Researchers have developed a dataset called ProBel for detecting propaganda in news articles. It includes labeled data for both Arabic and English languages, with annotations at various levels such as sentence-level decisions, technique classification, and span identification. The study evaluates different training methods using this dataset, finding that joint bilingual training yields the most stable results. --- Why it matters: This matters to AI researchers because it provides a new resource for propaganda detection, which is an important task in understanding and mitigating the spread of misinformation. It also explores the interaction between supervision levels when learned jointly across languages. Source: https://arxiv.org/abs/2608.22388