Enhancing Legal Argument Retrieval with Optimized Language Model Techniques

Understanding the meaning of legal concepts appearing in regulations is essential to both legal experts and lay people. The main way to explain these terms involves sifting through extensive case law and identifying court opinions regarding the interpretations of the concepts in certain scenarios. Automating this task can be achieved by using argument retrieval techniques. In this paper, we build on the extensive results in [1] and present six approaches for potentially improving the results. These approaches are then tested and compared on the same data set. By careful experiment design and extensive testing we were able to improve the NDCG@10 score by 12% points on the evaluation subset and by 13% points on the test subset, setting a new state-of-the-art result for the dataset of statutory interpretation.