Yes, this is an important challenge. There has been a lot of interest in the NLP community right now, particularly around QA tasks [1] Standard supervised model

Yes, this is an important challenge. There has been a lot of interest in the NLP community right now, particularly around QA tasks [1] Standard supervised models do it well, but zero-shot models still have trouble. 1. https://arxiv.org/abs/1806.03822 https://arxiv.org/abs/1806.03822