TaylorGAN: Neighbor-Augmented Policy Update for
  Sample-Efficient Natural Language Generation

      Chun-Hsing Lin Siang-Ruei Wu Hung-Yi Lee Yun-Nung Chen
            National Taiwan University, Taipei, Taiwan
  {jsaon92, raywu0}@gmail.com   hungyilee@ntu.edu.tw     y.v.chen@ieee.org

                         Abstract
       Score function-based natural language generation (NLG) approaches such as RE-
       INFORCE, in general, suffer from low sample efficiency and training instability
       ...