Impact of Rhetorical Roles in Abstractive Legal Document Summarization

This paper explores the relationships between rhetorical roles and the summarization of legal documents. By employing automated interpretation techniques, we segment legal judgment documents and identify rhetorical roles, treating it as a 13-class labeling problem. Using CRFs and a BiLSTM architecture, we extract rhetorical roles. Our study extends further by employing an ensemble summarizer to examine the impact of each rhetorical role on the summarization process and evaluate the same using ROGUE and BLEU scores. Through experimentation, we explore the impact of selecting individual rhetorical roles from the documents, thereby facilitating a comprehensive analysis of its effect on various scoring metrics. Our analysis reveals that certain roles, such as FAC and ANALYSIS, contribute significantly to summary quality, while others like ISSUE, have a less pronounced impact which interestingly also tend to be included in the summary almost verbatim, underscoring their inherent utility in the summarization process. By shedding light on this process, our research aims to equip researchers with valuable insights into streamlining summarization techniques, potentially reducing the volume of text data processed. Ultimately, these findings could pave the way for the development of more efficient summarization algorithms tailored to different roles in legal documents, enhancing accessibility and comprehension for professionals and researchers alike.