Double Trouble: How to not explain a text classifier's decisions using counterfactuals synthesized by masked language models?. (arXiv:2110.11929v4 [cs.CL] UPDATED)

A principle behind dozens of attribution methods is to take the prediction
difference between before-and-after an input feature (here, a token) is removed
as its attribution. A popular Input Marginalization (IM) method (Kim et al.,
2020) uses BERT to replace a token, yielding more plausible counterfactuals.
While Kim et al. (2020) reported that IM is effective, we find this conclusion
not convincing as the DeletionBERT metric used in their paper is biased towards
IM. Importantly, this bias exists in Deletion-based …

arxiv classifier decisions language language models synthesized text