SaudiArabicLaws/saudi-arabic-laws-and-regulations-corpus

Saudi Arabic Laws and Regulations Corpus

A structured, article-level Arabic legal corpus containing 22,593 records from 578 official Saudi legal documents, prepared for Arabic legal information retrieval, Retrieval-Augmented Generation (RAG), grounded generation, and LLM evaluation.

 

Release: 1.0.0 Language: Arabic Domain: Saudi laws and regulations Granularity: Legal article Total records: 22,593 Archival DOI: 10.5281/zenodo.21265180

Quick Start

The corpus is exposed as two Hugging Face configurations because the BOE and MOJ collections preserve source-specific schemas.

from datasets import load_dataset

REPO_ID = "SaudiArabicLaws/saudi-arabic-laws-and-regulations-corpus"

boe = load_dataset(REPO_ID, "boe")
moj = load_dataset(REPO_ID, "moj")

The BOE collection is the default configuration:

from datasets import load_dataset

dataset = load_dataset("SaudiArabicLaws/saudi-arabic-laws-and-regulations-corpus")

Streaming is also supported:

from datasets import load_dataset

stream = load_dataset(
    "SaudiArabicLaws/saudi-arabic-laws-and-regulations-corpus",
    "boe",
    split="train",
    streaming=True
)

Corpus Composition

The two collections are distributed separately to preserve source-specific metadata structures.

During cross-source deduplication, overlapping provisions were identified using normalized-text similarity. Where overlap was found, the BOE representation was retained as the primary version.

Dataset Statistics

Release-Level Profile

The two collections differ substantially in source structure and metadata density. For that reason, source-specific structural statistics are reported separately rather than forcing both configurations into a single descriptive schema.

BOE Article-Length Profile

Article length below is measured on the released article.text field in Unicode characters.

The gap between the mean and median reflects a corpus containing many concise provisions together with a smaller number of substantially longer statutory articles.

BOE Legal Status Distribution

BOE Structural Indicators

These values are derived directly from the released BOE xai_features fields.

BOE Sector Coverage

The five largest BOE sectors by article-level record count are:

These distributions describe the released corpus and should not be interpreted as the overall distribution of Saudi legislation.

BOE Document-Type Distribution

Deduplication Note

Cross-source deduplication was applied during corpus construction, and released BOE records are marked with dedup_applied = true.

A removal percentage is intentionally not reported because the released corpus does not provide a reliable pre-deduplication denominator from which such a percentage could be calculated.

Data Design

The corpus uses legal articles as retrieval units rather than arbitrary fixed-length chunks.

A single statutory article may contain a rule together with its conditions, exceptions, procedural requirements, qualifications, consequences, and cross-references. Fixed-length segmentation can separate these dependent elements and weaken the evidentiary value of retrieved text.

The core design principle is therefore:

One legal article = one evidence unit

This supports fine-grained retrieval while preserving the legal coherence of each provision.

Dataset Configurations

boe

File: data/BOE/boe_legal_articles_corpus.jsonl Source documents: 507 Article-level records: 16,473

The BOE representation includes, where available:

• —record and source identifiers
• —law title
• —legal status
• —document type
• —sector
• —article number and title
• —Arabic legal text
• —source HTML
• —citation information
• —sparse and dense retrieval representations
• —contextual and structural retrieval text
• —content and amendment indicators
• —explainability-oriented features

Detailed BOE field definitions, including the xai_features indicators, are documented in `data_dictionary.csv`.

moj

File: data/MOJ/moj_saudi_legal_corpus.jsonl Source documents: 71 Article-level records: 6,120

The MOJ representation includes, where available:

• —document identifiers and title
• —legislation type
• —category
• —legal status
• —Hijri and ISO issue dates
• —source URL
• —article number and numbering style
• —article sequence and structural context
• —Arabic legal text
• —amendment indicators, including added, modified, and canceled status

For exact field names, types, and descriptions, see `data_dictionary.csv`.

Repository Structure

.
├── README.md
├── data_dictionary.csv
├── license.txt
├── sources.csv
└── data/
    ├── BOE/
    │   └── boe_legal_articles_corpus.jsonl
    └── MOJ/
        └── moj_saudi_legal_corpus.jsonl

Each line in the two .jsonl files contains one JSON object representing one article-level legal record.

Provenance

The corpus was constructed from publicly accessible legal material published through official Saudi government portals.

Bureau of Experts at the Council of Ministers

Primary source for Saudi laws and regulations.

https://www.boe.gov.sa

Ministry of Justice

Complementary source for Saudi legal and judicial-regulatory material.

https://www.moj.gov.sa

A document-level source inventory is provided in `sources.csv`, including source URLs where available.

Construction Pipeline

1. Collection

Legal pages were collected from the BOE and MOJ portals. Playwright was used where dynamic JavaScript rendering required browser-based acquisition.

2. Parsing

HTML pages were parsed with BeautifulSoup and transformed into structured document-level and article-level representations.

3. Article Extraction

Individual legal provisions were identified from the source document structure and retained as discrete records.

4. Metadata Enrichment

Records were enriched with available document, hierarchy, numbering, status, category, source, and citation metadata.

5. Arabic Normalization

Retrieval-oriented representations were prepared using normalization procedures addressing:

• —diacritics and tatweel
• —Alef variants
• —letter-form variation
• —Arabic-Indic digits
• —punctuation
• —whitespace
• —article-reference expressions

6. Retrieval Preparation

The corpus includes representations suitable for lexical and semantic retrieval workflows, including use with:

• —BM25
• —dense embedding retrieval
• —hybrid retrieval
• —reranking
• —RAG evidence retrieval
• —citation retrieval
• —claim-to-evidence matching

7. Cleaning and Quality Control

Empty or invalid records and collection artifacts were addressed during dataset preparation.

8. Cross-Source Deduplication

Potential overlap between BOE and MOJ provisions was evaluated using normalized-text similarity. Where overlapping provisions were identified, the BOE representation was retained.

Schema Reference

The BOE and MOJ collections intentionally preserve different schemas because the underlying official portals expose different document structures and metadata.

The authoritative field-level reference is:

`data_dictionary.csv`

It documents the released fields, their source collection, data type, and intended meaning.

This separation avoids reducing source-specific legal metadata into a lossy common schema.

Research Applications

The corpus is intended to support research in:

• —Arabic Legal NLP
• —Saudi legal information retrieval
• —lexical retrieval
• —dense retrieval
• —hybrid retrieval
• —Retrieval-Augmented Generation
• —legal question answering
• —citation-aware retrieval
• —evidence-grounded generation
• —hallucination evaluation
• —faithfulness evaluation
• —claim-level verification
• —evidence attribution
• —explainable AI
• —legal document classification
• —corpus-level analysis of Saudi legislation

The dataset is model-agnostic and does not prescribe a particular retriever, embedding model, reranker, LLM, generation framework, or evaluation methodology.

Scope and Limitations

This release focuses on Saudi statutory, regulatory, and related legal material represented in the included BOE and MOJ collections.

It does not contain:

• —judicial decisions
• —court judgments
• —case law
• —legal outcome labels

The corpus should therefore not be interpreted as a case-law dataset or used for judicial outcome prediction.

Legal Currency

Law is dynamic. Provisions may be amended, replaced, suspended, or repealed after a dataset release.

Status and amendment information reflects the source material available during construction of this version and may not capture subsequent legal developments.

For applications where current legal accuracy is consequential, retrieved provisions should be verified against the corresponding official government source.

Source Coverage

The corpus represents material available through the included BOE and MOJ collections and should not be interpreted as a complete representation of every form of Saudi legal authority.

Source Heterogeneity

BOE and MOJ expose different metadata structures. Their schemas are intentionally preserved separately instead of being reduced to a lossy common representation.

Normalized Text

Normalized fields are primarily intended for retrieval and NLP use. Researchers requiring the closest representation of the published source should use the corresponding original or minimally processed fields where available.

Responsible Use

This dataset is a research resource and should not be presented or used as a substitute for professional legal advice.

Outputs generated by systems trained on or retrieving from this corpus should not automatically be treated as authoritative legal interpretation.

When legal accuracy or currency is consequential, information should be verified against the current official source.

License and Access

This dataset is released under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license.

The license applies to the compiled, structured, and processed dataset release and does not alter any rights or legal status associated with the underlying official legal texts.

The NC and ND terms place restrictions on some forms of reuse and redistribution. Users planning commercial use, redistribution, or publication of transformed versions should review the license terms and the dataset-specific clarification in `license.txt`.

Versioning

The archival record for this dataset is maintained on Zenodo.

The version DOI identifies this specific release.

The concept DOI represents the dataset across versions and resolves to the latest Zenodo release.

Zenodo serves as the archival citation record.

Authors

Author ordering follows the published Zenodo record.

Citation

If you use this dataset in academic or research work, please cite the archived Zenodo release:

@dataset{saudi_arabic_laws_2026,
  author    = {Almunyah, Alaa and
               Almehmadi, Aseel and
               Almaashi, Fatima and
               Sirdar, Hadeel and
               Alamri, Shaden},
  title     = {Saudi Arabic Laws and Regulations Corpus},
  year      = {2026},
  version   = {1.0.0},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.21265180},
  url       = {https://doi.org/10.5281/zenodo.21265180}
}

APA

Almunyah, A., Almehmadi, A., Almaashi, F., Sirdar, H., & Alamri, S. (2026). Saudi Arabic Laws and Regulations Corpus (Version 1.0.0) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.21265180

Archival Record

Zenodo — Version 1.0.0 https://doi.org/10.5281/zenodo.21265180

Zenodo — All Versions https://doi.org/10.5281/zenodo.21265179