OpenNMT Forum

Translating to new languages

This might sound like a basic question, but has anyone had any luck using OpenNMT or another tool to translate text to a new language? I have a large body of text that needs to be translated to newly-discovered languages for which there would likely be no existing models or texts to work from. I’m guessing I would need to translate enough text from a source language to the new language before I could even begin to use OpenNMT to translate more text, or is that wrong?

For example, I want to translate an existing book in English to a newly-discovered language in a remote village in Africa. How would I go about using OpenNMT to do this?

Thanks!

That’s correct. You need to provide some evidence to this new language during the training.

This is an active area of research. For example, search for “low-resource neural machine translation” on Google to find some possible approaches.

@poldon

Does this “newly-discovered language” share vocabulary and/or structure with other popular languages?

If so, you can use Domain Adaptation techniques to build your model, using two corpora, one in each language.

Let us know more about your language and your current corpus.

Kind regards,
Yasmin