I'm interested in understanding what work #NLP #MachineLearning researchers are doing on predicting whether text is factual or not. Quick googling shows attempts at classifying "fake news". What about work trying to predict citations from text, or to try to predict whether two pieces of text agree? I could imagine training these things on real Wikipedia vs generated text, with the assumption that Wikipedia and its citations mostly agree and generated text is mostly wrong. Do people do this?