Abstract
Identifying claims requiring verification is a critical task in automatedfact-checking, especially given the proliferation of misinformation on socialmedia platforms. Despite significant progress in the task, there remain openchallenges such as dealing with multilingual and multimodal data prevalent inonline discourse. Addressing the multilingual challenge, recent efforts havefocused on fine-tuning pre-trained multilingual language models. While thesemodels can handle multiple languages, their ability to effectively transfercross-lingual knowledge for detecting claims spreading on social media remainsunder-explored. In this paper, we introduce EX-Claim, an entity-awarecross-lingual claim detection model that generalizes well to handle claimswritten in any language. The model leverages entity information derived fromnamed entity recognition and entity linking techniques to improve thelanguage-level performance of both seen and unseen languages during training.Extensive experiments conducted on three datasets from different social mediaplatforms demonstrate that our proposed model significantly outperforms thebaselines, across 27 languages, and achieves the highest rate of knowledgetransfer, even with limited training data.