An Empirical Comparison on Imitation Learning and Reinforcement Learning for Paraphrase Generation

TL;DR

This paper empirically compares imitation learning and reinforcement learning for paraphrase generation, demonstrating that imitation learning consistently outperforms reinforcement learning and enhances state-of-the-art models.

Contribution

It provides the first direct empirical comparison between RL and IL in paraphrase generation, showing IL's superiority and its effectiveness in improving model performance.

Findings

IL outperforms RL in paraphrase generation

Pointer-generator with IL surpasses state-of-the-art methods

IL consistently yields better results than RL across datasets

Abstract

Generating paraphrases from given sentences involves decoding words step by step from a large vocabulary. To learn a decoder, supervised learning which maximizes the likelihood of tokens always suffers from the exposure bias. Although both reinforcement learning (RL) and imitation learning (IL) have been widely used to alleviate the bias, the lack of direct comparison leads to only a partial image on their benefits. In this work, we present an empirical study on how RL and IL can help boost the performance of generating paraphrases, with the pointer-generator as a base model. Experiments on the benchmark datasets show that (1) imitation learning is constantly better than reinforcement learning; and (2) the pointer-generator models with imitation learning outperform the state-of-the-art methods with a large margin.