CATCH-ME if you RAG: a dataset of Contextually Annotated multi-Turn Counterspeech against Hate and Misinformation Exchanges

The article introduces the CATCH-ME dataset, a large-scale, expert-curated collection of multilingual dialogues designed to address the intersection of hate speech and misinformation across multiple turns. This dataset includes document- and chunk-level span annotations and is anchored in verified external knowledge, making it suitable for Retrieval-Augmented Generation (RAG) systems. Its release is significant for practitioners as it provides high-quality, contextually rich examples to improve the generation of effective counterspeech models in diverse languages.