winogrande-multilingual
收藏资源简介:
该数据集包含德语(deu)和法语(fra)两种配置的验证集,用于多选问答任务。每个样本包含13个字段:唯一标识符(id)、基准名称(benchmark)、问题句子(sentence)、两个选项(option1/option2)、正确答案(answer)、错误选项缺陷描述(wrong_option_flaw)、种子ID(seed_id)、主题描述(topic_description)、推理类型(reasoning_type)、综合注释(synthesis_notes)、审核标记(flag_for_review)和审核原因(review_reason)。德语配置包含1,267个样本(1.02MB),法语配置包含100个样本(88.8KB)。数据格式为字符串(large_string)和布尔值(bool),特别包含对错误选项的逻辑缺陷分析和多种推理类型的标注,适用于问答系统评估和推理能力研究。
This dataset includes two validation set configurations for German (deu) and French (fra) languages, targeting multiple-choice question answering tasks. Each sample comprises 13 fields: unique identifier (id), benchmark name (benchmark), question sentence (sentence), two options (option1/option2), correct answer (answer), wrong option flaw description (wrong_option_flaw), seed ID (seed_id), topic description (topic_description), reasoning type (reasoning_type), synthesis notes (synthesis_notes), review flag (flag_for_review), and review reason (review_reason). The German configuration consists of 1,267 samples (1.02 MB), while the French configuration contains 100 samples (88.8 KB). The data formats include string (large_string) and boolean (bool). Specifically, the dataset features logical flaw analyses for incorrect options and annotations for diverse reasoning types, making it suitable for question answering system evaluation and reasoning capability research.




