as the term explainability is not defined in more detail by you, I wanted to ask, if only ad-hoc or also post-hoc explainable systems are accepted. I could imagine that, with post-hoc explainability being accepted, the most successful systems will likely be just based on pre-trained transformer LMs people apply LIME to, or something like that, which, in my opinion, is a bit pointless when the call explicitly mentions explainability. https://www.nature.com/articles/s42256-019-0048-x has some good arguments on why post-hoc explainability can be problematic in terms of real interpretability in multiple ways. For this reason, I wanted to reassure myself before I make a final decision on my approach.
We approach "explainable sexism detection" via the granularity of classification labels. This differs from other definitions of "explainability" you are referring to.
Please read our task description pages on this CodaLab for detail about the task, what format the data is in and how systems will be evaluated.
All the best,
HannahPosted by: hannah.rose.kirk @ Nov. 7, 2022, 2:34 p.m.