An Understanding of Learning from Demonstrations for Neural Text Generation

Abstract

In this blog post, we will go over the ICLR 2021 paper titled Text Generation by Learning from Demonstrations. This paper introduces a learning method based on offline, off-policy reinforcement learning (RL) which addresses two key limitations of a training objective used in neural text generation models: Maximum Likelihood Estimate (MLE). Goal of this blog post: Our main goal with this blog post is to provide researchers and practitioners in both NLP and RL with (1) a better understanding of algorithm presented in this paper (GOLD), and (2) an understanding of how RL is used for text generation.

Tasks

Reproductions