Distributional learning in Brazilian Portuguese: a computational study

In this paper, we address the problem of learning word categories during language acquisition. Our approach is computational: we built a model based on Redington et al. (1998) in order to investigate the informativeness of distributional information to the categorization of words. The data provided to the learner comes from two corpora of child-directed speech in Brazilian Portuguese. Specifi cally, the experiments presented here evaluate the informativeness of various contextual windows regarding a given target word, that is, which contexts are more or less informative of a word category. Our results show that local contexts are highly informative and that distributional information is useful as a source of categorial information