[ACL'25] Read it in Two Steps: Translating Extremely Low-Resource Languages with Code-Augmented Grammar Books

Preprint: https://arxiv.org/pdf/2506.01796

Important: Preventing Test Set Contamination We encrypted the source files of ZhuangRules in data.zip to prevent test set contamination. The password is zhuangrules. Please unzip the file and place the content into ./data directory.

List of files:

• zhuangrules_v0.json: The full ZhuangRules dataset.
• zhuangrules_v0_for_test.json: Test version of the dataset, generated via convert_to_test_format.py.
• zhuang_dictionaries.json: All Zhuang grammar rules used in the dataset.
• annotated_multi_rules_all.json: All annotated multi-rule test instances.
• code_grammars/: Code-based grammars proposed in our paper.
• igt/: Annotated Interlinear Glossed Texts (IGT).
• kalamang/: Extracted grammar rules from A Grammar of Kalamang.
• wals/: Annotated linguistic domains for each rule, based on the WALS taxonomy.

Use the scripts in ./scripts to run the LLMs and evaluate the results.

The license for the code and data is MIT.

@misc{zhang2025readstepstranslatingextremely,
      title={Read it in Two Steps: Translating Extremely Low-Resource Languages with Code-Augmented Grammar Books}, 
      author={Chen Zhang and Jiuheng Lin and Xiao Liu and Zekai Zhang and Yansong Feng},
      year={2025},
      eprint={2506.01796},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2506.01796}, 
}