Toward Diverse Text Generation with Inverse Reinforcement Learning
Zhan Shi, Xinchi Chen, Xipeng Qiu, Xuanjing Huang
TL;DR
This paper reformulates text generation as inverse reinforcement learning (IRL) to address exposure bias and reward sparsity observed in adversarial models. By learning a dense per-step reward function and employing an entropy-regularized policy gradient, it couples a reward approximator with a text generator that are trained in alternating steps. The approach mitigates mode collapse and improves generation quality and diversity, as shown on synthetic data, COCO captions, and IMDB reviews, with new BLEU-based metrics and human evaluation validating the gains. The work lays groundwork for applying IRL to a broader set of NLP tasks and provides practical guidance on training dynamics and evaluation.
Abstract
Text generation is a crucial task in NLP. Recently, several adversarial generative models have been proposed to improve the exposure bias problem in text generation. Though these models gain great success, they still suffer from the problems of reward sparsity and mode collapse. In order to address these two problems, in this paper, we employ inverse reinforcement learning (IRL) for text generation. Specifically, the IRL framework learns a reward function on training data, and then an optimal policy to maximum the expected total reward. Similar to the adversarial models, the reward and policy function in IRL are optimized alternately. Our method has two advantages: (1) the reward function can produce more dense reward signals. (2) the generation policy, trained by "entropy regularized" policy gradient, encourages to generate more diversified texts. Experiment results demonstrate that our proposed method can generate higher quality texts than the previous methods.