Research map / Large language models / Language, safety and interpretability
A Multi-Task Benchmark for Korean Legal Language Understanding and Judgement Prediction
Wonseok Hwang, Dongjun Lee, Kyoungyeon Cho, Hanuhl Lee, Minjoon Seo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pre-train on Precedent corpus (LCube-base vs LCube-base-zero) (ablation) | LCube-base · Case name | accuracy (EM): 79.6 -> 81.1 | Improved |
| pre-train on Precedent corpus (LCube-base vs LCube-base-zero) (ablation) | LCube-base · Statute | accuracy (EM): 85.8 -> 87.6 | Improved |
| pre-train on Precedent corpus (ablation) | LCube-base · Case name+ | accuracy (EM): 75.6 -> 78.9 | Improved |
| add domain adaptation on Precedent corpus (ablation) | LCube-base · Case name | accuracy (EM): 81.1 -> 82.7 | Improved |
| add domain adaptation on Precedent corpus (ablation) | KoGPT-2-base · Statute | accuracy (EM): 85.7 -> 89.4 | Improved |
| add domain adaptation (ablation) | LCube-base · LJP-civil Lv1 | accuracy (EM): 67.6 -> 60.9 | Hurt |
| add domain adaptation (ablation) | LCube-base · LJP-civil Lv2 | accuracy (EM): 58.8 -> 52.6 | Hurt |
| add domain adaptation (ablation) | mt5-small · LJP-civil Lv1 | accuracy (EM): 68.9 -> 69.1 | No clear effect |
| add domain adaptation (ablation) | LCube-base · LJP-criminal | F1 | No clear effect |
| remove facts from input (gist of claim only) (ablation) | LCube-base · LJP-civil Lv1 | accuracy (EM): 67.6 -> 58.5 | Hurt |
| replace facts with case names (ablation) | LCube-base · LJP-criminal Lv0 | F1: -3.7%, -7.2%, -6.9% (fine, imp. w/ labor, imp. w/o labor) | Hurt |
| add reason for sentencing section to input (oracle) (ablation) | LCube-base · LJP-criminal Lv0 fine | F1: 67.2 -> 77.5 | Improved |
Closest papers by meaning
- LexEval: A Comprehensive Chinese Legal Benchmark for Evaluating Large Language Models (NeurIPS 2024 · similarity 0.88)
- Lawma: The Power of Specialization for Legal Annotation (ICLR 2025 · similarity 0.87)
- SaulLM-54B & SaulLM-141B: Scaling Up Domain Adaptation for the Legal Domain (NeurIPS 2024 · similarity 0.84)
- Interpretable Long-Form Legal Question Answering with Retrieval-Augmented Large Language Models (AAAI 2024 · similarity 0.84)
- DELTA: Pre-Train a Discriminative Encoder for Legal Case Retrieval via Structural Word Alignment (AAAI 2025 · similarity 0.84)
- J&H: Evaluating the Robustness of Large Language Models Under Knowledge-Injection Attacks in Legal Domain (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
