The study addresses the lack of datasets by introducing CLSum, focusing on multi-jurisdictional common law court judgments. It employs LLMs for data augmentation to mitigate low data resources' impact and evaluates the quality of generated summaries comprehensively.
Previous research focused on civil law or specific jurisdictions, but this work targets all common law jurisdictions. The study emphasizes efficient content selection and integration to preserve key information in long judgment documents.
Models like LEDBase, Legal-LED, Vicuna7B, and Vicuna13B show competitive performance in zero-shot settings. Few-shot performance improves with training set size but diminishes as it increases.
Human evaluation indicates that Vicuna models outperform LEDLarge in informativeness across all subsets. Fleiss' kappa values suggest moderate agreement among annotators.
To Another Language
from source content
arxiv.org
ข้อมูลเชิงลึกที่สำคัญจาก
by Shuaiqi Liu,... ที่ arxiv.org 03-08-2024
https://arxiv.org/pdf/2403.04454.pdfสอบถามเพิ่มเติม