Team Tax-I Competed Successfully in International Legal-Tech Competition

Team Tax-I Competed Successfully in International Legal-Tech Competition

Team Tax-I won third place in the second task of the international legal-tech Competition on Legal Information Extraction / Entailment 2020 (COLIEE). Tax-I entered the competition as a bit of an outsider, as one of the few participants from the private sector competing against computer science university departments from as far as Canada, Japan and everywhere in between. Regardless, our winning performance once again validated the technology underlying Tax-I.

Our Journey

After coming across COLIEE, we realized this was the perfect opportunity to test whether our legal-ai functionality works in other domains. Timing an coordination were tricky as the competition was held in August and required scheduling around our not-always-overlapping holiday schedules. Thankfully there were four tasks in the competition and four of us, so we could work independently.

Through the development period, we deepened our knowledge of both AI-powered search and prediction functionalities. We are excited to integrate into Tax-I the knowledge we gained by both doing the competition and sharing our results with the legal-tech community at the conference.

Afterward the competition, our paper was published in the workshop proceedings and we had the pleasure of connecting with the other participants in the JURISIN 2020 conference. The organizers even asked for our input, as representatives of the private sector, for future tasks to be added to the competition. We are happy to provide such input and steer the field in our direction.

The Tasks

There were four tasks in the competition covering such tasks as search and entailment. While most people are familiar with search, entailment deserves some extra explanation. Entailment is a logical term regarding the relationship between two statements, A and B. If you read statement A and B and the contents of A are true given the contents of B, then B entails A and visa versa. For example, given the statements A: “Socrates only wears a toga” and B: “Socrates don’t wear pants”, we can say that B is entailed by A. Who wears a toga with pants anyway?

Search In Depth:

Tasks 1 and 3 were both search tasks, albeit in different domains. Task 1 focused on searching database of court cases. Think google, but instead of a search query we had to use the whole document. Task 3 focused on statute search, where given a question from the Japanese bar exam, we had to find the relevant articles from the Japanese civil code that you would need to consult to answer the question.

Entailment:

Tasks 2 and 4 were entailment tasks in the case law and statutory domains, respectively. Task 2 required us to determine which paragraph from a court case entailed a specific fragment of text. Task 4 required us to determine whether the question from the Japanese bar exam was true or false given a set of relevant articles from the Japanese civil code.

Our Solutions

In solving the COLIEE tasks, we encountered and overcame the limits of traditional methods of applying AI to text. Basically, text has to represented in a way that a machine learning algorithm can understand it. To learn more about how we overcame these obstacles, inventing new legal-specific methods of representing text, checkout our blog post which tackles this issue in depth.

By combining our home-grown recipes with state of the art functionalities, we created a system capable of identifying logical relations between paragraphs in a court case, answering questions to the Japanese bar exam, and finding the necessary civil code articles a human would need to consult if they were to answer the bar exam. A, in depth description of our methods can be found in our workshop proceedings paper found here.

If legal prediction or AI-powered case law search could save your team time and effort, don’t hesitate to reach out. Tax-I simplifies legal search, provides insightful analytics and reveals hidden insights that can make or break a case. We are open to collaborate and co-create.