ProPara

ProPara是由艾伦人工智能研究所创建的一个新型数据集，专注于理解描述过程的段落，如光合作用等科学过程。该数据集包含488个由人类编写的段落，共计81,345个关于实体状态变化的标注，旨在通过预测实体的位置和存在变化来理解文本。ProPara是首个包含自然文本描述变化世界的数据集，其创建过程涉及使用科学词汇生成提示，通过众包方式收集段落，并由专家标注实体状态。该数据集的应用领域包括过程执行和验证、效果预测等，旨在解决机器阅读理解中需要推理的难题。

ProPara is a novel dataset created by the Allen Institute for AI, dedicated to understanding paragraphs that describe processes such as scientific ones like photosynthesis. This dataset includes 488 human-written paragraphs, with a total of 81,345 annotations related to entity state changes, aiming to comprehend text by predicting entity location and existence changes. ProPara is the first dataset containing natural text that describes changes in the world. Its creation process involves generating prompts with scientific vocabulary, collecting paragraphs through crowdsourcing, and having experts annotate entity states. The application scenarios of this dataset cover process execution and verification, effect prediction, and more, aiming to address the challenging reasoning problems in machine reading comprehension.