othertales/uklegislation

UK Legislation Dataset

This directory packages scraped UK legislation into a layout that can be ingested directly with the Hugging Face datasets library. All documents live in JSON Lines format inside data/ with one piece of legislation per line. The schema captures both plain-text and XML renderings, along with document-level metadata and section breakdowns.

Repository Layout

• data/train.jsonl – full corpus of 175,515 documents ready for load_dataset.
• meta/ – auxiliary files that describe the scrape workflow and provenance.
• dataset_config.json – declarative configuration that mirrors the load_dataset arguments used during training.

Data Fields

Every record is a JSON object with the following keys:

• id (string): canonical legislation identifier, e.g. ukpga_2025_23.
• title (string): preferred title.
• type (string): legislation class (ukpga, uksi, etc.).
• year (int): legislation year.
• number (int): act or statutory instrument number within the year.
• text_content (string): consolidated plain-text content.
• xml_content (string): raw XML body as served by legislation.gov.uk.
• publication_date, enacted_date, coming_into_force_date (string): ISO dates when known.
• extent (array[string]): geographic extent codes.
• status (string): legislation status (In Force, Prospective, etc.).
• sections (array[object]): ordered section data with id, number, title, content fields per section.
• metadata (object): scraped metadata such as classification and title.
• url (string): source URL.
• checksum (string): SHA1 checksum of the XML payload.

Loading With datasets

from datasets import load_dataset

dataset = load_dataset(
    "json",
    data_files={"train": "data/train.jsonl"},
    features={
        "id": "string",
        "title": "string",
        "type": "string",
        "year": "int32",
        "number": "int32",
        "text_content": "string",
        "xml_content": "string",
        "publication_date": "string",
        "enacted_date": "string",
        "coming_into_force_date": "string",
        "extent": ["string"],
        "status": "string",
        "sections": [{
            "id": "string",
            "number": "string",
            "title": "string",
            "content": "string",
        }],
        "metadata": {
            "classification": "string",
            "title": "string",
        },
        "url": "string",
        "checksum": "string",
    },
)

You can further split the dataset with the usual train_test_split utilities or by sharding train.jsonl into additional files.

Provenance & Licensing

The dataset originates from legislation.gov.uk. Legislative content is generally covered by the UK Open Government Licence v3.0, but confirm applicability to your downstream use-case.