Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
The submission presents a carefully designed study of the message length distribution in emergent communication. Though the reviewers disagree on how surprising the finding is (e.g., R3 believes it directly follows from the reward structure), the paper appears uncontroversial, as all the reviewers agree that it will be interesting and important for anyone working on this problem. The author response was crucial in clarifying several points in the paper, and I hope that the authors will update the paper accordingly. I also agree with the reviewers that the assumption about the distribution of reference may be questionable, and hope that the authors will keep their promise and include results with the uniform input distribution.