Detecting misleading headlines through the automatic recognition of contradiction in Spanish

Misleading headlines are part of the disinformation problem. Headlines should give a concise summary news story helping reader to decide whether read body text article, which is why headline accuracy crucial element story. This work focuses on detecting misleading through automatic identification contradiction between and item. When detected, alerted lack precision or trustworthiness in relation text. To facilitate detection headlines, new Spanish dataset created (ES_Headline_Contradiction) for purpose identifying contradictory information its annotates semantic relationship by categorising texts as compatible , xmlns:xlink="http://www.w3.org/1999/xlink">contradictory xmlns:xlink="http://www.w3.org/1999/xlink">unrelated . Furthermore, another novel aspect this that it distinguishes different types contradictions, thereby enabling more fine-grain them. The was built via semi-automatic methodology, resulted cost-efficient development process. results experiments show pre-trained language models can be fine-tuned with dataset, producing very encouraging incongruency non-relation