Comparison of our ST model and other models. * is computed by C-F1 (100%) and R-F1 (100%) introduced by Eger et al. (2017). † calculates scores on the basis of label sets introduced by Kuribayashi et al. (2019) for comparison, where we map AAEC’s Claim:Against and Claim:For labels into Claim. We also map an MTC’s component labels to Claim or Premise and relation labels to Support or Attack.