Competência Sintática em Crianças Pequenas - Análise Cognitiva e Aquisição da Linguagem

ELSEVIER COGNITION Cognition 74 2000 209253 wwwelseviercomlocatecognit Do young children have adult syntactic competence Michael Tomasello Max Planck Institute for Evolutionary Anthropology Inselstrasse 22 D04103 Leipzig Germany Received 19 November 1998 received in revised form 1 September 1999 accepted 21 September 1999 Abstract Many developmental psycholinguists assume that young children have adult syntactic competence this assumption being operationalized in the use of adultlike grammars to describe young childrens language This continuity assumption has never had strong empirical support but recently a number of new findings have emerged both from systematic analyses of childrens spontaneous speech and from controlled experiments that contradict it directly In general the key finding is that most of childrens early linguistic competence is item based and therefore their language development proceeds in a piecemeal fashion with virtually no evidence of any systemwide syntactic categories schemas or parameters For a variety of reasons these findings are not easily explained in terms of the development of childrens skills of linguistic performance pragmatics or other external factors The framework of an alternative usagebased theory of child language acquisition relying explicitly on new models from CognitiveFunctional Linguistics is presented 2000 Elsevier Science BV All rights reserved Keywords Language Language acquisition Cognitive development Syntax 1 Introduction To become a competent speaker of a natural language it is necessary to be conventional to use language the way that other people use it To become a competent speaker of a natural language it is also necessary to be creative to formulate novel utterances tailored to the exigencies of particular communicative circum Tel 493419952400 fax 493419952119 Email address tomasevampgde M Tomasello 0010027700 see front matter 2000 Elsevier Science BV All rights reserved PII S0010027799000694 stances From the beginnings of modern cognitive science and further traceable at least back to Kant this paradoxical ability to be simultaneously conventional yet creative has been explained in terms of the human capacity to operate with abstract cognitive entities such as categories schemas structures or rules Interestingly young children show evidence of operating with at least some linguistic abstractions from very early in ontogeny Thus from the very beginnings of multiword speech children create novel utterances that they have never before heard for example the famous Allgone sticky as reported by Braine 1971 Based on this fact and on some logical arguments about learnability many researchers in the Generative Grammar Chomskian tradition have even gone so far as to posit that young children operate with adultlike linguistic competence The milder version of this continuity assumption states In the absence of compelling evidence to the contrary the childs grammatical rules should be drawn from the same basic rule types and be composed of primitive symbols from the same class as the grammatical rules attributed to adults in standard linguistic investigations Pinker 1984 p 7 The most extreme version of the continuity assumption asserts that young children from the beginning have essentially full linguistic competence A survey of recent influential contributions to the field Generative Grammer MT suggests that the proposal that the child embarks on grammatical development with a complete in some sense system of syntactic representation is widely supported Atkinson 1996 p 451 The continuity assumption is in many ways the fundamental theoretical postulate of generative approaches to language acquisition because it and only it enables linguists to describe young childrens language with adultlike formal grammars Recently however some new data have emerged that invite a new look at the continuity assumption in both its milder and more extreme forms Basically the data show that young childrens creativity productivity with language has been grossly overestimated beginning language learners produce novel utterances in only some fairly limited ways Specifically beginning language learners quite readily substitute nominals for one another and so generalize from such things as Allgone juice and Allgone paper to Allgone sticky sticky being conceived as a substance Such creativity is convincing evidence that these children have something like an abstract category of nominal perhaps limited to concrete objects people and substances from very early in development However beginning language learners are not creative or productive with their language in some other basic ways For example they do not use a verb in a sentence frame in which they have not heard it used Thus on the basis of hearing just The window broke and no other uses of this verb they cannot go on to produce He broke it or It got broken even though they are producing simple transitive and passive utterances with other verbs This lack of productivity suggests that young children do not yet possess abstract and verbgeneral argument structure constructions into which different verbs may be substituted for one another as needed but rather they are working more concretely with verbs as individual lexical items whose syntactic behavior must be learned one by one Overall childrens limited creativity with their early language calls into question the practice of describing their underlying syntactic competence in terms of abstract and adultlike syntactic categories schemas and grammars In this paper I do three things First I present some new data suggesting that young childrens early language is more concrete and itembased than is generally recognized Second I discuss the implications of these new data for generative Chomskian approaches to language acquisition which routinely make the continuity assumption and so use adultlike formal grammars to describe early language Third I attempt to spell out the general outlines of an alternative theory of language acquisition that does not attribute to young children adultlike syntactic competence This alternative theory is a usagebased theory inspired by the new models of linguistic competence from CognitiveFunctional Linguistics and it attempts to account for the new data in a very specific manner 2 Some new data on child language acquisition Most of childrens early language is grammatical from the adult point of view But there are at least two very different explanations for this fact One is that children are operating from the beginning with adultlike grammatical categories and schemas The other is that children are learning to use specific linguistic items and structures eg specific words and phrases in the way that adults are using them with the proviso that they can substitute nominals for one another relatively freely In other words young children may be using language like adults either because they have the same underlying linguistic competence as adults or because they are imitatively learning from them Given that childrens use of language in adultlike ways does not differentiate between these two explanations not even when they are able to meet Brown 1973 criterion of use of a grammatical structure in 90 of its obligatory contexts since this may still simply reflect reproduction of adult usage deeper analyses of childrens linguistic competence are needed The key requirement is to find some way to differentiate between utterances the child is generating on the basis of specific words and phrases and those she is generating on the basis of more abstract linguistic categories and schemas There are two basic methods both of which focus on childrens productivity that is their use of language in ways that go beyond what they have heard from adults The first method is the analysis of childrens spontaneous speech but with the stipulation that we look at all of a childs uses and most especially nonuses of a particular set of linguistic items or structures Thus a Spanishspeaking child might produce Te amo a thousand times correctly but a systematic analysis might also reveal that she uses this verb in none of its other forms for different persons or numbers If indeed there have been opportunities to use this verb in these other ways and there are no other external factors preventing such usage this limited facility with this verb tells us much about this childs overall syntactic competence The second method involves teaching children novel linguistic items and seeing what they do with them Berko 1958 For instance if we teach a Spanishspeaking child a novel verb ponzar for some novel madeup action the question would be Can she immediately use this newly learned verb in all of its persons numbers tenses and modalities or can she use it only in the way she has heard it used Like the experimentally introduced tracer elements used in medical diagnoses if the novel word is used in creative yet canonical ways the inference is that it has indeed been taken up by some kind of internal system in the current example abstract syntactic categories and schemas concerning verb person and number If it is not used in any creative ways but only in ways the child has already heard the inference is either that i there is no abstract system to take up the new element and the child is simply learning a specific linguistic item or structure or ii for some reason the existing abstract system is unable to take up the new element This latter possibility means for the most part the possibility that there are performance factors eg limited processing or memory skills that prevent the child from demonstrating her syntactic competence in the experiment Recent data collected by each of these two methods helps to specify which aspects of childrens language are generated on the basis of concrete linguistic items and structures and which aspects of their language are generated on the basis of abstract linguistic categories and schemas I first review the observational data and then the experimental data 21 Observational studies Even in the earliest modern analyses there were suggestions that young children were using at least some of their language in itemspecific ways that is that individual children were not showing great systematicity across different aspects of their early language development even when from an adult perspective they should have been For example a given child might use a lexical item like up in all kinds in interesting ways in all kinds of interesting combinatorial patterns but then use the very similar lexical items down and on only as single word utterances even when it would be to their communicative benefit to use them in word combinations Bowerman 1976 suggested that one of her two Englishspeaking children had many such itemspecific constructions MacWhinney 1978 suggested the same for at least some of his Hungarianspeaking children and Braine 1976 found many itemspecific patterns in the spontaneous speech of several children learning a number of different languages All of these researchers however concluded that most of the children also had some more general patterns as evidenced by the fact that they sometimes used semantically similar items in similar ways at a given developmental period for example a particular child might use the verbs eat and drink in similar ways at a given time The problem with this kind of data however is that we do not know if adults talking to this child used these particular lexical items in this same way and so we cannot know whether the childs similar use of these items is due to her abstract linguistic competence or to her imitative learning from adults Tomasello 1992 performed analyses aimed at these questions using diary data that for all practical purposes included all of the different ways his Englishspeaking child T used each of her verbs during the period from 15 to 24 months of age The advantage of continuous diary data over all other kinds of child language data is that they include information about what the child did not do an inference that is always extremely weak when periodic sampling is used eg one hour every two weeks as in most studies The major findings of this study may be summarized as follows Of the 162 verbs and predicate terms used almost half were used in one and only one construction type and over twothirds were used in either one or two construction types where construction type means verbargument configuration eg Mommy break and Daddy break are the same construction type whereas Break cup Mommy break cup and Break with stick are three additional construction types At any given developmental period there was great unevenness in how different verbs even those that were very close in meaning were used both in terms of the number and types of construction types used For example at 23 months of age the verb cut was used in only one simple construction type Cut whereas the similar verb draw was used in many different construction types some with much complexity eg I draw on the man Draw it by Santa Claus Where information on adult usage was available for a given verb there was a very good match with child usage see also DeVilliers 1985 Naigles HoffGinsberg 1998 There was also great unevenness in the syntactic marking of the same argument across verbs such that for example at a given developmental period one verb would have its instrument marked with with or by but another verb even when used in utterances of the same length and complexity would not have this marker Some verbs were used with lexically expressed subjects whereas others at the same time were not even though they were used in comparable construction types and in comparable pragmatic contexts eg T produced subjects for take and get but not for put Morphological marking on verbs was also very uneven with roughly twothirds of all verbs never marked morphologically for tense or aspect onesixth marked for past tense only onesixth marked for present progressive only and only 4 verbs 2 marked for both of these functions during the second year of life see Bloom 1992 Clark 1996 On the other hand within any given verbs development there was great continuity such that new uses of a given verb almost always replicated previous uses and then made one small addition or modification eg the marking of tense or the adding of a new argument By far the best predictor of Ts use of a given verb on a given day was not her use of other verbs on that same day but rather her use of that same verb on immediately preceding days The resulting hypothesis the Verb Island Hypothesis was that childrens early language is organized and structured totally around individual verbs and other predicative terms that is the 2yearold childs syntactic competence is comprised totally of verbspecific constructions with open nominal slots Other than the categorization of nominals nascent language learners possess no other linguistic abstractions or forms of syntactic organization This means that the syntagmatic categories with which children are working are not such verbgeneral things as subject and object or even agent and patient but rather such verbspecific things as hitter and hittee sitter and thing sat upon Using a combination of periodic sampling and maternal diaries Lieven Pine and Baldwin 1997 see also Pine Lieven 1993 Pine Lieven Rowland 1998 found some very similar results in a sample of 12 Englishspeaking children from 1 to 3 years of age In particular they found that virtually all of their children used most of their verbs and predicative terms in one and only one construction type early in language development suggesting that their syntax was built around these particular lexical items In fact fully 92 of these childrens earliest multiword utterances emanated from one of their first 25 lexicallybased patterns which were different for each child Following along these same lines Pine and Lieven 1997 found that when these same children began to use the determiners a and the in the 2 to 3 year period they did so with almost completely different sets of nominals ie there was almost no overlap in the sets of nouns used with the two determiners suggesting that the children at this age did not have any kind of abstract category of Determiner that included both of these lexical items A number of systematic studies of children learning languages other than English have found very similar results For example Pizutto and Caselli 1994 investigated the grammatical morphology used by 3 Italianspeaking children on their simple finite main verbs from approximately 15 to 30 years of age see also Pizutto Caselli 1992 Although there are six forms possible for each verb root firstperson singular secondperson singular etc 47 of all verbs used by these children were used in 1 form only and an additional 40 were used with 2 or 3 forms Of the 13 of verbs that appeared in 4 or more forms approximately half of these were highly frequent highly irregular forms that could only be learned by rote The clear implication is that Italian children do not master the whole verb paradigm for all their verbs at once but rather they only master some endings with some verbs and often different ones with different verbs In a similar study of one child learning to speak Brazilian Portugese at around 3 years of age Rubino and Pine 1998 found a comparable pattern of results including additional evidence that the verb forms this child used most frequently and consistently corresponded to those he had heard most frequently from adults That is this child produced adultlike subjectverb agreement patterns for the parts of the verb paradigm that appeared with high frequency in adult language eg firstperson singular but much less consistent agreement patterns in low frequency parts of the paradigm eg thirdperson plural For additional findings of this same type see Serrat 1997 for Catalan Behrens 1998 for Dutch Allen 1996 for Inuktitut Gathecole Sebastián Soto 1999 for Spanish and Stoll 1998 for Russian Finally in a study of 6 Hebrewspeaking children a language that is typologically quite different from most European languages Berman and ArmonLotem 1995 see also Berman 1982 found that childrens first 20 verb forms were almost all rotelearned or morphologically unanalyzed p 37 Of special note in spontaneous speech are socalled overgeneralization errors because presumably children have not heard such forms used in adult speech In the context of a focus on syntax the overgeneralizations of most interest are those involving argument structure constructions for example She falled me down or Dont giggle me in which the child uses verbs in syntactic constructions in noncanonical ways that seem to indicate that she has some abstract verbgeneral schema for such things as a transitive SVO construction Bowerman 1982 1988 in particular documented a number of such overgeneralizations in the speech of her two Englishspeaking children and Pinker 1989 compiled examples from other sources as well The main result of interest in the current context is that these children produced very few argument structure overgeneralizations before about 3 years of age and virtually none before 25 years of age see Pinker 1989 pp 1726 These dataintensive studies from a number of different languages together show a very clear pattern First young childrens earliest linguistic productions revolve around concrete items and structures there is virtually no evidence of abstract syntactic categories and schemas Second each of these items and structures undergoes its own development presumably based on individual childrens linguistic experience and other factors affecting learning in relative independence of other items and structures Third this pattern persists in most cases until around the third birthday at least for relatively large structures such as transitive SVO utterances and other verbargument constructions and so suggests that childrens earliest syntagmatic categories are lexically specific categories such as kisser kissee seer thing seen and so forth and so on In light of these findings the claim that young children possess abstract adultlike categories such as subject object agent or patient is tantamount to the claim that their naturally occurring language does not reflect their underlying syntactic competence Data from spontaneous speech by itself cannot decide the issue of course because we never know for certain what the child has and has not heard and so inferences about child productivity are always indirect ie they are based on what the child most likely has heard given typical adult usage Experimental observations on the other hand control the language that children hear and so can potentially remedy this weakness of natural observations for answering the basic question of child productivity 22 Experimental studies There is no question that young children learn and use the linguistic items and structures to which they are exposed with amazing facility Thus in their spontaneous speech young Englishspeaking children use canonical word order for most of their verbs including transitive verbs from very early in development Bloom 1992 Braine 1971 Brown 1973 In comprehension tasks children as young as two years of age respond appropriately to requests that they Make the doggie bite the cat reversible transitives that depend crucially on a knowledge of canonical English word order eg Bates MacWhinney 1989 Bates MacWhinney Caselli Devoscovi Natale Venza 1984 Chapman Miller 1975 DeVilliers DeVilliers 1973 Roberts 1983 Slobin Bever 1982 and successful comprehension is found at even younger ages if preferential looking techniques are used HirshPasek Golinkoff 1991 1996 But as noted earlier if we do not know what children have and have not heard adultlike production and comprehension of language is not diagnostic of the underlying processes involved The main way to test for underlying process is to introduce children to novel linguistic items that they have never before heard tracer elements and then see what they do with them For questions of syntax in particular the method of choice is to introduce young children to a novel verb in one syntactic construction and then to see whether and in what ways they use that verb in other nonmodeled syntactic constructions perhaps with some form of discourse encouragement involving leading questions and the like As in all behavioral experiments care must be taken to control factors other than those of direct interest In the current instance special care must be taken that external performance factors such as the memory and processing demands of the experimental task do not adversely affect childrens linguistic performance Experiments using novel verbs as tracer elements have demonstrated that by 35 or 4 years of age most children can readily assimilate novel verbs to abstract syntactic categories and schemas that they bring to the experiment For example with special reference to the simple transitive construction Maratsos Gudeman GerardNgo and DeHart 1987 taught children from 45 to 55 years of age the novel verb fud for a novel transitive action human operating a machine that transformed the shape of playdough Children were introduced to the novel verb in a series of intransitive sentence frames such as The dough finally fudded It wont fud and The doughs fudding in the machine Children were then prompted with either neutral questions such as Whats happening or more biasing questions such as What are you doing which encourages a transitive response such as Im fudding the dough see also Ingham 1993 Pinker Lebeaux and Frost 1987 used a similar experimental design except that they introduced children to the novel verb in a passive construction The fork is being floosed by the pencil and then asked them the question What is the pencil doing to pull for an active transitive response such as Its floosing the fork In both of these studies the general finding was that the vast majority of children from 35 to 8 years of age 23 or more of the sample in most cases could produce a canonical transitive utterance with the novel verb even though they had never heard it used in that construction These results suggest that children of this age come to the experiment with some kind of abstract verbgeneral SVO transitive construction to which they readily assimilate the newly learned verb simply on the basis of observing the real world situation to which it refers and in some cases hints from the way adults ask them questions about this situation Over the past few years my collaborators and I have pursued a fairly systematic investigation of Englishspeaking childrens ability to produce simple transitive SVO sentences with verbs they have not heard used in this construction but focusing mainly on children below the ages represented in these previous studies The focus on younger children is important because most theories of the acquisition of syntactic competence single out the age range from 2 to 35 years as especially important and indeed by virtually all theoretical accounts children of 35 years and older should possess much syntactic competence Reviewing these studies with children beginning at 20 years thus provides an opportunity to look for some kind of developmental trajectory in childrens earliest syntactic competence with novel verbs Indeed to anticipate the outcome of the review there does seem to be a gradual increase in childrens ability to perform in more adultlike ways in novelverb experiments during this early age range However care must be taken in reviewing these experiments to discriminate between the possibility that children are acquiring their abstract syntactic knowledge only gradually and the alternative possibility that they have such knowledge all along but must still learn how to display that knowledge in the context of different kinds of performance demands in both experimental and naturalistic contexts We must therefore pay serious attention to the control procedures used in these studies First and most simple was a study by Tomasello Akhtar Dodson and Rekau 1997 We were interested in what children just learning to combine words would do with novel verbs and also as a kind of control procedure with nouns Fifteen children from 16 to 111 identified as word combiners were exposed to multiple adult models of two novel nouns and two novel verbs in minimal syntactic contexts for the noun Look The wug and for the verb Look Meeking or else Look what Ernies doing to Big Bird Its called meeking this second type of verb model was an attempt to ensure that the children saw the event as transitive and also that they had heard adults name the participants involved Children were exposed to these words multiple times each day over a tenday period with opportunities to produce the words available continuously on each day Virtually all children produced each word at least once as a single word utterance mostly multiple times with an average of over 20 times per word per child and virtually all children responded appropriately on tests of comprehension to all words as well There was thus no difference in childrens learning of the nouns and verbs as lexical items However there was a very large difference in the way the children combined their newly learned nouns and verbs with other words They combined the nouns quite freely averaging 145 word combinations per child with a number of fully transitive utterances such as I see wug I want my wug I pushing wug and Wug did it On the other hand the children hardly combined their newly learned verbs with other words at all averaging only about 05 word combinations per child and there was only one token from one child of a transitive utterance I meeking it In a pair of similar studies with slightly older children 111 to 23 Olguin and Tomasello 1993 Tomasello and Olguin 1993 found very similar results with children producing over 6 novel combinations per child with nouns but only one child produced a novel transitive utterance with his newly learned verb 7 tokens of Ime gorp Working with older children still Dodson and Tomasello 1998 used the same basic methodology with children from 25 to 31 and found that only 3 of 18 children between 25 and 30 produced a novel transitive utterance with their newly learned verb Three of 6 children over 30 did so however suggesting the possibility that 3 years of age is an important milestone for many children Akhtar and Tomasello 1997 wanted to explore specifically whether childrens conservative use of newly learned verbs was due to some kind of performance factors Perhaps the children used verbs conservatively because that is what they thought was expected of them in the experiment although it is unclear why they did not also think this for nouns and perhaps they would show more skills in tests of comprehension In the first study we simply replicated Olguin and Tomasello 1993 but with older children from 29 to 38 Consistent with that study only 2 of the 10 children produced a novel transitive utterance with an appropriately marked agent and patient ie using canonical English word order In a second study we then tried to eliminate the possibility that the children might not understand what was expected of them in this experimental context Children at 29 and 38 heard us say This is called pushing as we enacted a pushing event They were then asked Whats happening and were encouraged through various kinds of modeling and feedback to respond with SVO utterances of the type Ernies pushing Bert We then trained them in exactly the same manner with a novel verb and action This is called gopping and then asked them Whats happening In response to this training all of the 20 children independently produced at least one canonical SVO utterance such as Ernies gopping Bert most children producing several The logic of this study was thus that children were trained in what would be expected of them in the test and they did not proceed to the test phase unless they first demonstrated an understanding of the task and an ability to master its performance demands For the test children were then introduced to another novel verb paired with a novel action This is called meeking followed by the question Whats happening In this final test sequence the 38 children were quite good with 8 of 10 children producing at least one productive transitive utterance of the form Shes meeking the car However only 1 of the 10 younger 29 children produced a novel transitive utterance with the test verb In a third study Akhtar and Tomasello 1997 also ran two different comprehension tests which would seem to have fewer performance demands than tests of language production In the first the children who had just heard This is called dacking for many models were then asked to Make Cookie Monster dack Big Bird All 10 of the children 38 were excellent in this task 9 or 10 correct out of 10 trials whereas only 3 of the 10 children at 29 were above chance in this task even though most did well on a control task using familiar verbs Because using the verb as single word utterance is a somewhat odd way for children to be introduced to a verb a theme to which I return shortly a second comprehension test was also conducted with children at 29 The children first learned to act out a novel action on a novel apparatus with two toy characters and only then their first introduction to the novel verb did the adult hand them two new characters and request Can you make X meek Y while placing the apparatus in front of them In this case childrens only exposure to the novel verb was in a very natural transitive sentence frame used for an action they already knew how to perform Since every child knew the names of the novel characters and on every trial attempted to make one of them act on the other in the appropriate way the only question was which character should play which role These under3yearold children were as a group at chance in this task with only 3 of the 12 children performing above chance as individuals See Fischer 1996 for some positive results using a slightly different methodology for children averaging 36 years of age As alluded to above one concern about these experiments is that when children hear things like Dacking or This is called dacking they do not really understand that the novel word is a verb My collaborators and I chose this socalled presentational construction because we felt that the full sentences used by Maratsos et al 1987 Pinker et al 1987 posed a different problem for children that is when children hear a verb used in an intransitive construction such as The top is spinning unaccusative and then are encouraged to produce Bill is spinning the top the child has to change the syntactic role being played by the top from actor subject to patient object and with passives the two roles must be interchanged Nevertheless the naturalness of these language models is an important advantage as they demonstrate for the child one way the novel verb may behave as a verb My collaborators and I therefore conducted three additional novel verb experiments with young children using these more natural models one with a passive model as in Pinker et al 1987 one with an intransitive model as in Maratsos et al 1987 and one with an imperative model We also followed the procedure of these previous studies in putting children under discourse pressure by asking them leading questions that encouraged particular types of responses Another issue involving performance demands is as follows Although in the earlier studies we compared childrens use of novel verbs to novel nouns and used other control procedures perhaps a more appropriate control is to teach children two novel verbs one in a transitive construction and one in some other construction then put them under discourse pressure to produce a transitive utterance with each of their newly learned verbs A transitive utterance would thus not be productive for the children who learned the new verb in a transitive construction but it would be productive for the children who learned the new verb in some other construction The three studies we have performed with passive intransitive and imperative models all had this kind of control condition Brooks and Tomasello 1999 exposed 20 children average age 210 to one novel verb in the context of a passive model such as Ernie is getting meeked by the dog and another novel verb in the context of an active transitive model such as The cat is gorping Bert each for a highly transitive and novel action in which an agent did something to a patient We then asked them agent questions of the type What is the AGENT doing This agent question pulls for an active transitive utterance such as Hes meeking Ernie or Hes gorping Bert which would be novel for meek since it was heard only as a passive but not novel for gorp since it was heard only as an active transitive Overall in two studies fully 93 of the children in the control condition who heard exclusively transitive models with the novel verb were able to use that verb in a transitive utterance On the other hand only 28 of the children at 210 who heard exclusively passive models with the novel verb were able to use that verb in an active transitive utterance Tomasello and Brooks 1998 exposed 16 children at 20 and 16 children at 26 to one novel verb in the context of an intransitive model such as The ball is dacking and another novel verb in the context of a transitive model such as Jim is tamming the car each for a highly transitive and novel action in which an agent did something to a patient We then asked them agent questions of the type Whats the AGENT doing Again this question pulls for a transitive utterance such as Hes dacking the ball or Hes tamming the car which would be novel for dack since it was heard only as an intransitive but not novel for tam since it was heard only as a transitive With the transitively introduced verb in the control condition 11 of the 16 younger children and all 16 of the older children produced a novel transitive utterance However with the intransitively introduced verb only one of 16 children at 20 and only 3 of 16 children at 26 produced a novel transitive utterance Lewis and Tomasello in preparation exposed 18 children at 20 26 and 30 to one novel verb in the context of both transitive and intransitive imperative models such as Dop the lion and Dop Lion and another novel verb in the context of both transitive and intransitive indicative models such as Jim is pilking the lion and The lion is pilking each for a highly transitive and novel action in which an agent is doing something to a patient We then asked them neutral questions of the type Whats happening With the indicatively introduced verb in the control condition 11 children at 20 11 children at 26 and 16 children at 30 produced either a transitive or intransitive utterance with subject as modeled However with the imperatively introduced verb never heard with a subject only 1 of 18 children at 20 2 of 18 children at 26 and 6 of 18 children at 30 produced a transitive utterance with a subject3 To my knowledge there are only two studies of children learning languages other than English that employ the novel verb experimental paradigm First Berman 1993 investigated young Hebrewspeaking childrens ability to use an intransitively introduced novel verb in a canonical transitive construction requiring them to creatively construct a special verb form a type of causative marker on the formerly intransitive verb as well as a special arrangement of the other lexical items involved Berman showed children of 29 39 and 80 oneparticipant pictures eg a ball rolling and described it with a novel verb in canonical intransitive form and then showed them a picture in which one participant acted on another eg a boy rolling the ball She then used a sentence completion task she started the sentence for them as in The boy in the hopes of eliciting novel transitive utterances The findings were as in the English reported above a steady increase in novel transitive utterances over age in this case from 9 at 29 to 38 at 39 to 69 at 80 a bit lower level of performance than Englishspeaking children of the same ages perhaps because the Hebrew children had to both change the verb morphologically and rearrange the order of some sentence elements Second Childers and Tomasello 1999 conducted a study in Chilean Spanish which designates subjects by means of special endings on verbs in the typical Romance paradigm with lexical subjects optional Children heard a number of utterances with one and only one form of a nonce Spanish verb either third person singular eg Mega or third person plural eg Megan and were then encouraged to produce the other form Results were that 4 of the 16 children at 26 and 6 of 16 children at 30 were able to produce the form they had not heard Despite the very different linguistic structures involved in this case all that was needed was a simple change of verb morphology with nothing to be added and no reordering of elements needed the Spanishspeaking children still had much trouble creatively producing the means for designating whodidwhattowhom with the novel verb All of these studies involve children producing or failing to produce canonical utterances that go beyond what they have heard from adults Their general failure to do so at early ages suggests that they do not possess the abstract structures that would enable this generativity However there is one recent study that may be of special importance because it succeeded in inducing children to follow adult models that were noncanonical English and so the children produced utterances that 3 The study of Naigles 1990 is sometimes taken to be discrepant with these findings Her study employed a preferential looking paradigm in which children simply had to look at the video scene that matched the adult language ie longer than at a mismatching picture However the two sentences that were compared in that study were The duck is glorping the bunny and The bunny and the duck are glorping with one picture depicting the duck doing something to the bunny and the other depicting the two participants engaged in the same parallel action The problem is that children might very well have been using the word and as an indicator of the parallel action picture Olguin Tomasello 1993 Pinker 1994 The similar study by Naigles et al 1993 using an actout task has a number of methodological problems see Akhtar Tomasello 1997 p 964 involved a different configuration of SVO than was typical in almost all of the speech they had previously heard or produced Akhtar 1999 modeled novel verbs for novel events with young children at 28 36 and 44 years old One verb was modeled in canonical English SVO order as in Ernie meeking the car whereas two others were in noncanonical orders either SOV Ernie the cow tamming or VSO Gopping Ernie the tree Children were then encouraged to use the novel verbs with neutral questions such as Whats happening Almost all of the children at all three ages produced exclusively SVO utterances with the novel verb when that is what they heard However when they heard one of the noncanonical SOV or VSO forms children behaved differently at different ages Only 1 of the 12 children at 28 and 4 of the 12 children at 36 consistently corrected the noncanonical adult word order patterns to a canonical English SVO pattern whereas 8 of 12 children at 44 did so Interestingly many of the younger children vacillated between imitation of the odd forms and correction of the odd forms to canonical SVO order indicating perhaps that they knew enough about English word order patterns to discern that these were strange utterances but not enough to overcome completely their tendency to imitatively learn and reproduce the basic structure of what the adult was saying A reasonable expectation is that if younger children were run in this experiment at 20 to 26 for example they would follow the adult models almost exclusively with little vacillation because they know even less about English SVO ordering than Ahktars youngest children 23 The developmental trajectory From these naturalistic and experimental studies it is clear that young children are productive with their early language in only limited ways Although there are data on a variety of structures in a variety of languages the results are strongest for the moststudied structure the English transitive construction Before 3 years of age only a few Englishspeaking children manage to produce canonical transitive utterances with verbs they have not heard used in this way We see this pattern when we look at their naturalistic utterances carefully and systematically including the various ways in which particular verbs are and are not used and we also see this same pattern when we look at their performance in a fairly diverse set of experimental paradigms in which they must a get to the transitive utterance from a variety of different constructions presentational intransitive passive imperative noncanonical and b they must do this in a variety of different tasks in a variety of types of discourse interactions with adults Explanations in terms of child production deficits and other syntactically extraneous factors are not a likely explanation for these experimental findings because of all of the control procedures used more extensive discussion below The general finding for the large majority of children under 3 years of age is thus always the same no matter the method they use some of their verbs in the transitive construction namely the ones they have heard used in that construction but they do not use other of their verbs in the transitive construction namely the ones they have not heard in that construction Many children 3 years of age and older however do show evidence that they posses an abstract transitive construction to which they can freely assimilate newly learned verbs and indeed a few children show such evidence at even younger ages Thus when all of the findings just reviewed are compiled and quantitatively compared we see a very gradual and continuous developmental progression see Fig 1 see Table 1 for key Fig 1 was constructed by computing a single number for the productivity of children at each age group in each of the experimental studies reported above see key to Fig 1 In the vast majority of cases including all of the studies by Tomasello and colleagues this number was simply the proportion of children who produced at least one novel and canonical transitive utterance despite the number of imitative utterances they produced which in some cases was quite high For a few studies this proportion could not be determined from published reports and so the overall proportion of childrens utterances that were productive was used instead mostly this was for older children and involved very high proportions in which case the two different ways of estimating productivity should correlate highly Even though each study has some unique qualities of experimental design and procedure many of which were detailed above nevertheless virtually all of the studies fall on a curve that slopes steadily upward from age 2 to 4 at which point the slope flattens a bit but still reaches 100 by 8 years of age But this developmental picture of the evergrowing abstractness of the transitive construction is obviously not the whole picture Children cannot just generalize all syntactic constructions to all verbs at will at some point they must constrain the generalization process so as to conform with adult usage I cannot give this difficult and important question all of the attention it deserves here but a brief look at some recent findings will perhaps be useful in the current context since these findings suggest once again that there is a gradual developmental process of constraint in which children are once again strongly influenced by the language they hear around them Pinker 1989 proposed that there are certain very specific and mostly semantic Table 1 Studies used How they are designated in figure Age of children What percentage of children or responses were productive The type of linguistic model used The type of elicitation question used Some notes on how the productivity score was calculated Study No in Fig 1 Age Productivity Linguistic model Eliciting question Scoring Tomas et al 1997 1 110 007 Presentational Neutral children Tomas and Brooks 1999 2 20 006 Intransitive Agent children Lewis and Tomas in prep 3 26 009 Imperative Neutral children 4 26 006 Intransitive Neutral children 5 30 013 Intransitive Neutral children 6 30 038 Intransitive Neutral children Olguin and Tomas 1993 7 21 013 Presentational Neutral children Dodson and Tomas 1998 8 210 025 Presentational Neutral children Brooks and Tomas 1999 9 210 020 Passive Agent children Studies 1 and 2 Akhtar and Tomas 1997 10 315 055 Passive Agent responses Ingham 1993 11 210 035 Presentational Neutral children 12 31 025 Presentational Neutral children 13 29 035 Intransitive Agent responses 14 38 010 Intransitive low freq English verbs Agent responses Img 080 Passive Agent responses P1 46 067 Passive Agent responses P2 310 038 Passive Agent responses P3 51 088 Passive Agent responses P4 61 088 Passive Agent responses action verbs Table 1 continued Study No in Fig 1 Age Productivity Linguistic model Eliciting question Scoring Maratsos et al 1987 P5 711 100 Intransitive Agent children 3 of 10 in 07 group M 50 075 SOV VSO Neutral children Akhtar 1999 A1 28 008 Intransitive HEBREW Sentence completion consistently correct responses fully correct A2 36 033 A3 44 067 A1 29 009 1st or 3rd Person Verb SPANISH Neutral children HI 38 034 H2 89 069 H3 30 025 S1 26 038 S2 30 constraints that apply to particular English constructions and to the verbs that may or may not be conventionally used in them For example a verb can be used felicitously with the English transitive construction if it denotes manner of locomotion eg walk and drive as in I walked the dog at midnight or I drove my car to New York but not if it denotes a motion in a lexically specified direction eg come and fall as in He came her to school or She falled him down How children learn these verb classes and they must learn them since they differ across languages is unknown at this time Two other factors involved in syntactic constraint have also been widely discussed entrenchment and preemption see Bates MacWhinney 1989 Braine Brooks 1995 Clark 1987 Goldberg 1995 First the more frequently children hear a verb used in a particular construction the more firmly its usage is entrenched the less likely they will be to extend that verb to any novel construction with which they have not heard it used Second if children hear a verb used in a linguistic construction that serves the same communicative function as some possible generalization they may infer that the generalization is not conventional the heard construction preempts the generalization For example if a child hears He made the rabbit disappear when she might have expected He disappeared the rabbit she may infer that disappear does not occur in a simple transitive construction since the adult seems to be going to some lengths to avoid using it in this way the periphrastic causative being a more marked construction In many cases of course both entrenchment and preemption may work together as a verb that is highly entrenched in one usage is not used in some other linguistic context but an alternative is used instead Two recent studies provide evidence that indeed all three of these constraining processes are at work that is entrenchment preemption and knowledge of semantic subclasses of verbs First Brooks Tomasello Lewis and Dodson in press modeled the use of a number of fixedtransitivity English verbs for children from 35 to 80 years verbs such as disappear that are exclusively intransitive and verbs such as hit that are exclusively transitive There were four pairs of verbs one member of each pair typically learned early by children and typically used often by adults and so presumably more entrenched and one member of each pair typically learned later by children and typically used less frequently by adults less entrenched The four pairs were comearrive takeremove hitstrike disappearvanish the first member of each pair being more entrenched The finding was that in the face of adult questions attempting to induce them to overgeneralize children of all ages were less likely to overgeneralize the strongly entrenched verbs than the weakly entrenched verbs that is they were more likely to produce I arrived it than I comed it 4 Second Brooks and Tomasello in press taught novel verbs to children 25 45 and 70 years of age They then attempted to induce children to generalize these 4 Bowerman 1988 1997 reports that her two daughters produced many overgeneralizations for some early verbs that should be highly entrenched such as go and come However precisely because these verbs are so frequent in childrens speech they have many opportunities to overgeneralize them it is thus difficult to know if these verbs are overgeneralized more often than other verbs on a proportional basis novel verbs to new constructions Some of these verbs conformed to Pinker 1989 semantic criteria and some did not Additionally in some cases experimenters attempted to preempt generalizations by providing children with alternative ways of using the new verb thus providing them with the possibility of answering Whats the boy doing with Hes making the ball tam which allows the verb to stay intransitive In brief the study found that both of these constraining factors worked but only from age 46 Children from 46 showed a tendency to generalize or not generalize a verb in line with its membership in one of the key semantic subclasses and they were less likely to generalize a verb to a novel construction if the adult provided them with a preempting alternative construction The details of these studies are not important for current purposes What is important is that these constraining influences on syntactic constructions emerge only gradually Entrenchment works early from 30 or before as particular verb island constructions become either more or less entrenched depending on usage Preemption and semantic subclasses begin to work sometime later perhaps not until 46 or later as children learn more about the conventional uses of verbs and about all of the alternative linguistic constructions at their disposal in different communicative circumstances Thus just as verbargument constructions become more abstract only gradually so also are they constrained only gradually Combining these findings on constraints with the findings depicted in Fig 1 we may create a developmental trajectory that includes both the growing abstractness of childrens constructions and also the factors that conspire to constrain the resulting generalization processes preventing children from using all constructions with all verbs The process may be illustrated as in Fig 2 with three verbs very similar in meaning laugh giggle and chortle this example being inspired by Bowermans childs famous overgeneralization at 30 Dont giggle me My hypothesis as illustrated graphically in Fig 2 is that laugh is not likely to be overgeneralized to the transitive construction because it is learned early and entrenched through frequent use as an intransitive verb only Chortle is also not likely to be overgeneralized but for a different reason even though it is not highly entrenched it is typically learned only after the child has begun to form verb subclasses and chortle belongs to one that cannot be used in the transitive construction and only after the child has also learned preempting alternative constructions such as That made me chortle with glee preserving its intransitive status Giggle is more likely to be overgeneralized because it is not so entrenched as laugh and it is learned before the child has formed verb subclasses or learned many alternative constructions that might preempt an overgeneralization that is it may be a verb that is learned in a high vulnerability window of developmental time Currently this model of how argument structure constructions and their associated verbs are constrained developmentally is speculative based on only two experimental studies but it is at least fairly explicit in the factors posited as causal and the ages at which they operate And of course at this point it is confined to the simple transitive construction in English Nevertheless although very little research has explored experimentally childrens productive use of constructions other than the English transitive construction there is some evidence from both naturalistic analyses and the experimental studies suggesting that the different verbargument constructions develop and are constrained in the same general manner as the transitive construction although each very likely has its own developmental timetable As just a hint at this variability among constructions Fig 3 plots childrens productive uses of intransitive utterances from Tomasello Brooks 1998 productive uses of imperative utterances from Lewis Tomasello in preparation and productive uses of passive utterances from Brooks Tomasello 1999 Pinker et al 1987 3 Implications for the continuity assumption The continuity assumption is arguably the core assumption of generative Chomskian approaches to language acquisition because only it permits the use of adultlike grammars to describe childrens early language The obvious problem made especially salient in the above review is that childrens early language looks very little like adult language To explain this discrepancy between childrens hypothesized adultlike competence and their actual childlike performance additional theoretical machinery is required There have been various proposals for such machinery over the past few decades mostly involving external factors that might conceal the childs true syntactic competence But the current data pose some much more serious and specific problems for the continuity assumption To illustrate the point I take the three major classes of generative acquisition theories as described by Clahsen 1996 and show in each case how the current data especially the experimental data present insurmountable problems I then show that all three of these approaches have also been seriously deficient in facing the very difficult problem of how children might link their preexisting universal grammars hypothesized by all generative approaches to particular pieces of particular natural languages the only serious attempt at solving this linking problem currently having no empirical support 31 Full competence plus external developments The first generative theory is a fairly straightforward application of Chomskys original competenceperformance distinction see Chomsky 1986 for an especially clear statement In Clahsens 1996 p xix formulation The first approach claims that young children when they begin to produce sentences already have full grammatical competence of the particular language they are exposed to and that differences between sentences children produce and adults sentences should be attributed to external factors ie to developments in domains other than grammatical competence The bestknown examples of external factors are memory and processing limitations eg Valian 1991 and pragmatic limitations Weissenborn 1992 The main problem in this case is that there have never been any serious attempts to actually measure and assess childrens performance limitations and so they are simply invoked whenever they are convenient There have been strenuous objections to this practice from generativists eg Roeper 1996 p 417 and nongenerativists eg Sampson 1997 alike But in addition from a more empirical point of view I would argue that in the experiments on the English transitive construction reviewed above a number of control procedures ruled out for all practical purposes performance limitations as a viable explanation for childrens lack of productivity with newly learned verbs Specifically the same children who failed to use newly learned verbs in transitive utterances were highly productive with novel nouns which rules out the possibility that children are simply reluctant to use newly learned words in novel ways in this experimental context performed conservatively when tested for their comprehension of novel transitive utterances which rules out many production factors since comprehension tasks pose fewer or at least different performance demands than production tasks produced transitive utterances both in their spontaneous speech and with novel verbs in the experiment if they had first heard an adult use those novel verbs in transitive utterances which rules out many other performance factors having to do with the difficulties of using newly learned verbs in transitive utterances It is not that children possess fully adultlike performance capabilities They clearly do not and any serious theory of language acquisition has to deal with childrens growing skills of linguistic performance that develop in tandem with their growing linguistic competence However in the experimental studies reviewed above all of the evidence suggests that the children were working within their performance limitations And they still showed no signs of possessing the kinds of abstract adultlike syntactic competence attributed to them by believers in the continuity assumption 32 Full competence plus maturation The second approach shares much with the first because despite the name the maturation occurs not in universal grammar itself but in aspects of linguistic competence considered peripheral to universal grammar In Clahsens 1996 p xix description The second approach assumes that UG principles and most of the grammatical categories are operative when the child starts to produce sentences Differences between the sentences of young children and those of adults are explained in terms of maturation The claim is made that there are UGexternal learning constraints which restrict the availability of grammatical categories to the child up to a certain stage and then are successively lost due to maturation Consider for example Wexler 1994 who argued that the feature TENSE matures at around the age of 25 and Rizzi 1993 who suggested that the constraint which requires all root clauses to be headed by CP in adult language is not yet operative in young children but that it matures at the age of approximately 25 The details of this theory are not important for current purposes not even the issue of what is considered internal and external to universal grammar The critical points are the same as for the previous theory even if it is posited that some aspects of universal grammar itself mature see Chomsky 1986 First like performance limitations maturation is basically an unconstrained fudge factor since any time new acquisition data arise it may be invoked without any consultation of genetic research or any independent assessment of this causal factor at all Braine 1994 Second and more empirically I would argue that in the experimental data reviewed above all possible factors that might be subject to maturation both internal and external to universal grammar were the same in the experimental and control conditions since everything in both conditions was focused on one and only one syntactic construction That is to say children who used the simple transitive construction with numerous verbs in their spontaneous speech and with novel verbs which they had heard in the transitive construction in the control condition would be presumed to have in place the required geneticmaturational bases for producing transitive utterances But then it is a total mystery why they did not use these same genetic bases to produce transitive utterances with novel verbs in the experiment given that performance limitations have essentially been ruled out see above 33 Lexicalism The third generative approach is a bit different and raises a new set of theoretical issues Again in Clahsens 1996 xx words The third approach shares with the two other views the assumption that all UG principles are available to the child from the onset of acquisition However the grammar of the particular language the child is acquiring is claimed to develop gradually through the interaction of available abstract knowledge eg about Xbar principles and the childs learning of the lexicon This view does not violate the continuity assumption The claim is thus that children must have a certain amount of linguistic experience with their own particular language before they can access certain aspects of their universal grammar eg Clahsen Eisenbeiss Penke 1996 Radford 1990 that is to say the process by which a particular language triggers different aspects of universal grammar is a bit more complicated than originally thought Because it makes reference to the particularities of particular languages and lexical items this approach at least holds out the promise of being able to account for the data reported above5 The main problem is that to account for the experimental data reported above this approach would have to claim that to assimilate a newly learned verb to those aspects of universal grammar involved in the transitive construction involving headdirection etc children must hear each specific verb used in that specific construction Generalized this would mean that to begin to participate in a productive system of generative grammar the child must hear each of her lexical items in each of its appropriate syntactic contexts see Hyams 1994 for a proposal very near 5 to this This theory can explain the data of course but it seems to me at the cost of the whole point of a generative account which classically posits that human beings possess and use linguistic abstractions early in ontogeny and independent of specific linguistic experiences other than a minimal triggering event If children have to hear each verb in each of its licensed syntactic constructions then the generativist account will be empirically indistinguishable from many usagebased accounts see below Perhaps even more seriously none of the proponents of this approach has attempted to work out precisely how the child goes about linking up itemspecific linguistic knowledge with universal grammar Atkinson 1996 pp 47374 in particular criticizes proponents of this view for not explicating how the linking process might take place But linking is a problem not just of this specific approach but rather of the generative paradigm as a whole 34 The problem of linking Pinker 1984 1987 1989 identified and explored the key problem for generative approaches to language acquisition The problem is how children link their universal grammar in whatever form that may exist to the particular language they are learning For example let us suppose that children are born with an innate idea of subject of a sentence or any other abstract linguistic entity How do they go about identifying this entity in the language they hear given that across languages sentence subjects seemingly do not share any distinctive perceptual features ignoring for current purposes the more difficult problem that many languages may not even have sentence subjects Foley Van Valin 1984 How does a child or an adult who hears an utterance in Turkish or Walpiri or Tagalog or English go about identifying a subject when not only are there no phonemes that consistently accompany subjects across languages there are also no other consistent features such as word order or case marking or coindexing on the verb that are the same across languages Dryer 1997 If we look for guidance to other behavioral systems in the biological world that have a strong genetic component we see the problem even more clearly Imprinting is one such system Thus some baby ducks are born with a builtin system for identifying and staying close to their mothers for obvious biological reasons But how does the duckling identify its mother in the first place Nature has built in a search image of mother constituted by the specific perceptual features that identify a mother in terms of what it is to look like how it is to move and what kind of noises it is to make But subject of a sentence cannot be specified in this way because children growing up in different cultures experience sentence subjects that are perceptually very different with basically no overlapping perceptual features So perhaps a better biological analogy is cognitive mapping and spatial cognition which involves more abstract conceptual entities Thus an individual mammal cannot be born knowing where in its local environment water and food and predators may be located but many mammals are born with the skill to create a cognitive map of their local environment on the basis of experience in that environment This would seem to be a more appropriate analogy to language acquisition since it involves an animal being pretuned to learn very quickly from local experiences with the local experiences only triggering the building of the cognitive map whose structure is not affected in any fundamental way by the specifics of the specific environment But even in the case of spatial cognition there are still experiential invariants in the animals visual world that serve to trigger the building up of the cognitive map in specific ways and these are the same in all different local environments such things as the distances angular relationships topological relationships and other Gibsoniantype higher order perceptual invariants of which all local spatial environments consist And so the question of whether language acquisition is like building a cognitive map reduces to the question of whether such things as subject of a sentence have some invariant perceptual or experiential features across languages Pinker 1984 1987 1989 recognized this fact and so he proposed the following a a list of key syntactic categories innately given to all human beings b a list of key experiential categories innately given to all human beings and c a set of innate linking rules to connect the two In the case of subject of a sentence the link was first to agent of an action or if there was no agent to such things as theme or goal the socalled linking hierarchy Thus if the child saw a dog bite a man and heard someone say The dog bit the man she would know on the basis of her general causal cognition that the dog is the agent of the action her innate linking rule would then connect agent to subject6 However in the specific case of sentence subject it is almost certain that Pinkers proposal is not correct First of all on general theoretical grounds it has been known for some time that in ergative languages the notion of subject does not operate like it does in English and other accusative languages and so a direct connection to agent is not possible Moreover even if there were some solution to this problem there are many languages that are socalled split ergative some of its constructions are ergative while others are accusative based on such things as person first and second person are accusatively structured whereas third person is ergatively structured or tense presentfuture is accusatively structured whereas past is ergatively structured DeLancey 1981 Van Valin 1992 In general terms Slobin 1997 has made a persuasive case that there is much too much variability across languages not to mention historical change within languages for any static and innate lookup table to function in the way it would need to solve the problem of linking see also Braine 1992 A second more empirical problem with Pinkers proposal is that at least two naturalistic analyses of early child language have failed to find any evidence for innate linking rules First Lieven et al 1997 analyzed the first sentences of 12 6 Because she notices the linguistic form associated with the subject the child can also now recognize future exemplars of sentence subject on the basis of this form alone say a particular word order configuration or a particular case marker even if they are not agents thus the Englishspeaking child will eventually have to deal with experiential subjects that are not agents as in John saw Mary and even passive sentences in which subjects are not agents and agents are not subjects Englishspeaking children and found that many early subjects came from such unremarkable utterances such as I like it Maria have it I see it and It has a hole in which there is no agent of an action at all see also Pye Loeb Redmond Richardson 1994 More strongly still Bowerman 1990 1997 found that it happens with some regularity in early child language that the subject hierarchy is violated totally that is arguments that are further down the linking hierarchy end up as subjects as in the utterance Pete hurt by car patient subject agent oblique reported by Tomasello 1992 for a child at 18 And so not only do innate linking rules run into difficulties crosslinguistically they also make wrong predictions for the order of acquisition of some structures within a language To my knowledge no generative theorist other than Pinker has proposed a systematic theory of how children might solve the linking problem One might suppose that positing parametric variation in different languages along with an acquisition mechanism involving parameter setting might help in solving the linking problem but it does not Parameter setting in fact depends on linking Thus Mazuka 1995 provided a detailed analysis of how children might set the hypothesized head direction parameter to either head first as in the Spanish la casa grande or head last as in the English the big house What she demonstrated was that parameter setting rests fundamentally on linking processes that is to set the head direction parameter in universal grammar a language learner must first be able to recognize clausal heads in the specific language she is learning Once this fundamental linking problem is accomplished the parameter setting is trivial indeed superfluous Setting a Head Direction parameter by analyzing the syntactic structure of the input involves a paradox The Head Direction parameter is supposed to determine the order in which the head and complement should appear in the language the child is acquiring But for a child to set this parameter she must first find out which units are the heads and the complements in the sentence she hears If her linguistic skills are sophisticated enough to know which are heads and complements she will also know which order they came in If she already knows which order the head and the complements come in a sentence there is no need to set the parameter Mazuka 1995 pp 2425 The hard part is thus recognizing heads and complements in a particular language and this difficulty is logically prior to any act of parameter setting 35 Summary My assessment of the continuity assumption is thus clear Neither of the two generative approaches invoking hypothesized but unmeasured factors that prevent 6 Because she notices the linguistic form associated with the subject the child can also now recognize future exemplars of sentence subject on the basis of this form alone say a particular word order configuration or a particular case marker even if they are not agents thus the Englishspeaking child will eventually have to deal with experiential subjects that are not agents as in John saw Mary and even passive sentences in which subjects are not agents and agents are not subjects children from displaying their adultlike syntactic competence either external performance factors or genetic maturation can explain the data presented above particularly the experimental data Lexicalist approaches may be able to explain the data but to do so they must invoke essentially the same kinds of local learning processes that generative approaches were designed to replace And then they are still stuck with the linking problem there is basically no answer to the question of how the language learning child might link up the linguistic items and structures she is learning locally with the hypothesized innate universal grammar 4 A usagebased account The continuity assumption thus has no empirical support and at least one very serious theoretical problem However most of the classic arguments in favor of the continuity assumption have actually been negative that is they are arguments against the possibility of a learningbased explanation since by hypothesis mature linguistic competence is so abstract and formal that children could not possibly learn or construct it Gleitman Wanner 1982 This argument may be summarized as You cant get there from here But many language acquisition theorists reject this argument eg Bates Goodman 1997 Bloom 1992 Budwig 1995 Lieven 1997 MacWhinney 1999 Slobin 1985 1997 Tomasello Brooks 1999 The basic issues are two First the adult endpoint of linguistic development does not have to be characterized in the abstract terms of a Chomskian universal grammar There is currently a new class of linguistic theories falling under the general rubric of Functional and Cognitive Linguistics that conceptualize adult linguistic competence in some new and more childaccessible ways Second there are also some new ways of thinking about how children learn and construct abstract cognitive entities Generativists typically make their impossibility arguments against outdated learning concepts from the 1950s such as simple association and blind induction But there are new ideas about cognitive development in the domain of language that go beyond these simplistic notions especially with respect to childrens very powerful skills of i intentionreading and cultural learning ii analogy making and iii structure combining My attempt here is to describe only very briefly and in general outline both the new way of looking at adult language and the three cognitive skills that help children to attain mature linguistic competence 41 The adult endpoint Generative grammar accounts of human linguistic competence are aimed at mathematical elegance Thus when an advance in the formalism is made as in the new minimalism Chomsky 1993 it is automatically assumed to characterized universal grammar with no empirical verification deemed necessary In using such a formalism to describe either adult or child language the attempt is not to account for all of human linguistic competence but only to explain core grammar the most abstract and systematic aspects of language use with lexical items idioms and quirky syntactic constructions all being consigned to the periphery the lexicon pragmatics etc The distinction between core and periphery forms the basis for several recent theoretical proposals to the effect that acquiring a natural language requires two distinct processes a abstract and a priori rules for the linguistic core and b normal processes of learning and memory for the linguistic periphery eg Chomsky 1980 Clahsen 1999 Pinker 1991 But in recent years a new paradigm in theoretical linguistics has arisen that attempts to determine the nature of human linguistic competence from a more psychological and less mathematical point of view It is called CognitiveFunctional Linguistics eg Bybee 1985 Croft 1991 Fillmore 1985 Givón 1995 Goldberg 1995 Lakoff 1987 Langacker 1987 1991 Talmy 1988 van Valin 1991 see papers in Tomasello 1998a In this view competence with a natural language consists of nothing more or less than the mastery of its various linguistic symbols and constructional schemas each of which consists of one or more linguistic forms signifier each with a communicative function signified Cognitivefunctional approaches attempt to explain all aspects of human linguistic competence from the highly canonical core to the highly idiosyncratic periphery Kay Fillmore 1999 Thus fluent speakers of English control both abstract morphological and sentencelevel constructions eg the regular past tense and the ditransitive sentence schema as well as very many concrete expressions based on individual words or phrases including everything from ritualized greetings Howyadoing to idioms Nothing ventured nothing gained to metaphors Shes dressed to the nines to noncanonical phrasal collocations I wouldnt put it past him Hes getting to me these days Hang in there That wont go down well with the boss She put me up to it etc see Benson Benson and Ilson 1997 for a dictionary of many of the tens of thousands of idiosyncratic English collocations see also Pawley and Syder 1983 Jackendoff 1996 In this view competence with linguistic symbols and constructional schemas are very general cognitive abilities that manifest themselves in many domains of human activity although they may take on some special characteristics in the domain of linguistic communication Because there is no mathematically elegant universal grammar guiding the process of acquisition there is no linking problem It is theoretically significant that the abstractness of a construction as evidenced by its productivity does not automatically mean that it is in the core from a generative grammar point of view Thus consider the incredulity construction Him be a doctor My mother ride a motorcycle Them come to the party This construction is highly abstract in the sense that it is not dependent on any particular word or phrase and it is highly productive in the sense that any fluent speaker of English can generate innumerable further exemplars But at the same time it is very odd from the point of view of the majority of English sentencelevel constructions because the subject is in the accusative case Him Them and the verb is nonfinite My mother ride without the agreement marker Another example of an abstract and productive yet idiosyncratic construction is the nominal extraposition construction Michaelis Lambrecht 1996 as in Its amazing the people you see here Its staggering the number of books that can pile up Its ridiculous how long it takes Constructions like these are important because they represent in essence existence proofs that human beings can master highly abstract and productive constructions that do not behave like any or many other constructions in the language In all linguistic theories including generative theories constructions such as these must be learned on their own Interestingly natural languages also contain some mixed constructions that is constructions that are in some ways abstract but that revolve around particular lexical items For example I wouldnt live in Boston let alone in New York She wont ride the stationary bike let alone lift weights She wont talk to let alone go out on a date with that swine This particular construction which has a very distinct communicative function involving some interesting pragmatic implicatures about the speakers attitude towards the entities involved is defined by a particular lexical item let alone even though it is otherwise fairly open to the entities that may be compared Fillmore Kaye OConnor 1988 This construction is thus distinctly reminiscent of the verb island constructions of twoyearold children There are of course highly canonical aspects of mature linguistic competence in the sense that many linguistic constructions are organized into inheritance hierarchies For example the English intransitive ditransitive and causative constructions are all instances of an even more abstract SubjectPredicate construction in English and so all three share its main characteristics Goldberg 1995 But the linguistic core is not a discrete entity indeed it is notoriously difficult to decide whether certain constructions eg mixed constructions are a part of core competence and constructions may differ from the core in many and diverse ways A plausible way to think of mature linguistic competence then is as a structured inventory of constructions some of which are similar to many others and so reside in a more corelike center and others of which connect to very few other constructions and in different ways and so reside more towards the periphery The proposal would thus be that the child initially learns individual itembased linguistic constructions eg verb island constructions and if there are patterns to be discerned among these different itembased constructions in adult usage she could then make abstractions and create inheritance hierarchies of constructions In this view of language and its acquisition therefore there is continuity not of structures adults control a more diverse and abstract set of constructions than do children but there is continuity of process in the sense that the processes of learning and abstraction are the same wherever and whenever they are applicable see below This general approach is usagebased in the sense that all linguistic knowledge however abstract it may ultimately become derives in the first instance from the comprehension and production of specific utterances on specific occasions of use With this redefinition of the endpoint of language acquisition in terms of linguistic constructions of varying degrees of complexity abstraction and systematicity it is now much easier to see how children might get from here to there especially if we also take into account some recent proposals about childrens surprising skills at i learning culturally ii making analogies and iii combining structures in their acquisition of language7 42 Intention reading and cultural learning There are over 5000 natural languages in the world each with its own inventory of symbols and constructions which change and develop over historical time and so what human children must be biologically prepared for most urgently is variation they must be prepared to acquire whichever set of linguistic symbols and constructions they encounter Levinson 1999 This means that a large part of the task of language acquisition must be accomplished by means of some form of social or imitative learning Classically imitation has been thought to play only a marginal role in language acquisition This is because imitation has been conceptualized as the child repeating verbatim something the adult has just said with little or no understanding This is a form of social learning that in another context I have called mimicking Tomasello 1996 and indeed it very likely plays only a minor role in language acquisition But in a more recent theoretical approach to social learning I have attempted to identify a subset of social learning processes called cultural learning one type of which is imitative learning Tomasello Kruger Ratner 1993 In the current context the key idea is this In cultural imitative learning as opposed to simple mimicking the learner understands the purpose or function of the behavior she is reproducing Thus Meltzoff 1995 found that 18monthold infants attempted to reproduce the intentional action they saw an adult attempting to perform even if that action was not carried through to completion and Carpenter Akhtar and Tomasello 1998a found that 16monthold infants attempted to reproduce an adults intentional goaldirected actions but not her accidental actions With regard to language in particular the child has to understand a special class of intentions known as communicative intentions Thus a child might hear her father exclaim Look A clown To fully understand his linguistic behavior with an eye toward reproducing it she must understand that her father intends that she share attention with him to a particular object that is to say understanding a communicative intention means understanding precisely how another person intends to manipulate your attention Tomasello 1998c in press a It is only by understanding the communicative intention behind these funny noises 7 It is important to note that many of the phenomena on which generative linguists have focused their attention and which are claimed to be unexplainable in other frameworks are currently being described and explained in alternative ways by CognitiveFunctional linguists As three concrete examples 1 the binding principles van Hoek 1997 2 the subjacency constraint van Valin 1991 1998 and 3 grammatical relations Langacker 1998 that the child can learn how to use a particular linguistic expression appropriately when she has the same communicative intention towards someones attention When we move beyond word learning to syntactic constructions the process becomes a bit more complicated but it is still essentially the same To comprehend the totality of an adults utterance the child must understand his overall communicative intention and also the communicative role being played by the various constituents of the utterance As a nonlinguistic example a child may see an adult use a stapler and understand that his goal is to staple together two pieces of paper In some cases the child may understand also that the goalfunction of placing the papers inside the staplers jaws is to align them with the stapling mechanism inside the stapler and that the goalfunction of pressing down on the stapler is to eject the staple through the two papers with both of these subactions being in the service of the overall goalfunction of attaching the two sheets of paper Of course even in the most optimistic interpretation the child does not understand many of the details of how this all works but still she may understand and attempt to imitatively reproduce some basic components In the case of linguistic constructions the child might hear an adult say I stapled your papers and also know the event to which he intends to draw her attention In addition she may also understand that in using the word stapled the adult intends to draw her attention to the type of activity he just performed and that in using your papers he intends to draw her attention to the items he just acted upon with your being used simply to help specify the papers involved The basic idea is thus that as the child hears a piece of language she attempts to read the speakers communicative intentions both at the level of the entire communicative act and at the level of its constituents Said another way in the process of cultural imitative learning the child is attempting to determine the communicative functions of the various linguistic items and structures she hears their communicative functions being the roles they play in the adults overall communicative intention The notion of communicative function is of crucial importance as it enables us to talk about such things as syntactic constituency your papers is a coherent constituent because it serves a single referential function and dependency your is a dependent element because it functions to identify which papers in the manner of functionalcognitive linguists eg Croft 1991 Givón 1995 Reconceptualized in this way to include intention reading my claim is that cultural imitative learning is more important in language development especially in the early stages than has traditionally been recognized This is clear in the data reviewed above which revealed that before their third birthdays children use individual verbs and syntactic constructions in just the way they have heard and understood them being used Interestingly this same very strong tendency toward imitative learning is also observed in young childrens social learning of tool use to the point that they sometimes copy adult tooluse behaviors even though this leads to undesirable results Nagell Olguin Tomasello 1993 This tendency is also apparent in childrens early symbolic play with objects as they almost always choose to do with toys and other objects what adults have demonstrated for them Tomasello Striano Rochat in press and also in their gestural communication as many parents invent and their children imitatively learn idiosyncratic gestures eg a specific mouth movement for fish Indeed human children in the ontogenetic period from 1 to 3 years of age are virtual imitation machines as they attempt to understand and reproduce virtually all of the activities they see in the cultural activities around them Carpenter Nagell Tomasello 1998b This early tendency towards imitative learning in both nonlinguistic and linguistic activities is perhaps best understood as the initial ontogenetic expression of the human organisms biological adaptation for culture Tomasello 1999 The strong tendency toward linguistic imitation in particular may be illustrated by two phenomena of child language that are often taken to be evidence against imitative learning but which are actually evidence for it if we look more precisely at what children do and do not hear First many young children say things like Her open it an accusative subject which they supposedly have not heard from adults But children regularly hear things like Let her open it or Help her open it and so they may just imitatively learn the end part of the sentence Her open it Very telling is the fact that children basically never make the complementary error direct object in nominative case Mary hit I or Jeff kissed she The reason they do not make this error is that they never hear adults say anything like this in any linguistic construction A similar account can be given for some of the findings going under the general rubric of optional infinitives in which children among other things sometimes fail to use subjectverb agreement markers appropriately Wexler 1994 Although there may be other factors at work in this case Leonard 1998 a major part of the explanation is very likely the large number of nonfinite verbs that children hear in various constructions in the language addressed to them especially in questions such as Should he open it and Does she eat grapes The child might then later say in partially imitative fashion He open it and She eat grapes It is important to stress that even in this new view cultural imitative learning by itself can support only limited forms of productivity Imitative learning is creative and productive only in the sense that it enables the use of particular linguistic symbols and constructions in novel communicative contexts But it does not enable children to produce novel utterances per se because it cannot create abstract linguistic categories or schemas This means that imitative learning cannot be the whole story of language development However the fact that it is not the whole story does not mean that it is not a very important part of the story It must be as all children learn the language to which they are exposed 43 Analogy making and structuremapping In the first constructivist formulations in the 1960s children began constructing linguistic abstractions very early in development One theory was that they did this on the basis of distributional regularities in the language they heard and produced leading to a Pivot Grammar based mostly on the positional patternings of words In both of these illustrations it is important to note that the child has imitatively learned the end portion of the adults utterance and its general function for indicating a real world situation but she has used what she has learned in a slightly different communicative context than the adult utterance demonstrating her immature understanding of all the constituents functional roles Braine 1963 Another theory was that children constructed linguistic abstractions on the basis of the semantic relations inherent in the utterances they heard and produced based on such things as AgentAction PossessorPossessed and the like eg Brown 1973 But these kinds of early abstractions are not the kind that adults make and so if children were to make them they would be headed down a developmental culdusac Gleitman Wanner 1982 The current theory avoids this problem by positing that in the beginning children make virtually no linguistic abstractions at all beyond something like concrete nominal only later attempting to zero in on adultlike linguistic categories and schemas But it is still a constructivist theory and the generativist objection to constructivist theories of language acquisition no matter when the construction process begins is that they founder on the problem of induction Induction it is said cannot create abstractions since to recognize similarities among different exemplars a child must already have the abstraction a priori that is innately Chomsky 1986 This view of induction goes back to Plato of course and very likely cannot be refuted on logical grounds just as other similar paradoxes cannot be refuted on logical grounds eg a dropped object never reaches the ground because it must first go half way and then half way again and so on ad infinitum However in the current case we can approach the problem more concretely in the following manner Undeniably English speakers create some kind of abstract schema for the nominal extraposition construction Its amazing the people you see here and just as undeniably this construction does not derive from an innate universal grammar since it is so idiosyncratic This very special construction is clearly learned and it is learned in such a way that it becomes quite productive presumably indicating abstractness The contention is simply that however it is done in the case of idiosyncratic yet abstract constructions such as this this is also how it is done in the case of canonical and abstract constructions such as the ditransitive construction and others The research that is most relevant for explicating the process by which children create abstract linguistic constructions is that concerning analogy and structure mapping see Gentner Markman 1997 for a recent review The basic idea is that human beings are capable of discerning similarities not only among objects on the basis of perceptual or functional features but also among relational and event situations on the basis of a common relational structure abstracted across the particular objects involved Gentner 1983 claims that when relations are mapped across situations the specific properties of objects are discarded the relations among objects are preserved and connected systems of relations such as causality are more likely to be transferred As one instance Brown and Kane 1988 taught children to use certain kinds of In both of these illustrations it is important to note that the child has imitatively learned the end portion of the adults utterance and its general function for indicating a real world situation but she has used what she has learned in a slightly different communicative context than the adult utterance demonstrating her immature understanding of all the constituents functional roles actions with a tool pull stack swing and then gave them transfer problems in which it was possible for them to use the same actions but with a different tool which they did reasonably well from 2 years of age This is of course exactly the kind of cognitive ability needed for children to create a verb island schema across different arrangements of object participants eg they learn X pushes Y and then transfer to M pushes N especially since so many early linguistic constructions revolve around highly salient intentional causal or spatial relations More abstract constructions can then be created by a similar relational mapping across different verb island constructions For example the several verb island constructions that children have with the verbs give tell show send and so forth all share a transfer meaning and they all appear in a structure NP V NP NP The specific hypothesis is thus that children based their constructional analogies on similarities of both form and function two utterances or constructions are analogous if a good structure mapping is found both on the level of linguistic form and on the level of communicative function Precisely how this is done is not known at this time and indeed a number of computational models are currently being explored for this and similar tasks Forbus Gentner Law 1995 There are also some proposals that a key element in the process might be some kind of critical mass of exemplars to give children sufficient raw material from which to construct their abstractions although the nature of this critical mass eg verb types versus verb tokens is not known at this time Marchman Bates 1994 It is also interesting that Gentner has found that higher order relations relations among relations begin to be learned during the late preschool period a very good match with the time period when children might be learning some higher order constructions eg the English SubjectPredicate construction based on the similarity of structure among many first order constructions such as intransitives causatives and ditransitives Some evidence for this proposal comes from the fact that the syntactic category subject in English which depends on generalizations across many first order constructions by all indications does not emerge until school age by which time children have mastered many types of first order constructions Braine Brooks Cowan Samuels TamisLeMonda 1993 I am thus proposing three stages of analogy making in the creation of abstract linguistic constructions given some stock of utterances using the verb push young children can by a process of structure mapping construct a verb island construction around the word push in much the same way they use the same imitatively learned action on different objects given some stock of verb island constructions similar to that used with push children can by a process of structure mapping construct something like a simple transitive construction given some stock of first order constructions such as the simple transitive and other similar constructions older children can by a process of second order structure mapping construct some higher order constructions such as the SubjectPredicate construction This proposal differs from previous constructivist proposals mainly in i its focus on the importance of sentencelevel argument structure constructions as the appropriate unit of analysis most previous proposals have focused on individual grammatical categories such as subject eg Schlesinger 1988 ii its emphasis on the extended period of child conservativeness denying the existence of early abstractions that go in nonadultlike directions iii its focus on both form and function as the basis for abstraction and iv the asymmetry posited for the abstraction of nominal categories versus verbbased constructional schemas Perhaps the most important aspect of these new features is that they enable us to make the explicit connection with research on analogy which has a similar focus on whole events and which posits a similar asymmetry between the objects and relations that constitute these events It is an empirical fact that young children make analogies in both their nonlinguistic and linguistic behavior there is no problem of induction It is important to stress that children cannot engage in these processes of analogy making and structure mapping unless they understand something of the functional structure of the utterances they have imitatively learned in terms of the constituency and dependency relations involved It is only with such an understanding based ultimately on an understanding of the communicative intentions of others that children can go on to align appropriately the corresponding constituents in the different constructions with respect to their similar functional roles what Gentner calls structural alignment The building of abstract linguistic constructions thus depends on a first step of imitative learning with some understanding of functional roles followed by a process of analogy making initially to get to first order constructions and then again later to get to some higher order constructions as specified by something like Gentners theory of structure mapping Exactly how this is done in the case of specific linguistic constructions what data are needed from adult language and the childs already mastered constructions and at what frequency is not known at this time 44 Structure combining At any given developmental moment children have at their command a relatively large number of linguistic constructions that vary from one another in both their complexity and their abstractness This characterization of childrens linguistic competence provides the basis for a very different way of thinking about their creative combining of linguistic structures Tomasello 1998b Children are not just combining words or isolated linguistic categories they are combining precompiled linguistic constructions of various shapes sizes and levels of abstractness As just one small example in Tomasello 1992 I looked closely at all of my daughters earliest utterances with 3 or more words which first emerged at around 1922 months of age One was See Daddys car said at around 19 months of age as she spied it coming Previously she had said things like See Maria See Daddy and See this on the one hand and also things like Daddys bread Daddys ball and Daddys salad on the other So my supposition is that she creatively combined something like a See verb island construction with a Daddys possessive a See Maria See Daddy See this See Daddys car b Examples from Sarah I think hes gone I think its in here I think my daddy took it Its a crazy bone I think I think I saw one I think dis is de bowl Examples from Nina See that monkey crying See Becca sleeping See that go See my hands are washed See he bite me See him lie down Mostly one subject type per verb Virtually no complementizers Virtually no nonpresent tenses Virtually no negations Fig 4 a One structure combining operation for the child of Tomasello 1992 b Some examples and general facts about earliest sentential complement sentences from Diessel and Tomasello in press construction see Fig 4a We do not know what this child actually heard or did not hear along the lines of See Daddys car previously in the discourse and so we do not know the extent to which this was a truly creative combining of constructions versus something suggested to her by an adult utterance But in any case even with an adult model the most natural supposition is that she had already mastered two twoword constructions that she then discovered or perceived how to combine to create a new meaning It is interesting to note in this regard that as independent constructions both See and Daddys had distinctive intonation contours See had a fairly neutral intonation whereas Daddys was said with a strong stress on Daddys The composite construction See Daddys car however had the neutral intonation contour of the See construction which is noteworthy since the See construction is syntactically dominant as well Assuming it was not simply mimicked from adults the intonation contour of the composite expression might then plausibly indicate this childs understanding of the syntactic dependency relations among the subconstructions involved As a more complex example of structure combining Diessel and Tomasello in press investigated childrens earliest complex sentences ie sentences with two or more verbs We looked at six children in the CHILDES database with a special focus on their complex utterances with sentential complements We found that virtually all early sententialcomplement sentences were composed of first as a main clause one of a handful of matrix verbs see also Bloom 1992 and second as a complement a simple sentence schema the child had already mastered The matrix verbs were of two types see Fig 4b First were epistemic verbs such as think and know In almost all cases children used I think to indicate their own uncertainty about something and they virtually never used the verb think in anything but this firstperson form ie no examples of He thinks She thinks etc This form was also virtually never negated no examples of I dont think virtually never used in anything other than the present tense no examples of I thought and virtually never with a complementizer no examples of I think that It thus appears that I think is a relatively fixed phrase meaning something like Maybe The child combines this fixed phrase with some full sentence but this combining does not amount to sentence embedding as it is typically portrayed in more formal analyses it is more like simple concatenation since the main verb think is not really acting as a verb Second children also use attentiongetting verbs like Look and See in conjunction with full sentences In this case they use them almost exclusively in imperative form again virtually no negations no nonpresent tenses no complementizers So again these early complex sentences do not appear to be abstract sentence embeddings but rather concatenations of a formulaic expression and a full sentence In all it seems that these early complex sentences are not abstract sentence embeddings as they are treated by generative theories but rather they are pastiches of welllearned linguistic patterns In particular they are combinations of a simple verbargument clause perhaps itembased perhaps more abstract juxtaposed with a specific itembased epistemic or attentiongetting expression such as I think You mean Look or See The key point for current purposes is that structure combining does not mean simply combining words but rather it means combining whole constructions that the child has previously mastered Children learn various kinds of constructions from early in development varying in both complexity and abstractness and so when they want to express some new meaning one thing they can do is to juxtapose or integrate those existing structures in some way The exact way this is done in specific cases for example the influence of different kind of source structures in the childs language the role of frequency of use of source structures and the role of different kinds of adult models is not known at this time 45 Summary This sketch of a possible alternative to generative theories of language acquisition was intended to make only one simple point The continuity assumption cannot be justified negatively that is by arguing that there must be continuity between child and adult linguistic competence since there is no way a child could get from concrete and itembased linguistic structures to the powerful abstractions that constitute adult linguistic competence The above account shows that there is a way if a we conceive of adult linguistic competence in more psychological and less mathematical terms and b we recognize that childrens skills of cognition and learning are more powerful than previously suspected especially with regard to intention reading and cultural learning analogy making and structure combining A usagebased theory of this sort also has the advantage that it does not have a linking problem since there is no universal grammar with which the childs local learning must ultimately link up It must also be stressed that in this view of language acquisition there is continuity of process the basic cognitive and learning mechanisms are the same at all developmental periods but there is discontinuity of structure Childrens concrete and itembased language early in development rests on lexically specific syntagmatic and paradigmatic categories thrower thing thrown etc not on the kinds of abstract and verbgeneral categories and schemas that characterize much of adult linguistic competence 5 Conclusion The modern study of child language acquisition began when developmentalists started to take Linguistics seriously But taking Linguistics seriously does not mean taking formal grammars written for adults and using them uncritically with children There is no doubt that formal grammars may be written for childrens language they may be written for just about any natural phenomenon including tonal music Jackendoff 1983 the human genome ColladoVides 1991 and dreaming Foulkes 1978 But the question is whether these formal grammars are psychologically real entities for young children Generative theories have simply assumed that they are and this continuity assumption has been used to justify the practice of taking an individual child utterance and describing it in essentially the same way that that utterance would be described if it were produced by an adult But all behavioral and cognitive scientists whether they study language or some other phenomenon know that similar behaviors may be produced by different underlying mechanisms In each case systematic research must be conducted to see whether indeed a childs adultlike or even partially adultlike behavior is underlain by adultlike mechanisms In the current case when we look at all of a given childs spontaneous language not only what she does but also what she does not do with particular words and phrases it is clear that childrens linguistic competence is much more concrete and itembased than adults Moreover when we give children novel verbs in controlled experimental situations they are initially conservative and only gradually show an increasing tendency with perhaps a special spurt at around 3 years of age to assimilate these novel verbs to abstract syntactic categories and schemas It is logically possible to argue that those abstract categories and schemas are present throughout early development and that at the younger ages children simply cannot effect the assimilation due to extraneous performance factors and the like But a number of control conditions and procedures in the experiments reviewed above effectively rule out that interpretation suggesting once again that at younger ages children simply do not possess the abstract syntactic competence characteristic of older children and adults From a more purely theoretical point of view the classical logical arguments of generative grammar against learningbased or usagebased theories simply do not hold when we a replace the mathematical view of language with a more psychologically based view of language and b replace the straw men typically used in these arguments simple association and blind induction with the more cognitively sophisticated learning and abstraction processes involved in intention reading cultural learning analogy making and structure combining And it must be emphasized that all theories of language acquisition of whatever type must posit local learning so that children can learn the particular structures of the particular languages into which they are born But in addition generative theories and only generative theories must also find a way for that local learning to link up with an innate universal grammar which so far they have not succeeded in doing There is no question that human children are biologically prepared to acquire a natural language in any number of ways involving basic processes of cognition social interaction symbolization and vocalauditory processing But this does not mean that they have to possess from the beginning the final adult syntactic structures in all of their complexity and abstractness Tomasello 1995 Indeed modernday linguistic constructions have taken many hundreds or perhaps thousands of years of social evolution to grammaticalize into the complex cognitive entities that exist today Recent research has demonstrated that when human beings communicate symbolically with one another in extended discourse interactions the stringing together of symbols begins to become grammaticalized for example content words such as nouns and verbs become function words or markers such as prepositions auxiliaries tense markers and case markers and loosely concatenated symbols acquire syntactic relationships involving constituency and dependency These transformations of linguistic structure occur as a result of socialinteractive processes in which i speakers try to abbreviate linguistic expression as much as they can and ii listeners try to make sure that speakers do not go so far in this direction that the message becomes incomprehensible Grammaticalization processes are wellattested in the written records of numerous languages in their relatively recent pasts and it is a reasonable assumption that the same processes were also at work in the earlier evolution of language turning loosely organized sequences of linguistic symbols into grammaticalized linguistic constructions which children then may learn Bybee Perkins Pagliuca 1994 Givón 1995 Traugott Heine 1991 It is an interesting hypothesis that the linguistic competence used to acquire these grammaticalized constructions is basically the same across all stages of human ontogeny But it is just that a hypothesis Continuity cannot be simply assumed without systematic investigation of the type that is conventional across the behavioral and cognitive sciences The research results reported here from both naturalistic obser vation and systematic experimentation suggest that the continuity hypothesis does not provide an accurate description of the early stages of childrens emerging syntactic competence Acknowledgements Thanks to the following people for helpful discussions andor comments on an initial draft of the paper Nameera Akhtar Shanley Allen Heike Behrens Patricia Brooks Jane Childers Gina ContiRamsden Holger Diessel Michael Israel Elena Lieven Julian Pine and Angelika Wittek Also thanks to Cognition reviewers Dan Slobin Eve Clark and an anonymous reviewer References Akhtar N 1999 Acquiring basic word order Evidence for datadriven learning of syntactic structure Journal of Child Language 26 339356 Akhtar N Tomasello M 1997 Young childrens productivity with word order and verb morphology Developmental Psychology 33 952965 Allen S 1996 Aspects of argument structure acquisition in Inuktitut Amsterdam John Benjamins Atkinson M 1996 Now hang on a minute Some reflections on emerging orthodoxies In H Clahsen Generative perspectives on language acquisition Erlbaum Bates E Goodman J 1997 On the inseparability of grammar and lexicon evidence from acquisition aphasia real time processing Language and Cognitive Processes 12 507587 Bates E MacWhinney B 1989 Functionalism and the competition model In B MacWhinney E Bates The crosslinguistic study of sentence processing Cambridge Cambridge University Press Bates E MacWhinney B Caselli C Devoscovi A Natale F Venza V 1984 A crosslinguistic study of the development of sentence comprehension strategies Child Development 55 341354 Behrens H 1998 Where does the information go Paper presented at MPI workshop on argument structure Nijmegen Benson M Benson E Ilson R 1997 The BBI dictionary of English word combinations Amsterdam John Benjamins Berko J 1958 The childs learning of English morphology Word 14 150177 Berman R 1982 Verbpattern alternation the interface of morphology syntax and semantics in Hebrew child language Journal of Child Language 9 169191 Berman R 1993 Marking verb transitivity in Hebrewspeaking children Journal of Child Language 20 641670 Berman R A ArmonLotem S 1995 How grammatical are early verbs Paper presented at the Colloque International de Besancon sur lAcquisition de la Syntaxe Besancon France November 1995 Bloom L 1992 Language development from two to three Cambridge Cambridge University Press Bowerman M 1976 Semantic factors in the acquisition of rules for word use and sentence construction In D Morehead A Morehead Normal and Deficient Child Language Baltimore University Park Press Bowerman M 1982 Reorganizational processes in lexical and syntactic development In L Gleitman E Wanner Language acquisition The state of the art Cambridge Cambridge University Press Bowerman M 1988 The no negative evidence problem How do children avoid constructing an overgeneral grammar In J A Hawkins Explaining language universals pp 73101 Oxford Basil Blackwell Bowerman M 1990 Mapping thematic roles onto syntactic functions Are children helped by innate linking rules Linguistics 28 12531289 Bowerman M 1997 Argument structure and learnability Is a solution in sight Berkeley Linguistics Society 22 Braine M 1963 The ontogeny of English phrase structure Language 39 114 Braine M 1976 Childrens first word combinations Monographs of the Society for Research in Child Development 41 Braine M 1992 What sort of innate structure is needed to bootstrap into syntax Cognition 45 77100 Braine M 1994 Is nativism sufficient Journal of Child Language 21 931 Braine M D S 1971 On two types of models of the internalization of grammars In D I Slobin The ontogenesis of grammar pp 153186 New York Academic Press Braine M D S Brooks P J 1995 Verb argument structure and the problem of avoiding an overgeneral grammar In M Tomasello W E Merriman Beyond names for things Young childrens acquisition of verbs pp 353376 Hillsdale NJ Lawrence Erlbaum Associates Braine M Brody R Fisch S Weisberger M Blum M 1990 Can children use a verb without exposure to its argument structure Journal of Child Language 17 313342 Braine M Brooks P Cowan N Samuels M TamisLeMonda C 1993 The development of categories at the semanticsyntax interface Cognitive Development 8 465494 Brooks P Tomasello M 1999 How young children constrain their argument structure constructions Language Brooks P Tomasello M 1999 Young children learn to produce passives with nonce verbs Developmental Psychology 35 2944 Brooks P Tomasello M Lewis L Dodson K 1999 How children avoid argument structure errors The entrenchment hypothesis Child Development Brown 1973 A first language The early stages Cambridge MA Harvard University Press Brown A Kane M 1988 Preschool children can learn to transfer Learning to learn and learning from example Cognitive Psychology 20 493523 Budwig N 1995 A developmentalfunctionalist approach to child language Hillsdale NJ Lawrence Erlbaum Associates Bybee J 1985 Morphology Amsterdam John Benjamins Bybee J Perkins R Pagliuca W 1994 The evolution of grammar Chicago University of Chicago Press Carpenter M Akhtar N Tomasello M 1998 14through 18monthold infants differentially imitate intentional and accidental actions Infant Behavior and Development 21 315330 Carpenter M Nagell K Tomasello M 1998 Social cognition joint attention and communicative competence from 9 to 15 months of age Monographs of the Society for Research in Child Development vol 63 Chapman R Miller J 1975 Word order in early two and three word utterances Does production precede comprehension Journal of Speech and Hearing Research 18 355371 Childers J Tomasello M 1999 Spanishspeaking childrens syntactic productivity with novel verbs Manuscript submitted for publication Chomsky N 1980 Rules and representations Behavioral and Brain Sciences 3 161 Chomsky N 1986 Knowledge of language Berlin Praeger Chomsky N 1993 A minimalist program for linguistic theory In K Hale S Keyser Eds A view from Building 20 Cambridge MA MIT Press Clahsen H 1996 Introduction In H Clahsen Generative perspectives on language acquisition Empirical findings theoretical considerations and crosslinguistic comparisons Amsterdam Benjamins Clahsen H 1999 Lexical entries and rules of language A multidisciplinary study of German inflection Behavioral and Brain Sciences Clahsen H Eisenbeiss S Penke M 1996 Lexical learning and syntactic development In H Clahsen Generative perspectives on language acquisition Empirical findings theoretical considerations and crosslinguistic comparisons Amsterdam Benjamins Clark E 1987 The principle of contrast A constraint on language acquisition In B MacWhinney Mechanisms of language acquisition Hillsdale NJ Erlbaum Clark E 1996 Early verbs event types and inflections In C Johnson J Gilbert Childrens language vol 9 Mahwah NJ Erlbaum ColladoVides J 1991 A syntactic representation of the units of genetic information Journal of Theoretical Biology 148 401429 Croft W 1991 Syntactic categories and grammatical relations The cognitive organization of information University of Chicago Press DeLancey S 1981 An interpretation of split ergativity and related patterns Language 57 626657 DeVilliers J 1985 Learning how to use verbs Lexical coding and the influence of input Journal of Child Language 12 587596 DeVilliers J DeVilliers P 1973 Development of the use of word order in comprehension Journal of Psycholinguistic Research 2 331341 Diessel H Tomasello M 1999 Why complement clauses do not include a thatcomplementizer in early child language Proceedings of the Berkeley Linguistic Society Dodson K Tomasello M 1998 Acquiring the transitive construction in English The role of animacy and pronouns Journal of Child Language 25 555574 Dryer M 1997 Are grammatical relations universal In J Bybee J Haiman S Thompson Essays on language function and language type Amsterdam John Benjamins Fillmore C 1985 Syntactic intrusions and the notion of grammatical construction Berkeley Linguistic Society 11 7386 Fillmore C Kaye P OConnor M 1988 Regularity and idiomaticity in grammatical constructions The case of let alone Language 64 501538 Fischer C 1996 Structural limits on verb mapping The role of analogy in childrens interpretation of sentences Cognitive Psychology 31 4184 Foley W Van Valin R 1984 Functional syntax and universal grammar Cambridge Cambridge University Press Forbus K Gentner D Law K 1995 MACFAC A model of similaritybased retir eval Cognitive Science 19 141205 Foulkes D 1978 A grammar of dreams New York Basic Books Gathecole V Sebastián E Soto P 1999 The early acquisition of Spanish verbal morphology AcrosstheBoard or piecemeal knowledge International Journal of Bilingualism Gentner D 1983 Structuremapping A theoretical framework for analogy Cognitive Science 7 155170 Gentner D Markman A 1997 Structure mapping in analogy and similarity American Psychologist 52 4556 Givón T 1995 Functionalism and grammar Amsterdam John Benjamins Gleitman L R Wanner E 1982 Language acquisition The state of the art In E Wanner L R Gleitman Language acquisition The state of the art New York Cambridge University Press Goldberg A 1995 Constructions A construction grammar approach to argument structure University of Chicago Press HirshPasek K Golinkoff R M 1991 Language comprehension A new look at some old themes In N Krasnegor D Rumbaugh M StuddertKennedy R Schiefelbusch Biological and behavioral aspects of language acquisition pp 301320 Hillsdale NJ Lawrence Erlbaum Associates HirshPasek K Golinkoff R M 1996 The origins of grammar Evidence from early language comprehension Cambridge MA MIT Press Hyams N 1994 Nondiscreteness and variation in child language Implications for Principle and parameter models of language acquisition In Y Levy Other children other languages Erlbaum Hillsdale NJ Ingham R 1993 Critical influences on the acquisition of verb transitivity In D Messer G Turner Critical influences on child language acquisition and development London Maximillian Press Jackendoff R 1983 A generative theory of tonal music Cambridge MIT Press Jackendoff R 1996 Twistin the night away Language 73 534559 Kay P Fillmore C 1999 Grammatical constructions and linguistic generalizations Language 75 133 Lakoff G 1987 Women fire and dangerous things What categories reveal about the mind University of Chicago Press Langacker R 1987 Foundations of cognitive grammar vol 1 Stanford University Press Langacker R 1991 Foundations of cognitive grammar vol 2 Stanford University Press Langacker R 1998 Conceptualization symbolization and grammar In M Tomasello The new psychology of language Cognitive and functional approaches Erlbaum Leonard L 1998 Children with specific language impairment Cambridge MIT Press Levinson S 1999 Language as nature and language as art Paper presented to the Pontifical Academy of Sciences Rome Lewis L and Tomasello M in preparation Young childrens productivity with imperative and indicative constructions Lieven E 1997 Variation in a crosslinguistic context In D Slobin The crosslinguistic study of language acquisition vol 5 Mahwah NJ Erlbaum Lieven E Pine J Baldwin G 1997 Lexicallybased learning and early grammatical development Journal of Child Language 24 187220 MacWhinney B 1978 The acquisition of morphophonology Monographs of the Society for Research in Child Development No 43 MacWhinney B 1999 The emergence of language from embodiment In B MacWhinney Emergentist perspectives on language acquisition Mahwah NJ Erlbaum Maratsos M Gudeman R GerardNgo P DeHart G 1987 A study in novel word learning The productivity of the causative In B MacWhinney Mechanisms oflanguage acquisition Hillsdale NJ Erlbaum Marchman V Bates E 1994 Continuity in lexical and morphological development A test of the critical mass hypothesis Journal of Child Language 21 339366 Mazuka R 1995 Can a grammatical parameter be set before the first word In J Morgan K Demuth From signal to syntax Hillsdale NJ Erlbaum Meltzoff A 1995 Understanding the intentions of others Reenactment of intended acts by 18monthold children Developmental Psychology 31 838850 Michaelis L Lambrecht K 1996 Toward a constructionbased theory of language function The case of nominal extraposition Language 72 215247 Nagell K Olguin K Tomasello M 1993 Processes of social learning in the tool use of chimpanzees Pan troglodytes and human children Homo sapiens Journal of Comparative Psychology 107 174186 Naigles L 1990 children use syntax to learn verb meanings Journal of Child Language 17 357374 Naigles L HoffGinsburg E 1998 Why are some verbs learned before others Journal of Child Language 25 95120 Naigles L Gleitman L Gleitman H 1993 Children acquire word meaning components from syntactic evidence In E Dromi Language and cognition A developmental perspective Norwood NJ Ablex Olguin R Tomasello M 1993 Twentyfive month old children do not have a grammatical category of verb Cognitive Development 8 245272 Pawley A Syder F 1983 Two puzzles for linguistic theory In J Richards R Smith Language and communication New York Longmans Pine J Lieven E 1993 Reanalysing rotelearned phrases Individual differences in the transition to multiword speech Journal of Child Language 20 551571 Pine J Lieven E 1997 Slot and frame patterns in the development of the determiner category Applied Psycholinguistics 18 123138 Pine J Lieven E Rowland G 1998 Comparing different models of the development of the English verb category Linguistics 36 440 Pinker S 1984 Language learnability and language development Cambridge MA Harvard University Press Pinker S 1987 The bootstrapping problem in language acquisition In B MacWhinney Mechanisms of language acquisition Hillsdale NJ Erlbaum Pinker S 1989 Learnability and cognition The acquisition of verbargument structure Cambridge MA Harvard University Press Pinker S 1991 Rules of language Science 253 530535 Pinker S 1994 How could a child use verb syntax to learn verb semantics Lingua 92 377410 Pinker S Lebeaux D S Frost L A 1987 Productivity and constraints in the acquisition of the passive Cognition 26 195267 Pizutto E Caselli C 1992 The acquisition of Italian morphology Journal of Child Language 19 491557 Pizutto E Caselli C 1994 The acquisition of Italian verb morphology in a crosslinguistic perspective In Y Levy Other children other languages Hillsdale NJ Erlbaum Pye C Loeb D Redmond S Richardson L 1994 When do children learn verbs In E Clark Proceedings of the 25th Annual Child Language Research Forum Stanford CSLI Publications Radford A 1990 Syntactic theory and the acquisition of English syntax Cambridge MA Blackwell Rizzi L 1993 Some notes on linguistic theory and language development Language Acquisition 3 371394 Roberts K 1983 Comprehension and production of word order in stage 1 Child Development 54 443449 Roeper T 1996 The role of merger theory and formal features in acquisition In H Clahsen Generative perspectives on language acquisition Empirical findings theoretical considerations and crosslinguistic comparisons Amsterdam Benjamins Rubino R Pine J 1998 Subjectverb agreement in Brazilian Portugese What low error rates hide Journal of Child Language 25 3560 Sampson G 1997 Educating Eve The language instinct debate London Casell Academic Schlesinger I 1988 The origin of relational categories In Y Levy I Schlesinger M Braine Categories and processes in language acquisition Hillsdale NJ Erlbaum Serrat E 1997 Acquisition of verb category in Catalan Unpublished Dissertation Slobin D 1985 Crosslinguistic evidence for the languagemaking capacity In D I Slobin The crosslinguistic study of language acquisition vol 2 Theoretical issues pp 11571256 Hillsdale NJ Lawrence Erlbaum Associates Slobin D 1997 On the origin of grammaticalizable notions Beyond the individual mind In D I Slobin The crosslinguistic study of language acquisition vol 5 Hillsdale NJ Erlbaum Slobin D Bever T 1982 Children use canonical sentence schemas A crosslinguistic study of word order and inflections Cognition 12 229265 Stoll S 1998 The acquisition of Russian aspect First Language 18 351378 Talmy L 1988 The relation of grammar to cognition In B RudzkaOstyn Topics in Cognitive Linguistics Amsterdam John Benjamins Tomasello M 1995 Language is not an instinct Cognitive Development 10 131156 Tomasello M 1992 First verbs A case study in early grammatical development Cambridge University Press Tomasello M 1996 Do apes ape In J Galef C Heyes Social Learning in Animals The Roots of Culture Academic Press Tomasello M 1998a The new psychology of language Cognitive and functional approaches Mahwah NJ Erlbaum Tomasello M 1998b The return of constructions Journal of Child Language 25 431488 Tomasello M 1998c Reference Intending that others jointly attend Pragmatics and Cognition 6 219234 Tomasello M in press Perceiving intentions and learning words in the second year of life In M Bowerman S Levinson Language acquisition and conceptual development Cambridge Cambridge University Press Tomasello M 1999 The cultural origins of human cognition Harvard University Press Tomasello M Brooks P 1998 Young childrens earliest transitive and intransitive constructions Cognitive Linguistics 9 379395 Tomasello M Brooks P 1999 Early syntactic development In M Barrett The development of language UCL Press Tomasello M Olguin R 1993 Twentythreemonthold children have a grammatical category of noun Cognitive Development 8 451464 Tomasello M Akhtar N Dodson K Rekau L 1997 Differential productivity in young childrens use of nouns and verbs Journal of Child Language 24 373387 Tomasello M Kruger A Ratner H 1993 Cultural learning Target Article for Behavioral and Brain Sciences 16 495511 Tomasello M Striano T Rochat P 1999 Do young children use objects as symbols British Journal of Developmental Psychology Traugott E Heine B 1991 Approaches to grammaticalization vols 1 and 2 Amsterdam John Benjamins Valian V 1991 Syntactic subjects in the early speech of American and Italian children Cognition 40 2181 Van Hoek K 1997 Anaphora and conceptual structure Chicago University of Chicago Press Van Valin R 1991 Functionalist theory and language acquisition First Language 11 740 Van Valin R 1992 An overview of ergative phenomena and their implications for language acquisition In D I Slobin The crosslinguistic study of language acquisition vol 3 Hillsdale NJ Erlbaum van Valin R 1998 The acquisition of Whquestions and the mechanisms of language acquisition In M Tomasello The new psychology of language Cognitive and functional approaches Mahwah NJ Erlbaum Weissenborn J 1992 The role of null subjects in early grammars In J Weissenborn H Goodluck T Roeper Theoretical issues in language acquisition Continuity and change in development Hillsdale NJ Lawrence Erlbaum Wexler K 1994 Optional infinitives head movement and the economy of derivation In D Lightfoot N Hornstein Verb movement Cambridge University Press