About NagaTranslate

What this site is

NagaTranslate provides free translation between English and the main languages of Nagaland — Nagamese, Ao (Chungli) and Sema (Sümi) — along with Hindi, Assamese, Bengali, Manipuri and Nepali as additional source and target languages. Alongside the translator there is a conversational practice tool, a searchable phrasebook, and a growing library of written guides to the languages themselves.

None of these three languages is supported by any major translation service. That is the gap this project exists to fill.

• Translator — text and voice, with audio output for Nagamese.
• NagaChat — conversation practice with explanations, 50 free messages per day.
• Phrasebook — categorised, searchable, contributed phrases.
• Learning guides — language overviews, full grammar references and comparative articles.
• Android app — the same translator, free on Google Play.

Who runs it

NagaTranslate is built and maintained by Agniva Maiti, an independent developer working on machine translation for low-resource Indian languages. The engineering, the site, the Android app and the written guides are all my work; the language data is not, and could not be.

I am not a native speaker of Nagamese, Ao or Sümi. Everything linguistic on this site rests on published reference material and on corrections from people who do speak these languages. Where a guide states a grammatical rule, that rule comes from a documented source or from a speaker correction, and the sources are listed at the bottom of every guide. Where I am uncertain, the page says so.

The code is open and can be inspected at github.com/AgnivaMaiti/nagatranslate.

Why it exists

Machine translation runs on text, and the amount of text a language has online determines whether it gets supported. Nagamese, Ao and Sümi have very little. There is no large Ao web corpus. Most digital Sümi text traces back to a handful of school readers and published books. Nagamese, despite being spoken by most of a state, has no official status and correspondingly little written material.

The practical consequence is that speakers of these languages get none of the tooling that speakers of larger languages take for granted, and that the languages become progressively less usable in exactly the digital contexts where more and more of life happens. That is the mechanism by which languages fall out of use — not dramatic suppression, but slow domain loss.

A small project cannot reverse that. What it can do is make the languages usable in one more place, and make the material needed to learn them findable by someone who searches for it.

The community part

The translator improves through corrections. When a speaker uses the Improve button on a translation, or submits a phrase through the contribute page, that pair enters the community dataset and conditions subsequent translations. There is no account required and no review queue that swallows things for weeks.

This has a much larger effect than the equivalent action would for a major language. Adding a contributed sentence to a language with hundreds of millions of sentences online changes nothing measurable. Adding one to a language with a few thousand does.

Ao and Sümi contributions are the most valuable, because those datasets are the smallest.

How it is paid for

NagaTranslate is free to use and has no paid tier. Running it costs money — model inference, speech synthesis and recognition, and hosting are all metered.

Those costs are covered by three things: my own money, voluntary contributions from users, and advertising. Contributions go to running costs first; any surplus is donated to children's charities. Advertising is disclosed in the privacy policy and, where it appears, is labelled.

Advertising has no influence on the content of the guides. Nothing on this site is sponsored, and no guide has been commissioned or paid for by a third party.

Getting in touch

Corrections, questions, partnership enquiries and complaints all go to hello@mynimalistic.com, or through the contact page. If you have found a linguistic error, the contribute form is faster and puts the fix directly into the dataset.