Reviews: Adversarial Ranking for Language Generation

For generative adversarial network (GAN) training, the paper presents an approach for replacing the binary classifier in the discriminator with a ranking based discriminator. This allows better training especially for problems which are not binary, such as (discrete) sequence generation, as e.g. in sentence generation. Strength: - Novel, well motivated approach - Extensive experiments on four datasets, a synthetic dataset, Chinese poem generation, coco image caption generation, and Shakespeare's plays generation. The paper shows the improvements w.r.t. Weaknesses: 1. Related Work: As the available space allows it, the paper would benefit from a more detailed discussion of related work, by not only describing the related works, but also discussing the differences to the presented work.