Language Acquisition: Critical Periods, Grammar, and Statistics

Children Learn Grammar Nobody Explicitly Teaches Them

By age four, the average English-speaking child produces and comprehends sentences they have never heard before — applying rules of morphology, syntax, and agreement that no adult has explicitly explained to them. A four-year-old who says "I goed to the store" (applying the regular past-tense rule to an irregular verb) demonstrates knowledge of the rule's existence, not ignorance of language. This productive, generative use of implicit grammatical knowledge sits at the center of one of cognitive science's deepest debates: is language learning a domain-specific biological endowment, or the product of domain-general learning mechanisms operating on statistical patterns in the linguistic environment?

The Critical Period Hypothesis

Eric Lenneberg, a Harvard linguist and neurologist, proposed the critical period hypothesis for language acquisition in his 1967 book Biological Foundations of Language. His argument: like birdsong acquisition in songbirds or visual cortex development in mammals, human language acquisition depends on a biologically constrained sensitive period tied to neural development. The critical period for native-level language acquisition extends from birth to approximately puberty (he suggested around age 13–14), after which the brain's language-learning plasticity declines sharply. Evidence Lenneberg cited included recovery patterns from childhood versus adult aphasia (children recover more completely), the observation that second language learners who begin after puberty rarely achieve native-level phonological fluency, and the maturational timing of the plasticity for other biological systems.

The most dramatic natural evidence for the critical period hypothesis came from a case study that remains both scientifically significant and deeply troubling. Genie — a pseudonym — was discovered in 1970 in Los Angeles at age 13, having been subjected to severe abuse and social isolation since infancy, deprived of nearly all language exposure. Despite intensive language therapy beginning immediately after discovery, Genie never acquired grammatically structured language. She developed vocabulary, could communicate needs, and showed some syntactic understanding, but her production remained agrammatic (noun-phrase strings without sentence structure) and never progressed to the level of a typically developing four-year-old. Susan Curtiss, who conducted Genie's longitudinal language assessment as a UCLA doctoral student, interpreted the findings as consistent with the critical period hypothesis — natural language acquisition capacity had expired before Genie received linguistic input.

• Alternative interpretations of Genie's case: severe early trauma and malnutrition may have impaired neural development independently of the critical period mechanism; Genie may have been intellectually disabled prior to isolation
• Second language acquisition studies (Flege, Birdsong, MacWhinney) show gradual rather than sharp discontinuities in acquisition outcomes with age of first exposure, suggesting a sensitive period with variable boundaries rather than a hard cut-off
• Phonological acquisition shows stronger critical period effects than lexical or syntactic acquisition — adult second-language learners can achieve native-level syntax and vocabulary but rarely native-level phonology for late-acquired languages
• Sign language studies with deaf individuals who received late access to a natural sign language show similar critical period patterns to spoken language, suggesting the critical period applies to language as a system, not specifically to auditory-oral modality

Chomsky's Universal Grammar vs. Tomasello's Usage-Based Grammar

Noam Chomsky's generative linguistics, developed from the 1950s onward, holds that the speed and success of language acquisition — despite the impoverished and error-filled input children receive — requires an innate, domain-specific language acquisition device (LAD) containing Universal Grammar (UG): abstract principles and parameters shared across all human languages, biologically specified in the genome. The poverty of the stimulus argument asserts that children know things about their language (e.g., they correctly apply island constraints to wh-movement without ever hearing examples of violations) that could not have been learned from input alone. Language acquisition, on this view, is less learning than parameterization of innate structure.

Michael Tomasello, then at the Max Planck Institute for Evolutionary Anthropology, developed an alternative framework from the 1990s onward: usage-based construction grammar. Tomasello's account holds that children begin with specific, lexically filled constructions (learned from direct experience with individual utterances) and gradually abstract across these instances to form more schematic grammatical representations. The mechanisms are domain-general: intention reading (understanding that utterances are communicative acts), pattern recognition, and analogical generalization. On this account, no domain-specific innate grammar is required — statistical regularities in the input, combined with general-purpose learning mechanisms, are sufficient.

The debate remains unresolved. Evidence for domain-specificity: certain grammatical universals appear across unrelated languages, and specific language impairment (SLI) dissociates language learning from general intelligence. Evidence for usage-based learning: children's early grammar is more item-specific than UG accounts predict, and computer models trained on child-directed speech acquire aspects of grammar without innate linguistic structure.

Statistical Learning: Saffran 1996

Jenny Saffran, Richard Aslin, and Elissa Newport's 1996 Science paper, "Statistical Learning by 8-Month-Old Infants," demonstrated that infants can extract statistical regularities from continuous speech streams without any explicit instruction or feedback. Infants listened to two minutes of a synthesized speech stream containing nonsense words (three-syllable sequences like "bidaku," "padoti") strung together without pauses or prosodic cues. The only information marking word boundaries was transitional probability — the probability that one syllable followed another. Syllables within words followed each other with high probability (1.0 within "bidaku"); syllables crossing word boundaries had lower probability (0.33). After only two minutes of exposure, infants listened longer to sequences of syllables that straddled word boundaries ("part-words") than to sequences corresponding to actual words — demonstrating they had implicitly extracted the statistical structure.

This finding established that the early stages of language acquisition — specifically, segmenting continuous speech into word-like units — could be accomplished by statistical learning mechanisms without language-specific innate knowledge. Subsequent research extended statistical learning to tone sequences (music), visual sequences, and abstract grammatical patterns, suggesting it is a domain-general mechanism that makes a substantial contribution to early language acquisition.

The Bilingualism Cognitive Reserve Debate

Ellen Bialystok, Fergus Craik, and colleagues published a landmark 2007 paper in Neuropsychologia arguing that lifelong bilingualism delays Alzheimer's disease symptom onset by approximately 4.1 years compared to matched monolingual controls. The proposed mechanism: the executive function demands of constantly managing two languages — selectively activating one while inhibiting the other — constitutes a form of cognitive exercise that builds reserve capacity against neurodegeneration. This "bilingual advantage" hypothesis generated enormous interest and drove policy discussions about multilingual education.

A 2018 Neuropsychologia meta-analysis by de Bruin, Treccani, and Della Sala found evidence of publication bias in the bilingualism advantage literature — studies finding no advantage were less likely to be published. The current scientific consensus has shifted substantially: bilingualism may contribute to cognitive reserve under some conditions for some populations, but the strong claims of a 4-year Alzheimer's delay and consistent executive function advantages across populations are not supported by the full literature. The experience of managing two languages is cognitively enriching; the magnitude of its protective effect is smaller and more variable than initially claimed.