Law Retrieval with Supervised Contrastive Learning Using the Hierarchical Structure of Law

Law Retrieval with Supervised Contrastive Learning Using the Hierarchical Structure of Law Jungmin Choi1,2, Ukyo Honda1,2, Taro Watanabe1, Kentaro Inui2,3, Hiroki Ouchi1,2 1 Nara Institute of Science and Technology 2 RIKEN 3 Tohoku University choi.jungmin.ce6@is.naist.jp, honda.ukyo.hn6@is.naist.jp, taro@is.naist.jp, inui@ecei.tohoku.ac.jp, hiroki.ouchi@is.naist.jp Abstract We study the information retrieval task to identify the relevant law articles for a query on a legal issue in when the legal system in question is statute law. In recent years, the mainstream approach has been to calculate the similarity between the query and each article using pre-trained language models. However, such methods have a weakness in retrieving relevant articles that have low n-gram simi- larity scores with the query. In this work, we show that in such hard cases, the articles tend to be of the same class as articles with high n-gram similarity scores in the hierar- chical structure of statute law, for instance, the Japanese Civil Code. From this observa- tion, we hypothesize that by making articles of same class close to each other in the fea- ture space, we could make it easier to retrieve the above mentioned hard articles. Our pro- posed method realizes this by supervised con- trastive learning using the hierarchical struc- ture. Experimental results show that the pro- posed method achieves higher performance in retrieving the correct articles with low n-gram similarity to the query. 1 Introduction Law is one of the domains where application of nat- ural language processing is expected to bring im- mense benefit to the society. According to a sur- vey conducted by Japan Federation of Bar Associ- ations1, nearly half of those who visited law firms 1https://www.nichibenren.or.jp/library/ ja/jfba_info/publication/data/shimin_ needs.pdf or legal support centers for consultation answered that they had hesitated consulting lawyers before visiting. Some of the most frequently cited reasons include unapproacheable image of lawyers, antici- pated difficulty communicating with them, and con- cerns about whether the issues will be taken seri- ously. This indicates that people often have psycho- logical obstacles to accessing legal services. It is an important societal task to lower the barrier to ac- cessing legal services and facilitate function of law throughout the society. A solution to this problem by natural language processing is to build an information retrieval sys- tem that suggests relevant laws to user-given queries regarding their legal issues. It will provide the user with an approximate idea about how their issues could be described in legal terms, which will enable them to further search for more refined information without necessarily having to consult legal profes- sionals or ask better informed questions when they seek legal support. When developing such a system, it is essential to take into account the characteristics of the legal system of interest. Depending on the legal system, laws are written in vastly different style and struc- ture. In this regard, legal systems can be categorized into two broad categories, case law and statute law. Case law, which includes the legal systems of the United Kingdom, United States, Canada, etc., is law that …

Content truncated to 500 words