How to train a model for another language? #5

Activity

1. yuanzh commented on Nov 4, 2019

   

   yuanzh

   on Nov 4, 2019

   Contributor

   More actions

   Can you please explain which language you want to train a model for? The one in PAWS-X or not?
2. maxbaluev commented on Nov 4, 2019

   

   maxbaluev

   on Nov 4, 2019

   Last edited by maxbaluev

   Author

   More actions

   @yuanzh No, I need Russian.
3. yuanzh commented on Nov 5, 2019

   

   yuanzh

   on Nov 5, 2019

   Contributor

   More actions

   You may follow the methods in the PAWS-X paper, i.e. machine translate training sentences to Russian, and then finetune a multilingual BERT on all languages.
4. toan-lt commented on Feb 18, 2023

   

   toan-lt

   on Feb 18, 2023

   More actions

   Hi @yuanzh, thanks for your suggestion. But let me clarify your guide so that anyone who wants to train a model for another languages could achieve it easily.

   Problem: I want to add a new language (does not exist in your PAWS-X dataset) i.e. Vietnamese. The purpose is that I want to check paraphrasing on Vietnamese-English.

   Approach:

   • Translate all English sentences pairs from PAWS-Wiki Labeled (Final) to Vietnamese sentences pairs.
   • Use BERT model to train the dataset (BERT model maybe brings the best accuracy) and merged method.

   That's it? May I miss something? Please let me if I'm missing important things. Thank you!