Hate speech (HS) in online spaces poses severe risks, including real-world violence and psychological harm to victims, necessitating effective countermeasures. Counterspeech (CS), which responds to hateful messages with opposing yet non-hostile narratives, offer a promising solution by mitigating HS while upholding free expression. However, the growing volume of HS demands automation, making Natural Language Processing a viable solution for the automatic generation of CS. Recent works have explored knowledge-driven approaches, leveraging external sources to improve the relevance and informativeness of responses. These methods typically involve multi-step pipelines combining retrieval and passage re-ranking modules. While effective, most studies have focused on English, with limited exploration of multilingual contexts. This paper addresses these gaps by proposing a multilingual, knowledge-driven approach to CS generation. We integrate state-of-the-art re-ranking mechanisms into the CS generation pipeline and evaluate them using the MT-CONAN-KN dataset, which includes hate speech, relevant knowledge sentences, and counterspeech in four languages: English, Italian, Spanish, and Basque. Our approach compares reranker-based systems employing multilingual cross-encoders and LLMs to a simpler end-to-end system where the language model directly handles both knowledge selection and CS generation. Results demonstrate that reranker-based systems outperformed end-to-end systems in syntactic and semantic similarity metrics, with LLM-based re-rankers delivering the strongest performance overall. This work is the result of our participation in the Shared Task on Multilingual Counterspeech Generation held at COLING 2025.

The proliferation of misinformation on social media platforms (SMPs) poses a significant danger to public health, social cohesion and ultimately democracy. Previous research has shown how social correction can be an effective way to curb misinformation, by engaging directly in a constructive dialogue with users who spread – often in good faith – misleading messages. Although professional fact-checkers are crucial to debunking viral claims, they usually do not engage in conversations on social media. Thereby, significant effort has been made to automate the use of fact-checker material in social correction; however, no previous work has tried to integrate it with the style and pragmatics that are commonly employed in social media communication. To fill this gap, we present VerMouth, the first large-scale dataset comprising roughly 12 thousand claim-response pairs (linked to debunking articles), accounting for both SMP-style and basic emotions, two factors which have a significant role in misinformation credibility and spreading. To collect this dataset we used a technique based on an author-reviewer pipeline, which efficiently combines LLMs and human annotators to obtain high-quality data. We also provide comprehensive experiments showing how models trained on our proposed dataset have significant improvements in terms of output quality and generalization capabilities.

Fighting misinformation is a challenging, yet crucial, task. Despite the growing number of experts being involved in manual fact-checking, this activity is time-consuming and cannot keep up with the ever-increasing amount of fake news produced daily. Hence, automating this process is necessary to help curb misinformation. Thus far, researchers have mainly focused on claim veracity classification. In this paper, instead, we address the generation of justifications (textual explanation of why a claim is classified as either true or false) and benchmark it with novel datasets and advanced baselines. In particular, we focus on summarization approaches over unstructured knowledge (i.e., news articles) and we experiment with several extractive and abstractive strategies. We employed two datasets with different styles and structures, in order to assess the generalizability of our findings. Results show that in justification production summarization benefits from the claim information, and, in particular, that a claim-driven extractive step improves abstractive summarization performances. Finally, we show that although cross-dataset experiments suffer from performance degradation, a unique model trained on a combination of the two datasets is able to retain style information in an efficient manner.