Style Transfer

pytorch implementation
Author : Saeran Park.
Unsupervised Text Style Transfer with Padded Masked Language Models (EMNLP 2020)[Paper]

Stage1. Train the model

I trained a single model instead of training two seperate models for source and target style.
It can be to add a domain embedding to each token emd as proposed by Wu et al.[Paper]

python train.py 

Stage2. Make a masked token for style transfer

python mask.py 

Stage3. Generate tokens

You can generate target style text for each sentence on generate.ipynb.