Easy to Decide, Hard to Agree: Reducing Disagreements Between Saliency Methods

Josip Jukić, Martin Tutek, Jan Snajder


Abstract
A popular approach to unveiling the black box of neural NLP models is to leverage saliency methods, which assign scalar importance scores to each input component. A common practice for evaluating whether an interpretability method is faithful has been to use evaluation-by-agreement – if multiple methods agree on an explanation, its credibility increases. However, recent work has found that saliency methods exhibit weak rank correlations even when applied to the same model instance and advocated for alternative diagnostic methods. In our work, we demonstrate that rank correlation is not a good fit for evaluating agreement and argue that Pearson-r is a better-suited alternative. We further show that regularization techniques that increase faithfulness of attention explanations also increase agreement between saliency methods. By connecting our findings to instance categories based on training dynamics, we show that the agreement of saliency method explanations is very low for easy-to-learn instances. Finally, we connect the improvement in agreement across instance categories to local representation space statistics of instances, paving the way for work on analyzing which intrinsic model properties improve their predisposition to interpretability methods.
Anthology ID:
2023.findings-acl.582
Volume:
Findings of the Association for Computational Linguistics: ACL 2023
Month:
July
Year:
2023
Address:
Toronto, Canada
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
9147–9162
Language:
URL:
https://aclanthology.org/2023.findings-acl.582
DOI:
10.18653/v1/2023.findings-acl.582
Bibkey:
Cite (ACL):
Josip Jukić, Martin Tutek, and Jan Snajder. 2023. Easy to Decide, Hard to Agree: Reducing Disagreements Between Saliency Methods. In Findings of the Association for Computational Linguistics: ACL 2023, pages 9147–9162, Toronto, Canada. Association for Computational Linguistics.
Cite (Informal):
Easy to Decide, Hard to Agree: Reducing Disagreements Between Saliency Methods (Jukić et al., Findings 2023)
Copy Citation:
PDF:
https://preview.aclanthology.org/remove-xml-comments/2023.findings-acl.582.pdf