FAI: Using AI to Increase Fairness by Improving Access to Justice

FAI：利用人工智能改善诉诸司法的机会来提高公平性

• 批准号: 2040490
• 负责人: Kevin Ashley
• 金额: $ 37.5万
• 依托单位: University of Pittsburgh
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-02-01 至 2025-01-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2040490&HistoricalAwards=false
• 关键词: FAI; Using; AI; Increase; Fairness

This project applies Artificial Intelligence (AI) to increase social fairness by improving public access to justice. Although many AI tools are already available to law firms and legal departments, these tools do not typically reach members of the public and legal service practitioners except through expensive commercial paywalls. The research team will develop two tools to make legal sources more understandable: Statutory Term Interpretation Support (STATIS) and Case Argument Summarization (CASUM). STATIS is an AI-based legal information retrieval tool to help users understand and interpret statutory terms. It helps them find sentences explicating the terms of interest and cases applying these terms. Inputs to the system are queries about a statutory term and the provision from which it comes. The system outputs a list of sentences retrieved from case law that mention the term in a manner useful for understanding and elaborating its meaning. CASUM summarizes case decisions in terms of legal argument triples: the major issues a court addressed in the case, the court’s conclusion with respect to each issue, and the court’s reasons for reaching the conclusion. Given a case text, it outputs simple argument diagrams graphically summarizing arguments in the decision. Ultimately, the tools will be deployed through legal information institutes (LIIs) that provide free access to the public. They will help the lay public to understand, as well as to access, legal source materials by making it easy for them to find sentences in legal cases that provide definitions, tests, examples or counterexamples of statutory terms and to see the issues, conclusions, and reasons a court addresses in a decision. The project applies the latest natural language processing approaches. Pre-trained legal language models will improve the performance of machine learning in identifying sentences in legal cases that explain statutory terms or state issues, conclusions, and reasons. Recent developments in extractive and abstractive summarization, text simplification, and argument mining will generate high quality legal information for diverse users. A legal language model will be pretrained on a large corpus of publicly available court decisions and fine-tuned to identify features that play a significant role in retrieving high value sentences explaining statutory terms. A prototype module for retrieving and ranking such sentences by explanatory value and a graphical user interface ultimately deployable via an LII website will be developed. Using the legal language model, techniques for matching annotated sentences from case summaries to the corresponding sentences in the full texts will be developed and fine-tuned to classify sentences in which a court identifies issues, conclusions, and reasons justifying the conclusions. Finally, a prototype module for graphically summarizing cases in terms of argument diagrams depicting legal argument triples will be developed and applied to summarizing cases that explain statutory terms. Planning will be done for a user interface suitable for integration with the LII websites.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

该项目应用人工智能（AI），通过改善公众获得司法公正来增加社会公平。尽管许多人工智能工具已经可供律师事务所和法律的部门使用，但除非通过昂贵的商业付费墙，否则这些工具通常不会接触到公众和法律的服务从业者。研究小组将开发两种工具，使法律的来源更容易理解：统计术语解释支持（STATIS）和案例论证摘要（CASUM）。STATIS是一个基于人工智能的法律的信息检索工具，帮助用户理解和解释法定术语。它帮助他们找到解释感兴趣的术语的句子和应用这些术语的案例。该系统的输入是关于法定期限及其所依据的条款的查询。该系统输出一个从判例法中检索到的句子清单，这些句子以有助于理解和阐述其含义的方式提到该术语。CASUM根据法律的论证三要素总结案件判决：法院在案件中处理的主要问题，法院对每个问题的结论，以及法院得出结论的理由。给定一个案例文本，它输出简单的论证图，以图形方式总结决策中的论证。最终，这些工具将通过向公众免费提供的法律的信息机构部署。它们将帮助普通公众理解和获取法律的原始材料，使他们能够轻松地找到法律的案件中提供法定术语定义、检验、示例或反例的句子，并看到法院在判决中所处理的问题、结论和理由。 该项目采用最新的自然语言处理方法。预训练的法律的语言模型将提高机器学习在法律的案件中识别解释法定条款或国家问题、结论和原因的句子的性能。最近在提取和抽象摘要、文本简化和论证挖掘方面的发展将为不同的用户生成高质量的法律的信息。一个法律的语言模型将在大量公开的法院判决语料库上进行预训练，并进行微调，以识别在检索解释法定术语的高价值句子方面发挥重要作用的特征。将开发一个原型模块，用于按解释价值检索和排列这些句子，并开发一个最终可通过法律信息基础设施网站部署的图形用户界面。使用法律的语言模型，将开发和调整将案例摘要中的注释句子与全文中的相应句子匹配的技术，以分类法院确定问题、结论和证明结论的理由的句子。最后，将开发一个原型模块，用于以描绘法律的论证三元组的论证图的形式，以图形方式总结案例，并将其应用于总结解释法定术语的案例。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

ArgLegalSumm: Improving Abstractive Summarization of Legal Documents with Argument Mining

ArgLegalSumm：通过论证挖掘改进法律文档的抽象摘要

• 发表时间: 2022
• 期刊: Proceedings of the 29th International Conference on Computational Linguistics
• 作者: Elaraby, Mohamed;Litman, Diane
• 通讯作者: Litman, Diane

Discovering Explanatory Sentences in Legal Case Decisions Using Pre-trained Language Models

使用预先训练的语言模型发现法律案件决策中的解释性句子

• DOI: 10.18653/v1/2021.findings-emnlp.361
• 发表时间: 2021
• 期刊: Findings of the Association for Computational Linguistics: EMNLP 2021
• 作者: Savelka, Jaromir;Ashley, Kevin
• 通讯作者: Ashley, Kevin

Accounting for Sentence Position and Legal Domain Sentence Embedding in Learning to Classify Case Sentences

• DOI: 10.3233/faia210314
• 发表时间: 2021-12
• 期刊: ArXiv
• 作者: Huihui Xu;Jaromír Šavelka;Kevin D. Ashley