The Slow Growth of Language in Children

Robbins Burling

University of Michigan

A surprisingly broad consensus has grown among scholars who are interested in the origin of language that the ability to speak was a relatively recent and quite sudden development (e.g. Noble and Davidson, 1996, p. 17). Hardly anyone would now go so far as to credit a single mutation for all of syntax, as Bickerton once did (1990; but see Bickerton, 1997 and Calvin and Bickerton, 2000), and it may be granted that some sort of pidgin-like proto-language could have served as bridge between absolutely no syntax at all and the full syntax of the languages we speak today. Nevertheless, full syntax is still widely presumed to have come quite late (40,000 to 150,000 years ago) and quite suddenly. One support for this presumption is the difficulty that some linguists have imagining a partial syntax. Syntax is presumed to be not only incredibly complex, but highly interconnected. No part, it is sometimes supposed, could exist without all the rest, and all of its extraordinary complexities are seen as following from a few simple principles (e.g. Berwick, 1998).

If one is to hold this view consistently, it is necessary to deny the existence of any form of partial or simple syntax in the varieties of natural language that we can observe today. Among other possible candidates for partial syntax is child language, for if some stage of child language exhibits partial syntax it would be difficult to insist that no system of partial syntax is possible. This means that anyone who denies partial syntax ought to be driven to interpret child language as jumping abruptly from a stage where it consists of little more than strings of poorly joined individual words (so-called "proto-language") to the stage of full syntax. No one has pursued the logic of this argument more forthrightly than Derek Bickerton (1990), and we owe him a considerable debt for presenting the best available evidence that children move abruptly from proto-language to full language, but the same view has to be implicit in the widely held presumption that syntax is a tightly integrated whole, all of whose parts depend upon one another in complex ways.

Unfortunately for those who hold this position, the facts of child language, as we can observe it each day in the behavior of our own children, do not fit an abrupt scenario. It is true that children frequently appear to achieve complex syntax with enviable speed, but they do not move from the single word stage, or even from some sort of proto-language, to full syntax in a few months, let alone in the single day that should be expected by anyone who is unable to imagine a partial syntax. I believe that the abrupt appearance of syntax in children is actually an illusion, but it is an illusion that is helped by our frequent failure to take account of both early and late learning. First, we may not notice the great amount of learning that takes place silently before active production of language even begins, and second, we may fail to recognize that the apparent mastery over syntax with which we sometimes credit five year old children is due in good part to their avoidance of complex constructions. By avoiding mistakes they may fool us into believing that they have learned it all. I will offer some examples of both early and late syntax, hoping to show that learning is spread across many years.

Such a conclusion, however, would be premature. At the age of two years and two months, when he said almost nothing, he was able to understand a great deal. He certainly understood many hundreds of words, quite plausibly thousands. When asked, he could point reliably not only to his eyes, nose and mouth, but to his chin, elbow, and knee. He could point reliably not only to a window or a door, but to the ceiling, wall, and floor. He knew the names of dozens of people. The occasion at which I was with him, at two years and two months, was a family reunion which collected a number of people whom Jamie had never known before, one of whom was the girl friend of my nephew, his father's cousin. She expressed astonishment, on one occasion, that he was able to comply, with no apparent difficulty, to the request "Can you give some popcorn to Cindy?" even though many people were present to whom the popcorn might, at that moment, have been passed. Cindy had not supposed that Jamie had seen enough of her to have learned her name. Of course he could use cues both from the sentence in which the name was embedded, and from the social context. These certainly could have helped him to interpret the request. The point is, however, that he knew enough language to help him in his interpretation. He must have known that every person has a name. Even if he selected Cindy because he knew the names of everyone else in the immediate vicinity, that would demonstrate his knowledge of all the other names, as well as his ability to pluck the name out from the syntactic context in which it occurred. Just how much syntax he could understand is very difficult even to speculate about, but he understood a lot of English.

More systematic studies of early language comprehension have been made than my admittedly causal observations of Jamie. Consider, for example, the work of Kathryn Hirsh-Pasek and Roberta Golinkoff (1991; 1996). They studied comprehension by playing recorded sentences to small children who could look at either of two TV monitors. One of these monitors illustrated the sentence and the other illustrated a contrasting situation. Hirsh-Pasek and Golinkoff compared the length of time in which the children looked at the two monitors. For example, one of the sentences that the children heard was "She is kissing the keys". While this sentence came over the loud speaker, one of monitors showed a woman kissing a bunch of keys that she held in one hand, while she held up a ball in plain view with the other hand. The second monitor showed the opposite situation: a woman kissing a ball, while she held up the bunch of keys.

Children between 13 and 15 months of age who had not moved past the one word stage and who had an average productive vocabulary of about 25 words, reliably focused for a longer time on the TV monitor whose picture corresponded to the sentence they heard than on the monitor that showed a contrasting scene. To be sure, not much syntax is needed to discriminate between the meanings, but since both pictures in this example show both kissing and keys, more is needed than simple receptive comprehension of the words. The child must, at least, be able to recognize that some sort of association connects the words kissing and keys in the sentence that they heard. This experiment suggests that children who are just beginning to talk, can already find associations between the words of the sentences that they hear.

Children who were just slightly older -- between 16 and 18 months -- but whose productive language had still not passed the one word stage, were able to respond correctly, with greater than chance accuracy, to sentences such as Where is Big Bird washing Cookie Monster? This sentence is "reversible" in the sense that either of the two characters might have been doing the washing, and the two monitors showed the contrasting actions. When the spoken sentence described Big Bird doing the washing, the children looked longer at the screen that depicted that action than at the screen where Big Bird was being washed. Their preference was reversed when the sentence they heard had the characters reversed. These children, none of whom used more than one word at a time in their production, were able to use word order to interpret the meaning of at least some of the sentences that they heard.

Additional evidence for the precedence of comprehension over production comes from imitation experiments conducted by LouAnn Gerken, Barbara Landau, and Robert E. Remez (1990). This group worked with children who were as young as 24 months, still in the early stages of productive language, but a bit older than the children studied by Hirsh-Pasek and Golinkoff These children were persuaded to imitate short phrases. Some of the phrases that they heard had four English morphemes arranged according to the following pattern: a verb, such as push, to which —es 'third singular' was suffixed, followed in turn by the article the and a noun such as truck. The resulting phrases had the form of pushes the truck, with a verb, two function morphemes and a noun. Other phrases were similar except that some of the English words and affixes were replaced by nonsense forms. In some, the content words, the verb and noun, were replaced by nonsense words, while in others it was the function morphemes, -ez and the, that were replaced. In still other examples all four morphemes were nonsense forms. As in real phrases, the verb base and noun base that framed the examples were given the full stress of content words, while the two morphemes that came in the middle were weakly stressed. The experimental results came from comparing the success of the children in imitating the content words and the function morphemes, both genuine and artificial, under a variety of conditions. One might suppose that children would more reliably imitate genuine English morphemes than nonsense substitutes, but the results were not so simple.

First, other things being equal, the children were notably more successful at imitating the content words (whether English or nonsense) than the function morphemes (whether English or nonsense). That is, they often said things like push truck instead of the full sequence that they heard. If we knew no more, this might be attributed to either of two things. First, the content words received greater stress than the function morphemes, as they do in normal speech, and this ought to have made imitation easier. On the other hand, the more reliable imitation of the English nouns and verbs could mean that the children had not yet learned the function morphemes as well as they had the content words.

More interesting was the fact that children omitted genuine English function words more often that they omitted nonsense substitutes for the function words. Instead of saying pushes the truck, some children said push truck. When imitating such sentences as pusheg le truck, where the nonsense forms -eg and le have been substituted for -es and the, more children imitated more accurately. They were more likely to include -eg and le than they were to imitate the genuine English suffix and article. Apparently they recognized some sort of difference between the familiar English function morphemes and their unfamiliar substitutes, and the very familiarity of the real morphemes made it possible to leave them out. At a minimum, the children reacted differently to the familiar morphemes than they did to the strange ones, and since they imitated the unfamiliar ones more accurately, their omission of the familiar form must have been the result of some sort of active editing out, not simply a failure to notice them.

In addition, the children imitated the content words, whether these words were genuinely English or nonsense substitutes, more successfully in the presence of genuinely English function morphemes than in the presence of nonsense substitutes. They could imitate push and truck more successfully when they heard pushes the truck than when they heard pusheg le truck. In other words, the genuine English function morphemes helped them to imitate the content words, even when they did not actually articulate the function morphemes. Children who failed to imitate the function morphemes, were still able to use these function morphemes to help in the identification and isolation of the content words. Gerken et al do not suggest that the children had a full adult understanding of the function morphemes, but it does seem that even at a stage when children are considerably less likely to imitate function morphemes than content words, the function morphemes can still be used to help them to extract the meaning, and even the grammatical structure, of the sentences. They used markers of syntax more reliably in comprehension than in their own production.

I have summarized the results of these experiments rather schematically, and I should point out that all of the results are statistical trends rather than absolute differences. The children were more likely to imitate content words than function morphemes but imitation of function morphemes did sometimes take place. Nevertheless, the overall trends are so clear that I find it impossible to doubt that comprehension of syntax comfortably precedes its production. It is much more difficult to study language comprehension than language production. When children appear to understand a bit of language, it is often impossible to know to what extent they have depended on words, on syntax, or on the nonverbal context within which the language is embedded, but studies like these show that the study of comprehension is not impossible. Any study that focuses exclusively on production, while failing to recognize the likelihood that comprehension of syntax, like comprehension of words, runs well ahead of production, is bound to give a distorted picture of a child’s linguistic skills. Producing a syntactic construction in the appropriate circumstances should be looked upon as only the final stage in a long developmental process.

The second source for the illusion of very rapid language learning comes with the late stages. It is sometimes asserted that five-year-old children have mastered their grammar (e.g. McNeill, 1966, p. 99; Rees, 1974, p. 255). By then, it has been supposed, children know all the syntax that they will ever need to know. It is true that five-year-olds have generally overcome the imperfect phonology of their earlier years, and those who are learning English, at least, no longer make many of the obvious morphological errors of younger children. Still, the illusion of syntactic mastery that five-year-old children give us is not achieved by knowing all there is to know about syntax, but by the simple expedient of avoiding complex syntactic constructions.

Much less attention has been given to 5 to 10 year-old language learners than to younger children, but those working with older children still point back to the pioneering work of Carol Chomsky (1969). In careful sessions Chomsky asked forty predominantly middle class 5 to 10 year-old children who were learning English, to manipulate dolls in ways that would demonstrate their understanding of various types of sentence. She found that five year olds still lacked mastery over several central aspects of syntax. Few five-year olds, for example, could recognize with any consistency that the pronoun he, in a sentence such as "He found out that Mickey won the race", has to refer to someone other than Mickey. In several other types of sentence, of course, the pronoun is ambiguous. Ambiguous sentences include those where the main clause comes second, as in After he got the candy, Mickey left, and those in which the pronoun refers back to an earlier main clause, as in Mickey said that he was hungry. In both these cases he can refer either to Mickey or to someone else. Only in sentences in which the pronoun is found in an initial main clause, does the pronoun refer unambiguously to someone other than the person mentioned in the second, subordinate, clause.

The children who were C. Chomsky’s subjects were asked to manipulate or point to two dolls, one of Mickey Mouse and one of Pluto. When given the sentence He found out that Mickey won the race and asked to point to the doll who found it out, most of the five year old children in the sample responded quite randomly. They were as likely to point to Mickey as to Pluto. Most children who were six or older had no trouble identifying the right doll. These older children, moreover, often explicitly recognized the ambiguity of the he in such sentences as Before he went out, Pluto took a nap. They made such comments as "It could be either one". Older children recognized the ambiguity of some sentences, and understood the other sentences correctly. Younger children never commented on the ambiguity and they responded randomly to sentences that an adult would regard as unambiguous. Clearly, five-year-olds still had some syntax to learn .

C. Chomsky investigated other bits of syntax that were learned even later. In the presence of a blindfolded doll, children were asked Is the doll easy to see or hard to see? Some children as old as seven and eight, and more who were five and six, replied that the doll was hard to see. Those who responded in this way were then asked Would you make her easy to see?. Logically and consistently, they then proceeded to remove the blindfold from the doll’s eyes. In discussing what they had done, some children even explained that the blindfold made her hard to see but that the difficulty could be solved by removing the blindfold. Children who called the blindfolded doll easy to see, were asked to make her hard to see and they responded in various ways. Some hid the doll under the table, some put something on top of her, some merely closed or covered their own eyes. As we would expect, a higher proportion of older children answered this question as adults would answer it, but there was considerable overlap in ages, some five year olds responding correctly, some eight year olds failing to do so. As Chomsky commented (1969, p. 32) "The fact that there are children of 7 and 8 who have not yet mastered this construction indicates that fairly basic syntactic learning is still going on considerably beyond the age at which it is generally considered to be complete".

We can be fooled into exaggerating children’s knowledge when they use forms in a way that appears superficially to conform to adult standards of syntax even though they still lack full comprehension of the way these forms are used. Annette Karmiloff-Smith has shown that French speaking children do not gain a full understanding of such basic function words as articles and possessives until they are about eight years old (1986). A difficult feature of articles and possessives is that they combine several functions, and French articles and possessives are even more complex on this score than are those of English. Les, for example not only shows definitness as does English the and totalization (in contrast to des 'some') but also pluralization (in contrast to le/la). A possessive pronoun such as mes 'my' indicates possession but, unlike English my, it also indicates plurality. More subtly, unless modified in some way, mes, like English my, implies all of the things mentioned, as in give me my books, or donnez moi mes livres, where the pronoun could not refer to only some of the books. At the same time mes and my denote a particular sub-class, so mes livres distinguishes 'my particular books' from all the books that might be under consideration.

French speaking children as young as three and a half years old use les and mes in ways that seem, superficially, to conform to adult standards, but among the children whom Karmiloff-Smith studied it was only those who were older who understood their full array of meanings. By five years of age, children easily used les to indicate plurality, but they did not recognize the totalizing sense of "all that are present". Both the possessive and pluralizing functions of pronouns like mes 'my' were understood well enough, but, once again the totalizing function was not. These words were taken to indicate a plural number, but not necessarily all of those under consideration.

Part of the evidence for the failure of young children to understand the full implications of les and mes, comes from children between five and eight years old. Unlike younger children they did, at least, demonstrate an awareness of the need to indicate totality but they more often accomplished this by adding a redundant tous 'all' than by exploiting the totalizing function of the plural article or possessive. Karmiloff-Smith summarizes in this way "Thus, whereas smaller children used les X in a given situation but only meant to convey pluralization, children of the second level [5 1/2 to 8 years old] added in the same situation (tous les X) in order to cover totalization also. It was not until the third level [8 to 12] that children used les X to mark pluralization and totalization simultaneously".

A number of studies have shown that children of five and even older, have considerable difficulty interpreting passives and some kinds of relative clauses. Sheldon (1974) for example has shown that five year old children, when asked to manipulate toy animals to show what a sentence describes, often misunderstand a sentence such as The dog bumps into the horse that the giraffe jumps over. Many children, some of them with great consistency, interpret such sentences as if it is the subject rather than the object of the main clause that is the object of the relative clause. In this example they make the giraffe jump over the dog rather than the horse. Children do considerably better with coordinate expressions that are synonymous with relative sentences. In other words, they are better, at interpreting The dog bumps into the horse and the giraffe jumps over the horse, than they are with the synonymous The dog bumps into the horse that the giraffe jumps over. This shows that it is the construction that is causing the difficulty, not the idea. Of course adults may also find the relative construction more difficult that the coordinate construction, but they are far better at understanding than five year old children. Five year olds simply do not have anything like an adult level of competence. Since five year olds rarely attempt to use constructions of this sort, their lack of adult competence will not be easily observed if we pay attention only to spontaneous production while ignoring comprehension.

The examples that I have given suggest that syntax starts to be learned well before it is exhibited in productive speech and that it continues to be refined for several years after five. Neither of these age periods has been the main focus for studies of syntactic development, however, and it is still the case that the best evidence for the progressive, step by step growth of syntax comes from the classical period for language acquisition, from about one and a half years old to five. Bickerton has argued that syntax develops very rapidly in children and he has recently (2000, p. 205) referred to "what some acquisitionists have called "the syntactic spurt". In my own reading of the literature on child language I have come across little discussion of any syntactic spurt, unlike the widely recognized "lexical spurt". Instead, I find a pervasive, though rarely explicitly stated, presumption of gradual and continuous syntactic growth. The question of whether or not syntax appears abruptly during this period has not really attracted much scholarly attention. Students of child language are more likely simply to take it for granted that syntax develops progressively through these years. A fine recent book-length survey of research in child syntax, for example, William O’Grady’s Syntactic Development (1997) never argues the point explicitly, but the examples give ample evidence that syntax develops progressively from simple beginnings through increasingly complex stages. Of course, the consensus among child language experts does not make it true, but the absence of attention to the question does present a few problems for someone like me who has a point to make. I will limit myself to one example that I find particularly persuasive.

Adele Goldberg (1995; 1999) has pointed out that linguists often presume that each verb brings with it the expectation that it will be associated with certain particular arguments such as a subject, direct object, locative and so forth. In other words, verbs are presumed to "project" their argument structures. Thus we suppose that the verb go in a sentence such as Pat goes down the street, requires a subject (Pat), and a prepositional phrase complement (down the street). We may also take it for granted that it is the verb go that leads us to interpret this sentence as describing motion (1999, p. 197). These presumptions run into problems with other verbs, however. We would hardly want to credit rumble in The truck rumbles down the street with the same attributes as go. Rumble does not, by itself, imply any sort of motion. Nor does rumble so strongly push us to use a prepositional phrase complement as does go. Nevertheless, upon hearing The truck rumbles down the street, we immediately understand that motion in involved. How do we do that?

Goldberg’s answer is that in addition to words, children learn constructions that have characteristic meanings and that they first learn these constructions with the help of what she calls "light" verbs. These are verbs that are very common, that have quite general meanings, and that typically are short, all characteristics that make them easy to learn. Examples of English light verbs, in addition to go, are do, make, give, and put. Light verbs such as these occur with very high frequently in the speech of children whose syntax is in its early stages. Sentences with light verbs act as prototypes for the later acquisition of other verbs, and they continue to act as prototypes for adults. When we hear the verb rumble in the same construction that we have learned for go, as in The truck rumbles down the street, we give it a parallel interpretation. Even though no word in this sentence by itself implies motion, we understand the sentence as to imply motion just as clearly as if the verb were go. For present purposes, the point is simply that learning the construction with a light verb, and subsequently learning to interpret other verbs in a parallel manner, takes time. The stages along the way are not difficult to observe in children once we know what to look for.

If syntax starts to be learned months before it appears in production, and if it continues to be learned as late as the age of ten, its acquisition is not as magically fast as has sometimes been supposed. Partial syntax of many levels of complexity can be observed in our own children. I have been baffled by the deficiency of imagination exhibited by the linguists who profess to be unable to imagine a partial syntax. Have these linguists never talked to a child?

If syntax can grow gradually in children throughout all the years of childhood, it could surely have grown gradually over many hundreds of thousands or even millions of years of phylogeny. I do not mean to imply by this that ontogeny recapitulates phylogeny. The sequence of steps though which children pass need not be the same sequence that was once taken by our species. Obviously, our species did not babble for a hundred thousand years while preparing for the language that was still to come. Nevertheless, the evidence of children's step by step acquisition of syntax ought to undermine the belief that syntax forms such a tightly integrated system that it could only have appeared suddenly in the species. At the very least, even someone who does believe that ontogeny recapitulates phylogeny ought not to use the evidence of children to support the idea that language developed suddenly, or even quickly, in phylogeny. Why not put the first feeble beginnings of syntax at two million years ago when the earliest members of the genus Homo exhibited the first substantial expansion of the brain beyond the hominid level? Of course, a few final touches could still have been added during the last hundred thousand years. Anything as complex as human syntax should have needed at least that long to develop to its present form.

Bibliography