In this paper we introduce a semantic rolelabeling system constructed on top of thefull syntactic analysis of text. The labelingproblem is modeled using a richset of lexical, syntactic, and semantic attributesand learned using one-versus-allAdaBoost classifiers.Our results indicate that even a simple approachthat assumes that each semantic argumentmaps into exactly one syntacticphrase obtains encouraging performance,surpassing the best system that uses partialsyntax by almost 6%.
展开▼