Early language development critically depends on the ability to form abstract representations of linguistic knowledge, and to generalize that knowledge to new situations. In verb knowledge, much generalization appears to be driven by various regularities between form and meaning, but it is difficult to assess how these factors interact in a complex learning environment. We extend a hierarchical Bayesian model to acquire abstract knowledge of verbs from naturalistic child-directed speech, then generalize these abstractions to novel verbs, simulating child behaviour. We use the syntactic alternation structure of a novel verb to infer aspects of its meaning, and use the meaning of a novel verb to predict its range of acceptable syntactic forms. The model provides a useful framework to investigate the interaction of complex factors in verb learning.