README.md

metadata

license: cc-by-4.0

Dataset for the paper: Deduction under Perturbed Evidence: Probing Student Simulation Capabilities of Large Language Models

You can find the paper here.

Note

The QID which ends in either math_dupe/nlp_dupe refer to math/nlp perturbations respectively.

The QID can be mapped to original strategyQA dataset which can be downloaded here.

Please note that the answer needs to be reversed for the facts. It refers to the original answer field in the strategyqa dataset.

Citation

If you use this dataset in your work, please cite:

@article{sonkar2023deduction,
  title={Deduction under Perturbed Evidence: Probing Student Simulation Capabilities of Large Language Models},
  author={Sonkar, Shashank and Baraniuk, Richard G},
  journal={arXiv preprint arXiv:2305.14507},
  year={2023}
}