Stable Style Transformer: Delete and Generate Approach with Encoder-Decoder for Text Style Transfer

Text style transfer is the task that generates a sentence by preserving the content of the input sentence and transferring the style. Most existing studies are progressing on non-parallel datasets because parallel datasets are limited and hard to construct. In this work, we introduce a method that follows two stages in non-parallel datasets. The first stage is to delete attribute markers of a sentence directly through the classifier. The second stage is to generate the transferred sentence by combining the content tokens and the target style. We evaluate systems on two benchmark datasets. Transferred sentences are evaluated in terms of context, style, fluency, and semantic. These evaluation metrics are used to determine a stable system. Only robust systems in all evaluation metrics are suitable for use in real applications. Many previous systems are difficult to use in certain situations because they are unstable in some evaluation metrics. However, our system is stable in all evaluation metrics and has results comparable to other models.