This paper advances the BPE dropout one step further by using both deterministic/probabilistic segmentations and adding a consistency loss to the predictions of them. Their method is shown to improve low-resource languages significantly but also improves English.
J(\theta) = \sum_{i=1}^n \big[ -\frac{1}{2} \underbrace{\text{log} \: p_\theta(y_i|\widehat{x}_i)}_\text{Det. Seg CrossEnt} - \frac{1}{2} \underbrace{\text{log} \: p_\theta(y_i|x'_i)}_\text{Prob. Seg CrossEnt} \\
+ \lambda \underbrace{D(p_\theta(y_i|\widehat{x}_i) \: || \: p_\theta(y_i|x'_i))}_\text{Consistency loss} \big]
Comments
- Idea is very simple but the analyses are very well written. I feel even if I’m able to come up this idea I just can’t write a long paper.
Rating
- 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.
0 投票者