DUTNLP System for the WMT2023 Discourse-Level Literary Translation
Anqi Zhao, Kaiyu Huang, Hao Yu, Degen Huang · 2023
This paper details the submission from the DUTNLP Lab for the WMT23 Discourse-Level Literary Translation in Chinese to English translation direction under unconstrained conditions.Our primary system aims to harness a large language model with various prompt strategies, allowing for a comprehensive exploration of the potential capabilities of large language models in discourse-level neural machine translation.Moreover, we apply detailed data preprocessing methods to filter bilingual data, which proves to be beneficial.Additionally, we assess a widely used discourse-level machine translation model, G-transformer, using different training strategies.In our experimental results, the method employing large language models achieves a BLEU score of 28.16, whereas the fine-tuned method scores 25.26.These findings indicate that selecting appropriate prompt strategies based on large language models can significantly enhance translation performance compared to traditional model training methods.