Дискурсты ескеретін марапаттар

AbstractIn this paper, we investigate the use of discourse-aware rewards with reinforcement learning to guide a model to generate long, coherent text. In particular, we propose to learn neural rewards …...