Метрики распределительного сходства

Previous work on paraphrase extraction and application has relied on either parallel datasets, or on distributional similarity met- rics over large text corpora. Our approach combines these two orthogonal sources …...