Abstract Learning to speak and understand a language is probably one of the most complicated tasks we ever do. Yet this learning task is regularly accomplished by very young children. Three fundamental components of this task structure our review: discovering the words (and other linguistic forms), entertaining concepts of the kind expressed by those linguistic forms, and mapping linguistic elements onto their meanings. First, children must analyze the stream of speech to find the particular linguistic forms used in their community. We review evidence that listeners, from infancy to adulthood, are exquisitely sensitive to the details and the statistics of their experience with spoken language. Second, children must be able to entertain concepts of the kind that are expressed in languages. Here we recruit evidence from the field of infant cognition to argue that pre‐linguistic infants form object categories and factor their knowledge of events into objects (e.g., cup, ball) and relations between them (e.g., containment). These structured concepts constitute the semantic building blocks of language. Striking evidence that usefully‐structured concepts precede language learning comes from the invented communication systems (Home Sign) of linguistically isolated deaf children. Third, children must accomplish a specific mapping between linguistic elements and the ideas they evoke. We propose that children learn a core vocabulary of concrete words (including many names for things) by observing which words systematically occur with particular elements in the world. This is no simple process—it depends on every aspect of the child's developing conceptual apparatus and nascent understanding of human intentions. This initial vocabulary, in turn, boosts the child into a better position to learn other words, and to begin to work out the phrase structure of the language. Even very young children use partial knowledge of syntax to guide sentence interpretation, and therefore to learn new words.
Stored only when both papers are in the corpus — 67 citations exist upstream.
Imported from OpenAlex (CC0) — nothing here is generated.