LLM-Based Text Style Transfer: Have We Taken a Step Forward?

Abstract

Text style transfer is the task of altering the stylistic way in which a given sentence is written while maintaining its original meaning. The task requires models to identify and modify various stylistic properties, such as politeness, formality, and sentiment. With the advent of Large Language Models (LLMs) and their remarkable performances for a variety of tasks, numerous LLMs have emerged in the past few years. This paper provides an overview of recent advancements in text style transfer using LLMs. The discussion is focused on LLM-based approaches commonly used for text generation and their adoption for text style transfer. The paper is organized around three main groups of methods: prompting techniques for LLMs, fine-tuning techniques for LLMs, and memory-augmented LLMs. The discussion emphasizes the similarities and differences among the discussed methods and groups, along with the challenges and opportunities that are expected to direct and foster further research in the field.