I started a project in chatbots for the normal regional languages chatting in english like below , I need a dataset to train my model in tensorflow ,Is any dataset default present in tensorflow or any framework available to do this . if so how to use it . I checked the awesome public data sets list, But i could not able to find it and also i verified in tensorflow sonnet related frameworks ,no chance , Please help with the solution ,I almost created the android app but very poor performance with the dataset available in hand

Tamil language written via english

"Neenga Epdi Irukenga?" ==> "How are you ?"

More Info

Keyboard Support is very limited for most of the regional languages ,To make the conversation in regional language ,people are typing regional langauges in english the way as like pronounce

Additional Info

Similar to transliteration but transliteration is for unique word ,the dataset needed for meaningful sentence

  • Have you seen Tatoeba? more info
    – philshem
    May 17, 2017 at 10:36
  • philshem, Thanks for the reply , Tatoeba is a translation data set if my undestanding is correct, Sorry I don't need the translation instead i need only english language but able to pronounce the regional language word in english(Typing tamil,French etc in english as based on the pronounciation :)), I updated the question as more info May 17, 2017 at 11:10
  • I think then the term you are looking for is Transliteration
    – philshem
    May 17, 2017 at 11:33

1 Answer 1


If you are indeed looking for transliteration, you may want to try a programming library like Python's Unidecode. It's not human tranlsations, but it takes unicode alphabets and transforms them to ASCII.

enter image description here

# -*- coding: utf-8 -*-
from unidecode import unidecode

t = u'வணக்கம். எப்படி இருக்கிறீர்கள்?'

print unidecode(t)

Gives me

> vnnkkm. epptti irukkirriirkll?

According to Google Translate, it should be something like this:

> Vaṇakkam. Eppaṭi irukkiṟīrkaḷ?

enter image description here

I've found this library to work quite well for most alphabets, but maybe Tamil isn't so supported.

From the library doc:

So a good rule of thumb is that the further the script you are transliterating is from Latin alphabet, the worse the transliteration will be.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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