UVA Students, Law Professor Creating Virtual Legal Researcher
Someday soon, thanks to techniques being refined at the University of Virginia, a computer search may open up the seemingly impenetrable world of statues, regulations, and court decisions to everyone, not just a small set of legal researchers and attorneys in the know.
Mauricio Guim, a doctor of juridical science candidate at the Law School, was recently named to the Presidential Fellows in Data Science Program at UVA for a multidisciplinary project he proposed with Faraz Dadgostari in the Department of Systems and Information Engineering. Their work, which attempts to create computational models for legal research, is being supervised by Professors Michael Livermore and Peter Beling (Systems and Information Engineering) and builds on Livermore’s work with researchers at Dartmouth College on mathematical models to analyze Supreme Court decisions.
“What we’ve done with the Dartmouth folks is create a mathematical model for legal search to mimic or model how a lawyer would navigate through legal texts,” Livermore explained.
The new project at UVA builds on that in significant ways, Livermore said. For one, it will expand on the underlying search parameters, incorporating more areas of law in the United States beyond Supreme Court decisions, as well as a data set from the Inter-American Court of Human Rights, which Guim collected when he worked as a scholar-in-residence at its headquarters in San Jose, Costa Rica, in 2015-16.
That data collection and input required a great deal of time and work. After spending the day working at the court, he would dedicate himself to the databases.
“At night, I read all the court’s decisions and started building a database of all the cases decided by the court from 1981 to 2016," he said. “I read several hundred cases and coded all the information that could be useful. The opinions are often hundreds of pages long, and not all of the pages are important, but you have to dive through them to find the important things.”
This data set is especially important because the court doesn’t have a state-of-the-art search engine and therefore the access to its relevant decisions is limited, "which is kind of problematic for a court that is supposed to set precedents in human rights cases for all Latin America," Guim said.
“The idea of precedent only works if you can easily search and find relevant cases, and there’s currently no tool for doing that,” said Guim, who is from Ecuador and practiced law there. “It is not easy to search and find relevant decisions. In Ecuador, for example, lawyers would have to go through all the court’s decisions, because there’s no tool to find relevant cases for a particular issue. And if the lawyers or the judge don’t know the precedent, the judge decides as if no precedent exists. You shouldn’t win or lose a case because there wasn’t access to legal information.”
Guim is looking forward to the opportunity to continue work on the project with his partners. It is, he explained, the first time he’s worked extensively with non-lawyers.
“It’s exciting because the Law School is separate from other departments,” he said. “We live in our own world. This is going to be a chance to work with people from different disciplines, different cultures. It’s an opportunity to learn something besides law.”
In Ecuador, Guim worked as a lawyer in several major free speech cases, defending a newspaper that was being sued by the country’s president, Rafael Correa, for libel, after a columnist wrote that the president could be sued for international crimes following a military operation that he ordered resulted in the deaths of civilians. Though the newspaper lost the case in Ecuadorian courts, the president eventually pardoned the columnist and the newspaper executives as a result of international pressure, particularly an injunction from the Inter-American Commission of Human Rights in Washington, D.C
Another major goal of the project is to further refine the search technique, with the end goal being an ever more accurate representation of human behavior in legal search.
“The Holy Grail is to say ‘I have a legal question,’ and then be able to interface with a natural language search engine that will find the relevant legal authority, and then go to it and identify answers or areas of uncertainty,” Livermore said. “Finding the relevant resources is a huge part of legal reasoning. So we want to both improve that and also broaden it.”
Livermore explained that the group will be doing preliminary work to compare its search algorithm to human legal researchers, and finding out what results the two return.
But, Guim clarified, the goal is decidedly not to replace humans.
“We’re trying to make legal research more efficient,” he said. “This can make a significant difference on the margins for judges and lawyers who don’t have the time and resources that we do. That’s why we need to do it.”