UGAN: Unified Generative Adversarial Networks for Multidirectional Text Style Transfer

Abstract

Recently, text style transfer has become a very hot research topic in the field of natural language processing. However, the conventional text style transfer is unidirectional, and it is not possible to obtain a model with multidirectional transformations through training once. To address this limitation, we propose a new task called multidirectional text style transfer. It aims to use a single model to transfer the underlying style of text among multiple style attributes and keep its main content unchanged. In this paper, we propose Unified Generative Adversarial Networks (UGAN), a practical approach that combines target vector and generative adversarial techniques to perform multidirectional text style transfer. Our model allows simultaneous training of multi-attribute data on a single network. Such unified structure makes our model more efficient and flexible than existing approaches. We demonstrate the superiority of our approach on three benchmark datasets. Experimental results show that our method not only outperforms other baselines, but also reduces training time by an average of 13%. INDEX TERMS Multidirectional text style transfer, generative adversarial networks, unified generative adversarial networks.

A Brief Survey on Natural Language Processing Based Text Generation and Evaluation Techniques

“…Combining both approaches has had a great effect on the overall results of the generated text. Yu et al [23] not only report better results but also show a decrease in the training time. While Yang et al [22] also reported a ROUGE-L score of 39.14 which is the best one among the reviewed studies.…”

“…Earlier this year Yu et al [23] created a model named Unified Generative Adversarial Networks (UGAN). This model unifies both the sequence-to-sequence and GANs architectures.…”

A review of unsupervised text style transfer based on deep learning

“…The traditional task of text style transfer is one-way, and only a single style one-way transfer model can be obtained through one training. In order to solve this limitation, YU et al 56 proposed multi-direction text style transfer, which enables the same model to establish multiple style mapping relationships through a single training, and keep the original content unchanged. This is to combine the target style vector with the generation of confrontation technology, and train multiple style data on a single network at the same time.…”