Reviews for this paper were originally quite mixed, and ultimately read as being truly borderline. The positive reviews highlighted some important concerns, but on the other hand the negative reviews didn't raise issues that appeared to be dealbreakers. Some of the main issues identified by reviewers center on a lack of clarity and polish. The positive reviews say the paper is "a good idea, but not ready" (R1), and that the claims need more nuance to be supported (R2). The negative reviews highlight a lack of clarity (R3) and weak baselines (R4), though in the authors' defense these issues are quite fixable. The authors submitted a very detailed response, and a discussion was initiated. After discussing some reviewers changed their reviews, both to increase the scores and to change the tone mostly in a more positive direction, after which I am willing to recommend acceptance.