Multi-LexSum: US Civil Rights Lawsuits with Multi-Level Summaries

Multi-LexSum contains between 1,000 and 10,000 US civil rights litigation cases featuring expert-generated summaries at three distinct granularities. Produced by the Allen Institute for AI in 2022, it serves as a benchmark for multi-document summarization within the legal domain.

Use Cases

• Training models to generate summaries of three different granularities for legal proceedings
• Evaluating multi-document summarization algorithms on civil rights litigation source texts
• Fine-tuning language models for legal domain-specific terminology and document structure

Strengths

• Expert-generated summaries provide high-quality ground truth
• Three levels of granularity allow for testing summary length control
• Multi-document source material provides a complex summarization challenge

Limitations

• Domain-specific focus on US civil rights litigation limits transferability to general text
• Monolingual English content restricts use for cross-lingual tasks

Provenance

Source

Allen Institute for AI (Arxiv 2206.10883)

Collection Method

Expert annotation of found legal documents

Freshness

Last updated August 2024; based on 2022 research.

Geography

United States