Reformulating Unsupervised Style Transfer as Paraphrase Generation

TL;DR

This paper redefines unsupervised style transfer as paraphrase generation, using fine-tuned pretrained models on paraphrase data, leading to superior results and insights into evaluation metrics and real-world applications.

Contribution

It introduces a simple, effective method for style transfer via paraphrase generation and provides a comprehensive analysis of evaluation metrics and a large real-world dataset.

Findings

Outperforms state-of-the-art style transfer systems in evaluations.

Existing automatic metrics can be easily gamed, and fixed variants are proposed.

A large dataset of 15 million sentences in 11 styles is collected for analysis.

Abstract

Modern NLP defines the task of style transfer as modifying the style of a given sentence without appreciably changing its semantics, which implies that the outputs of style transfer systems should be paraphrases of their inputs. However, many existing systems purportedly designed for style transfer inherently warp the input's meaning through attribute transfer, which changes semantic properties such as sentiment. In this paper, we reformulate unsupervised style transfer as a paraphrase generation problem, and present a simple methodology based on fine-tuning pretrained language models on automatically generated paraphrase data. Despite its simplicity, our method significantly outperforms state-of-the-art style transfer systems on both human and automatic evaluations. We also survey 23 style transfer papers and discover that existing automatic metrics can be easily gamed and propose fixed variants. Finally, we pivot to a more real-world style transfer setting by collecting a large dataset of 15M sentences in 11 diverse styles, which we use for an in-depth analysis of our system.