---
license: apache-2.0
pretty_name: AnTBench
---

# AnTBench

AnTBench accompanies **Can LLMs Spot Data Anomalies That Make Table QA Impossible?** It evaluates whether language models can detect anomalies in tables and recognize when those anomalies make a question unanswerable.

## Citation

If you find AnTBench useful, please cite:

```bibtex
@inproceedings{wei2026can,
  title={Can LLMs Spot Data Anomalies That Make Table QA Impossible?},
  author={Wei, Yucheng and Yang, Shenzhi and Wang, Ning and Zhang, Jinchang and Yang, Yuhang and Wang, Haobo and Zha, Liangyu and Ye, Wentao and Li, Liyao and Peng, Cheng and Chen, Gang},
  booktitle={Findings of the Association for Computational Linguistics: EMNLP 2026},
  pages={21952--21987},
  year={2026}
}
```

## Contact

For questions about the benchmark, code, or data, contact weiyucheng@zju.edu.cn.

## License

This project is released under the [Apache-2.0 License](LICENSE).