jvonrad/WIKI-FACT

--- language: - en - de - id - pt - ar - bn - sw - es - ru - fr - ja - zh task_categories: - multiple-choice pretty_name: WIKI-FACT (Multilingual MCQ Consistency Dataset using Wikipedia Factbase) configs: - config_name: default data_files: - split: train path: train.jsonl - split: validation path: validation.jsonl - split: test path: test.jsonl --- # Multilingual MCQ Consistency Dataset Multilingual factual multiple-choice QA dataset across 12 languages. ## Structure Each example contains: - `fact_id` - `property_id` - `subject`, `relation`, `object` - `langs`: dict of language-specific QA Each language entry: - `question` - `answer_text` - `options` ```json { "fact_id": "...", "property_id": "...", "langs": { "en": { "question": "...", "answer_text": "...", "options": [...] }, ... } } ``` ## Use case Designed for: - RL training (reward = correctness + cross-lingual consistency) - multilingual evaluation