Merger Agreement Understanding Dataset (MAUD)
收藏资源简介:
MAUD是一个由美国律师协会2021年公共目标交易点研究支持的专业标注阅读理解数据集,包含超过39,000个示例和47,000多个总标注。数据集由并购律师精心策划,用于训练模型理解和分类合同中的特殊法律语言。MAUD是唯一的专家标注并购协议数据集,适用于法律专业和NLP社区作为基准。数据集内容包括从152个英语公共并购协议中提取的法律文本,涉及多种并购交易点,旨在通过回答标准化多选题来评估模型的法律文本理解能力。
MAUD is a professionally annotated reading comprehension dataset supported by the 2021 Public Target Deal Points Study from the American Bar Association. It contains over 39,000 examples and more than 47,000 total annotations. Curated by M&A lawyers, the dataset is developed to train models to understand and classify specialized legal language in contracts. MAUD is the only expert-annotated dataset for merger agreements, serving as a benchmark for both the legal professional community and the NLP community. The dataset includes legal text extracted from 152 English-language public merger agreements, covering a wide range of M&A deal points, and aims to evaluate models' legal text comprehension capabilities via standardized multiple-choice questions.
Merger Agreement Understanding Dataset (MAUD)
概述
MAUD 是由 Atticus Project 策划的合并协议审查数据集,用于 2021 年美国律师协会公共目标交易点研究。
安装
- 安装支持 GPU 的 PyTorch:https://pytorch.org/get-started/locally/
- 运行
pip install -e . - 解压数据文件:
unzip data.zip
最佳超参数
最佳超参数及其对应的验证分数可在 best_found_hps/*.csv 文件中找到。
训练与评估
- 使用最佳超参数训练模型:运行
scripts/train.sh和scripts/train_multi.sh - 评估模型:运行
scripts/evaluate.sh
引用
如果 MAUD 对你的研究有用,请考虑引用:
@misc{wang2023maud, title={MAUD: An Expert-Annotated Legal NLP Dataset for Merger Agreement Understanding}, author={Steven H. Wang and Antoine Scardigli and Leonard Tang and Wei Chen and Dimitry Levkin and Anya Chen and Spencer Ball and Thomas Woodside and Oliver Zhang and Dan Hendrycks}, year={2023}, eprint={2301.00876}, archivePrefix={arXiv}, primaryClass={cs.CL} }




