Ontology-Based Information Extraction for Automatic Generation of Legal Arguments 
Advised by: Amal Shehan Perera, Nisansa de Silva
University of Moratuwa
Legal professionals dedicate a major part of their time to read previous court cases to gather legal precedents, understand the evolution of law etc. To the best of our knowledge, none of the existing computer applications related to legal domain provides information within court case transcripts in an intuitive manner. This study attempts to address these issues by coming up with information extraction mechanisms that will ultimately facilitate the task of representing information in court case transcripts in a well-structured, intuitive manner. We have carried out multiple information extraction tasks in the legal domain that utilize concepts and methodologies which span across machine learning, natural language processing, semantic analysis, sentiment analysis, and linguistics. Our end goal is to create a system that is to be assistance to legal officials in their practice, specifically, a system that can automatically identify legal arguments, facts, and citations for a given legal case. First, we developed a system combining a machine learning model and a rule-based component to determine relationships among sentences in legal cases. Identifying relationships among sentences can be considered as a fundamental and important task. It will enable a computer system to identify the information flow within a court case transcript and also to determine the facts, evidence which is related to a particular legal argument. On the process of developing a system to automatically identify relationships among sentences, we experienced drawbacks in the existing sentiment annotators when used for analyzing texts in the legal domain. One of the reasons for the issues was that none of the sentiment annotators were trained using texts from the legal domain. Having an accurate sentiment analyzer is a crucial aspect when it comes to Information Extraction. Therefore, we came up with a novel and fast approach to build a sentiment annotator for the legal domain using transfer learning. Extracting arguments from court case transcripts can be considered as another information task with significant importance. Given a sentence from a court case transcript, a system should have the capability to determine whether the sentence is an argument or not. After analyzing various methodologies, it was decided to use linguistic and rule-based approaches to detect argumentative sentences. Classifying an argument as to whether it is in favor of the plaintiff or the defendant can be considered as a crucial problem to be solved when it comes to information extraction from court case transcripts. In this study, we have demonstrated some approaches which can guide the process of solving this problem.
Keywords: Natural Language Processing | Machine Learning / Deep Learning | Ontologies | Law |