Neural language models are often trained with maximum likelihood estimation (MLE), where the next word is generated conditioned on the ground-truth word tokens. During testing, however, the model is instead conditioned on previously generated tokens, resulting in what is termed exposure bias. To reduce this gap between training and testing, we propose using optimal transport (OT) to match the sequences generated in these two modes. An extension is further proposed to improve the OT learning, based on the structural and contextual information of the text sequences. The effectiveness of the proposed method is validated on machine translation, text summarization, and text generation tasks.
@article{arxiv.2010.05994,
title = {Improving Text Generation with Student-Forcing Optimal Transport},
author = {Guoyin Wang and Chunyuan Li and Jianqiao Li and Hao Fu and Yuh-Chen Lin and Liqun Chen and Yizhe Zhang and Chenyang Tao and Ruiyi Zhang and Wenlin Wang and Dinghan Shen and Qian Yang and Lawrence Carin},
journal= {arXiv preprint arXiv:2010.05994},
year = {2020}
}