This repository hosts the code for our paper, Frankentext: Stitching random text fragments into long-form narratives.

🧟‍♂️ Frankentext is a new type of long-form narratives produced by LLMs under the extreme constraint that most tokens (e.g., 90%) must be copied verbatim from human writings. To produce Frankentexts, we instruct the model to produce a draft by selecting and combining human-written passages, then iteratively revise the draft while maintaining a user-specified copy ratio.

• [2025-05-23]: Dataset and prompts for Frankentext are now available! Pipeline code is coming soon...

(Coming soon...)

.
├── README.md
├── assets
├── data 
    ├── outputs
    └── inputs
├── prompts
└── scripts
    ├── pipeline
    ├── eval

• data:
  • inputs contains input data for each experiment.
  • outputs contains outputs from each experiment mentioned in the paper. Each subfolder represents outputs from an experiment.
• scripts contains code to obtain and automatically evaluate Frankentexts:
  • eval contains code to obtain metrics.
  • pipeline contains code to construct Frankentexts with models tested in the paper.
• prompts contains all prompts used in the paper.
  • llm_judge contains prompts to obtain coherence and relevance judgments with LLM.
  • pipeline contains prompt to generate and edit Frankentexts.

@misc{pham2025frankentextstitchingrandomtext,
      title={Frankentext: Stitching random text fragments into long-form narratives}, 
      author={Chau Minh Pham and Jenna Russell and Dzung Pham and Mohit Iyyer},
      year={2025},
      eprint={2505.18128},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.18128}, 
}