CourtPressGER: A German Court Decision to Press Release Summarization Dataset

TL;DR

This paper introduces CourtPressGER, a new dataset of German court rulings and press releases, enabling training and evaluation of LLMs for citizen-oriented judicial summary generation, highlighting model performance differences.

Contribution

The paper presents CourtPressGER, a novel dataset for judicial summarization, and benchmarks various LLMs, emphasizing the importance of hierarchical setups for smaller models and the high quality of human drafts.

Findings

Large LLMs generate high-quality summaries with minimal hierarchical loss

Smaller models need hierarchical setups for long judicial texts

Human-drafted press releases outperform model-generated summaries

Abstract

Official court press releases from Germany's highest courts present and explain judicial rulings to the public, as well as to expert audiences. Prior NLP efforts emphasize technical headnotes, ignoring citizen-oriented communication needs. We introduce CourtPressGER, a 6.4k dataset of triples: rulings, human-drafted press releases, and synthetic prompts for LLMs to generate comparable releases. This benchmark trains and evaluates LLMs in generating accurate, readable summaries from long judicial texts. We benchmark small and large LLMs using reference-based metrics, factual-consistency checks, LLM-as-judge, and expert ranking. Large LLMs produce high-quality drafts with minimal hierarchical performance loss; smaller models require hierarchical setups for long judgments. Initial benchmarks show varying model performance, with human-drafted releases ranking highest.