Upon an encoder-decoder transformer-architectured language backbone: On the encoder side, an extra graph neural network (GNN) is used to contextualize the graph nodes and edges with the language ...