The purpose of this note is to evaluate critically the early-formal-learning approach in the light of recent studies of the acquisition of gender systems in different languages, and the development of the mass/count distinction in English. The issue can be stated briefly. A recurring theme in theoretical writings in child language has been the question of the development of formal linguistic systems (eg Braine 1976, Gleitman 1981, Maratsos 1981, in press, Schlesinger 1982, Pinker 1984). The question is: can and do young children in the early phases of language learning attend to formal linguistic regularities and construct grammatical systems on the basis of formal cues, or are they predominantly dependent in their early learning upon the detection of correspondences between meaning and form? To use a well-known phrase, do young children treat language as a'problem space per se'(Karmiloff-Smith 1979)? That is, to what extent do they approach grammar as an independent system that relies on its own internal coherence? The crucial question has to do with the phase when formal learning begins, for it is rarely contested that adult grammars make use of arbitrary systems. In Slobin's words (1986), the question is what serves as' the opening wedge'for the acquisition of grammar—whether it is primarily content elements for which the child seeks verbal expressions, or whether he or she attends to formal distributional features in the verbal input as well, and uses them as clues to the discovery of grammatical elements. This question is not the same as the'bootstrapping'issue that is discussed in Grimshaw (1981) and Pinker (1984). Pinker rejects the possibility that a child's grammatical categories may be semantically defined. In his model, correspondences between meanings and ways of expressing them serve to'bootstrap'innately specified grammatical categories. Thus, the name of a person or a thing triggers the category noun; time of event triggers tense; action or change of state triggers verb etc.(Pinker 1984; 41). Since, in principle, a single instance of a given correspondence is enough to trigger