Do OpenNMT support machine transliteration?

Hi Everyone,

I am working on neural machine translation of English to Asian languages.

I have a question:

  • Can we apply machine transliteration in OpenNMT ? if yes, is there any tutorial or document ?

Hi,

Sure, just make sure the words you want to transliterate are correctly transliterated in your training data.

Thanks guillaumenkln for feedback.

Can you please provide a sample sentence. ?
I just want to know about the format of the transliteration in training data.

There is no specific format. If your parallel sentences contain transliterations, the model will learn to translate and transliterate.

Hi guillaumenkln, @guillaumekln
what happens in the case when I want to transliterate some verbs which I am using as noun. My model is definitely trained on translating that verb , but in some cases I want the transliteration of the word. Any work around for this? eg. “Winners Guide” it is a name of a book. My model is translating the word Winners and guide , but in my case since it is the name of the book, I want it to be transliterated and not translated.

Hi,

The easiest way is just to provide enough examples in the training data to translate it correctly (the context and casing will be important here).

actually in Hindi language we don’t have casing. That’s the real problem.

Have a look at polyglot transliteration. It should support Hindi.