Video Summary

The paper proposes a new method for generating text, which is based on the idea of reinforcement learning. The method is able to generate text that is both coherent and fluent, and it also achieves state-of-the-art results on several benchmark datasets.