Reproduce “EUROPA: A Legal Multilingual Keyphrase Generation Dataset” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Olivier Salaün, Frédéric Piedboeuf, Guillaume Le Berre, David Alfonso Hermelo, Philippe Langlais
Researchers from Université de Montréal introduced EUROPA, the first open multilingual keyphrase generation dataset for the legal domain, featuring 24 EU official languages and derived from over 17,000 Court of Justice of the European Union judgments. Baseline evaluations on this dataset reveal the necessity of generative models over extractive ones and demonstrate performance improvements with extended input context models for lengthy legal documents.
Key insights
- Legal KPG heavily relies on abstractive generation capabilities, as over 50% of reference keyphrases in the EUROPA dataset are not literally present within the source documents.
- Processing longer document contexts is critical for enhancing KPG performance in the legal domain, with models capable of handling extended input showing measurable improvements.
- Significant performance disparities exist across languages, indicating that KPG in low-resource languages remains a substantial challenge, particularly within a chronological data split.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.