查看文章

Natural languages are most often characterized as a combination of rulebased generalization and lexical idiosyncrasy. The English past tense is a familiar case, in which the irregular form went replaces the expected+ ed construction* goed. Baker (1979) notes that this is a relatively benign example for learners, since irregular forms are frequently encountered in the course of their linguistic experience. The experience of the form went may block* goed, if the learner assumes that verbs typically have a single past tense form—thus, an observed alternative form can serve as evidence that an absent regular form is not allowed in the language (see, for instance, the Competition model, MacWhinney 1989). Much more troubling are cases where an apparently legal construction is idiosyncratically absent, without any alternative. The dative shift in English is a welldocumented example:(1) John gave/donated a book to the library (2) John gave/* donated the library a book In such cases we can think of linguistic rules as being quasi-regular: they license the combination of some members of syntactic categories, but not others. The difficulty of learning such idiosyncratic absences from partial input and without negative evidence (as is the case with natural language) has become notorious in the language acquisition literature. In particular, given that only a finite set of sentences is ever heard, out of the infinite set of possible sentences in a natural language, it is clear that mere absence of