I am trying to figure out what are the options for building a natural language translation model for a language that is not yet supported by existing machine translations. The project is to build a system for translating a very limited subset of a small east african language into English (only one-way needed).
The language is to my knowledge not yet supported by any machine translation systems, but it is related to several other big African languages (putting sentences into Google Translate, the language is mostly auto detected as Kiswahili or Shona, and the English translation is sometimes usable).

I know building a translator for a new language is by no means trivial, but the problem domain is very small and I think it should be doable. Are there any features of the big cloud providers that support this, ML frameworks, or vendors that build these models as a service?

If this is not the right stack exchange for this question, kindly guide me to a better place.


I think your case would benefit from tunning a existing language to the new one but that is only a good approach if you plan to use it commercially. Also Google accepts help to improve their translation algorithm and you could petition for them to assemble a team for this and donate data.

Google uses a Neural Network that they call Google's NMT (Neural Machine Translation) for translation and it works as a encoder-decoder pair. You can read more on their paper.

Also, Google's NMT is available in tensorflow and can be trainned and improved. It is license under Apache License 2.0 and has a nice tutorial and explanation available on their GitHub:

Check on this GitHub folder here

  • $\begingroup$ Thanks for this, AutoML actually looks like it could work for our use case. Will try that. $\endgroup$
    – dschuld
    Apr 15 '19 at 9:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.