In 1957, a 28-year-old MIT linguist named Noam Chomsky published a slim monograph titled Syntactic Structures. The book argued that the behaviorist account of language acquisition — children learn to speak by reinforcement of correct utterances and extinction of incorrect ones — was radically inadequate to explain what children actually do: acquire a generative grammar that produces and parses infinitely many novel sentences they have never heard, on the basis of fragmentary input, in roughly the same developmental timeline across every culture studied. The poverty-of-the-stimulus argument implied that humans must come prewired with substantial linguistic structure — a language faculty specific to our species. Chomsky's generative grammar programme has become the dominant theoretical framework of linguistics and the foundation of cognitive science.
The language faculty is the species-specific cognitive endowment that makes natural-language acquisition possible. Every neurologically intact human acquires a natural language without explicit instruction; no other species does, despite decades of well-funded attempts on apes, dolphins, and parrots. Children across vastly different cultures hit similar developmental milestones, the input they receive is sparse and ungrammatical (the poverty of the stimulus), and the resulting grammar is rich and consistent across speakers. There is a critical period before puberty during which native competence comes easily and after which it almost never comes at all, as the case of Genie — the girl raised in isolation until thirteen — tragically demonstrated. Chomsky's theoretical positions have moved dramatically over six decades toward radical simplification. The Standard Theory of 1965 posited deep structures generated by phrase-structure rules and transformed into surface structures; Government and Binding (1981) replaced rule lists with universal principles parameterized for individual languages; the Minimalist Program from 1995 onward reduces the faculty to essentially one operation — Merge, the recursive combination of two elements into a single hierarchical structure — plus interfaces with meaning and phonology. Recursion is the core innovation that separates human language from animal communication. The neurological substrate has grown more complex than the Broca-and-Wernicke textbook schema, with imaging showing distributed left-lateralized networks; the KE family's FOXP2 mutation, which produces a heritable speech-and-grammar impairment, is the closest thing to a single-gene anchor for biological specialization.
The arrival of large language models has transformed the empirical landscape of the language-faculty debate. LLMs acquire substantial grammatical competence without any of the innate biases Chomskyan theory posits, purely from statistical learning over vast text corpora. On one reading, this vindicates the empiricist tradition. On another, it cannot settle the question, because LLMs are exposed to orders of magnitude more text than any human child (a typical 10-year-old has heard ~30 million words; GPT-4 was trained on trillions), and because LLMs do not learn the bidirectional connection between language and the sensorimotor interfaces Chomsky considers core. Chomsky himself remains sharply critical, calling LLMs plagiarism machines. Most working linguists treat the LLM era as a chance to ask which features of human language acquisition genuinely require innate scaffolding.