The original English ↔ Tetun translator.

Tetun (also spelled Tetum) is the national language of Timor-Leste, spoken by roughly a million people. Tetun Dili — also called Tetun Prasa — is the urban variety used in government, courts, media, and education. It is one of two official languages of Timor-Leste, alongside Portuguese, and borrows heavily from Portuguese, Malay, and Austronesian roots.

Tetun Terik is the older rural variety spoken outside Dili. This translator targets Tetun Dili, the standard used in official and public contexts.

In 2014, Tony Franklin was living in Timor-Leste and trying to learn Tetun. Frustrated by the absence of any decent online translator for the language, he built one — first as a simple test case, then as a website at translate.tetumdili.com.

Within two years it had spread through the Timorese population and was averaging around 700 daily users — roughly 57,000 total visitors by 2017 by the site's own analytics, and by web-traffic ranking at the time it reached about 22nd among websites in Timor-Leste.

An iOS version, Tetun Translator Pro, launched on 16 March 2019, and a major update in 2023 added photo-to-text and AI-enriched glossary lookups.

In 2026, the translator was rebuilt from the ground up with modern AI: Claude drafts each sentence, then the output is grounded in a 28,000+ row glossary and a 90,000+ sentence parallel corpus, and linted against 79 orthography and grammar rules before it is shown to the user. That is the version you are using now.

Most machine translation engines treat Tetun as a low-resource language and hallucinate freely — fluent-sounding output with wrong words in it. This translator is built to minimise silent guessing: every sentence is checked against real Tetun evidence before you see it, and the parts that are already known don't involve AI at all. It can still make mistakes, so review anything important before you rely on it.

First, the translation memory

The translator keeps a memory of human-verified English–Tetun sentence pairs. If your exact sentence is already in it, you get the stored human translation back verbatim — instantly, no AI involved. If a nearly identical sentence is known, the AI is instructed to minimally edit that trusted translation rather than draft from scratch — the same way a professional translator works from translation memory. Single words with one unambiguous dictionary answer are served straight from the glossary.

For everything else: draft, verify, repair

1. Draft. Claude drafts natural Tetun — phrased the way a Timorese writer would put it, not a word-for-word calque of the English — respecting your register (formal, neutral, casual) and domain (justice, health, government, tourism).
2. Verify the terminology. Key terms are checked against a 28,000+ row glossary built from the Dili Institute of Technology (DIT) dictionary, DIT Justice Sector, DIT Health & Medical, DIT Tourism, DIT Wordfinder and IT glossaries, and the INL Matadalan Ortográfiku ba Tetun-Prasa (2003).
3. Verify against real usage. The sentence is compared with 90,000+ parallel English–Tetun sentences from Timor-Leste government publications, DIT course materials, tetun.org, and Timorese civil-society writing. Where real human phrasing exists, it wins over anything invented.
4. Lint. The draft is checked against 79 grammar and spelling rules — Portuguese loan endings (-saun, -u), aspect markers, negation, apostrophes, common English-calque patterns — with DIT or INL orthography enforced.
5. Repair. If the linter still finds an error in the final text, an automatic repair pass fixes exactly the flagged spans — nothing else is touched. You will occasionally see a translation correct itself a moment after it appears; that is the repair pass working.

And it learns from Timor-Leste

Translations that keep being requested and pass every check are reviewed and promoted into the translation memory, where they serve future users instantly. The thumbs-down button — and the “know a better translation?” box behind it — feed a review queue, so corrections from fluent speakers make the translator better for everyone. The memory also grows from newly digitised bilingual sources: government text, DIT publications, and civil-society archives.

The linguistic rules are drawn from the Peace Corps Tetun Language Course (3rd ed., Catharina Williams-van Klinken, 2015), the DIT-TLPDP Tetun for the Justice Sector textbook (2015), and the INL orthography standard (Decree 1/2004). Machine translation still makes mistakes — for anything that will be signed, filed, or published, have a qualified human translator review the output.

Tetun has two standardised orthographies and this translator speaks both.

• DIT (default) — Dili Institute of Technology. Academic standard. Uses Portuguese-style nh and lh digraphs, drops most Portuguese accents. Examples: kompanhia, konhesimentu, milhaun.
• INL — Instituto Nacional de Linguística. Official government standard under Decree 1/2004. Uses ñ and ll, keeps Portuguese accents. Examples: kompañia, koñesimentu, millaun, fó, ne'ebé.
• Auto — outputs DIT by default and switches to INL automatically for government-register text.

The linguistic work behind this translator is not ours. It belongs to the people who built modern Tetun lexicography and language pedagogy:

• Dr Catharina Williams-van Klinken and the team at Dili Institute of Technology (DIT).
• Instituto Nacional de Linguística (INL), Universidade Nasionál Timor Lorosa'e.
• The United States Peace Corps — Tetun Language Course, 3rd edition (2015).
• Cliff Morris, A Traveller's Dictionary in Tetun–English (foundational).

This translator ingests their published data and pedagogical standards. It does not replace consulting the original works directly; it makes them reachable from a search box.