Note 001
Launching Indonesiaku
Indonesia holds 728 languages, and almost none of them exist inside the systems that increasingly mediate everyday life. Why we started, and what we are building.
728 living languages
Hundreds of voices. Preserved as bits.
Indonesiaku is a non-profit research lab using AI to document, translate, and preserve Indonesia's 728 regional languages, so that every one of them survives into the digital age.
Terima kasih
01 The stakes
A language falls silent somewhere in the world every two weeks. Indonesia holds a tenth of every tongue ever spoken, and most have never been written down, let alone computed.
When a language has no digital footprint, it is locked out of every tool the modern world runs on. We build the datasets, models, and open infrastructure that carry Indonesia's regional languages into the machines: field by field, sentence by sentence, speaker by speaker.728living languages spoken across 17,000 islands, the second-most linguistically dense nation on Earth
90%of Indonesia's regional languages projected to fall silent by 2124 without intervention
55MIndonesians who do not speak Bahasa Indonesia, a fifth of the country, underserved by technology
71%of Indonesians move between two or more languages every day. The archive must move with them
02 The work
01 · Dokumentasi
Collect
Field recordings and community-sourced translations, gathered with contributors across the archipelago and verified by native speakers.
02 · Pemodelan
Model
Machine translation tuned to the corpus, aimed at the registers, speech levels, and scripts that general-purpose systems miss.
03 · Preservasi
Preserve
Every dataset open-licensed, every model public, every note published. An archive built to outlive the lab that maintains it.
03 The engine
NusaMT-7B
A machine translation model built for the languages the large labs overlook, including the politeness registers, speech levels, and scripts that carry their culture.
| Status | In development · not yet serving |
|---|---|
| Target languages | 16 languages of Indonesia, from Acehnese to Tetum |
| Translation pairs | 120, fully bidirectional |
| License | Apache-2.0 · weights and data public |
Javanese · krama register
“Kula badhe tindak dhateng pasar.”
“Saya akan pergi ke pasar.”
Bahasa IndonesiaAdd your voice to the archive.
Translate five sentences in your language and a model learns from it forever. No credentials required, only fluency.
Become a contributor