french-legal-cases

---

annotations_creators: - no-annotation language: - fr language_creators: - found license: - cc-by-4.0 multilinguality: - monolingual pretty_name: French Legal Cases Dataset size_categories: - n>1M source_datasets: - la-mousse/INCA-17-01-2025 - la-mousse/JADE-17-01-2025 - la-mousse/CASS-17-01-2025 - la-mousse/CAPP-17-01-2025 task_categories: - text-generation - text-classification task_ids: - language-modeling - entity-linking-classification - fact-checking - intent-classification - multi-label-classification - multi-input-text-classification - natural-language-inference - semantic-similarity-classification - sentiment-classification - topic-classification - sentiment-analysis - named-entity-recognition - parsing - extractive-qa - open-domain-qa - closed-domain-qa - news-articles-summarization - news-articles-headline-generation - dialogue-modeling - dialogue-generation - abstractive-qa - closed-domain-qa - keyword-spotting - semantic-segmentation - tabular-multi-class-classification - tabular-multi-label-classification - document-retrieval - document-question-answering paperswithcode_id: french-legal-cases dataset_info: features: - name: id dtype: string - name: url dtype: string - name: case_number dtype: string - name: contenu dtype: string - name: sommaire dtype: string - name: sommaire_bis dtype: string - name: metadata dtype: string - name: dataset_source dtype: string - name: previous_text dtype: string - name: current_text dtype: string - name: next_text dtype: string - name: triplet_index dtype: int64 configs: - config_name: default data_files: - split: train path: data/train-*

---

Dataset Card for French Legal Cases Dataset

Dataset Description

• Homepage: https://huggingface.co/datasets/la-mousse/combined-fe-caselaw
• Repository: https://huggingface.co/datasets/la-mousse/combined-fe-caselaw
• Paper: N/A
• Point of Contact: [Your Contact Information]

Dataset Summary

This dataset combines French legal cases from multiple sources (INCA, JADE, CASS, CAPP) into a unified format with overlapping text triplets. It includes decisions from various French courts, processed to facilitate natural language processing tasks.

Supported Tasks and Leaderboards

• Tasks:
• Text Generation
• Legal Document Analysis
• Text Classification
• Language Modeling

Languages

The dataset is monolingual (French).

Dataset Structure

Data Instances

Each instance contains: - Document identifiers (id, url, case_number) - Original text content (contenu) - Summaries (sommaire, sommaire_bis) - Metadata (JSON formatted string) - Text triplets (previous_text, current_text, next_text) - Position tracking (triplet_index, window_index)

Example:

{
    'id': 'CASS12345',
    'url': 'https://...',
    'case_number': '12-34567',
    'contenu': 'Full text...',
    'sommaire': 'Summary...',
    'sommaire_bis': 'Additional summary...',
    'metadata': '{"date_decision": "2023-01-01", ...}',
    'dataset_source': 'CASS',
    'previous_text': 'Previous chunk...',
    'current_text': 'Current chunk...',
    'next_text': 'Next chunk...',
    'triplet_index': 0,
    'window_index': 0
}

Data Fields

• id: Unique identifier
• url: Source URL
• case_number: Case reference number
• contenu: Full text content
• sommaire: Primary summary
• sommaire_bis: Secondary summary
• metadata: JSON string containing additional metadata
• dataset_source: Origin dataset (INCA/JADE/CASS/CAPP)
• previous_text: Previous text chunk
• current_text: Current text chunk
• next_text: Next text chunk
• triplet_index: Position in sequence of triplets
• window_index: Window number for long texts

Data Splits

• Training split only

Dataset Creation

Curation Rationale

This dataset was created to provide a standardized format for French legal texts, with overlapping text chunks suitable for various NLP tasks.

Source Data

Initial Data Collection and Normalization

• INCA: Court of Cassation decisions
• JADE: Administrative court decisions
• CASS: Court of Cassation decisions
• CAPP: Court of Appeal decisions

Preprocessing

• Text chunking with 230-token chunks and 30-token overlap
• Sliding window approach for long texts
• Metadata preservation and standardization
• Token count verification
• JSON formatting for metadata

Quality Control

• Token length verification
• Chunk coherence checks
• Metadata validation
• Error logging and handling

Considerations for Using the Data

Social Impact of Dataset

This dataset aims to improve access to and understanding of French legal decisions, potentially benefiting legal research and analysis.

Discussion of Biases

The dataset may reflect inherent biases in the French legal system and case selection/publication processes.

Other Known Limitations

• Limited to published decisions
• Varying detail levels across sources
• Potential OCR errors in source texts

Additional Information

Dataset Curators

La-Mousse

Licensing Information

CC-BY-4.0

Citation Information

@misc{french-legal-cases-2024,
    title={French Legal Cases Dataset},
    author={[Your Name]},
    year={2024},
    publisher={HuggingFace},
    url={https://huggingface.co/datasets/la-mousse/combined-fr-caselaw}
}

Contributions

Thanks to @huggingface for the dataset hosting and infrastructure.