FLP logo

Important Opinions on CourtListener are Now Summarized by the Top Experts — Judges

Michael Lissner

Understanding legal opinions can be hard, so today we're launching a new tool that makes it exceptionally easy to understand what an opinion is about. For 370,000 opinions in CourtListener, we will now show at least one summary for the case.

Before we explain how it works, let's look at a few examples.

The most-cited opinion in CourtListener is Strickland v. Washington. For this case, our new system has 1,732 summaries that are organized into 214 high-level groups.

The top summary says the case is about:

Explaining that to establish prejudice the defendant must show “a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different”

Ah ha. It's the quintessential "ineffective counsel" case. Got it.

Let's look at another well-known case, Citizens United v. FEC. Its top summary says it's about:

Holding that the government may not, under the First Amendment, suppress political speech based on the speaker’s corporate identity

Spot on.

Another, Miranda v. Arizona:

Holding that custodial statements are inadmissible unless officers warn the individual of the following: (1) he has the right to remain silent, (2) any statement may be used against him, (3) he has the right to an attorney, and (4) he may waive those rights

Yep.

Altogether, we've gathered over a million summaries like the ones above, and we've associated them with 370,000 opinions. Every day we add more.

How is this possible?

This system works by finding citations that have been summarized by a court.

For example, the first summary above is drawn from a 9th Circuit case that has the following footnote:

A screenshot of a summary Strickland, as quoted above.
Strickland is summarized as above in a footnote of this 9th Circuit case.

Our process begins by using our citation parser, eyecite, to identify these summaries. Once that's completed, we filter out bad ones, organize good ones into meaningful groups, and rank the groups from best to worst.

Check out Miranda v. Arizona

The way we group and rank summaries takes into account a number of factors including the size of the group, the importance of the summarizing opinion, and the text of the summary itself. Between all this, we're able to identify the top summary for each group and the top groups for each opinion.

As you can see above, because these summaries are written by the courts, they tend to be exceptionally accurate. With this feature launched, hundreds of thousands of opinions in CourtListener now have expertly-written summaries.

So that's how we did it, but no project like this happens in a vacuum. We could never have done this without the pioneering work of Pablo Arredondo (who prototyped the concept at the Stanford Center for Legal Informatics) and Casetext (where the idea was first fully put into action), the tireless energy of Faiz Surani and Varun Iyer at University of California, Santa Barbara, who did all the heavy lifting in this project, and the Library Innovation Lab at Harvard, with whom we collaborated on eyecite.

What's next?

This new feature is so powerful, today has to be just the beginning. We have two more major components in the works.

First, we plan to export this data into a CSV so that researchers can work with it. This unique new legal dataset should be available in the coming days.

Second, we're working on making this data searchable. Instead of searching the full text of opinions, just search their summaries and find the ones that are on point.

If you think you might have ideas for how to use this data or how to make it searchable — or something else we haven't thought of — please get in touch. We're thrilled to take the next step on this journey.

© 2024 Free Law Project. Content licensed under a Creative Commons BY-ND international 4.0, license, except where indicated. Site powered by Netlify.