NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:5992
Title:Levenshtein Transformer

After considering the author feedback and other reviewers' comments, all of the reviewers agree that the paper is worthy of publication at NeurIPS. I believe that the paper does not quite convince us that the deletion operation is beneficial, or that Levenshtein Transformer has some benefits to offer in terms of translation quality in the context of recent papers such as iterative refinement (Lee et al., 2018), Constant time MT (Ghazvininejad et al, 2019), Insertion Transformer (Stern et al, 2019). I encourage the authors to try to answer these questions more directly if possible. Nevertheless, there are some novel aspects of the paper and the proposed technique that can help advance the field.