UDHR in every language: download all 533 translations as text and XML

The Universal Declaration of Human Rights (UDHR) is the United Nations (UN) statement of basic rights from 1948, the Office of the UN High Commissioner for Human Rights (OHCHR) lists 578 translations of it (OHCHR), and the UDHR page on this site hosts all 533 that are published as clean text, free to download.

Each language comes as a plain text file and an XML file, one at a time or all together in one 4.9 MB ZIP. Most also come as OHCHR's original PDF, and 90 more languages are available as PDF only. Every file has a SHA-256 checksum so you can prove your copy is intact.

What the UDHR is

The UN General Assembly proclaimed the UDHR in Paris on December 10, 1948, as resolution 217 A (UN, UDHR page). The text is a preamble and 30 articles.

OHCHR runs the translation project. OHCHR says the UDHR set a world record in 1999 as the most translated document (OHCHR).

How many translations exist

The count depends on who is counting and what format you need.

The 533 files are the 528 complete conversions plus 5 partial texts. The project marks 4 languages as having no known complete translation and 1 as partly converted. The other 82 index entries have not been converted yet (UDHR in XML, Translations).

Why the UDHR is the standard parallel text

A parallel text says the same thing in more than one language. The UDHR is the widest one in common use, for four reasons.

1. Same content, same order. Every translation is the preamble and 30 articles. The XML files tag each article with its number, so you can line up Article 1 in any 2 languages.
2. More languages than any other document. That is the 1999 world record OHCHR cites.
3. Uniform files. The UDHR in XML project converts the OHCHR texts into one format. All files are UTF-8, and the plain text is generated from the XML (Files).
4. Already used in tools and research. The Natural Language Toolkit (NLTK) ships a udhr corpus with the declaration in over 300 languages (NLTK book, chapter 2). A 2010 language identification study used it to cover 281 languages, with test strings of 5 to 21 characters (Vatanen et al., LREC 2010).

Typical uses: testing a language identifier, checking that a font or text pipeline handles a script, and comparing how many characters or tokens each language needs for the same content.

Limits to know

• Small. The English plain text file runs about 1,750 words. Good for tests and demos, far too small to train a translation model.
• Accuracy varies. OHCHR takes responsibility for accuracy only in the six official UN languages (OHCHR). Those are Arabic, Chinese, English, French, Russian and Spanish (UN, Official Languages).
• Some languages have several files. When a spelling reform changes a language, the project adds a new translation rather than updating the old one (Which text?).
• Names follow Ethnologue. Irish is listed as "Gaelic, Irish" (Language identification). Search by ISO 639-3 code if a name does not turn up.

How to download the UDHR in every language

1. Open the UDHR page.
2. For one language, type its name or ISO 639-3 code in the filter, then click .txt, .xml or .pdf.
3. For every language, click Download all. The ZIP is 4.9 MB and unpacks to about 18 MB in txt/ and xml/ folders. The PDFs (321 MB in all) are not in the ZIP.
4. Use index.xml or manifest.json to map each file to its language code, script, writing direction and OHCHR code.

Files are named by the project's key, for example udhr_eng.txt for English. Variants add a suffix, such as udhr_cmn_hans.txt for Mandarin in Simplified Chinese characters.

How to verify a download

Compute the file's SHA-256 (Secure Hash Algorithm, 256-bit) hash and compare it with the one on the UDHR page or in SHA256SUMS.txt.

All 64 characters must match. Windows prints capital letters, which does not change the hash.

To check the full set after unzipping, run shasum -a 256 -c SHA256SUMS.txt (macOS) or sha256sum -c SHA256SUMS.txt (Linux) inside the folder. The ZIP carries its own list.

The text and XML files are byte-for-byte copies of what the UDHR in XML project published, with nothing edited. A hash of the same file downloaded from that project will match. The PDFs are byte-for-byte copies of OHCHR's files, and manifest.json lists each one's original address.

Credit when you reuse it

OHCHR asks anyone who reproduces UDHR translations to link to its website as the source. Credit the UDHR in XML project for the text and XML files, and send corrections there (source on GitHub).

For more on telling languages apart in your own text, see language identification.

What to do next

• Download the full ZIP from the UDHR page and verify its SHA-256 before you use it.
• Load the XML, not the plain text, if you need to align languages article by article.
• Have a fluent reader check any translation outside the six official UN languages before you rely on it.
• Link OHCHR's translation project page whenever you publish or share the texts.

Azcuna's BABEL reads, translates, and searches foreign-language documents entirely on-device, with no cloud and no connection, and keeps every claim tied to the passage it came from.

Questions people ask

How many languages is the UDHR translated into?

The OHCHR translation project page says 578 different translations are available in HTML or PDF, and that the UDHR set a world record in 1999 as the most translated document. The UN's own UDHR page says over 500 languages. The UDHR in XML project has converted 533 translations to plain text and XML, covering 447 languages by ISO 639-3 code and 43 scripts.

Where can I download the UDHR in all languages?

The UDHR page on this site hosts all 533 translations published by the UDHR in XML project, as plain text and XML, one file at a time or as a single 4.9 MB ZIP. Each file has a SHA-256 checksum. The page also hosts 571 of OHCHR's own PDFs, 90 of them in languages with no text version yet.

Is the UDHR available as plain text in every language?

Not in every language yet. The UDHR in XML project lists 615 translations in its index. Of those, 528 are fully converted to XML, 1 is partly converted, 4 have no known complete translation, and 82 have not been converted. The plain text files are produced from the XML files and are encoded in UTF-8, according to the project's Files page.

Why is the UDHR used as a parallel corpus?

Every translation says the same thing in the same order: a preamble and 30 articles. That lets you compare languages article by article across hundreds of languages. The NLTK book describes its udhr corpus as the declaration in over 300 languages, and Vatanen and colleagues used it for language identification across 281 languages in an LREC 2010 paper.

Which UDHR translations are official?

OHCHR says it bears responsibility for the quality and accuracy only of the translations in the six official UN languages. The UN lists those as Arabic, Chinese, English, French, Russian and Spanish. Every other translation is the best one OHCHR could select or commission, with no guarantee of accuracy, so check a translation with a fluent reader before you rely on it.

How do I check that a UDHR download is not corrupted?

Compute its SHA-256 hash and compare it with the hash published on the UDHR page or in SHA256SUMS.txt. On macOS run shasum -a 256 followed by the file name. On Linux run sha256sum. On Windows run Get-FileHash in PowerShell. All 64 characters must match. Windows prints capital letters, which does not change the hash.

Sources

1. 1Universal Declaration of Human Rights · United Nations
2. 2Official Languages · United Nations
3. 3About the Universal Declaration of Human Rights Translation Project (Internet Archive capture, November 4, 2026) · Office of the United Nations High Commissioner for Human Rights
4. 4UDHR in XML · UDHR in XML project (Eric Muller)
5. 5UDHR in XML: Translations · UDHR in XML project (Eric Muller)
6. 6UDHR in XML: Which text? · UDHR in XML project (Eric Muller)
7. 7UDHR in XML: Language identification · UDHR in XML project (Eric Muller)
8. 8UDHR in XML: Files · UDHR in XML project (Eric Muller)
9. 9eric-muller/udhr · GitHub
10. 10Natural Language Processing with Python, Chapter 2: Accessing Text Corpora and Lexical Resources · NLTK Project (Bird, Klein, Loper)
11. 11Language Identification of Short Text Segments with N-gram Models · LREC 2010 (Vatanen, Väyrynen, Virpioja)

Where this shows up in our work

• BABEL, offline multilingual translation and search translates and searches foreign-language documents on-device, every fact tied to its source passage.