Are Large Language Models Actually Good at Text Style Transfer?

TL;DR

This paper evaluates large language models on text style transfer tasks across English, Hindi, and Bengali, showing that finetuning enhances performance significantly, especially in non-English languages, highlighting the need for dedicated datasets.

Contribution

It provides a comprehensive analysis of LLMs for multilingual text style transfer, comparing prompting and finetuning methods, and emphasizes the importance of specialized datasets for better results.

Findings

Prompted LLMs perform well in English but are average in Hindi and Bengali.

Finetuning significantly improves style transfer performance.

Dedicated datasets and models are crucial for effective multilingual TST.

Abstract

We analyze the performance of large language models (LLMs) on Text Style Transfer (TST), specifically focusing on sentiment transfer and text detoxification across three languages: English, Hindi, and Bengali. Text Style Transfer involves modifying the linguistic style of a text while preserving its core content. We evaluate the capabilities of pre-trained LLMs using zero-shot and few-shot prompting as well as parameter-efficient finetuning on publicly available datasets. Our evaluation using automatic metrics, GPT-4 and human evaluations reveals that while some prompted LLMs perform well in English, their performance in on other languages (Hindi, Bengali) remains average. However, finetuning significantly improves results compared to zero-shot and few-shot prompting, making them comparable to previous state-of-the-art. This underscores the necessity of dedicated datasets and specialized models for effective TST.