Unsupervised Induction of a Syntax-Semantics Lexicon Using Iterative Refinement

We present a method for learning syntaxsemantics mappings for verbs from unannotated corpora. We learn linkings, i.e., mappings from the syntactic arguments and adjuncts of a verb to its semantic roles. By learning such linkings, we do not need to model individual semantic roles independently of one another, and we can exploit the relation between different mappings for the same verb, or between mappings for different verbs. We present an evaluation on a standard test set for semantic role labeling.