| --- |
| task_categories: |
| - text-classification |
| language: |
| - pl |
| tags: |
| - legal |
| - long-context |
| - document-classification |
| - scotus |
| - polish |
| - translated |
| pretty_name: Polish SCOTUS-Dom |
| size_categories: |
| - 1K<n<10K |
| --- |
| # Polish SCOTUS-Dom |
|
|
| **Polish SCOTUS-Dom** is a long-document legal text classification dataset |
| derived from the publicly available |
| [SCOTUS](https://huggingface.co/datasets/pasinit/scotus) dataset. |
|
|
| The dataset contains Polish translations of U.S. Supreme Court opinions. |
| The task is an **11-class document classification problem** in which the goal |
| is to predict the legal issue area of a court case. |
|
|
| SCOTUS-Dom is part of the **LongContext** benchmark introduced with |
| [Polish ModernBERT](ARXIV_URL). |
|
|
| ## Dataset statistics |
|
|
| | Split | Examples | |
| |---|---:| |
| | Train | 7,413 | |
| | Validation | 912 | |
| | Test | 931 | |
| | **Total** | **9,256** | |
|
|
| The dataset contains particularly long documents: **74.5% of test examples |
| exceed 4,096 tokens** when tokenized with the Polish ModernBERT Base tokenizer. |
|
|
| ## Dataset construction |
|
|
| The original English court opinions were translated into Polish using |
| **GLM-4.6**. The original SCOTUS train, validation, and test splits and task |
| labels were retained. |
|
|
| ## Citation |
|
|
| If you use this dataset, please cite: |
|
|
| [CITATION] |
