An Empirical Comparison on Imitation Learning and Reinforcement Learning for Paraphrase Generation
This work provides a direct empirical comparison for NLP researchers to address exposure bias in sequence generation tasks, though it is incremental as it builds on existing methods.
The paper tackled the problem of exposure bias in paraphrase generation by comparing imitation learning (IL) and reinforcement learning (RL) using a pointer-generator base model, finding that IL consistently outperforms RL and achieves state-of-the-art results with a large margin on benchmark datasets.
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.