Abstract

We thank the reviewers for the thoughtful feedback! We are encouraged that all voted to accept, finding the paper clear / well-organized [R1]; our approach “very interesting” [R3] and novel [R2 R3]; our results significant and well-demonstrated [R1 R2]; and likely to be built on by the community [R1]. We are pleased they recognized the value of transferring visio-linguistic pretraining [R1 R2 R3] and the demonstrated benefits of our co-attentional two-stream model over a direct extension of BERT [R2 R3]. We respond to select comments below but will address all feedback.