d3b4g/maldivian-legal-corpus

Maldivian Legal Corpus (V1.1)

Open, structured corpus of Maldivian laws published on HuggingFace — sourced from the official MVLaw portal maintained by the Attorney General's Office of the Maldives.

Dataset summary

235 published Maldivian laws in Dhivehi (ދިވެހި), spanning from 1932 to 2025.

Stat	Value
Total laws	235 (215 Train / 20 Test)
Total sections	~14,000 (12,742 Train / 1,246 Test)
Estimated Dhivehi tokens	~3,000,000
Average chars per law	~50,000
Average sections per law	~59
Language	Dhivehi (dv)
Source	mvlaw.gov.mv
License	CC-BY-4.0

Why this dataset matters

Dhivehi is a critically low-resource language spoken by ~340,000 people in the Maldives. This dataset enables:

• LLM pretraining / fine-tuning on Dhivehi legal text.
• Legal RAG systems for Maldivian law (granular section-level chunking).
• Translation research — Dhivehi ↔ English legal translation.
• Legal NLP benchmarking for a low-resource language.
• Civic tech — searchable Maldivian law for citizens and lawyers.

Dataset structure

Configurations

1. default (Laws): Full legal documents (1 row = 1 Law).
2. sections (Granular): Broken down by Articles/Chapters (1 row = 1 Section).

Fields (default config)

Field	Type	Description
id	int	Unique legislation ID from MVLaw
title	string	Law title in Dhivehi
slug	string	URL-safe identifier
legislation_no	string	Official law number (e.g. 17/2021)
status	string	Always published in this release
implemented_at	string	Enactment date (ISO 8601)
pdf_url	string	Direct URL to official PDF
web_url	string	URL to law on MVLaw portal
categories	list[string]	Category tags (e.g. ["ޤާނޫނު", "ވޮލިއުމް VIII"])
text_content	string	Full law text in Dhivehi with [section heading] markers

How to use

1. Load full laws

from datasets import load_dataset

ds = load_dataset("d3b4g/maldivian-legal-corpus", split="train")
print(ds[0]["title"])

2. Load granular sections (for RAG)

from datasets import load_dataset

ds_sections = load_dataset("d3b4g/maldivian-legal-corpus", "sections", split="train")
print(ds_sections[0]["section_title"])

3. Load test split for evaluation

from datasets import load_dataset

ds_test = load_dataset("d3b4g/maldivian-legal-corpus", split="test")

Known limitations

• 18 laws excluded — scanned image-only PDFs (mostly pre-1970). OCR versions will be added in V2.
• 2 laws excluded — failed to fetch at scrape time; will be re-scraped.
• No English translations — An English translation is planned for V2.
• Arabic-numeral law numbers — some very old laws use Arabic-Indic numerals (e.g. .٨/ ٣ ١).

Data source & provenance

All laws are sourced from mvlaw.gov.mv, the official Maldivian laws portal maintained by the Attorney General's Office of the Republic of Maldives. Data was collected via web scraping in 2025.

License

This dataset is released under CC-BY-4.0. Underlying laws are public government documents. Please cite this dataset if you use it in research.

Citation

@dataset{maldivian_legal_corpus_2025,
  author    = {d3b4g},
  title     = {Maldivian Legal Corpus},
  year      = {2025},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/datasets/d3b4g/maldivian-legal-corpus},
  note      = {235 published Maldivian laws in Dhivehi, sourced from mvlaw.gov.mv}
}

Roadmap

• v1.1: Add sections config with one row per article (DONE)
• v2: Add English translations