README.md

---
license: mit
task_categories:
- text-classification
- text-generation
language:
- de
tags:
- legal
pretty_name: Clean Open Legal Data
size_categories:
- 100K<n<1M
configs:
- config_name: cases
  data_files:
  - split: main
    path: data/cases.jsonl.gz
---
<h1 align="center">Clean Open Legal Data</h1>
<h4 align="center">
   <p>
       <a href=#overview>Overview</a> |
       <a href=#dataset-structure>Dataset Structure</a>  |
       <a href=#key-fields>Key Fields</a>  |
       <a href=#example-entry>Example Entry</a> |
       <a href=#using-the-dataset-with-python>Using the Dataset with Python</a> |
       <a href=#license>License</a>
   <p>
</h4>

## Overview

This dataset is a comprehensive collection of open legal case records in JSONL format. It comprises **251,038** cases extracted and processed from the [Open Legal Data dump](https://static.openlegaldata.io/dumps/de/2022-10-18/) (as of _2022-10-18_). The dataset is designed for legal research, data science, and natural language processing applications. While the majority of entries (especially _tenor_, _tatbestand_, _gründe_, and _entscheidungsgründe_) have been carefully separated and extracted, approximately **1,176** (**0.47%** of total) cases have been flagged as "_messy_" due to extraction or formatting issues. These messy entries are included for full transparency, and their "_slug_" IDs are available in [`problematic_case_slugs.txt`](https://huggingface.co/datasets/harshildarji/openlegaldata/blob/main/problematic_case_slugs.txt) so users can decide whether to filter them out or work with them. In references, law references and case references are also separated, as shown in the [example entry](#example-entry).

## Dataset Structure

```
├── README.md
├── data
│   └── cases.jsonl.gz
└── problematic_case_slugs.txt
```

- **Language:** German
- **Format:** JSONL
- **Total Cases:** 251,038

## Key Fields

- **id:** Unique identifier for the record.
- **file_number:** Identifier for the case (e.g., `"1 A 2639/20"`).
- **slug:** URL-friendly unique identifier (e.g., `"ovgnrw-2022-03-25-1-a-263920"`).
- **ecli:** European Case Law Identifier.
- **date:** Date of the decision in `YYYY-MM-DD` format.
- **court:** JSON object with court details (e.g., _name_, _city_, _state_, _jurisdiction_).
- **type:** Type of legal decision (e.g., "Beschluss").
- **tenor:** List of summary statements of the decision.
- **tatbestand:** List of factual background details.
- **gründe:** List of reasoning or legal arguments provided.
- **entscheidungsgründe:** Detailed decision reasons.
- **references:** Contains references to laws and related cases.

## Example Entry

Below is an example entry from the JSONL file:

```json
{
    "id": 344319,
    "file_number": "1 A 2639/20",
    "slug": "ovgnrw-2022-03-25-1-a-263920",
    "ecli": "ECLI:DE:OVGNRW:2022:0325.1A2639.20.00",
    "date": "2022-03-25",
    "court": {
        "id": 823,
        "name": "Oberverwaltungsgericht Nordrhein-Westfalen",
        "slug": "ovgnrw",
        "city": "Unspecified",
        "state": "Nordrhein-Westfalen",
        "jurisdiction": "Verwaltungsgerichtsbarkeit",
        "level_of_appeal": null
    },
    "type": "Beschluss",
    "tenor": [
        "Der Antrag wird abgelehnt.",
        "...",
        "Der Streitwert wird auch für das Zulassungsverfahren auf 7.500,00 Euro festgesetzt."
    ],
    "tatbestand": [],
    "gründe": [
        "Der Antrag des Klägers auf Zulassung der Berufung hat keinen Erfolg.",
        "...",
        "Das angefochtene Urteil ist nunmehr rechtskräftig, § 124a Abs. 5 Satz 4 VwGO."
    ],
    "entscheidungsgründe": [],
    "references": {
        "law": [
            "§ 124a Abs. 4 Satz 4 VwGO",
            "§ 124a Abs. 5 Satz 4 VwGO",
            "..."
        ],
        "case": [
            "1 A 5162/05",
            "1 KR 87/11",
            "..."
        ]
    }
}
```

## Using the Dataset with Python

Below is an example of how to load and explore the dataset using Python with the [🤗 Datasets](https://huggingface.co/docs/hub/datasets-usage) library:

```python
import json

from datasets import load_dataset

# Load cases
cases = load_dataset("harshildarji/openlegaldata", "cases", split="main")

# Total cases
print(len(cases))

# View first case
print(json.dumps(cases[0], indent=4, default=str, ensure_ascii=False))
```

## License

This dataset is released under the MIT license, same license as the [Open Legal Data platform](https://github.com/openlegaldata/oldp).