LEXTREME: A Multi-Lingual and Multi-Task Benchmark for the Legal Domain

TL;DR

LEXTREME is a comprehensive multilingual and multi-task benchmark for legal NLP, designed to evaluate models across diverse languages and datasets, highlighting the field's challenges and progress.

Contribution

The paper introduces LEXTREME, the first multilingual and multi-task legal NLP benchmark, with aggregate scoring and open resources for evaluation.

Findings

Best baseline (XLM-R large) scores 61.3 on aggregate metrics.

Benchmark remains challenging with significant room for improvement.

Provides a diverse, multilingual dataset covering 24 languages.

Abstract

Lately, propelled by the phenomenal advances around the transformer architecture, the legal NLP field has enjoyed spectacular growth. To measure progress, well curated and challenging benchmarks are crucial. However, most benchmarks are English only and in legal NLP specifically there is no multilingual benchmark available yet. Additionally, many benchmarks are saturated, with the best models clearly outperforming the best humans and achieving near perfect scores. We survey the legal NLP literature and select 11 datasets covering 24 languages, creating LEXTREME. To provide a fair comparison, we propose two aggregate scores, one based on the datasets and one on the languages. The best baseline (XLM-R large) achieves both a dataset aggregate score a language aggregate score of 61.3. This indicates that LEXTREME is still very challenging and leaves ample room for improvement. To make it easy for researchers and practitioners to use, we release LEXTREME on huggingface together with all the code required to evaluate models and a public Weights and Biases project with all the runs.