Children's ability to learn aspects of their language in the absence of supportive evidence is discussed. Specifically, the learnability of null subjects in languages in which they appear is examined when indirect negative evidence is present. It is concluded that parameters such as the null subject parameter may not generate languages, strictly speaking, but may only fix the bounds within which languages can be realized. It is also suggested that children may be sensitive to the nonoccurrence of certain types of evidence, and that the view of learnability outlined here may be empirically plausible only to the extent that the theory of principles and parameters in language is empirically plausible. (MSE)