Understanding Legal Documents: Classification of Rhetorical Role of Sentences Using Deep Learning and Natural Language Processing

Automatically extracting the patterns of reasoning from complex legal documents can make legal systems more effective by increasing on-time case processing and the case clearance rate. The crucial task to achieve this is to automatically classify sentences in legal documents into categories based on their content. In this paper, a deep learning model is proposed that breaks down legal documents and classifies the rhetorical types of sentences. A hypothesis is tested, that using a small set of labelled data, deeper and more accurate models for the processing of legal documents can be constructed. This will automate the processing of legal documents hence decreasing, and ultimately eliminating the backlog that currently exists throughout various legal systems. This work can be generalized for legal appeals cases in diverse fields. The various configurations used to train the Bidirectional Long Short-Term Memory (Bi-LSTM) model will be compared along with a variety of embeddings. The experiments obtained their highest accuracy using the Bi-LSTM model with 300-dimensional GloVe embeddings compared to the previous techniques used for the semantic understanding of law related documents.