Abstract
<jats:p>The universal human language faculty is a cognitive system that delimits the class of possible human languages, together with an acquisition process that characterizes the relation between linguistic data available to the learner and particular grammars. It is common to infer properties of this system from observed generalizations across languages concerning phenomena which occur commonly or not at all. This chapter argues that many such inferences are insecure at best. Possible but rare diachronic changes may pose counterexamples to otherwise valid generalizations, and some such regularities may actually represent the frequency of common paths of change rather than properties of the space of possible grammars. Examples are drawn from phonology, morphology, and morphosyntax.</jats:p>