Usage-based theory is the account of language acquisition that treats grammar as an emergent product of language use rather than the unfolding of an innate biological endowment. Most closely associated with Michael Tomasello in developmental psycholinguistics and with Joan Bybee, Nick Ellis, and Adele Goldberg in linguistics, the position offers a direct alternative to nativism and Universal Grammar. Its core slogans, drawn from Tomasello, are that meaning is use and structure emerges from use.
Tomasello (2003) argues that the human child brings two general cognitive capacities to the acquisition task, neither of them language-specific:
| Capacity | What it does |
|---|---|
| Intention-reading | Inferring the communicative goals of others, including joint attention to objects and events |
| Pattern-finding | Statistical and analogical detection of recurring form-function pairings across utterances |
Intention-reading explains how the child links sound patterns to meaning. Pattern-finding explains how those item-based pairings gradually generalise into productive constructions. Neither mechanism requires innate grammatical knowledge.
In the usage-based account, children do not start with adult-like grammatical categories and combine them productively. They start with concrete utterance schemas tied to specific verbs ("X kicks Y", "X sees Y"), then gradually abstract over them as similar patterns accumulate. Productivity is graded and slow. A child might use "go" creatively long before generalising the same flexibility to "depart". The fundamental linguistic unit in this view is the construction: a learned pairing of form and meaning at any level of complexity, from morphemes to idioms to fully abstract syntactic templates. This claim is shared with construction grammar in linguistics and gives the framework its theoretical backbone.
Usage-based SLA, developed especially by Nick Ellis and Diane Larsen-Freeman, takes the same principles and applies them to second-language learning. Several claims follow directly:
The connectionist computational tradition supplies many of the formal models that operationalise these claims. Networks trained on realistic input distributions reproduce many features of L1 and L2 acquisition without any innate grammatical structure.
Usage-based theory rejects three commitments inherited from generative linguistics:
The theory replaces these with a unified view in which what is learned, how it is learned, and what is used are governed by the same domain-general principles.
Geoff Jordan, writing in 2026, packages the standing generative-leaning objections in three points. First, the storage-and-retrieval challenge: storing hundreds of thousands of exemplars and comparing them in real time strains plausible models of short-term and long-term memory. Second, the analogy-selection problem: without prior structure, an exemplar system has no principled way to decide which features matter for generalisation, and so cannot avoid spurious similarities. Third, the poverty-of-the-stimulus gap: usage-based theories give no satisfactory account of how learners reliably acquire abstract constraints that are under-represented or absent in input, why low-frequency or irregular forms are sometimes acquired early, or why some high-frequency patterns never fully consolidate. Jordan's preferred alternative is Carroll's Autonomous Induction Theory, and his complaint is that emergentist accounts trade one explanatory gap (where rules come from) for another (where systematic generalisation comes from).
Usage-based theory undergirds a recognisable cluster of pedagogical commitments: extensive input over explicit rule teaching, attention to meaningful constructions over isolated grammar points, the importance of exposure to varied exemplars to drive abstraction, and scepticism about teaching grammar as an autonomous system separated from its functions. Lexical and pattern-based approaches to teaching, including the lexical approach and corpus-informed materials, draw their justification largely from this framework.