Automatic learning of textual entailments with cross-pair similarities