. W. (1972). Approaches to a theory of language acquisitionwith examples from German children. Unpublished doctoraldissertation, Harvard University.

Roeper, T. W. (1986). How children acquire bound variables. In B.Lust (Ed.), Studies in the acquisition of anaphora, Volume I.Dordrecht, Holland: D. Reidel Publishing Company.

Solan, L. (1983). Pronominal reference: Child language and thetheory of grammar. Dordrecht, Holland: D. Reidel PublishingCompany.

Solan, L., &; Roeper, T. W. (1978). Children's use of syntacticstructure in interpreting relative clauses. In H. Goodluck &; L.Solan (Eds.), Papers in the Structure and Deuelopment of Child

Language. UMASS Occasional Papers in Linguistics (Vol. 4, pp.105-126).

Tavakolian, S. L. (1978). Children's comprehension of pronominalsubjects and missing subjects in complicated sentences. In H.Goodluck &; L. Solan (Eds.), Papers in the Structure andDeuelopment of Child Language. UMASS Occasional Papers inLinguistics (Vol. 4, pp. 145-152).

Tavakolian, S. L. (1981). The conjoined-clause analysis of relativeclauses. In S. Tavakolian (Ed.), Language acquisition and linguistictheory (pp. 167-187). Cambridge, MA: MIT Press,.

de Villiers, J. G., &; de Villiers, P. A. (1986). The acquisition ofEnglish. In D. Slobin (Ed.), The crosslinguistic study of languageacquisition Volume I: The dJlta. Hillsdale, NJ: Lawrence ErlbaumAssociates.

Wexler, K" &; Chien, Y. (1985) The development of lexicalanaphors and pronouns. Papers and Reports on Chl7d LanguageDeuelopment. Stanford, CA: Stanford University.

FOOTNOTES·Language and cognition: A deoelopmental perspectiue (in press).Norwood, NJ: Ablex.

tAlso University of Connecticut, Storrs.ttAlso Graduate Center, City University of New York.lSusan Carey has pointed out (Boston University Conference,1988) that the linguistic maturation hypothesis predicts thatknowledge of a linguistic principle should correlate withgestation age rather than with birth age in children bornprematurely. Unfortunately, variability is probably such thatno clear correlation could be expected to show itself by age 4 or5, when passiVes and other relevant syntactic constructions areclaimed to emerge.

2For full details of procedure and results of this experiment, andof all other studies reported in this paper, we refer readers tothe original publications.

3As far as is known at present, no natural language exhibits thisblanket prohibition against backward pronominalization; seediscussion in Lasnik and Crain (1985). This suggests that it isnot a possible constraint in a natural language grammar, inwhich case it should not be entertained by children at any ageor stage of acquisition (unless one assumes linguisticmaturation).

4The awkwardness of the prosodic contour for (18), with itsheavy juncture before the final word, may indicate that thiskind of construction is also an unnatural one for the sentenceproduction routines.

SIn reViewing the literature on relative clauses, de Villiers andde Villiers (1986) suggest that if earlier work had counted theassertion-only response as correct, children would have beenseen to perform better there too. This objection is unwarranted,for two reasons. First, responses of this type did not appear inother studies, presumably because these studies failed to meetthe presuppositions of the restrictive relative clause. Moreimportant, in the Hamburger and Crain study this responsewas not evinced by any of the 3-year-old children, andaccounted for only 13% of the responses of the 4-year-olds.Nevertheless, even the 3-year-olds acted out sentences withrelative clauses at a much higher rate (69%) of success than inearlier studies.

6These results should be interpreted with caution due to thesmall and unequal number of subjects in each SUbgroupleading to rather uneven data. For example, the older F groupperformed relatively poorly compared to the younger F group,

136 Crain and Fodor

though closer analysis reveals that this is due to the poorperformance of just one child (4;4) in the F group.

7There were no main-clause-only errors in the Fl grol;lP' For theF group, 12 of the 16 main-clause-only errors (out of 168responses) are due to one child

8An A-chain is the association of a trace with a moved nounphrase in an A-pOSition (= argument position such as Subject).For example, in The bagel was ellten by Bill there is an A-chainconsisting of the bagel and its associated trace after eaten.

9The proper reversal of underlying subject and object orderoccurred even when the task was complicated by animplausible scene to be described. For example, the sentenceOne dinoSllur's being eated from the ice cream cone was used todescribe a situation in which the dinosaur was indeed beingeaten by the ice cream, not vice versa.

lOIn preliminary evaluation of the audio tapes, we have found itunexpectedly easy to distinguish children's contracted andnon-contracted forms in most cases.