Hi @LilyNLP ,

Thank you for releasing your code. I had a couple of questions regarding reproducing the results shown in the paper.

• Your codebase only has code for TextFooler attack. Is it possible to release codes for other attacks as well (PWWS etc.) so that I can reproduce your results?

• The codes for SAFER and DISP do not support MNLI (text pair classification), which means you would have adapted those to your codebase. Is it possible to release those codes as well so I can exactly reproduce results in your paper?

• The paper says that you use TextAttack for implementation, but I don't see it being used in the current implementation.

We use TextFooler† (Jin et al., 2019) as the major attack model for AWS. 
Moreover, we implement (Ren et al., 2019) and 
GENETIC (Alzantot et al., 2018) based on the TextAttack (Morris et al., 2020) 
code base to further verify the efficiency of our proposed method.

Thank you,