Bi-SimCut: A Simple Strategy for Boosting Neural Machine Translation

TL;DR

Bi-SimCut is a straightforward training strategy that enhances neural machine translation by combining bidirectional pretraining and unidirectional finetuning with a simple regularization method, achieving strong results without extra data or large models.

Contribution

The paper introduces Bi-SimCut, a novel training approach that leverages SimCut regularization for improved NMT performance, serving as a strong baseline for future research.

Findings

Achieves high BLEU scores across multiple translation benchmarks.

Does not require extra datasets or large pretrained models.

Provides a simple yet effective training strategy for NMT.

Abstract

We introduce Bi-SimCut: a simple but effective training strategy to boost neural machine translation (NMT) performance. It consists of two procedures: bidirectional pretraining and unidirectional finetuning. Both procedures utilize SimCut, a simple regularization method that forces the consistency between the output distributions of the original and the cutoff sentence pairs. Without leveraging extra dataset via back-translation or integrating large-scale pretrained model, Bi-SimCut achieves strong translation performance across five translation benchmarks (data sizes range from 160K to 20.2M): BLEU scores of 31.16 for en -> de and 38.37 for de -> en on the IWSLT14 dataset, 30.78 for en -> de and 35.15 for de -> en on the WMT14 dataset, and 27.17 for zh -> en on the WMT17 dataset. SimCut is not a new method, but a version of Cutoff (Shen et al., 2020) simplified and adapted for NMT, and it could be considered as a perturbation-based method. Given the universality and simplicity of SimCut and Bi-SimCut, we believe they can serve as strong baselines for future NMT research.