Yes transformers are a class of graph NN but there can still be value in dealing directly with both text and knowledge graphs (for RAG-style applications, presu

Yes transformers are a class of graph NN but there can still be value in dealing directly with both text and knowledge graphs (for RAG-style applications, presumably) because the graph enforces additional constraints. It is like the difference between concrete and abstract syntax. LLMs frequently generate code that won't compile, since they predict tokens not AST nodes. They are underconstrained for the task. How to address? You can train a single model to handle both, as the authors did, or you can manually enforce constraints while decoding.