How does BERT’s attention change when you fine-tune? An analysis methodology...

This paper is the first to analyze the effect of fine-tuning on each attention head. It proposed nothing but a basic and premature analysis method which simply examines the attention matrices before and after fine-tuning. The only surprising finding is that “fine-tuning does not focus on improving the attention heads that are already good at the problem”, despite drawing such a bold conclusion only on a simple task, is unpersuasive to me.


  1. Section 2.2, “not” should be “no”.


  1. In section 5.1, what is the precise definition of “a distribution close to the empirical in-scope and out-of- scope distribution”. Is this distribution univariate or multivariate?
  2. Since the downstream control task is so easy, how can it be claimed as a control task?

Overall Recommendation

  • 5: Transformative: This paper is likely to change our field. It should be considered for a best paper award.
  • 4.5: Exciting: It changed my thinking on this topic. I would fight for it to be accepted.
  • 4: Strong: I learned a lot from it. I would like to see it accepted.
  • 3.5: Leaning positive: It can be accepted more or less in its current form. However, the work it describes is not particularly exciting and/or inspiring, so it will not be a big loss if people don’t see it in this conference.
  • 3: Ambivalent: It has merits (e.g., it reports state-of-the-art results, the idea is nice), but there are key weaknesses (e.g., I didn’t learn much from it, evaluation is not convincing, it describes incremental work). I believe it can significantly benefit from another round of revision, but I won’t object to accepting it if my co-reviewers are willing to champion it.
  • 2.5: Leaning negative: I am leaning towards rejection, but I can be persuaded if my co-reviewers think otherwise.
  • 2: Mediocre: I would rather not see it in the conference.
  • 1.5: Weak: I am pretty confident that it should be rejected.
  • 1: Poor: I would fight to have it rejected.