コンピューティング科学研究領域のNGUYEN, Minh Le 教授のチームが法律文書の情報抽出および含意関係認識を行う国際コンペティション、Competition on Legal Information Extraction/Entailment (COLIEE) 2023のTask2、Task3、Task4においてhighest performanceを達成しました。

　COLIEE 2023は、国際的な法律文書処理コンテストです。判例検索問題に対処するためのAI技術を見出すことを目的としており、今回で10回目の開催となりました。成文法と判例法からそれぞれ2つずつ、計4つのタスクが出題され、参加者はAIを用いて問題を解き、その手法と自動解答の実験結果について論文を発表、NGUYEN教授のチームはそのうち3つのタスクで1位を獲得し、最高成績を収めました。
　COLIEE-2023は、AIと法分野の主要会議であるICAIL 2023と併催で行われました。

※参考：Competition on Legal Information Extraction/Entailment (COLIEE) 2023
　　　　ICIAL2023

■受賞年月日
　令和5年6月19日

■チーム名・メンバー
タスク2、タスク3
CAPTAIN：Chau Nguyen, Phuong Nguyen, Thanh Tran, Dat Nguyen, An Trieu, Tin Pham, Anh Dang and Le-Minh Nguyen

タスク4
JNLP：Quan Bui, Truong Do, Khang Le, Hien Nguyen, Nguyen Hiep, Trang Pham, and Le-Minh Nguyen

■論文のタイトル
タスク2、タスク3
Efficient Methods for Legal Information Retrieval and Entailment Tasks

タスク4
Data Augmentation and Large Language Model for Legal Case Retrieval and Entailment.

■手法の概要等
タスク2、タスク3
The Competition on Legal Information Extraction/Entailment (COL- IEE) is held annually to encourage advancements in the automatic processing of legal texts. Processing legal documents is challenging due to legal language's intricate structure and meaning. This paper outlines our strategies for tackling Task2, Task3, and Task4 in the COLIEE 2023 competition. Our approach involved utilizing appropriate state-of-the-art deep learning methods, designing methods based on domain characteristics observation, and applying meticulous engineering practices and methodologies to the competition. As a result, our performance in these tasks has been outstanding, with first places in Task2 and Task3 and promising developments in Task4.

タスク4
Legal case retrieval involves identifying relevant cases that share similarities with a given case, while entailment requires assessing whether a legal statement can logically follow from another. These tasks are challenging due to the intricate nature of legal language and the vast quantity of legal documents. To overcome these difficulties, we propose implementing data augmentation techniques to produce additional training data and employing a large language model such as BART or T5 to capture the nuances of legal language. Specifically, we augment the provided dataset by generating synthetic cases that exhibit similar attributes to the original cases. We subsequently train a large language model on the augmented dataset and employ it to retrieve pertinent cases and determine entailment. Our findings also reveal that specific large language generative models, such as the Flan model, have demonstrated potential for performing exceptionally well on the COLIEE task4 dataset. Notably, the Flan model achieved state-of-the-art results on the COLIEE2023 and 2022 task4 test sets.

■受賞にあたって一言
COLIEE 2023 is a prestigious competition in the field of Legal AI. This year, the Nguyen Lab achieved the highest overall performance in the competition, winning first place in three out of the four tasks. We are incredibly honored to receive this award and would like to express our deep gratitude to JAIST for providing our team with an excellent research environment. This recognition serves as an encouragement to our students and all members of the lab, motivating us to continue conducting research and strive for more significant achievements in the future.

令和5年6月30日