Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Better Chinese-centric Neural Machine Translation for Low-resource Languages

Bin Li, Yixuan Weng, Fei Xia, Hanjun Deng

TL;DR

This work targets Chinese-centric low-resource multilingual MT by addressing data sparsity and quality through monolingual word-embedding data augmentation, a novel Incomplete-Trust loss to mitigate noisy supervision, bilingual curriculum learning to leverage related languages, and contrastive re-ranking via MBERT. The approach uses a strong mBART backbone with data preprocessing and post-processing steps to enhance translation quality. Ablation results show cumulative BLEU improvements, culminating in top performance against multiple baselines, and the authors publicly release data and code to advance Chinese-centric low-resource MT research. The study demonstrates practical gains for low-resource translation and provides a reproducible framework for future work in this niche.

Abstract

The last decade has witnessed enormous improvements in science and technology, stimulating the growing demand for economic and cultural exchanges in various countries. Building a neural machine translation (NMT) system has become an urgent trend, especially in the low-resource setting. However, recent work tends to study NMT systems for low-resource languages centered on English, while few works focus on low-resource NMT systems centered on other languages such as Chinese. To achieve this, the low-resource multilingual translation challenge of the 2021 iFLYTEK AI Developer Competition provides the Chinese-centric multilingual low-resource NMT tasks, where participants are required to build NMT systems based on the provided low-resource samples. In this paper, we present the winner competition system that leverages monolingual word embeddings data enhancement, bilingual curriculum learning, and contrastive re-ranking. In addition, a new Incomplete-Trust (In-trust) loss function is proposed to replace the traditional cross-entropy loss when training. The experimental results demonstrate that the implementation of these ideas leads better performance than other state-of-the-art methods. All the experimental codes are released at: https://github.com/WENGSYX/Low-resource-text-translation.

Towards Better Chinese-centric Neural Machine Translation for Low-resource Languages

TL;DR

This work targets Chinese-centric low-resource multilingual MT by addressing data sparsity and quality through monolingual word-embedding data augmentation, a novel Incomplete-Trust loss to mitigate noisy supervision, bilingual curriculum learning to leverage related languages, and contrastive re-ranking via MBERT. The approach uses a strong mBART backbone with data preprocessing and post-processing steps to enhance translation quality. Ablation results show cumulative BLEU improvements, culminating in top performance against multiple baselines, and the authors publicly release data and code to advance Chinese-centric low-resource MT research. The study demonstrates practical gains for low-resource translation and provides a reproducible framework for future work in this niche.

Abstract

The last decade has witnessed enormous improvements in science and technology, stimulating the growing demand for economic and cultural exchanges in various countries. Building a neural machine translation (NMT) system has become an urgent trend, especially in the low-resource setting. However, recent work tends to study NMT systems for low-resource languages centered on English, while few works focus on low-resource NMT systems centered on other languages such as Chinese. To achieve this, the low-resource multilingual translation challenge of the 2021 iFLYTEK AI Developer Competition provides the Chinese-centric multilingual low-resource NMT tasks, where participants are required to build NMT systems based on the provided low-resource samples. In this paper, we present the winner competition system that leverages monolingual word embeddings data enhancement, bilingual curriculum learning, and contrastive re-ranking. In addition, a new Incomplete-Trust (In-trust) loss function is proposed to replace the traditional cross-entropy loss when training. The experimental results demonstrate that the implementation of these ideas leads better performance than other state-of-the-art methods. All the experimental codes are released at: https://github.com/WENGSYX/Low-resource-text-translation.
Paper Structure (21 sections, 6 equations, 4 figures, 3 tables)

This paper contains 21 sections, 6 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related works
  3. Neural machine translation
  4. The expansion of parallel corpus
  5. The use of other data
  6. Contrastive learning
  7. Task introduction
  8. Dataset
  9. Evaluation metric
  10. Method
  11. System achitecture
  12. Data preprocessing
  13. Backbone baseline introduction
  14. In-trust loss function
  15. Contrastive re-ranking
  16. ...and 6 more sections

Figures (4)

  • Figure 1: Architecture of the proposed NMT system.
  • Figure 2: Noise samples in bilingual translation texts.
  • Figure 3: Overview of the contrastive re-reanking.
  • Figure 4: Overview of bilingual curriculum learning.