Leveraging VerbNet to build Corpus-Specific Verb Clusters

In this paper, we aim to close the gap from extensive, human-built semantic resources and corpus-driven unsupervised models. The particular resource explored here is VerbNet, whose organizing principle is that semantics and syntax are linked. To capture patterns of usage that can augment knowledge resources like VerbNet, we expand a Dirichlet process mixture model to predict a VerbNet class for each sense of each verb, allowing us to incorporate annotated VerbNet data to guide the clustering process. The resulting clusters align more closely to hand-curated syntactic/semantic groupings than any previous models, and can be adapted to new domains since they require only corpus counts.