M3TQA: Massively Multilingual Multitask Table Question Answering

Tabular data is a fundamental component of real-world information systems, yet most research in table understanding remains confined to English, leaving multilingual comprehension significantly underexplored. Existing multilingual table benchmarks suffer from geolinguistic imbalance-overrepresenting Indo-European and high-resource languages-and insufficient scale for rigorous cross-lingual analysis. To address these limitations, we introduce a complete framework to enhance the massively multilingual multitask table question answering by introducing the instruction and reinforcement learning dataset M 3 TQA-INSTRUCT, a large-scale benchmark spanning 97 languages across diverse language families, including underrepresented and lowresource ones. Further, we construct M 3 TQA by curating 50 real-world tables in Chinese and English, then applying a robust, six-step LLM-based translation pipeline powered by DeepSeek and GPT-4o, achieving high translation fidelity (median BLEU: 60.19) as validated by back-translation. The benchmark includes 2,916 professionally annotated questionanswering pairs across four tasks designed to evaluate nuanced table reasoning capabilities. Experiments on state-ofthe-art LLMs reveal critical insights into cross-lingual generalization, showing that synthetically generated, unannotated QA data can significantly boost performance-especially for low-resource languages. M3T-Bench thus establishes a new standard for multilingual table understanding, offering both a challenging evaluation platform and a scalable methodology for future research.