VAE based Text Style Transfer with Pivot Words Enhancement Learning

TL;DR

This paper introduces VT-STOWER, a VAE-based method for text style transfer that leverages pivot words learning and external style embeddings to improve performance in unsupervised scenarios.

Contribution

It proposes a novel VAE framework with pivot words enhancement and style strength control, advancing unsupervised text style transfer techniques.

Findings

Outperforms state-of-the-art on sentiment, formality, and code-switching tasks.

Effectively learns style and content with limited non-parallel data.

Demonstrates flexible style strength adjustment in transfer tasks.

Abstract

Text Style Transfer (TST) aims to alter the underlying style of the source text to another specific style while keeping the same content. Due to the scarcity of high-quality parallel training data, unsupervised learning has become a trending direction for TST tasks. In this paper, we propose a novel VAE based Text Style Transfer with pivOt Words Enhancement leaRning (VT-STOWER) method which utilizes Variational AutoEncoder (VAE) and external style embeddings to learn semantics and style distribution jointly. Additionally, we introduce pivot words learning, which is applied to learn decisive words for a specific style and thereby further improve the overall performance of the style transfer. The proposed VT-STOWER can be scaled to different TST scenarios given very limited and non-parallel training data with a novel and flexible style strength control mechanism. Experiments demonstrate that the VT-STOWER outperforms the state-of-the-art on sentiment, formality, and code-switching TST tasks.