Large-Scale Extraction of FineGrained Semantic Relations between Verbs

Abstract. Broad-coverage repositories of semantic relations between actions could benefit many NLP tasks, as well as tasks related to reasoning and infer-ence. We present an automatic method for extracting fine-grained semantic rela-tions, addressing relations between verbs. We detect similarity, strength, an-tonymy, enablement, and temporal relations between pairs of verbs with high mutual information using lexico-syntactic patterns over the Web. On a set of 26,118 strongly associated verb pairs, our extraction algorithm yielded 56.5% accuracy1. On the relations strength and similarity, we achieved 79.6 % and 66.7 % accuracy respectively.