A Hybrid Embedding Approach for XLM to Dialect Neural Machine Translation

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/86853

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/86853

Title:	A Hybrid Embedding Approach for XLM to Dialect Neural Machine Translation
Authors:	黃嘉銘;Ka Ming, Wong
Contributors:	資訊工程學系
Keywords:	無監督神經機器翻譯;深度學習;低資源;Unsupervised Neural Mechine Translation;Deep Learning;Low Resource
Date:	2021-11-25
Issue Date:	2021-12-07 13:20:12 (UTC+8)
Publisher:	國立中央大學
Abstract:	粵語是漢語的變體。在中國南方地區得到廣泛應用。此外，它在世界各地有很多演講者。雖然粵語和普通話的詞系統和大部分詞義相同，但由於語法和用詞的不同，這兩種方言不能相互理解。因此，為這些語言創建翻譯模型是一項重要的工作。無監督神經機器翻譯是應用於這些語言的最理想方法，因為並行數據很少。在本文中，我們提出了一種方法，該方法結合了改進的跨語言語言模型，並對無監督神經機器翻譯進行了逐層注意。在我們的實驗中，我們觀察到我們提出的方法確實將粵語到中文和中文到粵語的翻譯提高了 1.088 和 0.394 BLEU 分數。此外，我們發現訓練數據的領域和質量對翻譯性能有巨大影響。來自社交網絡，尤其是論壇（LIHKG 連登）的粵語數據解析，不是用於方言翻譯的理想資源。;Cantonese is a variant of Chinese. It has been widely used in the southern part of China. Also, it has lots of speakers around the world. Although Cantonese and Standard Chinese share the same word system and most of the word meaning, due to the difference in grammar and use of words, these two dialects are not mutually intelligible. Therefore, creating a translation model for these languages is a significant work. Unsupervised Neural Machines Translation is the most ideal method to apply to these languages because parallel data is scarce. In this paper, we proposed a method that combined a modified cross-lingual language model and performed layer by layer attention on unsupervised neural machine translation. In our experiments, we observed that our proposed method does improve the Cantonese to Chinese and Chinese to Cantonese translation by 1.088 and 0.349 BLEU score. Also, we discovered the domain and quality of the training data has a huge impact on translation performance. Cantonese data parses from the social network, especially from forums(LIHKG 連登), is not an ideal resource to use in dialect translation.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	112	View/Open

社群 sharing

Loading...