Transplant Then Regenerate: A New Paradigm for Text Data Augmentation
Guangzhan Wang Hongyu Zhang Beijun Shen Xiaodong Gu, [4pt] Shanghai Jiao Tong University · 2025
Read Transplant Then Regenerate: A New Paradigm for Text Data Augmentation by Guangzhan Wang Hongyu Zhang Beijun Shen Xiaodong Gu, [4pt] Shanghai Jiao Tong University (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-14723.txt.
Description
Guangzhan Wang, Hongyu Zhang, Beijun Shen, and Xiaodong Gu propose a new way to multiply training text for AI models. Standard tricks mostly rephrase words, yielding copies with the same meaning and shape. Their 2025 method, LMTransplant, aims for richer variety while keeping the original's core traits.
The seed text is first planted into a wider context written by a large language model. The model then regenerates a fresh variant grown from that expanded setting. Because the new text grows out of new surroundings, it varies in content, not just wording. Tests across several text tasks beat older augmentation methods. Gains also held as the volume of added data grew, which suits large training runs.
Catalog record
| CREATOR | Guangzhan Wang Hongyu Zhang Beijun Shen Xiaodong Gu, [4pt] Shanghai Jiao Tong University |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 53 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-14723