Generative Adversarial Networks for Text Generation -- Part 2: RL

Before we dive in, I would like to point out that all the basic ideas in this article are based on the paper -- "SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient (Lantao Yu et. I have left out details that aren't necessary for understanding the intuition behind the approach and its working. However, I would definitely recommend that you read the paper as well. Let's leave the whole idea of neural network based text generators aside for a moment and think of our text generator as an RL agent. What would its states and actions be?