Writing Like the Best: Exemplar-Based Expository Text Generation

---

This repository contains source code and data for our ACL 2025 Paper Writing Like the Best: Exemplar-Based Expository Text Generation.

How to Run

Step 1: Fill in your secret keys in "secret_keys.sh".

Step 2: Run main experiments

You may change the Wikipedia to RoleEE or USNews for corresponding datasets, or change GPT4 to LLaMA3 for different backbone LLM, and change false to true for debug mode. You may change the corresponding config files for detailed settings, which is under the /configs folder.

bash experiments.sh Wikipedia GPT4 false

Step 3: Run Evaluations

You may change baseline to ablation for different evaluation modes.

bash evaluations.sh

Data

You may download our datasets here and save in /data.

Reference

If you use our work, please consider citing our work:

@inproceedings{liu-chang-2025-writing,
    title = "Writing Like the Best: Exemplar-Based Expository Text Generation",
    author = "Liu, Yuxiang and Chang, Kevin Chen-Chuan",
    editor = "Che, Wanxiang and Nabende, Joyce and Shutova, Ekaterina and Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.acl-long.1250/",
    doi = "10.18653/v1/2025.acl-long.1250",
    pages = "25739--25764",
    ISBN = "979-8-89176-251-0",
}