Mohammad Amin Ghasemi - Department of Mathematics andComputer ScienceAmirkabir University of TechnologyTehran, Iran
Mehdi Ghatee - Department of Mathematics andComputer ScienceAmirkabir University of TechnologyTehran, Iran

Language style transfer is the task of altering the stylistic characteristics of a text while preserving its original context. This paper focuses on style transfer tasks aimed at reducing author-specific characteristics and enhancing readability for contemporary readers. Our study centers on the transformation of Shakespearean texts for this purpose. Prior research has demonstrated that back-translation, a two-step process involving translation to an intermediate language and translating back to the original language, can be effective in achieving this objective. Leveraging translation engines, particularly their propensity to generate text resembling everyday spoken English due to their training data, we employ this approach to create a dataset for training a Sequence-to-Sequence (Seq۲Seq) model enhanced with attention mechanism. Our results confirm the effectiveness of directly applying the back-translation. Furthermore, in some cases where back-translation yielded suboptimal results, the existence of similar training examples where it performed well aided our model in generating improved text compared to the target text.

Style transfer, Back-Translation, Seq۲Seq models, Attention mechanism