Effective Data Augmentation Methods for CCMT 2022

Jing Wang, Lina Yang

Abstract: The purpose of this paper is to introduce the specific situation in which Guangxi University participated in the 18th China Conference on Machine Translation (CCMT 2022) evaluation tasks. We submitted the results of two bilingual machine translation (MT) evaluation tasks in CCMT 2022. One is Chinese-English bilingual MT tasks from the news field, the other is Chinese-Thai bilingual MT tasks in low resource languages. Our system is based on Transformer model with several effective data augmentation strategies which are adopted to improve the quality of translation. Experiments show that data augmentation methods have a good impact on the baseline system and aim to enhance the robustness of the model.