Brown and Frazer The Acquisition of Syntax

The Acquisition of Syntax Author(s): Roger Brown and Colin Fraser Source: Monographs of the Society for Research
in Child Development, Vol. 29, No. 1, The Acquisition of Language: Report of the Fourth Conference Sponsored by the Committee on Intellective Processes Research of the Social Science Research Council (1964), pp. 43-79 Published by: Blackwell Publishing on behalf of the Society for Research in Child Development Stable URL: http://www.jstor.org/stable/1165754 Accessed: 12/11/2009 05:41
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=black. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Blackwell Publishing and Society for Research in Child Development are collaborating with JSTOR to digitize, preserve and extend access to Monographs of the Society for Research in Child Development.
http://www.jstor.org
III THE ACQUISITION OF SYNTAX'

FRASER2 ROGER BROWN and COLIN Harvard University
What is done in a developmentalstudy of behaviordepends upon the conceptionof the terminalstate, the outcome of the developinvestigators' ment. Normal adults speakingtheir native languageseem to us to possessa set of rules of word constructionand sentenceconstructionwhich enables them to go beyondthe speechthey have actuallyheard and practicedto the creationof lawful novelties.If new monosyllables created,speakersof are will agree that stug is "betterEnglish"than ftug. Probablythis is English because they have a shared implicit knowledge of the initial consonant clustersthat are acceptable English. If this new word is to be pluralized, in they will agree that stug/-z/ is betterthan stug/-s/. Probablythis is because they have sharedknowledge of a rule of regular English inflection.If the new word is first heardin the sentence:"Here is some stug" they will agree that a secondsentence:"The stug is there"is more likely to be grammatical than a secondsentence:"A stug is there."Probably this is becausethey have sharedknowledgeof the syntacticrules for the employmentof mass nouns. The constructionrules of which speakershave implicit knowledge are, in their explicitform, the grammarof a language.As these rules have been writtendown in traditional grammars, they constitutea collectionof largely unrelatedstatementsabout such matters as the parts-of-speech, paradigms of conjugationand declension, markingof gender,and the agreementof the to systematizetraditionalgrammarinto a mechanismfor the generationof all the sentencesof a language that are grammaticaland none that are Grammarbecomes a theory for a range of phenomenaungrammatical. the sentencesof a language-and also a programfor generatingsentencesa programthat might be followed by an electronicdevice (Yngve, I96I).
1 The work describedin this paper was supported by the National Science Foundation by a grant administered through the Center for Communication Sciences, Massachusetts Instituteof Technology. From Verbalbehaviorand learning: problemsand processes,edited by N. Cofer and B. S. Musgrave. Copyright, I963. McGraw-Hill Book Company. Used by permission. 2 Now at the University of Exeter, Exeter, England.
adjectives and nouns. Chomsky (I957)
has shown that it may be possible
43
44
ACQUISITION OF LANGUAGE
It is the developmentin childrenof this kind of sentence-generating grammar that we are trying to study. The child growing up hears, from his family, his friends, and the television, a large sampleof the sentencesof a language,and we think that he induces from the regularitiesin this sample an implicit grammar.Firstlanguage learning, so conceived,reminds us of two other operationswith language: that of the linguist in the field and that of the adult learning a second language. The descriptivelinguist trying to work out the structure of an unfamiliartongue begins by collectinga large set of utterances-his "corpus."From regularitiesin the corpus and from inquiries of a native informant,he inducesrules that predictbeyond what he has observed.One check on the adequacyof his rules is their abilitycorrectlyto anticipatenew utterances. Anothercheck is the ability of the rules to duplicatethe distinctions made by a native informantwho has been asked to judge of each of a collectionof utteranceswhether it is or is not a well-formedsentence. It for may be that the linguisticprocedures discoveringsyntaxin distributional facts are a good model for the child's learning of his native languagewith the differencethat the linguist works deliberately and aims at explicit formulationwhereas the child works unwittingly and arrives at implicit formulations.The child's syntax is made explicit for him in "grammar" school, but we suggest that he operateswith syntax long before he is of schoolage. In learninga secondor foreign language,it does not seem to be possible to memorizea list of sentencesthat is long enough to providethe right one when you need it. Somehow the situationis never exactly right for any of the sayings one has rehearsed.To be effective in a second language, it is sentences,and there are two techniques necessaryto be able to "construct" for giving the student this ability. The traditionalmethod is explicit instructionin the rules of grammar.With these rules and a stock of words, one puts togetherthe sentenceto suit the occasion.A difficultyis that deliberate constructionis a slow business,and the boat will have sunk before you can properlycall for help. Some modern instructiontreats the secondlanguagelearnerlike a child and has him practiceagain and again the same set of sentences.The sentencesmay be deliveredto an entire group by film strip, or the studentmay pace himself with one of the Richardsand Gibson pocketbooks.Eventually the student finds himself the creator of a new sentence-one not practicedbut somehow implied by what has been practiced. Second-language learning by sentence rehearsalrelies on this step into automatic construction,though nothing much seems to be known about how to contrivesets of examplesthat will facilitateits occurrence. It has seemed to us, then, that first-language learning must have much in commonwith second-language and also with scientifictechniques learning for the discoveryof linguisticstructure. The sharedcharacteristic is the that ground of the analogy is the necessityin all three cases of inducing general
ROGER BROWN and COLIN FRASER
45
constructionrules from sets of sentences.Of course the analogy suppresses those featuresof first-language learningthat are not to be found in the other two processes, and it has taken considerable pressurefrom reality to bring them to our attention. This paper is divided into three sections.The first reviews some studies with inventedlinguisticmaterialswhich show that childrendo indeed have rules of word constructionand of sentence construction.The second discussestechniquesby which an investigator might induce a child'sgenerative grammarfrom a large collectionof the child's utterancesand the, possibly parallel,techniquesby which the child could have induced that grammar from a large set of parentalutterances. Most of the discussionin this section makes use of materialsfrom a recordof four hoursof speechfrom one child of 25/2 months. The third section discussessome substantiveresults from the recordsof I3 childrenbetween2 and 3 yearsof age; these are the results that forced us to recognize that there are differencesbetween childrenand either linguisticscientistsor adult studentsof a second language.
EVIDENCE THATCHILDREN HAVECONSTRUCTION RULES
In the natural situation of the child with his family the best evidence that he possessesconstructionrules is the occurrenceof systematicerrors. So long as a child speakscorrectly,it is possiblethat he says only what he has heard.In generalwe cannotknow what the total input has been and so cannot eliminatethe possibilityof an exact model for each sentencethat is put out. However, when a small boy says "I digged in the yard"or "I saw some sheeps"or "Johnnyhurt hisself," it is unlikely that he is imitating. his Furthermore, mistake is not a random one. We can see how he might have made it by overgeneralizing certain existent regularities.Many verbs ending in voiced consonantsform the simple past with /-d/ and many nouns ending in voicelessconsonantsform the plural with /-s/. The set of forms me, my, myself and you, your, yourself strongly suggests he, his, hisself. As it happens,actual English usage breakswith these simple regularitiesand prefersdug, sheep,and himself. By smoothingthe languageinto a simpler system than it is, the child reveals his tendencyto induce rules. Guillaumemade this point in 1927 and illustratedit with a rich collection of French children'ssystematicerrors. A Study of Morphological Rules Smith (1933) attempted to do a study of the development Although of the rules of inflectionby simply waiting for the emission of erroneous forms, it is more economicalto invent nonsense syllablesand try to elicit inflections.This is what Berko did for English in her doctoral research (1958).
46
OF ACQUISITION LANGUAGE
This is a wug.
Now there is another one. There are two of them. There are two
methodfor eliciting inflections. FIGURE 1-Illustration Berko's of A child is shown the small animal of Figure I and told: "This is a wug. ." The experimenter holds Now thereare two of them. There are two her voice up to signal the child that he is to completethe sentence;he will usuallysupply wug/-z/. For a differentanimal the word is bik and the correct plural bikls/. For a third animal it is niss and the plural niss/-az/. Printed English uses the letter "s" for all of these endings, but, as the phonemicnotation shows and as attentionto your own pronunciationwill reveal,the endings are distinct.The rule in English is: A word ending in a voicelessconsonant forms its plural with the voiceless sibilant /-s/ as in cats, cakes,and lips; a word ending in either a vowel or a voiced consonant forms its plural with the voiced sibilant /-z/ as in dogs, crows, and ribs; a word ending in the singular with either /s/ or /z/ forms its plural with neutralvowel as in classesand poses.We all follow /-z/ plus an interpolated these rules and know at once that a new word like bazooka will have, as its plural, bazooka/-z/, even though most speakersof English will never know the rule in explicit form. Berko invented a set of materialsthat providesa completeinventoryof the English inflectionalsystem:the pluraland possessiveendings on nouns;
ROGER BROWNand COLINFRASER
47
the simple past, the 3rd person present indicative,and the progressiveon verbs; the comparativeand superlativeon adjectives.She presentedthese materialsas a picture-bookgame to children of preschool,first-, second-, and third-gradeage levels and worked out the developmentof the rules with age. The productivityof the regular inflectionsfor children seems to be greaterthan it is for adults. Both kinds of subjectswere shown a picture of a man swinging somethingabout his head and told: "This is a man who knows how to gling. He glings every day. Today he glings. Yesterday ." Adults hang suspendedbetweengling, glang, glung, and even he but childrenpromptlysay glinged. Berkoalso testedto see whether glought children who generalize the regular inflection would correctly imitate irregularforms or would assimilatethem to the rules. She showed a picture and said, for instance,"Here is a goose and here are two geese. There are ." Most of her subjectssaid gooses and performedsimilarlywith two other irregularforms. These observationssuggest that rules of great generalitymay surviveand overridea numberof counterinstances. Knowledgeof the paradigmsof inflectioncould take a child beyond his of corpusto the correctconstruction new forms. It will not be necessaryto hear each new noun in its plural and possessiveforms; these can be anticipated from the regularparadigmand, except for an occasionalsheep and alumni, the anticipationwill be correct.The rules of inflectionare rules of morphology,i.e., rules of word constructionratherthan rules of syntax or sentenceconstruction.In English, though not in Russian and many other languages,inflectionis a rathertrivial grammaticalsystem and knowledge of inflectioncannot take a child very far beyondhis corpus.There is much greaterpower in syntax. A Study of SyntacticRules The fundamentalnotion in linguistic syntax is that the words of any naturallanguagecan be grouped into classeswhich are defined by the fact that the members of a class have similar "privileges of occurrence"in sentences.Certain very large and rough syntacticclasses are traditionally called the parts-of-speech. English, count nouns like house, barn, table, In and fence are wordsthat can be plugged into such sentencecontextsas: "See the is new"; "This is "; "I own a "; "The mine." If a child has learned to organize words into such classes,to enter them on mental lists of syntacticequivalents,he will have a very powerful means of getting beyondhis corpus. Hearing car as a new word in the sentence:"See the car" a child could use this context as a basis for listing car with count nouns and so be prepared to hear and say such additionalsentencesas: "I own a car"; "The car is new"; "This car is mine." And a multitudeof others. Of course the particularsentenceuttered on a given occasionwould depend on semantic
48
and motivationalfactors,but the populationof sentencesfrom which the particularcould be drawn would be establishedby the syntactic kinship linking car with house, barn,table,and fence. What evidencedo we have that a child acquires,with increasingexperience of his native language, implicit rules of syntax? Brown and Berko (I960) inventeda game for childrenthat utilizes nonsensesyllablesbut not, as in the inflectiongame, for the purposeof eliciting endings. For one problem the child was asked: "Do you know what a wug is?" He was then shown a pictureof a little girl and told: "This is a little girl thinking abouta wug. Can you make up what that might mean?"The picturewas included only to engage the child's attention-it did not portraythe referentof the new word but only someone thinking about it. The new word, wug, had been introducedin two sentences.In both cases it was precededby the indefinite article;it functionedonce as a noun complementand once as the The positionsof wug in these two contextsserve to objectof a preposition. it as a singular count noun. An adult speakerof English would identify have expected such additional sentencesas: "Wugs are good" and "That Brown and Berko were interestedin seeing wug is new" to be grammatical. whether young children would answer the question: "Can you make up what that might mean?"with a flow of sentencesemployingwug as a count noun. In the complete study 12 nonsense syllableswere used and they were placed in sentencesidentifying them as belonging to one of six parts-ofspeech.Where wug was to be identifiedas a transitiveverb, the investigator said: "This is a little girl who wants to wug something."As an intransitive verbthe same sentencewas used with the omissionof something.With wug as a mass noun the little girl would be "thinkingabout some wug." Wug becamean adjectiveby having the girl think of "somethingwuggy" and an adverbby having her think of "doingsomethingwuggily." Childrenin the first, second, and third grades all went on to make up sentences using their new words,but they did not alwaysuse them correctly. They did betteras they got olderand betterat all ages with the count noun, adjective,transitiveand intransitiveverbs than with the mass nouns and adverbs.For the purposesof the presentargument3the importantresult is that children showed an ability, increasingwith age, to constructgrammaticallycorrectsentencesusing new words.
3 The same children were given a word-association test employing familiar English words belonging to the same six parts-of-speechinvolved in the nonsense syllable task. The principal finding was that the frequency of paradigmatic word associations (i.e., a response word that belongs to the same part-of-speechas the stimulus word: house-barn; run-walk; milk-water) was related, across the various parts-of-speech and age groups, to the tendency to make correct syntactic use of the nonsense syllables. This result was taken to mean that paradigmaticword associationsreflect the developing organization of vocabularyinto syntactic classes.
49
The eliciting of speech with standardinvented materialsbrings some very desirablecontrolto the study of grammaracquisition.It left us, however,with the suspicionthat we had only been chippingat the problem.The things we thought of doing were largely suggested by fragmentaryfacts about adult grammarand guessesas to what childrenwould have to learn. If we were to collecta large numberof utterances from an individualchild, would it be possibleto subjectthis collectionto the kind of distributional analysis that a linguist applies to an unfamiliarlanguage and thereby to discoverthe child's total generativegrammar?Could one write grammar programsfor childrenat differentages and describelanguage development as a sequenceof progressively complicatingprograms?
INDUCINGA GRAMMARFROMA CORPUS
Before collecting any data, we studied English grammar in its more traditionalforms (Jespersen, I938; Smart, I957) and also in the recent works of Francis (I958), Hockett (1958), Chomsky (I957), and Lees (I960). The traditionalworks supply most of the substantiveknowledge about word classes-knowledge which has been reinterpretedand systematized in recent works. The substanceis often more effectivelytaught theorists by the earliergrammarssince this is knowledgethat contemporary often take for granted.The generativegrammarusing constituentanalysis and transformation rules has been worked out for a part of English, but thereis not yet a completegrammarof this kind for any language. Rules of grammarare culturalnorms;like other norms they are descriptive of certainregularitiesof behaviorwithin a community,and they also are prescriptivein recommendingthis behavior to new members of the community.In general,the studentof culturecan discovernorms either by observingbehaviorand inducing regularitiesor by asking participantsin the culture (informants) to tell him what kinds of behavior are "right" (proper, correct) and what kinds are "wrong" (improper,incorrect). For example, if we were interestedin the rules of etiquette of the American middle class we might ask informantswhat a seated gentlemanshould do when a lady entersa room and what a hatted gentlemanshould do when a lady entersan elevator.In addition,we might observethe behaviorof seated and hatted gentlemenin the two situations.When the cultural rules to be discovered linguistic,the possibleapproaches the same: directinquiry are are of informantsas to what it is proper to say and what it is not proper to say; direct study of what is, in fact, said. Studentsof grammarhave varied in the degreeto which they have reliedon one procedure ratherthan another. The partial generativegrammarsfor adult speech that have thus far been writtenhave been writtento meet the test of the grammarwriter'sown delicate sense of what is and what is not a well-formedsentence in his
50
nativelanguage.This is a specialcase of the techniqueof directinquiryof a native informantconcerningright and wrong behavior;in the presentcase the investigatoris his own informant.Of coursethe linguist working out a generativegrammarbelieveshis personaljudgmentsof the "grammaticality" of utterances a represent communityconsensus.The evidenceso far reported for judgmentsmade by informantswho are not linguists (Hill, 1961; Maclay and Sleator,I960) suggeststhat there is some consensuson grammaticality in English but that the consensusis not perfect. While we have adoptedthe generativemodel for grammar,we have not been able to use the method of the linguists who have written generative grammars. Clearlywe ought not to rely on our own senseof grammaticality in writing the grammarsof very young children.In addition,however,we have not found a way to make directinquiriesof native informantsbetween 24 and 36 months of age as to what they regard as well-formedsentences. For older children,judgmentsof grammaticality may possiblybe elicited in terms of what it is "right"and "wrong"to say. For the younger children, we have so far workedentirelyfrom obtainedbehavior:a sampleor corpus of what has actuallybeen said. It is by no means certainthat the direct study of obtainedspeech is an alternativeand equivalentapproachto the eliciting of judgmentsof grammaticality.Chomsky(1957) certainlydoes not suggestthat the notion of the "well-formed" "possible"sentence can be operationallytranslatedinto or the "obtained"sentence. Common observationshows that adult speakers of Englishoften produceverbalsequencesthatare not well-formed sentences. In the case of such anotherset of normsas the rules of etiquette,behavioral The practice might depart rather radicallyfrom ideal recommendations. truth is that the relationbetween grammaticalnorms and verbal behavior is quite unexploredboth theoretically and empirically.Without waiting for clarification this relationfor adult speakers, have proceededto explore of we the possibilityof writing rules for the actual verbal behaviorof children. Chomsky (i957) has argued that there are no really adequatemechanical for procedures discoveringan adequategrammar,and our experiencecauses us to agree with him. Still there are some helpful tips, and Harris (I95I) is the best sourceof these. In addition to reading about grammars,we practiced working them out from a speechcorpuslong beforewe collecteddata of our own. As practicematerialswe used the recordscollectedby Barker and Wright (I954) for their Midwest studies. If anyone has a taste for word games, the grammardiscoverygame is an engrossingone, and, not surprisingly,it yields more understandingthan can be obtained from the works. readingof theoretical We decidedto begin work in the age range from 24 to 36 months.The time at which most childrenbegin producing youngerage is the approximate two-wordutterances(McCarthy,1954). Some preliminary work showed us
51
that by 3 yearsmany childrenhad aboutas complexa grammaras we were able to describe.We located 13 children in this age range whose parents were willing to have us spenda large part of one day at their homes recordvaried from ing the child's speech.For the first seven cases our procedures one to another,but for the last six they have been reasonably uniform. The families were of the professional, class, and it is college-educated in likely that the linguisticbehaviorof these childrenwas "advanced" terms of age norms for the American population.Only two of the 13 children were acquaintedwith one anotherand so, with the exceptionof these two, the speechof any one child in the samplecould not have been directlyaffecting the speechof any other.Many of the familieshad first arrivedin Boston just seven months earlier,and the speechof these parentshad been learned in severaldifferentpartsof the country. We hoped to get as much speechas possiblein as little time as possible and to have examplesof the full varietyof sentencetypes the child could produce.There were those who warned that the child would be shy and in this speechless our presence; was not the case. Motherstold their children that visitors were coming and, in general, we were eagerly welcomed, shown a paradeof toys and games, and talked to rathersteadily.It became clearthat the child expecteda guest to put in some time as a playmate,and so the recordingwas a two-man job with one of us taking data and the other prepared play cowboy,horsie,blocks,coloring,trains,and the mule to in "Kickthe mule."Severalof the early recordswere made by Fraseralone, but for the last six cases both of us were always there. We found that by about noon we needed a rest and so went away for lunch, returningabout two; the child took his nap in the interval.About half of each recordwas made beforelunch and abouthalf after lunch. Much of the time the child was occupiedwith his normalroutineof play, talking with his mother, washing, and eating. So long as these activities involved a reasonableamount of speech, we took no active part beyond When the operantlevel was very deliveringsignalsof attentionand approval. low, we sometimestried to raiseit. In the firstdays we did the sortof verbal promptingthat is anyone'sfirst notion of a technique for eliciting speech from children. You ask what something is called and this brings out vocabularyitems-which are not useful for a study of grammar.Or you ask a "yes-no" questionsuch as: "Is that your horsie?"to which the answer is either "yes"or "no."We eventuallylearnedthat it is easier to "inspire" speech by doing something interestingthan to elicit it with questions. If the adult "playmate" startsa game that is simple, repetitious,and destructive, the child will usually join him and start talking. A universalfavorite is to build (painstakingly)an unsteadytower of blocksand registerchagrin when the child sends it crashingdown. A simple game involving implicit rules-such as the green blocks belong to me and the red ones to you-
52
createsa situationin which negative sentencescan be elicited. If the adult rule and moves one of the child's blocks,he playmatebreaksthe established is usuallytold that he is not to do that. recorded?After trying severaldifferent How were the child's utterances to we found that it was not restrictive ask the motherto limit activithings, ties to one floorof the house and a small numberof adjoiningrooms.Then and long extensioncords we were able, with two Wollensaktape recorders in the last six cases, to get almost everythingon tape. The machineswere and handledby the "playmate," the other memberof the team made a continuous transcriptof the child's utterances.This on-the-spottranscriptwas later checkedagainstthe tape and correctedinto a final best version.Going over the tape takes about four hours to each hour of recordingtime. What is the level of detail in the transcription?It is neither phonetic nor phonemicbut only morphemic.It is, in short, as if we were to write down in conventionalEnglish spelling what an adult seemed to be saying in an interview.Of coursethe intelligibilityof speechin the youngest children was not very good. We found it helpful to do no writing for the first half hour and to have the motherinterpretfor us everythingher child said in that period. In this time we learned to allow for the child's phonetic peculiaritiesand sometimesfound an initial near-complete unintelligibility crucialpoints giving way to about75 percent intelligibilty.At grammatically our generalrule was to creditthe child with the regularadult contrastif he made any sort of appropriate phonetic distinction.For instance,the emergence of the modal auxiliarywill in a sentencelike I will get my book is not at first markedwith a well articulated/wil/ but probablyonly with a shift of the vowel formantsin the I toward a back vowel like /u/. If we could hear a differencebetween this I and the way I sounded in I got my book, the child was creditedwith will. For the last six cases we were ultimately able to transcribe fully an averageof 78 per cent of the total utterances on the tapes; this is a degree of successquite similarto that reported in previousstudies (McCarthy,I954). Where we were uncertainabout the the accuracyof transcription, material was placed in brackets.Utterances were also markedwith the following symbols:I for a functionalimperative; ? for an interrogative; for an utterancethat mimics an immediatelypreM ceding utterancefrom anotherperson;R for an utterancethat is a response to a question. One difficultywe had anticipateddid not materialize-only one. We had thought that division of the flow of speechinto utterancesmight be an uncertainbusiness.In fact, however,the usual criteriaof either a prolonged pause or a shift of speakersworked very well. There were few instancesof uncertainutterancedivision. For the last six caseswe aimed at and obtaineda minimumof 500 different utterancesfrom each child. Since the utterance rate of the younger
53
children is lower than that of the older children, we spent more time with the younger ones.
Methodsfor Discoveringa ProvisionalGrammar

The process of grammar discovery has two facets. It is, most immediately, the technique of the investigator who is trying to describe the grammatical apparatus of a particular child. The investigator induces from obtained utterances a probable generative mechanism. Since, however, the child is also presumed to have built this internal mechanism by processing obtained utterances, it follows that the investigator's procedure may be a good model of the child's learning. Of course the child has not induced his grammar from his own sentences but rather from the somewhat more varied and complex sentences heard from adult speakers. A comparison of the recorded speech of mother-to-child with the speech of child-to-mother shows that the grammars induced by children from adult speech are not identical with the adult grammars that produced the sentences. The investigator of a child's speech, on the other hand, hopes to find the very grammar that produced the original utterances. Even so the similarity in the tasks of investigator and child is very great-to get from sentences to a grammar-and so while acting as investigators we shall want to consider whether the child may have carried out operations similar to our own. For the kind of grammar we are trying to write the fundamental problem is to discover the syntactic classes.4 Members of a common class are supposed to have similar privileges of occurrence, and these privileges are supposed to be different from one class to another. In addition to syntactic classes our grammars will involve rules of combination describing the ways in which members of the various classes may be put in sequence. Generally the rules of combination will get simpler as the syntactic classes get larger and fewer. If a and an in English were words having identical privileges of occurrence, our grammar would be simpler than it is. As it stands, however, the two forms must be separated, and a rule of combination written that requires an before count nouns with an initial vowel and a before count nouns with an initial consonant. As a basic technique for the discovery of syntactic classes we might undertake to record for each different word in the corpus all of the utterance contexts in which that word occurs; in short we might make a concordance. The contexts of each word could then be compared with the contexts of each
4 The generative grammar, as Chomsky (1957) has described it, is more than a set of sequence rules for syntactic classes; it also provides several levels of appropriateconstituent analysis. We have so far not accomplished this result for the speech of children because we have not been able to invent appropriatebehavioral tests of the child's sense of constituent structure. The requirement that a grammar make appropriate structural analyses does of course help greatly in the evaluation of adult grammars, and it is desirableto meet this requirement wherever it is possible to do so.
TABLE I
Total Contexts of Four Words in the Record of Adam

Total Contexts of "Mum" Here it is, Mum. Here, Mum. Here (the) coffee pot broken, Mum. More sugar, Mum. There it is, Mum. What's that, Mum. Mum, (where is the cards)? Mum, (where's the rags) ? Want coffee, Mum.* Want apple, Mum. Want blanket, Mum. Want more juice, Mum. I want blanket, Mum. I want (it), Mum. I want paper away, Mum. (The) pan, Mum. I want apple, Mum. I want blanket, Mum. I want blanket now, Mum. I want juice, Mum. Mum, I want some, Mum. Popeye, Mum? I wanta do, Mum. I wanta help, Mum. I found, Mum. I do, Mum. I don't, Mum. I get it, Mum. (Gonna) dump, Mum. Fall down, Mum. Fall, Mum. An apple, Mum. Total Contextsof "Dad" See paper, Dad. Want coffee, Dad.* I want cream, Dad. See, Dad?* Dad, want coffee? Some more, Dad? Total Contexts of "Here" Here (a car). Here all gone. Here (block). Here brick. Here chairs. Here coffee is. Here comes Daddy. Here flowers. Here goes.t Here is.t Here it goes.t Here it is.t Here it is, Mum.t Here's it here. Here light. Here (mail) more paper. Here more. Here more bricks. Here more blocks. Here more firetruck. Here more toys. Here more truck.t Here Mum.t Here Mummy. Here my bricks. Here not a house. Here stars. Here (the) coffee pot broken, Mum. Here the card. Here the cards. Here the cheese. Here (the) flowers. Here the paper. Total Contexts of "There" There goes.t There more block. There more truck. There (he) goes. There more nails. There is.t There Mum.t There it goes.t There my house. There it is.t There it is, Mum.t There my nails. There Noah. There kitty. * Identifies contexts common to "Mum"and "Dad." t Identifiescontexts common to "Here" and "There." I wanta put (it) right there . . . (under) the couch. Me see (in there). Blanket in there. In there. Right there.t Here (we go). See the bolt here, see? That block here. That one here. That one right here. I put bucket here. Come here. Do here. Leave that block here. Put it here. Here not a house. Right here.t Over here. Over here, Mum. Now here. Work, Dad? Hi, Dad.* Apple, Mum. Again, Mum? Out, Mum? Salad, Mum? See, Mum?* Coffee, Mum? Turn, Mum? No, you see, Mum? No help, Mum. Won't help, Mum. Coffee, Mum. Hi, Mum.* O.K., Mum. Here, Mum. Over here, Mum. Enough, Mum. Silver spoons, Mum.
55
other word and tentativesyntacticclassesset up so as to put togetherwords words having few or having many contextsin commonand so as to separate no contextsin common. To illustratethe method of shared contexts, we have taken from the recordof Adam5 (2812 months) the word: here, there Mum, and Dad. In adult English here and there are locative adverbialswhile Mum and Dad are animatenouns. Will the patternof sharedcontextstaken from Adam's recordsuggestthe assignmentof here and there to one class and Mum and Dad to a differentclass? In Table i all of the utterancescontainingthese four words are listed and the sharedcontextsindicated.The upper limit to the numberof contextsthat two words can sharein a given recordis set by the number of contexts obtained for the less frequent of the two. Here occursin 48 differentcontextsand there in 19. It would be possiblefor the two words to share 19 differentcontexts;in fact they shareeight or 42 per cent of the possiblecontexts.Mum occursin 49 differentcontextsand Dad in eight; they share 38 per cent of the possible contexts. Here shares no contextswith Mum and none with Dad, and the same is true for there.The pattern of shared contexts suggests the class break that operatesin adult English. If one has read about the notion of syntactic equivalence but never actuallylined up the contextsfor sets of words, it is startlingto find such small numbersof identicalcontexts;especiallystartlingsince here-there and Mum-Dad must be as near to perfect syntacticequivalenceas any pairs of words in the language and, in addition, the short sentences and small of of vocabulary a child maximizethe probability repetition.Is a 38 per cent for us to assumethat the membersof a pair are interchangeoverlapenough able? It may be. We have taken the word here.alone and set down all its differentcontexts in the first half of Adam's recordand also all of its differentcontexts in the second half of the record (Table 2). A context that has alreadyapin pearedin the first half is listed again on its first appearance the second half. Here in the first half is the same word as here in the secondhalf, and so we know that these two heres are syntacticequivalents.In the first half the word occursin 33 differentcontexts;in the secondhalf in 19. There are four sharedcontextsor 2I per cent of the possible-a lower value than the value obtainedfor here-there and Mum-Dad. Perhapsit would be possibleto set an exact percentage-of-shared-contexts criterionfor the assignmentof two words to the same class, the criterionto be empiricallydeterminedfrom percentages sharedcontextsof identical of words at various levels of absolute and relative frequency in two time periods. Clearly,the obtained percentageswill be very unstable for small numbersof occurrences. less frequent words only a mammoth speech For samplewould serveand the whole thing becomesa job for a machine.The
5 The names used in this report for identifying child subjects are not the actual names of the children who were studied.
56
ACQUISITION
O1 LANGUAGE
TABLE 2
Contextsof "Here"in the First and Second Halves of the Record of Adam
First Half Here all gone. Here (block). Here brick. Here chairs. Here coffee is. Here comes Daddy. Here is.* Here it goes.* Here it is, Mum. Here light. Here (mail) more paper. Right here. Here more.* Here more bricks. Here more firetruck. Here more toys. Here more truck. Here Mum. Here my bricks. Here not a house. Here (the) coffee pot broken, Mum. Here the card. Here the cards. Second Half Here (a car) Here flowers. Here goes. Here is.* Here it goes.* Here it is. Here (we go). Here's it here. Here more.* Here more blocks. Here Mummy. Here stars. Here (the) flowers. See the bolt here, see. In here? (Over here)? Over here, Mum. Over here.* Now here. Here the cheese. Here the paper. That block here. That one here. That one right here. Come here. Leave that block here. (Put it) here. Here not a house. Over here.*
* Identifies contexts common to first and second halves of record.
basic problem of setting a criterion for that machine to use or of writing the program for working out a criterion may also be a problem in the child's learning of syntax. Once the classes have been established by some shared-contexts criterion, some contexts will turn out to be more "criterial" or distinctive for a given class than will others. After the in English one can have count nouns in the singular, count nouns in the plural, and mass nouns (which are used in the singular only). After a one can have only count nouns in the singular, and so this context is, for singular count nouns, the more criterial of the two. The introduction of new words to a child in highly criterial "tracer" contexts " for personal names) should be the best guarantee of sub(like "Hi sequent correct usage. A corpus of 500 utterances is not large enough to take us very far with a mechanical shared-contexts procedure. However, a generally adequate mechanical procedure for the discovery of grammars has not been worked out in any case. Grammar writing is for the present like theory writing in science, an undertaking for which there are some guides, clues, and models but not a set of guaranteed procedures. Let us therefore take one record, the utterances of Eve (25/2 months) and, allowing ourselves a very free use of inductive reasoning, see what we can make of it.
57
It is a useful first step to restrictourselvesto the simplestutterancesin the record,those of just two words, and, among these, the utterancesin which the initial word occursat least twice. In Table 3 the recurrentinitial
TABLE 3
Two-Word Utteranceswith RecurrentInitial Words from Eve Second

Word bear ...... bird ...... block ..... boat ...... Bobby .... book ..... bowl ..... boy ...... broken .... candle .... car ....... carriage ... chair ..... cricket ... cookie .... cow ...... Daddy .... dimple .... dirty ..... do ....... dog ...... doggie .... doll ...... dollie ..... Dru ...... eye ....... fall ....... fuzzy ..... Gale ...... girl ....... go ....... goes ...... going ..... honey ..... horsie ..... is ........ kitty ..... man ...... meatball men ...... mike ..... Mummy nurse ..... pea ....... A
Daddy Mummy
's
See
That
The
There
Two
+ +
+ +
+
+ +
+ + +t +
+ +
+ q-
4-
+ + + + +t
+
+++ + + 4-
++
+
+++++4
+
+4-
+ + +
+-
4-3
( Table continued on next page )
58
TABLE
3 (continued)
Two-Word Utteranceswith RecurrentInitial Words from Eve

SeconI Second A Word peas ...... Peter ..... picture .... pillow .... potty ... pretty..... puff ...... puppy .... radio ..... Rayma.... reel ...... rocker .... rug ....... sun ....... that ..... 'tis ....... wire ......
N
Daddy Mummy
's
See
That +
The +
There
Two
+ + + ++ + + + + + + + + + + + +
whistle ....
+ indicates that utterancesof this type occurred.
NOTE.-"-+"
words are listed at the heads of columns, and each row represents a second word that occurs in at least one utterance after one of the recurrent first words. The second word is given a + in a column if it occurs after the word that heads the column, and so a filled square represents an obtained twoword utterance defined by the column and row headings. Table 3 is simply a technique for making a first inquiry into shared contexts, an inquiry limited to the most frequently repeated and so most informative contexts. By comparing the filled slots for any two rows, one can determine the number of shared contexts for two second words. By comparing the filled slots for any two columns one can determine the number of shared contexts for two initial words. Table 3 is itself a descriptive grammar for an obtained set of two-word sentences. The table is a description that is only very slightly more economical than actual listing of the sentences; the single slight economy is accomplished by writing the recurrent first words just once instead of repeating them on each appearance. The descriptive grammar of Table 3 can easily be turned into a generative grammar. Table 4 presents the generative version. It reads: "In order to form an utterance: select first one of the initial words; select, secondly, one from the class of words that are permitted to follow the initial you have chosen." A machine that can go through this program would produce all-and-only the obtained set of utterances.
BROWNand COLINFRASER ROGER TABLE 4 Obtained All A Grammar Describing andOnlythe 89 Utterances
A Daddy Mummy 's See That The There + Ci + C2 + C3 + C4 + C5, a + C6 + C7 + C8
59
Utterance ->
Two
C1 -> C2 bCa -
C9
block, book, candle, cricket, dog, fall, kitty, meatball, nurse, pillow, reel bear, book, honey bear, dimple, do, go, puff C4 -> bird, Daddy, picture C5 -> boy, eye, Mummy, radio, rocker, that bird, boat, book, bowl, boy, broken, car, cookie, cow, Daddy, dirty, doggie, fuzzy, C6 going, horsie, kitty, Mummy, Peter, pretty, puff, Rayma, rocker, sun, wire C7 --> bird, book, girl, horsie, kitty, mike, peas, puppy, reel, rug, whistle C8 -- bird, boat, book, boy, carriage,chair, Daddy, doll, dollie, Dru, go, goes, is, kitty, man, Mummy, pea, potty, radio, reel, 'tis C9 -- Bobby, chair, Gale, men, reel NorE.-{ } means a choice of one of the contained sequences.
Table 3 constitutesa list of differentutterances:each is listed just once. varied in their frequencyof repetition. In the originalrecordthe utterances One could write a generativegrammarthat would do a betterjob of mimicking the original record by setting probabilitiesat the various choicefor points: the route"That-broken," instance,ought to be taken more often A than the route "That-Rayma." printedrecordturnedout by this machine from actual obtained records,indistinguishable would be indistinguishable in as a list of sentencesbut perfectlydistinguishable the life situationfor the reasonthat the machinedescribedtakes no accountof semantics.It is quite of capableof greetingthe appearance Daddy with "That doggie" or "That cookie." We have a grammarmachine, not a machine that is a complete model for human sentenceproduction. Now we want to go beyondthe obtainedsentencesto the syntacticclasses they suggest. Is there any ground for consideringall of the initial words to be membersof a single class? Considerfirst the possibilityof a sharedcontexts criterion,of the type discussed,with referenceto here-thereand Mum-Dad.
60
The previousdiscussionwas greatlysimplifiedby restrictingthe problem to single pairs of words. However, the number of candidatesfor membership in a common syntacticclass is generally greater than two, and that is the case with the initial words of Table 3. Each single word that is a with each other candidatefor inclusioncan be comparedfor shared-contexts word in the presumptiveclass. The criteria for class identificationmust then prescribeminimal levels of overlapfor the full set. This is a complicated problemfor which many differentsolutionscan be imagined. In the presentcase the words that and there are the most frequent in the set. They share seven secondword (bird, boat, book, Daddy, kitty, and Mummy) which is 33 per cent of the number of contexts that could be shared.Of the other words in the set 's, see, the, a, Mummy, and Daddy all share at least one context with both that and there, while two shareswith close pair that and there there but not that. One might use the syntactically as touchstonesand count as members any words that share at least one context with either critical word. That rule would put all of the initial words into the same syntacticclass (class I). What evidenceis there that the entries in the rows of Table 3 constitute a second syntacticclass? The words bird, book, kitty, and reel are most frequent.Since the contextsare the words of class i, and so are few, these frequent words have a high degree of mutual overlap.Using the four as most of the words enteredin rows could be entered in class 2 touchstones, on the criterionof having at least one context in common with one of the A touchstones. few, such as dimple and do, would be left out. Most of the words in class 2 are count nouns, and if we had a largercorpus,we should probablyfind that dimple belongs here for Eve while do does not. Our criterionis probablynot good enough to separateout just the few words that do not fit. The membersof a syntacticclass are creditedwith identical privileges of occurrence. Table 3, the initial words are far from having identical In of actual occurrence, and the second words are also far from this patterns pattern. If the actual occurrenceswere identical in either set, then the assignmentto a common class might be confidentlymade, but the assignment would not take us beyondwhat we have obtained.If we do go beyond what has been obtainedin our descriptionof privilegesof occurrence, then, of course,we will not have "proved"our description.This is the familiar problemof the impossibilityof proving a generalizationthat goes beyond what has been examined-the very sort of generalizationthat is most valuable and which the human mind is most bent on making. We are, in the setting up of syntacticclasses,trying to move from a partial similarityof actual occurrence an identity of potentialoccurrence. to This is the process of induction-uncertain but powerful. To hold that the membersof each of our presentclasseshave identical privilegesof occurrenceis to hold that all the empty positionsin Table 3
ROGER BROWNand COLINFRASER
61
may properlybe filled-they are utteranceswe should expect to hear from Eve. The generativegrammarchanges to the form of Table 5. This grammar will turn out more sentencesthan Table 4, and it ought to be very much easier to remember since the sequential contingenciesare greatly simplified.
TABLE 5
A Grammarthat Results from Filling In the Blanks in Table 3

Utterance -> C1 -C2 -) C1 + C2
A, Daddy, Mummy, 's, See, That, The, There, Two bear, bird, block, boat, Bobby, book, bowl, boy, broken, candle, car, carriage, chair, cricket, cookie, cow, Daddy, dimple, dirty, do, dog, doggie, doll, dollie, Dru, eye, fall, fuzzy, Gale, girl, go, goes, going, honey, horsie, is, kitty, man, meatball, men, mike, Mummy, nurse, pea, peas, Peter, picture, pillow, potty, pretty, puff, puppy, radio, Rayma, reel, rocker, rug, sun, that, 'tis, whistle, wire
NOTE.-This grammarpredicts the 89 utterancesobtained plus 469 others.
the model of Table 4 with only the obtained secondaries occurring and one on the model of Table 5 with any member of class 2 likely to occur after any initial word. I think we can foresee that for the first set of materials it would take a long time to learn the exact set of words that can go after each initial word. With the second set it should not take very long to learn the principle that any second word can follow any initial word, but it would take some time to memorize all of the second words. But after how many pairs, of what kinds, will subjects move to the generalization of Table 5? We think it would be interesting to try paired-associate learning of this kind. Refinements on the provisional grammar. Turning to utterances of three or more words, we find evidence to indicate that not all of the members of class i have identical privileges. Consider first the words the and a. Table 7a presents a set of utterances in which the or a occurs after members of class I and before members of class 2. However, the and a do not occur
to do what the grammardoes-respond to each initial word with all acceptable second words. There could be two sets of materials: one made up on
The utterances of Table 4 and Table 5, since they are two-word utterances, can very well be thought of as a set of paired associates. It is usually the case in paired-associate learning that initial words and second words occur always and only in their pair relation. In the present case some initial words and some second words occur in a variety of combinations. If these utterances were to be made up as experimental materials, what task could we set the subject? If he were simply asked to respond to each initial word with some one of the acceptable second words, then he could meet the requirements by learning single associates. Suppose, however, he is required
62
after all members of class i; they do not occur after the, a, Mummy, or Daddy. There are no utterances like "The the doggie" or "The a horsie." These facts suggest that the and a should be withdrawn from class I and set up as a new class 3. The grammar would then be rewritten as in Table 6a.
TABLE 6
GrammarsSuggested by Three-WordUtterances
a. Eve's Grammarwith the Articles SeparatedOut Utterance --> C1 -) C2 C3 -) (C) +- (C8) + C2
Daddy, Mummy, 's, See, That, There, Two bear, bird, block, boat, etc. a, the
b. Eve's GrammarAllowing for Possessives Utterance -'s, See, That, There, Two C1-bear, bird, block, boat, etc. C2 C3 -a> , the, plus human terms
NOTE.-( ) means that selection of the enclosed is optional.
(C1) + (C3) + C2
The optional markings in Table 6a indicate that a class can be completely bypassed. It is necessary to make class 3 optional so as to get sentences like "That bird" and "There boat." Making class I optional makes it possible to obtain "The boat," "A book," and the like. If neither optional class is bypassed we get such utterances as "That the cup" and "See the reel." If both optional classes are bypassed, we get such one-word utterances as "Dolly," "Book," "Reel," and utterances of this kind were very numerous in Eve's record. We cannot get from this grammar such utterances as "The the horsie," "The that bird," "A see that" and other radically un-English sequences. The grammar of Table 6a does turn out some kinds of utterance for which Eve's record provides no models. It is possible, for instance, to obtain "Two the bear" or "Two a bird." Eve produced nothing like these, and for adults they are definitely ungrammatical. It is quite likely that two should go with the and a in class 3 so as to yield: "That two bear" and "See the bird." However, Eve's record gives us no utterances like these, and indeed no three-word utterances including two, and so we may as well leave two in class i. We have identified a point at which more evidence is needed in order to choose between formulations.
63
There is another sort of utterancethat the grammarof Table 6a will produce for which Eve's record provides no model: "Mummy the bird" or "Daddya book."These sentencesdo not use Mummy and Daddy in the vocative; the vocative would involve a distinctive juncture (pause) and intonation which is suggested in print by a comma (e.g., "Mummy, the
TABLE 7
Three-Word Utterances that Refine the Original Grammar

a. Utteranceswith the or a in Middle Position 's a man 's a house 's a Daddy There the kitty See the horsie See the radio See the reel See a boat That the bowl That the cup That a horsie
b. Utteranceswith Human Terms in Middle Position See Evie car See babyeyebrow That Mummy book That Mummy paper That Mummy hair That Mummy spoon That Daddy car That Daddy honey That man car That baby bed That Evie book That Evie dish That Evie pillow That Evie spoon There Evie car There man coat
bird'). An adult might form the vocative version as a kind of ellipsis of "Mummy,see the bird,"but he would not form the nonvocativeutterance that the grammarof Table 6a produces. Thus far, Mummy and Daddy are in the same limbo as two; there is reasonfor withdrawingthem from class I, but we do not yet know where to put them. Eve's recordprovidesno three-wordutteranceswith two, but it does provide such utteranceswith Mummy and Daddy and these are listed as Table 7b. We find Mummyand Daddy in middle positionbetween that (a class I word) and the class 2 words: book, hair, paper,spoon, car, and honey. Furthermore,we find in this same position the words: Evie, man, and baby. All of these middle-positionwords are names of human beings, and so we may hypothesizethat such terms constitutea syntactic class. The difficultynow is that the middle position between class i and class 2 alreadybelongs to class 3 which is composedof the and a. This means that there are three clear options: (a) to include the human terms with the and a in class 3; (b) to set up the human terms as a separate class 4 and put this class in the selectionsequenceahead of class 3; (c) to class4 in sequenceafterclass3. Here now are examples put the independent of utterances that are predictedby the three versionsof the grammar: (a) "That Mummy book"; (b) "That Mummy the book"; (c) "That the
64
Mummy book." Our interim decision is fairly clear. Since the record provides instances of (a) but not of either (b) or (c), we will put the human terms in class 3 with the and a as indicated in Table 6b. However, if a larger corpus were drawn, we would have to be on the alert to correct this decision. It may be worthwhile to consider the very complicated rules required for adult English at an analogous grammatical point. Utterances like "Mummy book" and "Man car" and "Evie dish" are, we know from the situational contexts, Eve's version of the possessive; she omits the inflection. In forming possessives with human terms, the adult speaker will use an article before such generic human terms as man ("the man's car"), and so these terms cannot be classified as alternatives to the and a. With personal names like Evie the adult will form possessives but, with personal names, articles are never used, and so we will say "Evie's dish" but neither "The Evie's dish" nor "Evie's the dish." This means that personal names cannot be classified as alternatives to either the articles or the generic human terms. A final exasperating fillip is provided by the fact that some morphemes, like Mummy and Daddy, can serve either as personal names ("Mother's purse") or as generic terms ("the mother's role") and so would have to be listed in two syntactic classes. Even the simple grammar we have now produced for Eve involves double syntactic listing for some forms-the human terms. They must be in class 3 because of such utterances as "That Mummy book," but they must also remain in class 2 because of such utterances as "That Mummy." The combinational formula predicts "That Mummy Mummy" (Grandmummy?) which is reasonably probable but also "That Evie Evie" which would not
TABLE 8
Utteranceswith RecurrentSecond Members

Initial Word S
E C
O N
O R
Baby ................... Bird* .................. Carriage* ............... Chair* ................. Doggy* ................. Dollie* ................. Eyebrow ................ Kitty* .................. Microphone ............. Mummy* ............... Reel* ................... Rocker* ................. Something ..............
tired all gone broken broken broken all gone all gone all gone all gone broken broken fall down tired fall down fall down tired
* A word already identified as a member of C2.
65
be producedby an adult. As to whether Eve would say it, we cannot tell from the presentmaterials. Expansionof Eve's grammar. Recurrenceof a context is always what is needed for the identificationof a syntacticclass. We startedwith recurrence of initial members,but recurrenceat the end of an utterancecan be equally useful. In Eve's record there is a set of terminal recurrences(all gone, broken,fall down, tired) which appearsin Table 8. The first thing to note is that most of the initial membersof these utteranceshave already been identifiedas membersof class 2. Their appearance here with overlapof occurrencebefore all gone, broken,fall down, and tired ping privileges is a valuableconfirmationof the preliminaryanalysis. Since the initial terms are alreadyregardedas one class, we are inclined to assume that they have identical privileges of occurrencewhich means that emptycells could be filled. This makesa new class4 of all gone, broken, fall down, and tired, and we now have a second type of sentence.The full to grammar (Table 9a) involves two major routes corresponding the two kinds of sentences. There is a slight addition to be made to the grammarof Table ga. The addition is suggestedby three sentences:"The chair broken"; "The baby tired"; and "The book fall down." The initial word the has alreadybeen placedin class 3, and so it is a good guess that any word from class 3 can appearin first position in this new sentence type, and Table 9b presents this version.Class3 must be enteredas optionalin the grammarsince most of the obtained sentencesomit it. It is not certain that our analysisof the limited materialsin Table 8 is correct.The word tired, for example, does not actually overlap in the obtained distributionwith any of the other second members. Our analysis, relying on the prior identificationof the first members as class 2, assumes that the overlap would occur if more sentenceswere gathered.However, it should be noted that tired occursonly after names of animate beings-baby, doggy, Mummy-whereas the other second membersare not restrictedin this way. This is a fragment of evidence pointing to the eventual separationof animate nouns from the general syntacticclass of nominal expressions.This separationis a necessity for adult grammar; certainverbs can only take animate nouns as objects. We can say: "It surprisedJohn"or "It surprisedthe dog" but not: "It surprisedthe chair."In our discussionof the possessivewe consideredthe posout sibilityof separating a class of generic human terms. The distributional facts of adult English will show that this class has to be broadenedinto animatenouns (we can have "the dog's house"as well as "the man's car") and is the same class hinted at by the restrictions tired. However, Eve's on utterancesare so few that we cannot be sure these restrictionswould be
maintained.
The reciprocal tired in Table 8 is broken,for brokenonly occursafter of inanimatenouns-carriage, chair,dolly, rocker,something.In adult English
66
broken would be restrictedin this way, but we cannot be sure that it will continue so for Eve. All gone is interesting because Eve uses this form where an adult speakerwould not-after count nouns. We would not say "The microphoneis all gone" or "A kitty is all gone." For us all gone goes after pluralcount nouns ("The microphonesare all gone") or mass nouns ("The sand is all gone"). Eve does not give us any plural count nouns in this record and does not make a syntacticdistinction between mass and count nouns, and she uses all gone without any discoverable restrictions.
TABLE9 GrammarsSuggested by the Terminal Recurrenceof Table 8
a. Eve's Grammarwith Two Basic Sentence-Types { Utterance C1 C2 C3 -'s, See, That, There, Two bear, bird, block, boat, etc. a, the, plus human terms (Cl) (C3) + C2
C4 -> all gone, broken, down,tired fall

b. Eve's Grammarwith Articles Added to the New Sentence-Type
Utterance C1 -- 's, See,That,There,Two C2 -> bear,bird,block,boat,etc.
(Ce) (C3) + (C2) +0 C4
+ (Cs) + C2
C3 -> a, the, plus human terms C4 -) all gone, broken, fall down, tired NOTE.- { } means a choice of one of the contained sequences.
The grammar of Table 9b does not represent the distributional distinctions among the class 4 members for the reason that we have so very little evidence on this point. Still, as we shall see in the next section, there are reasons for preferring to represent the distinctions rather than to assume identical privileges. If we were forced to freeze the grammar on the present evidence, it would probably be better to write the more differentiated version. The sentences we have been discussing all begin with a class 2 word (bird, book, etc.) with a prior class 3 (the, a, etc.) being optional. There are many other sentences answering this description in Eve's record, but they do not have the recurrence of second members that made the sentences in Table 8 useful. These additional sentences are all unique and they vary
67
in length: "Daddy fix it"; "Eve listen to tick-tock"; "The reel go round and round"; etc. The best thing to do with these sentences,at present,is to make a cut after the class 2 words (Daddy, Eve, reel) and call all of the remaindersecond members.These second membersare, from the point of view of adult English, complex predicates.With our present inadequate materialswe can only add them to all gone, broken,fall down, and tired as additional membersof class 4. There is not enough material for a finer breakdown.The fact that some of these complex predicatescontain class 2 words and some do not hints at an eventual distinctionbetween transitive and intransitiveverbs.
TABLE 10
of Utterances Previously Identified Added Intonation Typeswith Interrogative

Utterancesthat Have Occurredin DeclarativeForm* 's Daddy, That daddy? Book? Doggie?
Utterancesthat Have Been Predictedin DeclarativeForm but Have Not Occurred See babyeyebrow? Daddy fix this? See babyhorsie? Daddy work? 's a Daddy? See cow? 's cow? That block? 's Mummy? That Evie spoon? Daddy? *Utterances that have occurred without rising interrogativeintonation.
There remains a collection of utterancesbelonging to types already identifiedbut modified by the addition of a rising interrogativeintonation (?), and these appearin Table Io. Since instancesof both kinds of utterances allowed by our grammaroccur with the interrogativeintonation,it seems reasonableto suppose that any sentence allowed by that grammar can be so modified. The result is the revised formula of Table Ii which at one swoop doublesthe number of sentencesthe grammarcan generate. Table II is the completeversion of our provisionalgrammar. There are severalthings worth noting about the grammarof Table I. It will producemost of the 500 utterancesactuallyobtainedfrom Eve and will produce many thousandsbeyond the obtained.The only classes that are not optionalin the grammarare the reference-making words of class 2
(bear, bird, etc.) and class 3 (all gone, broken, etc.). These are, by our observation, the forms that are used as single-word "mands" and "tacts" (Skinner, 1957) for some time before the two-word utterance begins. The generative grammar written out as a formula suggests a distinction between lexical meanings and structural or syntactic meanings. The class
68
TABLE I
Complete ProvisionalGrammarfor Eve

C)+ (C) + (CO)
(CO3) + (C2 +
Utterance
>
-C4
C1 ->
C2 --
's, See, That, There, Two

bear, bird, block, boat, etc.
C3 ->
C4 -)
a, the, plus human terms all gone, broken, fall down, tired, fix it, listen to tick-tock, etc.
listings are the lexicon,and an utteranceinvolvesa shift of lexical meaning wheneverthere is a change in the selectionwithin a class (e.g., from "That bird"to "Thatbear").Structural meaningsmay perhapsbe definedas those that shift wheneverthere is a change in the selectionswithin the formula. Thus if we shift from the first major sentence route to the second, there seemsto be a shift from demonstrative naming ("Therebird"or "Seeboat") to predication("Bird all gone" or "Bearfall down"). If we shift from not selecting ? to selecting it, we shift from a declarationto an interrogation. It is more difficultto say what the changesof meaningcould be that go with selectingor not selecting the variousoptional classes. Testing the Adequacyof a Grammar In our discussionof techniquesfor the discoveryof a grammarwe have repeatedlypointed to the existence of equally reasonablealternativedecisions. The arbitrariness choice could be somewhat reduced by taking a of larger corpus. However, we are sure that the best-foundedgrammarwill not be uniquely determinedbut will only be a good provisionaltry. The same thing is surely true for the rules a child induces from adult speech; they will be hypothesesabout the form of future speech events. Is there any way to check such hypotheses? Very roughly the test of a grammarmust be the same as the test of any theory of empiricalevents-the ability to make correctpredictions.In the present case this means the ability to anticipate sentences that have not of enteredinto the construction the grammar.One might, for example,have tested our grammarfor Eve by taking a second large collection of utterances on the following day or days and seeing how many of those that occur were predicted. Three different kinds of successful prediction may be distinguished. There is, in the first place, occurrenceof an utterancethat has alreadyoccurred in the corpus from which the grammarwas induced. This sort of occurrence does not help to validatethe grammaras a set of generalizations
69
that go beyond the data; the occurrenceof a familiar sentencecould have been predictedfrom a simple list of sentencesin the original corpus.There is, in the second place, the occurrenceof an utterancenot included in the originalcorpusbut allowed for by the generalizedrules. For example,Eve's recorddid not include the sentence "There horsie" though it did include "That horsie" and "The horsie." By adopting the general rule that any class 2 word can follow any class i word, the utterance"There horsie"is predicted.If it occurs,that fact would increaseour confidencein the generalization. Suppose that, in a second corpus collected from Eve, we obtained the utterance"That lion." The word lion did not occur in the original corpus, was not predicted.It is not to be expectedthat a speech and so the utterance sample of only 500 utteranceswould exhaust a lexicon, and it would be unreasonable permit incompletenessof lexicon to discredita grammar. to It would be betterto say that an utterancedoes not come within the province of a grammarunless all of its words are includedin the lexicon of the grammar-the lexicon being the lists of forms belonging to each syntactic class. Therefore, the sentence "That lion" counts neither for nor against the grammar.However, the occurrenceof this sentenceenables us to add
lion to the class
2
list since the criterion for inclusion is occurrence after
a class i word and that is such a word. Subsequentutterancessuch as "There lion" and "The lion" could be counted as evidence for the grammar since they are predictedonce lion has been added to the lexicon. We cannotthink of any way to set a definitecriterionas to the number of sentences, of a collectionof obtainedsentences,that must be predicted out for a grammarto be judged acceptable. is clear,however,that the number It of successfulpredictionscan be increasedby simply writing the grammar in the most generalterms possible.We found, you recall,that the forms all after class2 words (Table 8). gone, broken,fall down, and tired all occurred more closelyat Table 8, we found that the word brokenoccurred Looking only after the names of inanimatethings and tired only after the names of animate things. This fractionof the grammarcould have been written in either of the two forms presentedin Table I2. The less general form predicts 39 utterances,and the more general predictsthese 39 plus 13 others. The more general will make all of the successfulpredictionsmade by the less general and has the possibilityof making some that the less general cannot make. The way to write a "successful" grammar, if "success"is simplymeasuredby the predictionof obtainedevents,is to write a grammar that predictsevery conceivablesequenceof obtainedforms. The job of a grammar,however,is to predictsentencesthat are possible while not predictingsentencesthat are impossible.This second part of the job gives an additional criterion of evaluation that will prevent us from always preferringthe most general grammar.That grammaris to be preferredwhich predictswhat occurswhile predictingas little as possiblethat
70
TABLE 12
Alternative Forms for a Fragment of Eve's Grammar Form a. TheMoreGeneral

Utterance -)
C2
- C4
Mummy, chair,doggie,dollie,eyebrow, kitty,microphone, C2 -) baby,bird,carriage, reel,rocker, something fall.down,tired C4 - all gone,broken,

b. The More RestrictiveForm Utterance ->
+ C5 C2 + tired Can. C2 inan. + broken
chair, doggie, dollie, eyebrow,kitty, microphone, -) baby,bird, carriage, reel, something Mummy, rocker, an -- baby,bird,doggie,dollie, kitty,mummy C2 reel, chair,eyebrow, microphone, rocker,something C2 nan.- carriage, C2
C5
all gone, fall down
does not occur. Therefore, the less general grammar will be better than the more general if it is equally successful in predicting what happens. A grammar should stay as close to the obtained materials as is consistent with generalizing beyond the obtained materials so as to predict future events. Only by invoking both criteria can we hope to obtain distinct grammars representing a developmental sequence. Our next step in this research program will be an empirical and logical study of techniques for evaluating grammars in terms of their predictive powers. (See Brown, Fraser, and Bellugi.) In sum, the most that we are now able to conceive by way of grammarevaluating techniques is a set of rough criteria for preferring one version to another. This problem of evaluation probably exists in implicit form for the child learning to speak. How does he judge the adequacy of his inductions? He can see how well they anticipate what others say. He can construct new utterances and see whether others appear to find them acceptable and comprehensible. He can ask direct metalinguistic questions: "Mother, can one say 'The carriage is tired'?"
OF CHILD GRAMMARAS A REDUCTION ADULT GRAMMAR
The provisional grammar we have written for Eve is not a grammar of adult English. Most of the utterances it generates are not, for us, gram-
71
matical sentences.We cannot accept "That horsie"or "Kitty all gone" or "ThereMummyhair"or "Two chair."Words that seem to be syntactically equivalentfor Eve are not so for us. For example, the words see and that are membersof class i for Eve and equivalent in that they can precede class2 wordssuch as doggie, dollie, etc. For the adult who might say "That is a doggie"and "I see a doggie,"that and see are syntactically unlike. For Eve the word two is roughly equivalent to the and a, but that is because she does not pluralizecount nouns. It is clearthat Eve does not speak adult English, and so we cannot use an adult English grammarto describeher speech. It would be quite misleading to list Eve's that as a demonstrative pronounand her see as a transitiveverbthough that is a correctclassification of these words in adult English. Since the speech of the younger children in our set of 13 is clearlynot the English, it is necessaryto "discover" grammarthat is implied by their utterances.We have not in this first try collectedlarge enough corpusesto write convincinggrammars. However, for the youngerchildrenit does look as if we might manage with about 1500 differentutterancesto get a good provisional grammar. The prospect of "discovering"grammars for the older children is not nearly so good. Since the older children have larger vocabularies longer utterances, and there is less of the recurrence that is the basis of a structuralanalysis. The anthropological linguist who wants to describeadult usage in an unfamiliarlanguage is likely to try something like i,ooo hours of speech,and there are times when we fear that not very much less than this would sufficefor children of about 3 years. The speechof the children in our collectionwho are nearly 3 years old is mostly English. They produceacceptable simple sentences.Furthermore, these simplesentencescorrespond very well with the set of English sentences for which a satisfactory generativegrammarhas been written by Chomsky (1957) and Lees (I960). For these records,then, it does not seem necessary to "discover" grammar; it is quite reasonable analyze them in terms the to of the syntacticcategoriesof the adult grammar.We have done this job, classifying utterancesunder headings: copular sentences, transitive verb sentences,intransitiveverb sentences,sentenceswith modal auxiliaries,sentences with progressives, sentenceswith preliminary verbs,imperatives, negYes-No interrogatives,and unclassifiablefragatives, Wh-interrogatives, ments. When this analysisin terms of an adult grammarhad been done for the most advancedrecords,we realized that it could, with some additional assumptions,be extended to the recordsof all the other children. For the striking fact about the utterancesof the younger children, when they are from the vantagepoint of adult grammar,is that they are almost approached all classifiable grammatical as sentencesfrom which certainmorphemeshave been omitted. You may have noticed that while Eve's sentences are not or grammatically"complete"they are somehow intelligible as abbreviated
72
telegraphicversionsof familiar constructions. "Mummyhair" and "Daddy car" seem only to omit the possessiveinflection.Both "Chairbroken"and "That horsie" become acceptablecopular sentencesif we leave the wordorder intact and fill in is and a or the. We have, therefore,analyzed all of the recordsin terms of English simple sentencesby assigning each utterance to the sentence-type most closely approximates. it We have not yet hit on any good techniquesfor summarizingour compendious data, but one illustrativetable (Table I3) will make a few imTABLE13
of Estimates MeanUtterance on Grammatical Lengthand Reports Selected Formsfor 13 Children

Name of Child Andy ........ Betty ........ Charlie ...... Adam ....... Eve ......... Fanny ....... Grace ....... Helen ....... Ian .......... June ........ Kathy ....... Larry ........ Jimmy ....... Age in Months 26
3I /2 22 28 /2
MeanNumber Morphemes*
2.0 2.1 2.2 2.5
Be in Progressive no no no no no yes yes yes yes yes yes yes yes
Modal Auxiliaries will or can no no no no no no yes yes yes yes yes yes yes
25/2 31 z
27 30 35/2 30 / 35/2
32
2.6 3.2 3.5 3.6 3 3.8 4.5 4.8 4.8 4.9
* Mean count from 0oo consecutive utterances.
portantpoints. For each child we made an estimateof the mean length of utterancesby counting morphemesfrom Ioo consecutiveutterancesoccurring in midmorning.From this count we thought it wise to omit the singleword rejoindersand exclamations:No, Ok, Yeah, and Oh since these are sometimesemitted many times in successionby speakersof any age, and if the small utterancesample happenedto include such a repeatingcircuit the estimate would be very unrepresentative. The estimates appear in Table 13, and the children'snamesappearin an orderof increasingaverage utterancelength. This order is related to the chronologicalage order but not identicalwith it. An age-relatedincreasein the mean length of utterances is, of course, one of the best establishedfacts in the study of child speech (McCarthy, 1954). There is, however, no reason to expect perfect with age. We conceive of developmentalsequence as a correspondence Guttman scale with performancesfollowing an invariant order but not
TABLE 14
Imitations of Spoken Sentences ModelSentence I. I showed you the book. 2. I am very tall......... 3. It goes in a big box ... 4. Read the book. ....... 5. I am drawing a dog ... 6. I will read the book ... 7. I can see a cow. ...... 8. I will not do that again. 9. I do not want an apple. II. Is it a car? ...........
12. Where does it go? ....
Eve, 25 /2 I show book. (My) tall. Big box. Read book. Drawing dog. Read book. See cow. Do again. I do apple. 't car? Where go? Go?
Adam, 28 /z (I show) book. I (very) tall. Big box. Read book. I draw dog. I will read book. I want see cow. I will that again. I do a apple. I read books? Is it car? Go?
Helen, 30 C I very tall. In big box. I drawing dog. I read the book. C I do that.
lan, 31 /z
I show you the bo I'm very tall. Read (a) book. Dog. I read the book. Cow. I again. I read book? That a car? Where do it go? C
It goes in the box.
I do not want app I read books? Car? Does it go?
Io. Do I like to read books? To read book?
13. Where shall I go? .....
NOTE.-C means imitation correct. -
indicates that no intelligible imitation was obtained. ( ) i
74
pegged to particularages. The sequencecan be coveredat varying rates of speed; the rate would be a function of intelligence and learning opportunities. When the analyzed records are ordered by mean utterancelength, it becomes apparentthat children all "reduce"English sentences in similar fashion. If the utterances one child are, on the average,shorterthan the of utterancesof anotherchild, then, of course,the first child will be omitting some morphemesthat the second child is producing.However, utterances could be shortenedby omitting any morphemesat all; thereare many ways to abbreviate and that sentences, it is perfectlyconceivable individualchildren would hit on differentways. They do not do so. We have checkedall the records for progressive constructions (e.g., "I am going to town"). Without exchildren whose mean utterancelength is below 3.2 form this conception, structionby omitting the forms of the verb to be (e.g., "I going"). We have checkedsentencesin which the verb would, for an adult, ordinarilyrequire such a modal auxiliaryas will or can (e.g., "I will park the car"). Children whose mean utterancelength is below 3.5 invariably form these sentencesby modals ("I park the car"or "I go outside"or "I make a tower"). omitting In general, it appearsthat children whose speech is not yet English are using grammarswhich are systematicderivativesof adult grammar and that the particularfeaturesof the derivativegrammarare predictablefrom the mean length of utterance. The abbreviation effect can be more directly studied in the utterances a child produceswhen he is asked to repeat back a sentence said by an adult. For the last six cases we presented 13 simple sentencesof various
TABLE 15
of Imitations Summary Resultswith Sentence

Number of Morphemes Imitated Cor- M rectly in Each of Three Serial Positions Mean Lengti on Imitations Initial Middle (in morphemes) Final 4 9 6 9 12
12
Name and Age of Child .................... Eve, 25 Adam, 28 2 ................... Helen, 30 .................... Ian, 31 / .................... Jimmy, 32 .................... June, 35/2 ...................
5 3 8 7
1
13 64%
I2 12 9 12 13 13 96%
2.2 3.1 3.7 3-5 5.0 4.9

5.2
Per cent correctlyimitated ....... Mean length of model sentences ..
70%
NOTE.-The difference between initial and final positions is significant by sign test (two-tailed) with p = .032; the difference between middle and final positions is significant with p = .062; the difference between initial and middle positions is not significant.
75
grammaticaltypes and ask the subject to "Say what I say." Half of the sentenceswere presentedat the end of the morning session and half at the end of the afternoonsession.For each child a differentrandom order was presented.The sentenceswere spoken slowly and carefullyand always by the same investigator.The child had the microphonedirectly in front of him, and so good recordingswere made. Table 14 presentsthe first efforts of each child, and Tables 15 and I6 a summaryof the main results.
TABLE 16
PercentagesCorrectlyImitated of Morphemesin Various SyntacticClasses

SyntacticClass ClassesHaving Few Membersin English Inflections ................................................ Pronouns .................................................. Articles .................................................. Modal auxiliaries ......................... Copular verbs (to be) ....................................... ClassesHaving Many Membersin English Nouns ................................................... Adjectives ................................................ Main verbs ............................................... Imitated Correctly 44% 72 39 56 33 100 92 85
............
NoTE.-Using tests for differences between two percentages, the percentage correct in each of the classes with many members is significantlygreater than the percentagecorrect in any of the classes with few members (p < .ooI, 2-tailed test).
With increasingage, children produce more imitations that are morphemically identical with the original. With increasing age, the imitative utterances produced include larger numbers of morphemes-approaching the numbers in the model sentences. The morphemes produced are invariably in their original order. Omissions do not appear to be random or idiosyncratic. On the contrary, it looks as if, across children and across sentences, there is a consistent tendency to retain one kind of morpheme and drop another kind. The two sorts of morphemes contrast on several correlated dimensions. The morphemes most likely to be retained are: morphemes that occur in final position in the sentence; morphemes that are referencemaking forms; morphemes that belong mainly to the large and expandable noun, verb, and adjective parts-of-speech;6 morphemes that are relatively unpredictable from the context; and morphemes that receive the heavier stresses in ordinary English pronunciation. The morphemes least likely to
6Aborn and Rubenstein (1956) have published evidence that, in six-word sentences, nouns are most frequent in final sentence position whereas function words tend to be most frequent in the fourth and fifth positions. This tie between position and part-ofspeech is found also, and not by our intention, in the sentences we provided for imitation.
76
be retained are: morphemes that occur in intermediatepositions in the sentence; morphemes that are not reference-making forms; morphemes that belong to such small-sizedgrammatical categoriesas the articles,modal auxiliaries,and inflections;morphemesthat are relativelypredictablefrom context and so carry little information; and morphemesthat receive the weaker stressesin ordinaryEnglish pronunciation. This does indeed seem to be telegraphicEnglish. There is substantialsupportfor our findings in the resultswith sentencerepetitionfor Ioo childrenat ages 2 and 3 obtained by Gesell and his associates(I940) and also in work of Stutsman(1926). Let us suppose that very young children speak a rather uniform telegraphicEnglish. How do they come by it? It is conceivablethat they hear it from adults, that they are imitating a "babytalk" which is an adult invention. We have on our tapes a large quantity of speech from mother to or child, and, while this materialhas not yet been transcribed analyzed, it is quite clear that Eve's mother and the mothers of the other children do not usually use telegraphicEnglish. The very young children are exposed to a much more complicated grammarthan they use and the older children to a somewhatmore complicatedgrammarthan they use. While it seems safe to say that children do not learn their telegraphic English from adults, it is probablyalso safe to say that the average adult can do a good job of producingtelegraphicEnglish if he is asked to talk like a baby.We have even heard a 3-year-old drop his more mature grammar in speaking to a 2-year-oldand produce a very good version of the 2-year-old'sspeech. It is an old observationof linguists that the "baby talk" version of a language is very uniform from one adult to another. We can see good reasonswhy this should be so. If there is somethingabout the operationof the child mind that causeseach child to "reduce"English in a similarform, then adults everywhere could learn the same sort of baby talk from their own children. It is even possible,of course, that baby talk in all languagesshows certainstablefeatures,e.g., omissionof low-informaforms. If, in addition,baby talk is a systematictransformation, predictable tion of adult simple sentencesaccomplished the omissionof certainkinds by of words,then an adult should be able to throw some simple mental switch that activatesthe baby grammar. A basic factor causing the child's reductionof adult sentencesis surely an upper limit on some kind of immediatememory span for the situation in which the child is imitating and a similar limit of programmingspan for the situationin which the child is constructingsentences.A comparison of the mean lengths of utterancesproduced as imitations (presented in Table I5) with the mean lengths of spontaneouslyproduced utterances from the same children (presented in Table 13) shows that the paired values are very close and that neither is consistentlyhigher. An increasing span for random digits is so reliably related to increasingage as to be a test partof the Stanford-Binet (e.g., a span of two digits at 30 months,three
77
at 36, four at 54 months are the norms). Munn summarizesthe resultsof many studiesby saying: "That the memory span of children increaseswith age has been shown for all kinds of materialinvestigated"(1954, p. 413). We know from Blankenship'sreview of memory span work (1938) and from Miller'sdiscussionof "the magical number seven" (1956) that it is not yet possibleto reduce the variousmeasuresof span to a common unit (bits) and therebyreconcilevariationsin the data. But there does seem to be ample indicationthat one-or-more memoryspans show a steady increase in early childhood. It is this limitation of span in children for which the work of the descriptive linguistprovidedno parallel,and our obsessionwith linguistic technique long diverted us from recognizing the systematically derivativenatureof child speech. Span limitation is probablythe factor compelling children to reduce adult sentences,but it does not, of course, accountfor the systematictendency to drop one sort of morphemeand retain another sort. Because the two sorts of morphemesdiffer in numerous correlatedrespects,there are many conceivableexplanationsfor the child's selective performance,and this will be so until the variablesare experimentally separated.Here are a few of the many ways in which the storycan now be put together.Perhaps the human mind operateson an unlearned"recency" principle,and English sentences(maybe also sentencesin other languages) are nicely adaptedto this principle in that the least predictable,most informative words fall usually into the final position. Perhaps,on the other hand, the "recency" effect in human serial learning is an acquired tendency to pay particular attention to material in final position, a tendency acquired from the fact that sentencesare so constructed to place in that position words carrying as a lot of information.Perhapsit is differentialstress that selects what the child will reproduceand sentencesare nicely adaptedto this predilectionin that the heavierstressesfall on the less predictable forms. Or perhapsit is some combinationof these ideas.
CONCLUSION
This paper began with an argument that the correctEnglish sentences producedby a child are not good evidence that he possessesconstruction rules for the reason that we can never be sure that a correct sentence is not directly copied from a model. It seemed to us that systematicerrors and manipulationof inventedwords were betterevidence,as a child is not likely to have had exact models for these. In the secondsectionof the paper we discussed techniques for inducing constructionrules or a generative grammarfrom the child'snaturalspeech.Since this speech,for Eve at least, is not good English, it can be argued that she had no models for it and so it is legitimateto infer rules from such data. In the third section,however, we have seen that child speech can be rather well characterized a sysas
78
tematic reductionof adult speech, and so, after all, there were models for Eve's sentences.She could have learnedmost of them by selectiveimitation if not by imitation per se. Eventuallychildren must do more than imitate and memorize if only becausethere is not enough time for them to learn as particularverbalresponsesall the sentencesthey will be able as adults to produceand evaluate as to grammaticality. (For a detailedstatementof this argument,see Bruner,
1957, p. I56; or Miller, Galanter, and Pribram, I960, pp. 146-147.) In addi-
tion to the logical argument that children must learn constructionrules in view of their terminal linguistic achievement,there is much empirical evidence that children older than Eve do, in fact, learn constructionrules. Some of this evidence is availableto every parent in a child's systematic errors(sheeps,I bringed,etc.), and some of it has been collectedin a controlled fashion by Berko (1958) and by Brown and Berko (I960). While children must and do eventuallyinduce constructionrules, it is not necessary that they do so from the very earliestage at which words are combined.Eve, after all, is not yet preparedto producean infinite set of sentences,nor, so far as we know, is she able to distinguishall grammatical sequencesfrom ungrammatical sequences.It is possiblethat for the earliest one linguistic accomplishments sort of learning theory will serve-a theory developedlargely from the study of animal behavior-while for later accomplishmentsa completely different theory will be necessary-a theory permitting the inductive formation of syntacticclasses that generalize far beyondobtainedinformation.However, it is also possiblethat the induction of rules goes on from the very first. Eve, for example,producessuch utterances as "Two chair" and "Kitty all gone" which could conceivablybe either direct or selectiveimitations,but might very easily be constructions rules for the use of two and all gone. For a resultingfrom overgeneralized we will show how Eve might be forming construction rules single example, at the same time that she is practicingselectiveimitation without this constructionprocessbeing clearlyrevealedin the presentdata. Selective reductionmight cause Eve to imitate "That is a doggie" as "That doggie"; "That is a horsie"as "That horsie";and "See the doggie" as "See doggie." At this point induction could operateand she might, beassume "is sharedby "doggie"and "horsie," causethe context"That that the othercontext"See "can be sharedand so form the utterance "See horsie."It is quite possiblethat the child reducesfirst, then forms inductive generalizations,and makes new utteranceson the model of his reductions. It will often happen if the above suggestionis correctthat a child will find forms syntactically equivalentthat are not so for the adult. SupposeEve heard the sentence "I see the man" and reduced it to "See man." The the original demonstrates nonequivalenceof see and that since see occurs after a pronoun, and, in addition, the original version does not justify
BROWNand COLINFRASER ROGER
79
adding man to the list started with doggie and horsie. The reduced sentence, on the other hand, leaves see equivalent to that and suggests that man belongs with doggie and horsie. If a child induces a grammar from its own reduced sentences,it should generallylose the distributional detail provided by such morphemesas an, the, will, do, and /-z/. The result would be syntactic classes not identical with those of adult speech, few in number and large in size. In addition there would be loss of numerous semantic distinctions,e.g., the difference between "See the man" and "I see the man." It is importantto note, however,that the gross sense of a sentencewill usuallybe retained; e.g., in "See man" as in "I see the man" it is clearthat man is the objectof seeing. The crude sense of the sentenceis generallyrecoverable from the child's reductionfor the reason that one profound dimension of English grammar is perfectly preserved in telegraphese, and that dimensionis word order. For the present, then, we are working with the hypothesisthat child speech is a systematicreduction of adult speech largely accomplishedby omitting function words that carry little information.From this corpus of reduced sentences we suggest that the child induces general rules which govern the constructionof new utterances.As a child becomes capable (through maturation and the consolidation of frequently occurring sequences) of registeringmore of the detail of adult speech,his originalrules will have to be revisedand supplemented. the generativegrammargrows As more complicated and more like the adult grammar,the child's speechwill become capableof expressinga greatervariety of meanings.
EXPLORATIONS IN GRAMMAR EVALUATION1

ROGERBROWN,COLIN FRASER,2
and URSULA BELLUGI

HarvardUniversity
In the I96I edition of Gleason'sAn Introductionto Descriptive Linguistics the following paragraph appears: The difference of betweena description a corpusand of a languageis of sentences. A partlya matterof scope.A corpusconsists a few thousand as of language mightbe considered consisting a very largenumberof sentences-all those,eitheralready spokenor not yet used,whichwouldbe ac1 The work describedin this paper was supported by the National Science Foundation by a grant administered through the Center for Communication Sciences, Massachusetts Institute of Technology. 2 Now at the University of Exeter, Exeter, England.

Brown and Frazer The Acquisition of Syntax

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Brown and Frazer The Acquisition of Syntax

Uploaded by

Copyright:

Available Formats

The Acquisition of Syntax Author(s): Roger Brown and Colin Fraser Source: Monographs of the Society for Research

III THE ACQUISITION OF SYNTAX'

adjectives and nouns. Chomsky (I957)

has shown that it may be possible

ROGER BROWN and COLIN FRASER

ROGER BROWNand COLINFRASER

ROGER BROWN and COLIN FRASER

ROGER BROWN and COLIN FRASER

ROGER BROWN and COLIN FRASER

Methodsfor Discoveringa ProvisionalGrammar

Total Contexts of Four Words in the Record of Adam

ROGER BROWN and COLIN FRASER

* Identifies contexts common to first and second halves of record.

ROGER BROWN and COLIN FRASER

Two-Word Utteranceswith RecurrentInitial Words from Eve Second

( Table continued on next page )

Two-Word Utteranceswith RecurrentInitial Words from Eve

ROGER BROWNand COLINFRASER

A Grammarthat Results from Filling In the Blanks in Table 3

NOTE.-This grammarpredicts the 89 utterancesobtained plus 469 others.

ROGER BROWN and COLIN FRASER

Three-Word Utterances that Refine the Original Grammar

Utteranceswith RecurrentSecond Members

* A word already identified as a member of C2.

ROGER BROWN and COLIN FRASER

C4 -> all gone, broken, down,tired fall

Utterance C1 -- 's, See,That,There,Two C2 -> bear,bird,block,boat,etc.

(Ce) (C3) + (C2) +0 C4

ROGER BROWN and COLIN FRASER

of Utterances Previously Identified Added Intonation Typeswith Interrogative

Complete ProvisionalGrammarfor Eve

's, See, That, There, Two

ROGER BROWN and COLIN FRASER

list since the criterion for inclusion is occurrence after

Alternative Forms for a Fragment of Eve's Grammar Form a. TheMoreGeneral

Mummy, chair,doggie,dollie,eyebrow, kitty,microphone, C2 -) baby,bird,carriage, reel,rocker, something fall.down,tired C4 - all gone,broken,

all gone, fall down

ROGER BROWN and COLIN FRASER

of Estimates MeanUtterance on Grammatical Lengthand Reports Selected Formsfor 13 Children

Be in Progressive no no no no no yes yes yes yes yes yes yes yes

2.6 3.2 3.5 3.6 3 3.8 4.5 4.8 4.8 4.9

* Mean count from 0oo consecutive utterances.

It goes in the box.

I do not want app I read books? Car? Does it go?

Io. Do I like to read books? To read book?

13. Where shall I go? .....

NOTE.-C means imitation correct. -

indicates that no intelligible imitation was obtained. ( ) i

of Imitations Summary Resultswith Sentence

2.2 3.1 3.7 3-5 5.0 4.9

Per cent correctlyimitated ....... Mean length of model sentences ..

ROGER BROWN and COLIN FRASER

PercentagesCorrectlyImitated of Morphemesin Various SyntacticClasses

ROGER BROWN and COLIN FRASER

BROWNand COLINFRASER ROGER

EXPLORATIONS IN GRAMMAR EVALUATION1

and URSULA BELLUGI

You might also like