Counterfactual Probing Intergroup Bias for affect and specificity

Code, data and analysis related to the paper Counterfactual Probing for the influence of affect and specificity on Intergroup Bias, to be published at Findings of ACL 2023

Data

Data used in the experiments can be found in the data/ folder, with annotations for affect, and specificity scores. In addition the columns in our earlier dataset, the following columns are added (for details on annotation question, refer to the paper):

• Feeling: annotation for feeling question
• Behavior: annotation for behavior question
• Specificity: specificity score from specificityTwitter

Code

The scripts collect_states.py and train_inlp.py sample hidden state embeddings and train INLP classifiers against our property of interest. run_alter.py performs the actual counterfactual intervention, using the learned matrices from INLP.

Citation

@inproceedings{venkat-probing-2023,
    title = {Counterfactual Probing for the influence of Affect and Specificity on Intergroup Bias},
    author = {Govindarajan, Venkata S and Atwell, and Beaver, David I. and Mahowald, Kyle and Li, Junyi Jessy},
    booktitle = {Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics},
    month = july,
    year = {2023},
    address = {Toronto, Canada},
    publisher = {Association for Computational Linguistics},
    url = {http://arxiv.org/abs/2305.16409}
}