[[Word Embeddings]] ## Questions - Can you make embeddings hierarchical e.g. an embedding for the whole doc + 1 embedding per paragraph + 1 embedding per sentence - e.g. how might you relate word embeddings to sentence embeddings - could take the view that the tensor product of all word embeddings forms the tensor product space of potential sentences we can make with those words - how does that relate to positional encodings - what sort of topological object / space does that give use - how could you feed that back in as model input?