SimpleStyle: An Adaptable Style Transfer Approach

Attribute Controlled Text Rewriting, also known as text style transfer, has received significant attention in the natural language generation community due to its crucial role in controllable natural language generation systems. In this work we present SimpleStyle a minimalist yet effective approach for attribute controlled text rewriting based on a simple mechanism composed of two ingredients. controlled denoising and output filtering. Despite the simplicity of our approach, which can be succinctly explained with just a few lines of code, it is competitive with previous state-of-the-art methods both in automatic and in human evaluations. Additionally, we demonstrate the practical effectiveness of our system, by applying it to real-world data from social networks. Additionally, we introduce a soft masking sampling technique that further improves the performance of the system. We also show that feeding the output of our system into a text-to-text student model can produce high-quality results without the need for additional filtering. Finally, we suggest that our method can solve the fundamental missing baseline absence that holding progress in the field by offering our protocol as a simple, adaptive and very strong baseline for works wish to make incremental advancements in the field of attribute controlled text rewriting.