Evaluate Prompt-based Text Style Transformation

• Tech Stack: Python, Natural Language Processing
• Github URL: Project Link

Text style transfer is the task of transferring a text from one written style to another while preserving content. Previous researchers have made use of encoder-decoder architectures to paraphrase a given text into a normalized sentence, then performed stylization on a fine-tuned inverse-paraphraser to insert a specific style. While promising, the method is limited by having to fine-tune separate models for each desired style. Recent work has explored prompt-based text style transfer with pre-trained large language models, allowing more complex transformations to arbitrary styles. We propose that prompt-based text style transfer could be good enough to allow for arbitrary styles while providing reliable prompts. By experimenting with straight prompting as well as two-step paraphrase prompting methods, we benchmark prompting on large language models on standard text style transfer methods.