Text style transfer is an important task in natural language generation, which aims to control certain attributes in the generated text, such as politeness, emotion, humor, and many others. It has a long history in the field of natural language processing, and recently has re-gained significant attention thanks to the promising performance brought by deep neural models. In this project, we study two methods and architectures employed for this task, compare their generated results and the results of our attempt at reproducibility. In some places, we have also attempted to analyse how our modifications to these architectures compare to the ones designed by the authors. We end with some insights on how the task of text style transfer can be approached, with possible extensions to our modifications that can make it work better.
Paper by Ke Wang, Hang Hua and Xiojun Wan
Our attempt at reproducibility with some architecture modifications is in this folder.
Paper by Ning Dai, Jianze Liang, Xipeng Qiu, Xuanjing Huang
Our attempt at reproducibility with some architecture modifications is in this folder.
A demo of the style transfer task using our best trained model can be found in this python script.
- Wang et. al model: trained on amazon, yelp
- Dai et. al model: multi-class
https://drive.google.com/file/d/1SYTfO6GON6mF3fRq9uBd8rnooW4l5CYs/view?usp=sharing