Abstract

Recently, large language models such as GPT-2 have shown themselves to beextremely adept at text generation and have also been able to achievehigh-quality results in many downstream NLP tasks such as text classification,sentiment analysis and question answering with the aid of fine-tuning. Wepresent a useful technique for using a large language model to perform the taskof paraphrasing on a variety of texts and subjects. Our approach isdemonstrated to be capable of generating paraphrases not only at a sentencelevel but also for longer spans of text such as paragraphs without needing tobreak the text into smaller chunks.

 

Quick Read (beta)