How to manage multi-edges between two nodes with Biaffine classifier?

[RiverGao]

Hi there! Your work is very impressive, and I have a question here:

In my experiments, I find that there exist a few multi-edge situations (whereas there is more than one edge between a pair of nodes) in the CAMR dataset. Also, this tends to happen when, for example, there is a self-pointing action in the sentence.

According to your paper, you use the bi-affine classifier to classify relations between nodes, which only provides one prediction between one pair of nodes. So, how do you manage multi-edges?

Looking forward to your reply, thank you!