CONDAQA

CONDAQA是一个专注于否定推理的英语阅读理解数据集，由卡内基梅隆大学创建。该数据集包含14,182个问题-答案对，涉及超过200种独特的否定提示，旨在评估模型处理否定语句的能力。数据集通过众包方式收集，工人们需要对包含否定提示的段落进行三种编辑：重述否定语句、改变否定范围和反转否定。CONDAQA的应用领域是自然语言处理，特别是提高模型对否定语句的理解和推理能力。CONDAQA is an English reading comprehension dataset focused on negation reasoning, developed by Carnegie Mellon University. This dataset contains 14,182 question-answer pairs, covering more than 200 unique negation cues, and is designed to evaluate models' ability to process negated statements. The dataset is collected via crowdsourcing, where annotators are required to perform three types of edits on paragraphs containing negation cues: rephrasing negated statements, adjusting the scope of negation, and reversing negation. The application field of CONDAQA is natural language processing, specifically aimed at improving models' understanding and reasoning capabilities regarding negated statements.