Four knowledgeable reviewers support acceptance for the contributions.Reviewers find the paper well-written and appreciate the strong empirical results as well as the theoretical analysis. Thr reviewers indicated that the method does not depend on domain-specific data augmentation (e.g., mixup). Therefore, UDA can be applied to various domains, such as vision and language, as experimented in this paper and it will attract the attention from many researchers of wide area. Therefore, I also recommend acceptance. However, please consider revising your paper to address all the concerns and comments from the reviewers.