ABSTRACT
Languages like generalizations and children like generalizations, a perfect harmony, but of course not an accidental one. An obvious solution to the overgeneralization problem is to deny that children are limited to positive information about the target language. Every theory must assume that humans are innately equipped with a mental 'metalanguage' for natural language. A purely formal theory of innate linguistic knowledge assumes that this metalanguage is all that is innately specified, and that this is sufficient, in interaction with the learner's data, to constrain severely the set of grammars that he will contemplate. Explicit constraints on grammatical structures or derivations are postulated by Government Binding Theory, by Lexical Functional Grammar, and by Generalized Phrase Structure Grammar. The formal/substantive distinction is relevant to the puzzle about the extent of generalization in acquisition because the innate constraints that are commonly invoked as an antidote to the learner's tendency to hypothesize overgeneral rules can be shown to constitute substantive knowledge.
