29.05.03 · psychology / cognition-intelligence

Language acquisition: critical period, Chomsky's nativism vs. connectionist accounts

stub3 tiersLean: nonepending prereqs

Anchor (Master): Chomsky, N. — Aspects of the Theory of Syntax (1965)

Intuition Beginner

Every normal child learns language on a predictable schedule, without formal instruction or correction. Babies coo and babble by around six months, produce their first recognizable words near their first birthday, combine two words by about eighteen months, and speak in increasingly complex sentences between three and five years of age. This sequence unfolds across every culture and every language on the same rough timetable, even though the sounds, words, and grammatical rules being mastered differ entirely.

Noam Chomsky argued that this feat is impossible to explain by imitation alone. Children routinely produce sentences they have never heard an adult say, and they overgeneralize rules they were never taught — saying "goed" instead of "went," or "two mouses" before learning the exceptions. Chomsky's explanation is that children are born with an innate capacity for grammar, a Language Acquisition Device, that needs only ordinary exposure to activate. This is the nativist position.

Eric Lenneberg proposed the critical period hypothesis: a biological window, closing around puberty, during which a first language is acquired naturally and effortlessly. After the window closes, mastering grammar becomes far harder. The most tragic natural experiment was Genie, a child isolated and abused until age thirteen. Despite years of later intervention, she acquired vocabulary but never mastered the grammar that children normally acquire in their second year.

Visual Beginner

Figure: The language-acquisition landscape. Top: the acquisition timeline from cooing to complex sentences on a universal schedule. Left: the critical-period curve — ease of acquisition peaks in infancy and declines through puberty; Genie, exposed at thirteen, falls past the drop. Right: the two competing mechanisms — Chomsky's innate Language Acquisition Device versus domain-general statistical learning — both feeding the same child. Bottom: Saffran's word segmentation, with transitional probabilities of syllables marking word boundaries in continuous speech.

Worked example Beginner

The wug test

In 1958 Jean Berko showed a young child a drawing of a small blue bird and said, "This is a wug." She then showed two of the birds and asked, "Now there are two of them. There are two ___?" Children who had never heard the invented word "wug" answered "wugs" — adding the plural ending without hesitation.

The result matters because no adult had ever said "wugs" to these children. They were not imitating. They had extracted the English plural rule from the speech around them and applied it to a brand-new word. This productivity — generating forms you have never encountered — is the central evidence against a pure imitation account of language learning.

Check your understanding Beginner

Exercise (easy, multiple choice).

A child deprived of normal human contact and language exposure until age thirteen later learns many single words but never produces grammatical sentences. Which hypothesis does this outcome most directly support?

A. The babbling stage is innate B. The critical period hypothesis C. Universal Grammar D. Statistical learning

Hint

Ask what the timing of exposure — not the content — tells you about language learning.

Answer

Option B. The critical period hypothesis predicts that a first language must be acquired within a biologically bounded window closing near puberty. Genie's profile — vocabulary without grammar, despite intensive teaching after age thirteen — fits a window that had already closed. The other options speak to mechanism, not to the timing constraint the case demonstrates.

Formal definition Intermediate

The vocabulary of language acquisition is standardised across the anchor texts ^{[source pending]}. These terms identify empirically dissociable stages and mechanisms, not mere labels.

Acquisition stages

Child language acquisition proceeds through a robust, cross-linguistically replicated sequence. The ages are modal; the ordering is invariant.

Stage	Age	Characteristic behaviour
Cooing	0-6 months	Vowel-like vocalisations
Canonical babbling	7-10 months	Reduplicated consonant-vowel syllables ("bababa")
Variegated babbling	10-12 months	Mixed consonants and vowels
One-word (holophrastic)	12-18 months	Single words carrying whole-utterance force
Two-word	18-24 months	Minimal pivot sentences ("more milk")
Telegraphic	2-3 years	Multi-word utterances, function words sparse
Complex sentences	3-5 years	Embedded clauses, grammatical morphology mastered

Babbling is not random: canonical babbling is dominated by CV syllables drawn from a near-universal phonetic inventory, and it is produced even by deaf infants (manual babbling in sign), which marks it as a motor-cognitive milestone rather than echoic imitation ^{[source pending]}.

Chomsky's nativism

Universal Grammar (UG) is Chomsky's claim that the human capacity for language includes a set of innate principles and parameters shared by all natural languages. Principles are invariant structural constraints (e.g. structure dependence — grammatical rules operate over hierarchical phrase structure, not linear order); parameters are the dimensions along which languages vary and are fixed by exposure (head direction, pro-drop, verb-raising).

The Language Acquisition Device (LAD) is the innate faculty that takes linguistic input and outputs a grammar. It is a functional description, not an anatomical claim.

The poverty of the stimulus argument holds that the input children receive — fragmented, error-ridden, and finite — is too thin to determine the grammar they acquire, unless supplemented by innate structure. The textbook illustrations are structure dependence (children never entertain linear-order hypotheses such as "invert the first auxiliary you encounter") and the production of novel well-formed sentences. If the child knows more than the input could teach, the surplus must be innate ^{[source pending]}.

Parameter setting models acquisition as the selection among pre-existing options rather than induction from scratch. Exposure to head-initial English sets the head-direction parameter; exposure to pro-drop Italian sets the null-subject parameter. The mechanism is fast and uniform across children precisely because the hard structural work is already specified.

The critical period

Eric Lenneberg's critical period hypothesis ^{[source pending]} proposes that there is a biologically bounded window — opening in infancy and closing near puberty — during which a first language is acquired naturally and without explicit instruction. After the window closes, the capacity for natural acquisition declines sharply.

The converging evidence is substantial. The Genie case (Curtiss 1977) — a child deprived of normal interaction until age thirteen — yielded vocabulary acquisition but no productive grammar, the profile predicted by a closed critical window. Deaf children with delayed exposure to sign language (Newport 1990) show that early learners achieve native-like syntax while late learners plateau regardless of years of subsequent use. Second-language learning shows the same gradient: performance declines as a function of age of arrival, with a sharp inflection near puberty (Johnson & Newport 1989). On the input side, Patricia Kuhl's work on perceptual narrowing shows that infants under twelve months discriminate phonetic contrasts across all languages, but by twelve months they tune to their native language's contrasts and lose sensitivity to foreign ones — evidence that the perceptual system itself is shaped within an early window ^{[source pending]}.

Statistical learning

Jenny Saffran and colleagues ^{[source pending]} demonstrated that eight-month-old infants track the transitional probabilities between syllables in continuous speech. The probability that syllable B follows syllable A, written $P (B ∣ A)$ , is higher within words than across word boundaries. Infants segment the speech stream at the local minima of transitional probability, recovering word-like units from two minutes of unbroken, unpunctuated syllables. This finding established that a powerful distributional-learning mechanism is available in infancy and operates without meaning, feedback, or social cues.

Alternative accounts

Three non-nativist frameworks challenge or complement Universal Grammar.

Connectionist (neural-network) models demonstrate that simple recurrent networks can learn tasks — past-tense formation, reading aloud — that nativists had argued require symbolic rules. The past-tense debate (Rumelhart & McClelland 1986; Pinker & Prince 1988) is the locus classicus: a network that learns regular and irregular mappings from input alone, without a pre-installed rule component.

Usage-based / construction grammar (Michael Tomasello) holds that children build grammar bottom-up from concrete, item-based constructions that they gradually generalise by analogy. There is no Universal Grammar; there are general-purpose cognition (intention-reading, pattern-finding, statistical learning) and a long developmental accumulation of constructions ^{[source pending]}.

The social-pragmatic account (Jerome Bruner) stresses that acquisition unfolds in interaction. Joint attention, caretaker scaffolding, and the LASS (Language Acquisition Support System) format the input so that the child can extract structure. Bruner's point is not that social interaction replaces learning mechanisms but that the input is structured by interaction in ways nativist models ignore ^{[source pending]}.

Key mechanism Intermediate

How the child gets from input to grammar

The central mechanistic question is what transforms the stream of speech a child hears into a productive grammatical system by age five. Two families of answer have dominated for sixty years, and they differ on whether the child brings dedicated linguistic structure to the task.

The nativist mechanism. On Chomsky's account, the child is born with Universal Grammar — a set of principles and parameters common to all human languages ^{[source pending]}. Acquisition is not induction from scratch but the setting of a small number of binary parameters on the basis of minimal input. The mechanism is fast and uniform across children because the structural work is already done; exposure selects among pre-existing options. The engine is the LAD, which takes linguistic input and outputs a grammar. The key predictions are universality (all human grammars share core properties), uniformity (children acquire on the same schedule everywhere), and early productivity (children produce forms they have never heard). The poverty-of-stimulus argument is the load-bearing empirical premise: if the input cannot determine the output, the missing structure must be supplied innately.

The empiricist mechanism. On the connectionist and usage-based accounts, the child brings no dedicated linguistic machinery — only domain-general learning: distributional statistics, analogy, chunking, and intention-reading ^{[source pending]}. Grammar is constructed from patterns in the input, not triggered. Saffran's statistical learning shows that the raw distributional machinery is powerful enough to segment speech in infancy. Connectionist models show that rule-like behaviour can emerge from weighted associations without explicit rules. Tomasello's construction grammar shows how item-based patterns generalise into abstract constructions through graded analogy. The mechanism is slower, more error-prone, and more variable across children than the nativist story predicts — which the empiricists count as evidence for their view, since real acquisition is gradual, errorful, and variable.

The evidence each side cites. Nativists cite the wug test, overregularisation, structure dependence, the critical period, and cross-linguistic universals — all of which, they argue, require innate structure to explain. Empiricists cite Saffran's statistical learning, connectionist simulations, the gradual and item-based character of real acquisition, and cross-linguistic variation in the order and form of early constructions — all of which, they argue, show that general-purpose learning suffices and dedicated innate grammar is unnecessary.

Current state of the debate. The poverty-of-stimulus argument has been challenged empirically (Pullum & Scholz 2002 — the input is richer and more regular than nativists assumed) and computationally (Bayesian models can acquire substantial structure from realistic input). The critical-period evidence supports a biological constraint but does not, by itself, decide between a dedicated language faculty and a general-learning account whose window closes for other maturational reasons. The mature view, held by most working researchers, is hybrid: children bring both biases (some plausibly language-specific, such as sensitivity to hierarchical structure) and powerful general learning, and the open question is the division of labour between them, not the victory of one side.

Exercises Intermediate

Exercise 1 (medium, multiple choice).

Children learning English form yes/no questions by inverting the subject's auxiliary, not the first auxiliary in the sentence. Given "The boy who is crying is hungry," they ask "Is the boy who is crying hungry?" rather than "Is the boy who crying is hungry?" This is the textbook case for which nativist claim?

A. Parameter setting B. Structure dependence C. The critical period D. Pro-drop

Hint

The child ignores linear position (the first "is") and operates on grammatical structure (the main-clause auxiliary). What does that reveal about the form of the rule?

Answer

Option B. The rule children apply is structure-dependent — it refers to phrase structure, not to linear order. A purely statistical learner exposed only to surface strings has no obvious basis to prefer the structural hypothesis over the linear one, and the structural hypothesis is never entertained and rejected; it is present from the start. Chomsky takes this as direct evidence for innate structural principles.

Exercise 2 (medium, multiple choice).

In Saffran's word-segmentation experiment, eight-month-old infants heard two minutes of continuous syllables with no pauses, stress, or meaning. Afterward they preferred to listen to novel three-syllable sequences with low rather than high internal transitional probability. This demonstrates:

A. That infants require social interaction to learn B. That infants track transitional probabilities between syllables C. That Universal Grammar is present at eight months D. That the critical period has already closed

Answer

Option B. Infants treated high-transitional-probability sequences as familiar (habituated) and low-transitional-probability sequences as novel (preferred, i.e. longer looking). This shows they extracted statistical structure — the conditional probabilities between adjacent syllables — from unsegmented input, without meaning, feedback, or social cues. The result established a powerful domain-general learning mechanism available in early infancy.

Exercise 3 (medium, short answer).

Newport (1990) studied deaf signers of ASL who were first exposed to sign at different ages. Early learners (exposed from infancy) achieved native-like syntax; late learners (first exposed after puberty) plateaued regardless of decades of subsequent use, and the decline was graded with age of first exposure. In 3-4 sentences, explain why this supports a critical period rather than merely a "practice makes perfect" account.

Hint

If the only variable were total practice time, what would you expect for late learners who have signed for decades?

Answer

A pure practice account predicts that more years of use should produce more native-like performance, regardless of when exposure began. Newport's late learners had decades of daily ASL use yet never reached native-like syntax, so cumulative practice cannot be the operative variable. What predicted performance was age of first exposure, with a graded decline through childhood and a sharp inflection near puberty. That signature — sensitivity to timing rather than to total input — is the defining mark of a biologically bounded critical period.

Exercise 4 (hard, comparison).

Contrast the nativist and the usage-based (Tomasello) accounts of how a child acquires the English past tense. In 4-5 sentences, state what each predicts about (a) the order of acquisition of regular versus irregular forms, (b) the existence and shape of overregularisation errors, and (c) what kind of evidence would count against each view.

Hint

The two accounts differ on whether the regular rule exists before exposure (nativist) or is abstracted from patterns after exposure (usage-based). Trace what each implies for errors and for dissociation under atypical input.

Answer

The nativist (dual-route) account holds that regular past tense is generated by a symbolic rule applied to any stem, while irregulars are retrieved from an associative memory. It predicts that once the rule is acquired it applies productively (hence overregularisation of irregulars like "goed"), and that regulars and irregulars can dissociate under neural insult or developmental disorder. Evidence against it would be a failure to find such a dissociation, or a single connectionist network that reproduces the full error profile.

The usage-based (single-route) account holds that both regular and irregular forms are learned by the same statistical/analogy mechanism operating over stored exemplars. It predicts graded, frequency-dependent performance on both regulars and irregulars, and that overregularisation reflects the growing weight of the regular pattern rather than the activation of an innate rule. Evidence against it would be a clean dissociation — selectively preserved regular inflection when exemplar memory is destroyed — which the neuropsychological literature (aphasia, SLI) partially provides. The debate remains open because each model accommodates the other's signature findings with added assumptions.

Advanced results Master

The Minimalist Program: merge and recursion

Chomsky's later work — the Minimalist Program (1995 onward) — recasts Universal Grammar around a single combinatorial operation, merge. Merge takes two syntactic objects and combines them into a new set: merge(X, Y) = {X, Y}. Applied recursively, merge generates unbounded hierarchical structure, which is the formal property distinguishing human language from animal communication systems. External merge combines distinct objects (building phrases); set-merge and agree handle displacement phenomena such as movement and feature checking ^{[source pending]}.

The empirically bold claim is Hauser, Chomsky & Fitch (2002): that recursion is the only computational operation unique to the human language faculty, and that everything else — the sensory-motor interface, the conceptual-intentional interface, much of the lexicon — is shared with other animals or derivable from general cognition. This "thin" version of UG narrows the innate endowment to a single operation and shifts the explanatory burden: if recursion alone is species-unique, the vast surface diversity of languages must fall out from how recursion interfaces with other systems, not from a rich set of innate parameters. The claim is contested (Pinker & Jackendoff 2005 argue that the language faculty is substantially broader than recursion alone), but it sets the terms of the modern debate ^{[source pending]}.

FOXP2 and the KE family

The KE family (Hurst 1990) is a three-generation pedigree in which about half the members present a severe speech and language disorder inherited in a Mendelian pattern. The gene was mapped to a point mutation in FOXP2 (Lai et al. 2001) on chromosome 7. FOXP2 encodes a transcription factor — a regulatory protein that switches other genes on and off during neural development — not a "grammar gene." The affected members show deficits in sequencing orofacial movements and in morphosyntax, but the disorder is broader than language alone, affecting motor coordination and certain nonlinguistic sequencing tasks ^{[source pending]}.

The significance is double-edged. On one hand, FOXP2 is the first identified gene whose disruption specifically impairs language, lending credibility to a biological substrate for the language faculty. On the other hand, FOXP2 is conserved across mammals and is regulatory rather than language-specific, which fits a picture in which language builds on general neural-development machinery rather than on a dedicated genetic module. The gene has been under positive selection in the human lineage since the split from chimpanzees, but the same variant is found in Neanderthals, complicating any simple story linking it to the emergence of modern language.

Critical-period evidence in detail

Three lines of evidence pin down the critical period with increasing precision.

Newport (1990), deaf children and ASL. Deaf signers first exposed to ASL at different ages provide the cleanest test, because the confound of hearing-language input is absent. Early learners (exposed by age three) achieve native-like syntax; those first exposed after puberty plateau at a non-native asymptote regardless of decades of daily use. The decline is continuous with age of first exposure through childhood — a maturational gradient rather than a sharp cutoff — which has led most researchers to speak of a sensitive period rather than a hard critical window ^{[source pending]}.

Johnson & Newport (1989), late second-language learning. Native Korean and Chinese speakers who immigrated to the United States at different ages were tested on English grammar. Performance declined linearly with age of arrival through childhood and then levelled off at a lower, highly variable plateau after puberty. The pre-puberty decline is the critical-period signature; the post-puberty plateau reflects individual differences in motivation, aptitude, and input that the closed window no longer fully compensates.

Kuhl, perceptual narrowing and native-language attunement. Infants under six months discriminate phonetic contrasts from languages they have never heard; by twelve months they have lost sensitivity to non-native contrasts and retained only their native language's distinctions. This perceptual magnet effect — native prototypes pull perceptual judgments toward category centres — shows that the phonetic perceptual system itself is restructured within the first year, long before word production begins. The critical period thus operates earliest at the phonetic level and later at the syntactic level, consistent with a cascade of overlapping sensitive windows rather than a single global window ^{[source pending]}.

The poverty-of-stimulus debate

The poverty-of-stimulus argument is the linchpin of nativism, and it has been the focus of the sharpest empirical and computational challenge.

Pullum & Scholz (2002) re-examined the canonical examples — auxiliary inversion, structure dependence — and argued that the actual input children receive is richer, more structurally regular, and more directly informative than nativists typically assume. If the relevant evidence is present in the input, the stimulus is not poor, and the inference to innate structure is undermotivated ^{[source pending]}.

Bayesian learning models (Pearl, Goldwater, and others) show that rational statistical learners with realistic prior distributions can acquire substantial grammatical structure from child-directed speech — including the structure-dependence that nativists had claimed requires innate principles. The priors in these models are themselves a form of innate bias, so Bayesian work does not eliminate innateness, but it relocates it from a rich UG to a smaller set of general-purpose learning priors. The division of labour between dedicated linguistic bias and general statistical learning is exactly the open empirical question ^{[source pending]}.

Computational grammar induction remains the acid test. No current system acquires a human-like grammar from raw child-directed speech with no built-in linguistic bias, which nativists count as evidence that the bias is necessary. Connectionists count the steady progress of these models as evidence that the required bias is shrinking. Both readings are defensible; the matter is unresolved.

Cross-linguistic variation

The nativist prediction that all languages share a small set of core properties is tested against the diversity of actual grammars. Several phenomena stress-test the claim.

Differential object marking. In Spanish and Hebrew, animate or specific direct objects receive an overt marker ("a" in Spanish) that inanimate or non-specific objects omit. The marking is governed by animacy and definiteness hierarchies that recur across unrelated languages, suggesting a shared principle even where the surface morphology differs.

Ergative languages (Basque, many Australian and Mayan languages) align their core grammatical relations differently from the nominative-accusative pattern familiar from Indo-European: the subject of an intransitive groups with the object of a transitive (the absolutive), not with the transitive subject (the ergative). Children acquire ergative alignment as readily as accusative alignment, which constrains claims about which alignments are "natural" to the language faculty.

Classifier languages (Mandarin, Japanese) lack grammatical number on nouns and instead use classifiers — nominal classifiers, verbal classifiers — selected by semantic class. The obligatory expression of number in English and its near-absence in Mandarin show that even basic grammaticised categories vary substantially, while classifier semantics recur across unrelated languages.

The empirical upshot is that languages vary more in surface form than early generative grammar predicted, but that the variation is constrained — it is not arbitrary. Whether the constraints come from UG or from general cognition (working memory, categorisation, communicative function) is contested.

Linguistic relativity revisited

The Sapir-Whorf hypothesis — that the language one speaks shapes the way one thinks — has been tested with increasing rigour since the 1990s. The strong ("linguistic determinism") version is false: no language renders its speakers unable to think a thought expressible in another language. The weak ("linguistic relativity") version is empirically supported in restricted domains.

Colour terms. Languages with different basic colour-term inventories (Russian distinguishes two blues, "siniy" and "goluboy", that English collapses) produce correspondingly faster categorical discrimination at the category boundary — even in pre-attentive, perceptual tasks (Winawer et al. 2007).

Spatial frames of reference. Some languages (Guugu Yimithirr, Tzeltal) use an absolute (geocentric) frame of reference exclusively, where English speakers use a relative (egocentric) frame. Speakers of absolute-frame languages encode direction in memory and reasoning in ways English speakers do not, with measurable effects on nonlinguistic spatial memory ^{[source pending]}.

Number systems. Languages without exact number words above small integers (Piraha, Munduruku) constrain their speakers' performance on exact-quantity tasks while leaving approximate numerical cognition intact — supporting a specific, language-mediated effect on exact arithmetic (Gordon 2004; Pica et al. 2004).

The mature view is that language biases thought in measurable but bounded ways, largely in domains that are themselves linguistic or that recruit language online (memory, categorisation). It does not determine thought.

The nontrivial upshot

Language acquisition is the cleanest case in psychology of a complex, structured capacity emerging rapidly, uniformly, and without instruction. Three findings survive every theoretical re-analysis: children acquire grammar on a universal schedule; a biologically bounded window governs the ease of acquisition; and no current model — nativist, connectionist, or Bayesian — fully accounts for the trajectory. The honest conclusion is that the child brings both substantial innate structure and powerful general learning, and that the central open problem is not whether either exists but how they divide the work. The debate between Chomsky and the empiricists has narrowed from a question of whether there is innate structure to a question of how much and what kind — a genuine, if incremental, scientific advance.

Connections Master

Cognition and intelligence 29.05.01 is the direct prerequisite and parent unit. The treatment of language here presupposes the modular-versus-general-processing framing introduced there, and language acquisition is the test case on which the modularity debate is most fully fought out.
Decision-making 29.05.02 pending connects through framing: the wording of a prospect sets its reference point, and the heuristics-and-biases program supplies the empirical foundation for how linguistic framing shapes judgment. The two units share the broader question of how internal representations interact with external symbolic structure.
Neuroscience [29.02.NN] (pending) supplies the substrate. Broca's and Wernicke's areas, the FOXP2 regulatory network, and the left-perisylvian language network are the neural realisations of the capacity whose acquisition this unit describes, and the critical-period evidence is itself a maturational claim about the brain.
Learning and memory 29.04.01 supplies the general-learning machinery — classical and operant conditioning, implicit statistical learning, and the memory systems — that the empiricist accounts of acquisition invoke. Saffran's transitional-probability learning is a species of the implicit learning catalogued there.
Developmental psychology [29.06.NN] (pending) is the natural home of the acquisition timeline. Attachment, joint attention, and the caretaker-infant interaction format (Bruner's LASS) are developmental phenomena that structure the input the child learns from.
Statistical reasoning 29.01.03 pending supplies the formalism for statistical learning. Transitional probabilities are conditional probabilities, and the Bayesian models that challenge poverty-of-stimulus are applications of the Bayesian framework formalised there.
Cross-cultural and indigenous psychology 29.12.01 connects through linguistic relativity and cross-linguistic variation. The Sapir-Whorf evidence, classifier systems, and absolute spatial frames of reference are cross-cultural phenomena, and the universality claims of nativism must hold against the full range of human languages catalogued there.

Historical & philosophical context Master

The modern study of language acquisition begins with Noam Chomsky's 1959 review of B. F. Skinner's Verbal Behavior. Skinner had offered the most ambitious behaviorist account of language: children learn language the way they learn any behavior — through reinforcement, imitation, and stimulus-response association. Chomsky's review demolished the behaviorist account on its own terms, arguing that it could not explain productivity (novel sentences), structure dependence, or the speed and uniformity of acquisition. The review is among the most influential papers in the cognitive sciences, not because it settled how language is acquired but because it removed the dominant empiricist framework and made innate structure the default hypothesis ^{[source pending]}.

The nativism Chomsky proposed has deep roots in the rationalist tradition — Descartes's innate ideas, Leibniz's "innate" dispositions, the Port-Royal grammar — against the empiricist line running from Locke through Skinner. The poverty-of-stimulus argument is a modern restatement of a rationalist claim: the mind contributes structure that experience alone cannot supply. This places language acquisition at the centre of a centuries-long philosophical argument about the relative contributions of nature and nurture, and it explains why the empirical details of child language have borne such disproportionate theoretical weight.

Eric Lenneberg's Biological Foundations of Language (1967) supplied the second pillar, the critical period, by reframing language as a biologically timed maturational process analogous to other developmental milestones. The Genie case (discovered 1970, reported by Curtiss 1977) provided the tragic natural experiment that made the hypothesis vivid, though its evidential value is debated — a single case with confounds of abuse and deprivation cannot, by itself, establish a critical window. The converging evidence from deaf signers (Newport) and second-language learners (Johnson & Newport) did the heavier lifting.

The empiricist counter-revival began in the 1980s. Rumelhart and McClelland's (1986) connectionist model of past-tense learning showed that a network with no explicit rules could reproduce the U-shaped error profile of children, provoking Pinker and Prince's (1988) defence of the dual-route symbolic account. Jenny Saffran's (1996) statistical-learning experiments shifted the ground further by showing that infants extract distributional structure from speech with extraordinary efficiency, lending credibility to general-learning accounts. Michael Tomasello's Constructing a Language (2003) consolidated the usage-based alternative: children build grammar from concrete, item-based constructions by analogy and intention-reading, with no dedicated innate grammar.

The current state is less polarised than the founding rhetoric. Few researchers now defend either a rich Universal Grammar or a pure blank-slate empiricism. The live questions are quantitative — how much innate structure, of what kind, interacting with which learning mechanisms — and they are addressed by computational modelling, infant experimentation, and cross-linguistic comparison rather than by philosophical argument. The philosophical weight remains, because the answer bears on what kind of cognitive animal we are, but the science has moved the disagreement onto terrain where evidence can, in principle, decide it.

Bibliography Master

Chomsky, N. — "A review of B. F. Skinner's Verbal Behavior", Language 35, 26-65 (1959).
Chomsky, N. — Aspects of the Theory of Syntax (MIT Press, 1965).
Chomsky, N. — The Minimalist Program (MIT Press, 1995).
Lenneberg, E. H. — Biological Foundations of Language (Wiley, 1967).
Berko, J. — "The child's learning of English morphology", Word 14, 150-177 (1958).
Curtiss, S. — Genie: A Psycholinguistic Study of a Modern-Day "Wild Child" (Academic Press, 1977).
Saffran, J. R., Aslin, R. N. & Newport, E. L. — "Statistical learning by 8-month-old infants", Science 274, 1926-1928 (1996).
Johnson, J. S. & Newport, E. L. — "Critical period effects in second language learning: the influence of maturational state on the acquisition of English as a second language", Cognitive Psychology 21, 60-99 (1989).
Newport, E. L. — "Maturational constraints on language learning", Cognitive Science 14, 11-28 (1990).
Tomasello, M. — Constructing a Language: A Usage-Based Theory of Language Acquisition (Harvard University Press, 2003).
Bruner, J. S. — Child's Talk: Learning to Use Language (Norton, 1983).
Hauser, M. D., Chomsky, N. & Fitch, W. T. — "The faculty of language: what is it, who has it, and how did it evolve?", Science 298, 1569-1579 (2002).
Pinker, S. & Jackendoff, R. — "The faculty of language: what's special about it?", Cognition 95, 201-236 (2005).
Pullum, G. K. & Scholz, B. C. — "Empirical assessment of poverty of stimulus arguments", Theoretical Linguistics 28, 9-50 (2002).
Rumelhart, D. E. & McClelland, J. L. — "On learning the past tenses of English verbs", in Parallel Distributed Processing, vol. 2 (MIT Press, 1986).
Pinker, S. & Prince, A. — "On language and connectionism: analysis of a parallel distributed processing model of language acquisition", Cognition 28, 73-193 (1988).
Kuhl, P. K. — "Early language acquisition: cracking the speech code", Nature Reviews Neuroscience 5, 831-843 (2004).
Lai, C. S. L., Fisher, S. E., Hurst, J. A., Vargha-Khadem, F. & Monaco, A. P. — "A forkhead-domain gene is mutated in a severe speech and language disorder", Nature 413, 465-466 (2001).
Pinker, S. — The Language Instinct (HarperCollins, 1994).
Winawer, J., Witthoft, N., Frank, M. C., Wu, L., Wade, A. R. & Boroditsky, L. — "Russian blues reveal effects of language on color discrimination", Proceedings of the National Academy of Sciences 104, 7780-7785 (2007).
Gordon, P. — "Numerical cognition without words: evidence from Amazonia", Science 306, 496-499 (2004).
Myers, D. G. & DeWall, C. N. — Psychology, 13th ed. (Worth, 2021), Ch. 9.
Gleitman, H., Gross, J. & Reisberg, D. — Psychology, 8th ed. (Norton, 2011), Ch. 9.