Using Grammar Rule Clusters for Semantic Relation Classification
Emily Jamison · 2011
Automatically-derived grammars, such as the split-and-merge model, have proven helpful in parsing (Petrov et al., 2006). As such grammars are refined, latent information is recovered which may be usable for linguistic tasks besides parsing. In this paper, we present and examine a new method of semantic relation classification: using automaticallyderived grammar rule clusters as a robust knowledge source for semantic relation classification. We examine performance of this feature group on the SemEval 2010 Relation Classification corpus, and find that it improves performance over both more coarse-grained and more fine-grained syntactic and collocational features in semantic relation classification. 1