I am working on developing a transliteration model to convert Romanized Tamil (Tamil written in the Latin alphabet) text into its native Tamil script. I’ve come across several projects that use transformer-based models like T5, BART, mT5, and mBART for translation tasks. I understand that mT5 and mBART models have been trained on Tamil language data. Romanized Tamil can be consider as another language known as “Tanglish”.

My question is: How can I use one of these models to translate Romanized Tamil (Tamil written in the Latin alphabet) into the native Tamil script? Note that, in this translation task, the input and output are expected to have a similar number of words. Any guidance on how to approach this problem would be greatly appreciated. Tamil is an Indic language and I used this language here as an example.