basic approaches.tex (587B)
1 \begin{frame}{Initial Approaches} 2 Use both pieces of information jointly, linguistic and topological.\\ 3 ``The more features, the better.'' 4 5 \bigskip 6 7 \begin{block}{Nonparametric WL} 8 By using the Wasserstein definition of \(\operatorname{sim}_\text{topo}\). 9 10 We can use \(\operatorname{sim}_\text{topoling}\) as-is, without fine-tuning. 11 \end{block} 12 13 \pause 14 15 \begin{block}{MTB GCN} 16 Train \bertcoder{} and a GCN to learn that parallel arcs should have high \(\operatorname{sim}_\text{topoling}\). 17 18 This enable MTB to use neighborhood for prediction. 19 \end{block} 20 \end{frame}