Learning to Deceive with Attention-based Explanations

Code & associated data for the following paper.

Learning to Deceive with Attention-Based Explanations

Danish Pruthi, Mansi Gupta, Bhuwan Dhingra, Graham Neubig, Zachary C. Lipton

The 58th Annual Meeting of the Association for Computational Linguistics (ACL-20).

For dependencies, please check the environment.yml file in src directory. To create the same conda environment you can run conda env create -f environment.yml (You might have to edit the prefix in the last line in the file.) The code also uses a library called log for colorful printing (available here).

Please refer to README files for our experiments on classification tasks, and sequence-to-sequence tasks.

The examples alongside attention-based explanations used for the human-subject experiment:

• Organic attention
• Manipulated attention from Wiegreffe & Pinter, 2019
• Attention from our manipulation scheme

Bibtex

@article{pruthi2020learning,
  title={Learning to Deceive with Attention-Based Explanations},
  author={Pruthi, Danish and Gupta, Mansi and Dhingra, Bhuwan and Neubig, Graham and Lipton, Zachary C},
  booktitle = {The 58th Annual Meeting of the Association for Computational Linguistics (ACL)},
  address = {Seattle, USA},
  month = {July},
  year = {2020}
}