CEAID: Benchmark of Multilingual Machine-Generated Text Detection Methods for Central European Languages

Source code for replication of the experiments in the paper accepted to the Findings of the ACL 2025 conference.

CEAID dataset covering machine-generated and human-written texts in news and social-media domains for 7 Central European languages is available here. An extension of the dataset for evaluation of adversarial robustness is available here.

Cite

If you use the data, code, or the information in this repository, cite the following paper, please (also available on arXiv).

@misc{macko2025ceaidbenchmarkmultilingualmachinegenerated,
      title={{CEAID}: Benchmark of Multilingual Machine-Generated Text Detection Methods for Central European Languages}, 
      author={Dominik Macko and Jakub Kopal},
      year={2025},
      eprint={2509.26051},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2509.26051}, 
}

TBA