LexAbSumm: Aspect-based Summarization of Legal Decisions

TL;DR

LexAbSumm introduces the first aspect-based summarization dataset for legal judgments drawn from ECtHR materials, addressing the need for targeted summaries over generic ones. It evaluates several long-document abstractive models (e.g., LED, PRIMERA, LongT5, SLED, Unlimiformer) with aspect-conditioned inputs and assesses them using ROUGE and BERTScore. Findings show that while abstractive models outperform baselines, conditioning on specific aspects remains challenging, especially for unseen aspects and across input variants (facts, law, whole). The work provides a valuable resource for developing better aspect-aware legal summarization and highlights directions for model design to improve aspect conditioning and reduce non-aspect language generation.

Abstract

Legal professionals frequently encounter long legal judgments that hold critical insights for their work. While recent advances have led to automated summarization solutions for legal documents, they typically provide generic summaries, which may not meet the diverse information needs of users. To address this gap, we introduce LexAbSumm, a novel dataset designed for aspect-based summarization of legal case decisions, sourced from the European Court of Human Rights jurisdiction. We evaluate several abstractive summarization models tailored for longer documents on LexAbSumm, revealing a challenge in conditioning these models to produce aspect-specific summaries. We release LexAbSum to facilitate research in aspect-based summarization for legal domain.

Figures (3)

• Figure 1: Example of an aspect-judgement-summary triplet from the fact sheet PDF file.
• Figure 2: Distribution of input and output tokens in three LexAbSumm splits.
• Figure 3: Distribution of tokens in the aspects of LexAbSumm.