Multilingual approaches to extractive question answering in political texts

This work contributes to multilingual extractive question answering (QA) by presenting QA-specialized versions of ConfliBERT for English, Spanish, and Arabic, language models designed for analyzing political conflict and violence. A cross-lingual QA framework is proposed, including curated datasets and the Spanish translation of an English QA corpus to mitigate the scarcity of annotated resources for low-resource languages. The models are fine-tuned specifically for extractive QA and benchmarked against general-purpose BERT variants, showing consistent gains across all target languages. By addressing language gaps in high-stakes domains, the study underscores the potential of multilingual QA systems to support both research and decision-making in political contexts.