The first Workshop on Natural Legal Language Processing (NLLP) is taking place today in Minneapolis, MN, where it is co-located with NAACL 2019. The workshop was organized by BLAW Machine Learning engineer Leslie Barrett and AI researcher Daniel Preoţiuc-Pietro, as well as NLP Architect Amanda Stent and data scientist David Rosenberg from Bloomberg’s Office of the CTO, and three academic collaborators. This timely article looks at how our teams are leveraging AI & NLP in Bloomberg Law’s products to transform and speed up how lawyers do their legal research.
Most legal work, especially litigation, requires significant research. Navigating through data can be time-consuming for both litigation and transactional associates. Bloomberg Law (BLAW) is a fully-integrated legal technology platform that provides the resources legal professionals need to grow their business and advise their clients – all in one platform. It can also be accessed by running BLAW OUT <GO> on the Bloomberg Terminal.
The BLAW platform creates strategic research through a number of products that leverage technology, creating new innovative ways for its users to conduct legal research and inform their strategy, including Smart Code; Docket Key, Litigation Analytics and Points of Law (winner of the 2018 AAAL New Product Award). Moving forward, the focus is on leveraging AI and machine learning technologies to build workflow solutions aimed at making attorneys more efficient in completing their everyday legal tasks.
BLAW’s products leverage AI and utilize Bloomberg’s expansive primary legal, financial and market content to deliver workflow solutions to help clients save time and money on non-recoverable costs. AI is essential in creating more efficient and streamlined processes and providing meaningful insight and information to BLAW users.
Lawyers conduct research for many different reasons when preparing for litigation or beginning to draft a transactional document. When defending a case, litigation attorneys often use BLAW’s Litigation Analytics product to research the opposing party (in most cases, a company) and their law firms, because any prior interactions between those parties and a judge can influence their litigation strategy.
For example, before appearing in front of a judge, litigation attorneys will research the judge handling the case and review their prior court opinions to ascertain the judge’s rulings in similar cases. BLAW’s Litigation Analytics provides all the requisite information about that judge, including how long cases before the judge generally take, the judge’s reversal rate, as well as statistics about how that judge has previously ruled in similar cases.
“We’re leveraging AI to extract judges’ names from court documents, determining the identity of the judge using Named Entity Recognition and Disambiguation (NER/NED). A judge’s name is extracted using a method called Conditional Random Fields (CRF), which can be used to predict sequences in text and for pattern recognition. Decision Tree and Random Forest models are used to link the text to the entity in the People profiles database. Analytics Engine then aggregates this data to provide insights,” explained Fulya Erdinc, Team Lead of the BLAW Machine Learning Engineering team.
“For example, if an attorney is preparing to file a motion in front of a judge, they might want to see how many times the judge has denied or granted such a motion and the details of those cases, so the lawyer can take action based on those findings. We provide lawyers with information they need on legal issues and let them make the decisions by themselves.”
Docket Key is another product on the BLAW platform that uses AI to identify and classify entries on a docket sheet. A docket is a document that tracks cases through their lifecycle. After being filed, a case is entered into the court system and tracked on the docket, which contains the entire history of a case from when it was filed through the judge’s decision. This all begins with the initiating paper (or complaint) and, as a case goes on, there are many filings, such as answers, briefs, and transcripts.
“There can be lots of entries, and attorneys can be looking for a specific type of filing,” said Erdinc. “Doing a regular search on documents, the keyword may match on many dockets. By classifying those entries to the type of filing they’re looking for, we can accelerate this process. It’s a classic example of text classification. The challenge is that courts can differ in how they format the entries and the included text. For example, Federal Courts in California and Texas differ from other Federal Courts in how the entries are captured. As a result, we developed separate models for those courts.”
There are many different approaches these tools can use, and Bloomberg’s Global Data team plays the important role of product validation as legal experts who help Engineering understand an attorney’s workflow. Global Data’s role also includes generating the training data that the models rely on, as well as analyzing and evaluating the models from a user’s perspective to assess whether the output will meet client expectations.
“Attorneys are generally interested in legal arguments and not necessarily procedural statements like why the court is allowed to hear the case,” said Valery Richman, Team Leader of the Legal Data Analysis team in Global Data. “We help engineers understand what and how attorneys look for the relevant content.”
In addition, once a suggested content model is deployed, the Global Data regularly re-evaluates the model based on guidelines that capture product and client expectations. These results are then provided to Engineering, who use the data to inform changes they make to further improve the models.
This tight partnership between Engineering, Global Data, and the BLAW team is crucial to the successful development of products for the legal sector which use AI. BLAW’s products are an example of what the teams can create collaboratively, even when the subject area is as nuanced as the law.
The traditional way lawyers conduct research is by navigating through content on the BLAW platform or via keyword searches. This can often be difficult when looking for a relevant precedent for a very narrow issue. It requires complex keyword searches and significant time invested in reviewing the results, as they are often over-inclusive. Ask any lawyer and they will tell you it is often a very inefficient and tedious process.
BLAW’s focus moving forward is to improve that process and make lawyers more efficient in completing their tasks. Utilizing AI and leveraging the Points of Law product and its several machine learning models, BLAW is well on its way to building tools that will transform how lawyers do their work.