English

Confidence-Aware Scheduled Sampling for Neural Machine Translation

Computation and Language 2021-07-23 v1

Abstract

Scheduled sampling is an effective method to alleviate the exposure bias problem of neural machine translation. It simulates the inference scene by randomly replacing ground-truth target input tokens with predicted ones during training. Despite its success, its critical schedule strategies are merely based on training steps, ignoring the real-time model competence, which limits its potential performance and convergence speed. To address this issue, we propose confidence-aware scheduled sampling. Specifically, we quantify real-time model competence by the confidence of model predictions, based on which we design fine-grained schedule strategies. In this way, the model is exactly exposed to predicted tokens for high-confidence positions and still ground-truth tokens for low-confidence positions. Moreover, we observe vanilla scheduled sampling suffers from degenerating into the original teacher forcing mode since most predicted tokens are the same as ground-truth tokens. Therefore, under the above confidence-aware strategy, we further expose more noisy tokens (e.g., wordy and incorrect word order) instead of predicted ones for high-confidence token positions. We evaluate our approach on the Transformer and conduct experiments on large-scale WMT 2014 English-German, WMT 2014 English-French, and WMT 2019 Chinese-English. Results show that our approach significantly outperforms the Transformer and vanilla scheduled sampling on both translation quality and convergence speed.

Keywords

Cite

@article{arxiv.2107.10427,
  title  = {Confidence-Aware Scheduled Sampling for Neural Machine Translation},
  author = {Yijin Liu and Fandong Meng and Yufeng Chen and Jinan Xu and Jie Zhou},
  journal= {arXiv preprint arXiv:2107.10427},
  year   = {2021}
}

Comments

Findings of ACL-2021, code at https://github.com/Adaxry/conf_aware_ss4nmt

R2 v1 2026-06-24T04:25:00.597Z