StAtutory Reasoning Assessment dataset (SARA)

SARA数据集是由约翰斯·霍普金斯大学创建，专注于美国税法中的法定推理评估。该数据集包含从美国国内税收法典（IRC）提取的规则和自然语言问题，共计376条记录。数据集的创建过程涉及从IRC中选择相关章节，简化并确保规则的内部一致性，以及手动创建测试案例。SARA数据集的应用领域主要集中在法律领域的自然语言理解和推理，旨在通过机器学习模型解决法律文本理解和应用中的挑战。

The SARA dataset, developed by Johns Hopkins University, is dedicated to evaluating statutory reasoning in U.S. tax law. It contains 376 records in total, composed of rules and natural language questions extracted from the United States Internal Revenue Code (IRC). The construction of the dataset entails selecting relevant sections from the IRC, simplifying the rules while ensuring their internal consistency, and manually creating test cases. Primarily applied in the domain of legal natural language understanding and reasoning, the SARA dataset aims to address the challenges in legal text comprehension and application via machine learning models.