Existing paraphrase datasets are too easy due to the low lexical overlap between sentence pairs. This paper presents a methodology to create high lexical overlap pairs through constrained sequence generation and back translation.
Specifically, a language model is used to generate a sentence with the same BOW vector of the original sentence. The generation is constrained by part-of-speech and NER tags to reduce the search scope. This procedure generates paraphrases and non-paraphrases with 1:4 ratio (judged by human). Another procedure is to back translate a sentence using beam size 5 and aggressively filter out easy ones by calculating the word-order inversion rate and BOW similarity, which instead mostly generates paraphrases.
These two procedures can generate negative pairs which will be labeled by human. Non well-formed sentences will be corrected too.
The final step is to balance the label in this human-labeled dataset by applying a set of rules on each sentence with their labeled counterpart from the aforementioned two procedures.
- Their methodology is very sounding, each detail is carefully considered.
- Their experiments on DIIN is very interesting. A small DIIN model performs on par with BERT.
- 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.