This paper proposes LMTransplant, a novel text augmentation paradigm leveraging large-scale language models (LLMs). LMTransplant aims to generate diverse and creative transformations at the content level by leveraging the knowledge of LLMs, rather than simply transforming at the lexical level like conventional backtranslation. This is achieved through a "transplant-regeneration" strategy: integrating the source text into the context augmented by the LLM and then having the LLM generate the transformed text. Experimental results demonstrate that LMTransplant outperforms existing methods and demonstrates excellent scalability as the augmented data size increases.
