Comparing Ref-Vectors and word embeddings in a verb semantic similarity task

This work introduces reference vectors (Ref-Vectors), a newkind of word vectors in which the semantics is determined by the prop-erty of words to refer to world entities (i.e. objects or events), ratherthan by contextual information retrieved in a corpus. Ref-Vectors arehere compared with state-of-the-art word embeddings in a verb semanticsimilarity task. The SimVerb-3500 dataset has been used as a benchmarkto verify the presence of a statistical correlation between the semanticsimilarity derived by human judgments and those measured with Ref-Vectors and verb embeddings. Results show that Ref-Vector similaritiesare closer to human judgments, proving that, within the action domain,these vectors capture verb semantics better than word embeddings.