TAADpapers

Why we included this project

Textual adversarial attack and defense research is scattered across a lot of venues, and this list does the organizing work for you. It collects about 155 papers, grouped by whether they attack at the sentence, word, or character level, by defense approach, and by whether the work covers certified robustness or benchmark evaluation. Entries link straight to the PDF and, when the authors released code, to the repository, so moving from reading to reproduction is quick. Attack papers also get annotated by how much knowledge the attacker is assumed to have about the victim model, which is handy when you are trying to match a technique to your own threat model. This is a reading list, not software you deploy, but for anyone scoping a robustness project or writing a survey it saves a lot of digging through ACL, EMNLP, and NAACL proceedings.