“Here Are the Rules: Ignore All Rules”: Automatic Contradiction Detection in Spanish

This paper tackles automatic detection of contradictions in Spanish within the news domain. Two pieces information are classified as compatible, contradictory, or unrelated information. To deal with task, ES-Contradiction dataset was created. contains a balanced number each three types The novelty research is fine-grained annotation different dataset. Presently, four covered contradiction examples: negation, antonyms, numerical, and structural. However, future work will extend all possible contradictions. In order to validate effectiveness dataset, pretrained model used (BETO), after performing experiments, system able detect F1m 92.47%. Regarding type contradictions, best results obtained negation (F1m = 98%), whereas structural obtain lowest 69%) because smaller examples, due complexity generating them. When dealing more generalistic such XNLI, our fails most properly, size both datasets very only covers contradiction. using classification leads us conclude that there highly complex need external knowledge be properly detected this avoid for them previously exposed system.