As seen in the Pitt Research November 2021 Newsletter
Kevin Ashley (left) and Diane Litman
Searching for information about legal decisions presents many barriers for non-experts. Within thousands of words documenting a case, it is difficult to identify the issue or legal reasoning behind a decision. Pitt professors Kevin Ashley and Diane Litman are working to make legal information more accessible through an artificial intelligence framework summarizing essential language in cases. The two have received a $600,000 grant from the joint Amazon and National Science Foundation Fairness in Artificial Intelligence program to support their project, Using AI to Increase Fairness by Improving Access to Justice.
Ashley and Litman were also awarded a 2021 Pitt Momentum Funds Teaming Grant for a related project, the Center for Text Analytic Methods in Legal Studies, in collaboration with Daqing He and Rebecca Hwa, professors in the School of Computing and Information, and James Anderson at the RAND Institute for Civil Justice.
“Making aspects of a complex decision available to the ordinary user is an important part of Increasing access,” explains Ashley, professor in the School of Law, senior scientist at the Learning Research and Development Center (LRDC), and faculty member in the Intelligent Systems Program (ISP). “We want to make information available to the public without paywalls and the need and expense of entering into a subscription agreement.”
A core idea is identifying case sentences that explain statutory terms using a machine learning tool to search through vast repositories of information. Ashley and Litman’s team is building the tool using data from digital compendiums of legal statutes and court decisions with a goal of fielding the resulting tool in the open source Canadian Legal Information Institute (CanLII) and Cornell LII.
“We are developing short, three-sentence summaries of case opinions based on three questions: what is the issue, what is the conclusion, and what is the reason behind the decision. Those as the three main things I would want to know about a case,” Ashley explains.
Litman, professor in the School of Computing and Information, LRDC senior scientist, and ISP faculty member, characterizes her interest as simplification of language.
“We are changing a technical vocabulary into a more accessible vocabulary, changing really long sentences into shorter sentences,” she describes. “Legal language is written in a very complicated way. We try to make it accessible, but to keep it accurate and not oversimplify. The way we use this data set begins with examples of humans – attorneys or law students – summarizing these long decisions. They are not instructed to simplify, but one could look at a summary and see if they are extracting important sentences or rewriting.”
The development of the machine learning tools has been done primarily by Jaromir Savelka, an ISP PhD graduate, now a post-doc at Carnegie Mellon University, and Huihui Xu, a current ISP graduate student. New Computer Science graduate students Mohamed Salem Elaraby and Yang Zhong have also joined the team. Using the LII data, law students first annotate summaries of cases. The machine learning program maps sentences from those summaries back to the full text of cases to identify the gist of the case expressed as Issues, Conclusions, and Reasons.
“We hope to create a public resource,” Ashley stresses. “If a user is interested in the meaning of a particular statutory term, the program could identify sentences from cases that are good explanations and provide a quick summary of cases pertaining to that term. We want to learn if this is useful.”