Two-year-olds assign appropriate interpretations to verbs presented in two English transitivity alternations, the causal and unspecified-object alternations (Naigles, 1996). Here we explored how they might do so. Causal and unspecified-object verbs are syntactically similar. They can be either transitive or intransitive, but differ in the semantic roles they assign to the subjects of intransitive sentences (undergoer and agent, respectively). To distinguish verbs presented in these two alternations, children must detect this difference in role assignments. We examined distributional features of the input as one possible source of information about this role difference. Experiment 1 showed that in a corpus of child-directed speech, causal and unspecified-object verbs differed in their patterns of intransitive-subject animacy and lexical overlap between nouns in subject and object positions. Experiment 2 tested children's ability to use these two distributional cues to infer the meaning of a novel causal or unspecified-object verb, by separating the presentation of a novel verb's distributional properties from its potential event referents. Children acquired useful combinatorial information about the novel verb simply by listening to its use in sentences, and later retrieved this information to map the verb to an appropriate event. (Contains 2 footnotes, 4 tables, and 2 figures.)