Citation-Based Summarization of Landmark Judgments

TL;DR

The paper addresses automatic extractive summarization of landmark judgments by leveraging contextual citations from citing judgments in a Common Law setting. It introduces CB-JSumm, an unsupervised three-phase pipeline that uses InLegalBert embeddings to form a cosine similarity matrix $ S$ between citances and target judgment sentences and a set of scoring strategies (CiSumm, Additive, CD) to produce compact summaries of length $l$, while preserving the original order. The authors curate two Indian datasets, IN-Jud-Cit and IN-Ext-Cit, and demonstrate that the Citation Diversity (CD) scoring method often yields the best performance in terms of ROUGE and semantic similarity, outperforming several baselines. The work contributes a novel algorithm, two openly available datasets, and evidence that citation-context-driven extractive summarization can effectively capture key issues and arguments in landmark judgments, with potential to aid legal research and jurisprudential analysis.

Abstract

Landmark judgments are of prime importance in the Common Law System because of their exceptional jurisprudence and frequent references in other judgments. In this work, we leverage contextual references available in citing judgments to create an extractive summary of the target judgment. We evaluate the proposed algorithm on two datasets curated from the judgments of Indian Courts and find the results promising.

Figures (2)

• Figure 1: A target landmark judgment $\mathbf J$, set $\mathbf C$ of citing judgments, and the corpus of extracted citation sentences $\mathbf S$
• Figure 2: Pipeline for proposed CB-JSumm Algorithm