We perform

We thank the reviewers for their thoughtful comments. Below we try to provide as many of them as we could accomplish within the tight time frame. R1, R3, R4: Where does the performance gain come from--the transition system or the GNN? R4: Does the choice of GNNs matter? The 3 most frequent categories are "PP Attachment", Compared to Mrini et al., our method has more "PP Attachment" (342 vs. 320) R1, R3: Compare with existing parsers in terms of efficiency. About 50% of the time is spent on the XLNet encoder, which is the same computation for all three methods.