Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
This is important and timely work. As the reviewers mention, this is likely to be a "buzzy" paper. The primary objection to the paper is that it risks slipping between "fake text" (generated by computers but with no judgements about truth) and "fake news" (untrue statements written by humans or computers). There's an important difference between an auto-generated summary of a basketball game that reports accurate events and a hostile article written to misinform. I'd like to see more discussion in the paper about what the systems are actually identifying. Currently I (and R2) find it difficult to interpret what is being distinguished and why. I'm more optimistic than R2 about the feasibility of adding some results on adding other generators like BERT.