StyleFlow: Disentangle Latent Representations via Normalizing Flow for Unsupervised Text Style Transfer

TL;DR

StyleFlow introduces a novel normalizing flow-based model for unsupervised text style transfer, effectively disentangling content and style to enhance content preservation and achieve state-of-the-art results.

Contribution

It proposes a disentanglement-based approach with attention-aware coupling layers and a data augmentation method using normalizing flow, differing from typical encoder-decoder models.

Findings

Effective content preservation demonstrated by experiments

Achieves state-of-the-art performance on key metrics

Robustness improved through data augmentation

Abstract

Text style transfer aims to alter the style of a sentence while preserving its content. Due to the lack of parallel corpora, most recent work focuses on unsupervised methods and often uses cycle construction to train models. Since cycle construction helps to improve the style transfer ability of the model by rebuilding transferred sentences back to original-style sentences, it brings about a content loss in unsupervised text style transfer tasks. In this paper, we propose a novel disentanglement-based style transfer model StyleFlow to enhance content preservation. Instead of the typical encoder-decoder scheme, StyleFlow can not only conduct the forward process to obtain the output, but also infer to the input through the output. We design an attention-aware coupling layers to disentangle the content representations and the style representations of a sentence. Besides, we propose a data augmentation method based on Normalizing Flow to improve the robustness of the model. Experiment results demonstrate that our model preserves content effectively and achieves the state-of-the-art performance on the most metrics.