Comment by treyd

Has there been research into some hierarchical attention model that has local attention at the scale of sentences and paragraphs that feeds embeddings up to longer range attention across documents?