Legal Document Summarization Using Ripple Down Rules

Abstract

This paper presents an approach for legal document summarization using Ripple Down Rules(RDR). RDR is an Artificial Intelligence(AI) based approach and an alternative technique to Machine Learning(ML) algorithms for incrementally building the knowledge base. In this implementation, we have used RDR to develop an improving and increasing knowledge base of classification rules for assigning rhetorical role labels to the sentences in a legal document. The RDR rules for classification are developed using a set of syntactic, semantic and statistical features at word, sentence and document level. For each sentence in the legal document we have assigned an rhetorical role with the help of Ripple Down Rule. We have generated the final summary using the identified thirteen rhetorical roles. The proposed system is evaluated using 50 legal documents from four different domains. Experiments demonstrate that the RDR based Legal Document summarization approach has advantages over supervised and unsupervised ML techniques such as, independence from the need of annotated dataset and continuous updation of classification rules for rhetorical role labeling with the help of expert knowledge.