Paper 4 essay — how children acquire grammar

The proposition that children construct their native grammar rather than being taught it lies at the heart of modern developmental linguistics. To assess it honestly we must distinguish between two senses of "teach" — explicit correction by caregivers, and the passive exposure to linguistic input that all infants receive. The evidence, accumulated since the mid-twentieth century, strongly supports the idea that children build grammar from impoverished input, drawing on cognitive resources that no parent ever consciously gives. This essay argues the construction account is correct in its essentials, while saying the role of social interaction has been underestimated by its strongest proponents.

The classical argument for grammar-as-construction comes from Noam Chomsky's idea of a Language Acquisition Device (LAD). Chomsky said that the linguistic input children receive — the "primary linguistic data" of everyday parental speech — is full of false starts, broken sentences, and ambiguous structures, yet by age five every neurotypical child gets a stable, rule-governed grammar of considerable abstract complexity. The poverty-of-the-stimulus argument, as it became known, said that children must come with some innate scaffolding — a Universal Grammar — that makes of possible languages, allowing them to fix the parameters of their native tongue from limited evidence.

Empirical work on grammatical errors gives some of the strongest support for this view. Roger Brown's famous longitudinal study of Adam, Eve, and Sarah, conducted at Harvard in the 1960s, showed that children's early grammatical errors are not random reproductions of adult speech but regular. A child who says "I goed" or "two foots" has manifestly not heard these forms from caregivers; she has made an underlying rule (past tense = stem + /-d/; plural = stem + /s/) and applied it to verbs and nouns that happen to be different. These so-called virtuous errors are the clearest empirical signature of construction: the child has made novel forms that fit an internally derived rule rather than to the surface data she has been exposed to.

The classic experimental confirmation of this constructive capacity is Jean Berko Gleason's 1958 "wug test." Gleason showed pre-school children with a drawing of a single cartoon bird and the novel word "wug," then showed a picture of two such birds and asked the child to complete the sentence "Now there is another one. There are two ___." Children, including pre-schoolers who could not yet read, said "wugs" with the correctly voiced /z/ plural suffix. They had never heard the word "wug" in their lives. They could not, in any meaningful sense, have been "taught" its plural. What they had clearly internalised was not the lexical fact about specific nouns but the morphophonological rule that makes English plurals from arbitrary stems. The wug test, used across dozens of languages and decades, is perhaps the single most economical demonstration of the constructive view: the child applies a rule to data she has never encountered, which is the operational definition of generative competence.

Cross-linguistic evidence broadens the case. Children acquiring languages as different from English as Mandarin, Inuktitut, and Hungarian go through the same developmental stages — single-word, two-word, telegraphic, complex — despite the radically different grammatical resources their target languages provide. Mandarin-speaking children, for instance, get the use of classifier morphemes around the same developmental window in which English-speaking children get the articles "a" and "the," even though classifiers and articles serve only loosely analogous functions and the relevant input is structurally different. If the acquisition mechanism were primarily imitative, we would expect the developmental trajectory to vary unpredictably with surface input. We observe instead a regularity that strongly suggests a shared underlying process — the maturation of a structure-seeking capacity that comes beneath the level of the particular features being learned.

Against this innatist position stands the older behaviourist account associated with B. F. Skinner, who said in Verbal Behavior (1957) that language is acquired through operant conditioning — that is, through reinforcement of correct utterances and extinction of incorrect ones. Skinner's view, attacked by Chomsky's 1959 review, fails. First, it cannot explain virtuous errors: a child who has been positively reinforced for saying "went" should, on a strict reinforcement model, never produce "goed" at all. Second, careful observation of caregiver speech shows and, when it occurs, often gets ignored. Roger Brown's seen example of a mother trying to correct her son's "Nobody don't like me" eight times, only for the child to produce the same construction on the ninth attempt, makes the limits of reinforcement vividly.

Yet a strictly Chomskyan account is also incomplete. From the 1990s onwards, the usage-based tradition associated with Michael Tomasello has said children acquire grammatical constructions through close attention to communicative intention, not through the activation of an abstract innate module. On Tomasello's account, the child first learns concrete "item-based" constructions ("More milk," "Where Daddy?") and only gradually gets the abstract grammatical patterns underlying them. The construction is real, but it is reconstructed from richer input than Chomsky said — input that includes joint attention, gaze direction, and the large social scaffolding said by Jerome Bruner as a Language Acquisition Support System (LASS).

Empirical work on the critical period — the idea that language must be acquired before a developmental window closes, typically taken to be around puberty — adds a further dimension to the construction account. The famous case of Genie, the American girl kept from linguistic input until age thirteen, showed that despite intensive instruction in adulthood she never got the morphosyntactic competence of even a young child. If grammar were simply learned from input, no biological deadline should exist. The critical period suggests, instead, that constructing a maturation-bound neural readiness that runs out once the relevant brain regions complete their developmental window.

Comparable evidence comes from studies of simultaneous bilingual acquisition. Children raised with two languages from birth — say, Cantonese at home and English at preschool — typically make both grammars in parallel and reach native competence in each by the same age window as monolingual peers, despite receiving roughly half the input per language. A behaviourist account predicts the opposite: less input should mean slower or weaker acquisition. The empirical record says the opposite story, which is again consistent with a model on which the child's grammatical competence is constructed from input rather than transferred from it. What the child needs is not more input but enough input to fix the parameters of the language she is building.

What follows for the original claim? The construction view is correct in its central contention: children are not taught grammar in the way they are taught arithmetic. Caregivers do not list rules; they do not make stronger correct forms; they do not fix errors. And yet children acquire, in three to four years, a structurally rich grammar that takes years. The evidence is consistent only with a model on which the child is an active hypothesis-tester, made with cognitive and probably innate machinery for getting rules from data.

But the strongest formulation of the claim — that grammar is constructed without help — is too big. The usage-based and interactionist traditions have shown, that joint attention and pointing make the early stages of word learning, and that children deprived of conversational partners (as in cases of severe neglect or unaddressed deafness) do not develop normally. The slogan "children are not taught, they construct" is best understood as a corrective to the behaviourist view, not as a denial that input has a role.

In conclusion, the proposition deserves qualified endorsement. Grammar is not taught; it is built from the data she encounters, with the help of capacities — probably innate, certainly developmentally constrained — that come before any specific cultural input. What caregivers provide is not a curriculum but an environment: rich, made, and emotionally engaged, but not didactic. To call this combination "teaching" would miss the active, hypothesis-generating role of the child; to call it "no teaching at all" would miss the scaffolding without which the constructive process cannot proceed. The most sensible account, then, treats the child not as a passive recipient and not as a solitary architect, but as a structure-seeker whose work depends on a structured social world.

The proposition that children actively construct their native grammar rather than being explicitly instructed in it lies at the heart of modern developmental linguistics. To assess it rigorously we must distinguish between two senses of "teach" — explicit correction by caregivers, and the incidental exposure to linguistic input that all infants receive. The evidence, accumulated since the mid-twentieth century, overwhelmingly supports the view that children reconstruct grammar from impoverished input, drawing on cognitive resources that no parent ever consciously imparts. This essay maintains that the construction account is correct in its essentials, while conceding that the role of social interaction has been underestimated by its strongest proponents.

The classical argument for grammar-as-construction comes from Noam Chomsky's postulate of a Language Acquisition Device (LAD). Chomsky observed that the linguistic input children receive — the "primary linguistic data" of everyday parental speech — is full of false starts, fragmentary sentences, and ambiguous structures, yet by age five every neurotypical child attains a stable, rule-governed grammar of considerable abstract complexity. The poverty-of-the-stimulus argument, as it became known, inferred from this mismatch that children must arrive equipped with some innate scaffolding — a Universal Grammar — that narrows the hypothesis space of possible languages, allowing them to fix the parameters of their native tongue from limited evidence.

Empirical work on grammatical errors provides some of the strongest support for this view. Roger Brown's famous longitudinal study of Adam, Eve, and Sarah, conducted at Harvard in the 1960s, documented that children's early grammatical errors are not random reproductions of adult speech but systematic overgeneralisations. A child who says "I goed" or "two foots" has manifestly not heard these forms from caregivers; she has inferred an underlying rule (past tense = stem + /-d/; plural = stem + /s/) and applied it to verbs and nouns that happen to be irregular. These so-called virtuous errors are the clearest empirical signature of construction: the child has generated novel forms that conform to an internally derived rule rather than to the surface data she has been exposed to.

The classic experimental confirmation of this constructive capacity is Jean Berko Gleason's 1958 "wug test." Gleason presented pre-school children with a drawing of a single cartoon bird and the novel word "wug," then showed a picture of two such birds and asked the child to complete the sentence "Now there is another one. There are two ___." Children, including pre-schoolers who could not yet read, produced "wugs" with the correctly voiced /z/ plural suffix. They had never heard the word "wug" in their lives. They could not, in any meaningful sense, have been "taught" its plural. What they had clearly internalised was not the lexical fact about specific nouns but the morphophonological rule that generates English plurals from arbitrary stems. The wug test, replicated across dozens of languages and decades, remains perhaps the single most economical demonstration of the constructive view: the child applies a rule to data she has never encountered, which is the operational definition of generative competence.

Cross-linguistic evidence broadens the case. Children acquiring languages as typologically distant from English as Mandarin, Inuktitut, and Hungarian go through broadly comparable developmental stages — single-word, two-word, telegraphic, complex — despite the radically different grammatical resources their target languages provide. Mandarin-speaking children, for instance, acquire the use of classifier morphemes around the same developmental window in which English-speaking children acquire the articles "a" and "the," even though classifiers and articles serve only loosely analogous functions and the relevant input is structurally incommensurable. If the acquisition mechanism were primarily imitative, we would expect the developmental trajectory to vary unpredictably with surface input. We observe instead a regularity that strongly suggests a shared underlying process — the maturation of a structure-seeking capacity that operates beneath the level of the particular features being learned.

Against this innatist position stands the older behaviourist account associated with B. F. Skinner, who argued in Verbal Behavior (1957) that language is acquired through operant conditioning — that is, through reinforcement of correct utterances and extinction of incorrect ones. Skinner's view, famously demolished by Chomsky's 1959 review, fails on two counts. First, it cannot account for virtuous errors: a child who has been positively reinforced for saying "went" should, on a strict reinforcement model, never produce "goed" at all. Second, careful observation of caregiver speech reveals that explicit correction of children's grammar is rare and, when it occurs, frequently goes unheeded. Roger Brown's oft-cited example of a mother trying to correct her son's "Nobody don't like me" eight times, only for the child to produce the same construction on the ninth attempt, illustrates the limits of reinforcement vividly.

Yet a strictly Chomskyan account is also incomplete. From the 1990s onwards, the usage-based tradition associated with Michael Tomasello has argued that children acquire grammatical constructions through fine-grained attention to communicative intention, not through the activation of an abstract innate module. On Tomasello's account, the child first masters concrete "item-based" constructions ("More milk," "Where Daddy?") and only gradually extracts the abstract grammatical patterns underlying them. The construction is real, but it is reconstructed from richer input than Chomsky allowed — input that includes joint attention, gaze direction, and the substantial social scaffolding described by Jerome Bruner as a Language Acquisition Support System (LASS).

Empirical work on the critical period — the hypothesis that language must be acquired before a developmental window closes, typically taken to be around puberty — adds a further dimension to the construction account. The famous case of Genie, the American girl isolated from linguistic input until age thirteen, demonstrated that despite intensive instruction in adulthood she never attained the morphosyntactic competence of even a young child. If grammar were simply absorbed from input, no biological deadline should exist. The critical period suggests, instead, that the construction of grammar depends on a maturation-bound neural readiness that fades once the relevant brain regions complete their developmental window.

Comparable evidence comes from studies of simultaneous bilingual acquisition. Children raised with two languages from birth — say, Cantonese at home and English at preschool — typically construct both grammars in parallel and reach native competence in each by the same age window as monolingual peers, despite receiving roughly half the input per language. A behaviourist account predicts the opposite: less input should mean slower or weaker acquisition. The empirical record tells the opposite story, which is again consistent with a model on which the child's grammatical competence is constructed from input rather than transferred from it. What the child needs is not more input but enough input to fix the parameters of the language she is building.

What follows for the original claim? The construction view is correct in its central contention: children are not taught grammar in the way they are taught arithmetic. Caregivers do not enumerate rules; they do not systematically reinforce correct forms; they do not reliably correct errors. And yet children acquire, in three to four years, a structurally rich grammar that adults take decades to approximate when learning a second language formally. The evidence is consistent only with a model on which the child is an active hypothesis-tester, equipped with cognitive and probably innate machinery for extracting rules from data.

But the strongest formulation of the claim — that grammar is constructed without environmental assistance — is too strong. The usage-based and interactionist traditions have established that the quality and quantity of caregiver input matters, that joint attention and pointing scaffold the early stages of word learning, and that children deprived of conversational partners (as in cases of severe neglect or unaddressed deafness) do not develop normally. The slogan "children are not taught, they construct" is best read as a corrective to the behaviourist view, not as a denial that input has a role.

In conclusion, the proposition deserves qualified endorsement. Grammar is not transmitted by explicit instruction; it is constructed by the child from the data she encounters, with the assistance of capacities — probably innate, certainly developmentally constrained — that antedate any specific cultural input. What caregivers provide is not a curriculum but an environment: rich, attuned, and emotionally engaged, but never didactic. To call this combination "teaching" would elide the active, hypothesis-generating role of the child; to call it "no teaching at all" would underrate the scaffolding without which the constructive process cannot proceed. The most defensible account, then, treats the child not as a passive recipient and not as a solitary architect, but as a structure-seeker whose work depends on a structured social world.