Instance-Based Domain Adaptation for Improving Terminology Translation

Proceedings of Machine Translation Summit XIX, Vol. 1: Research Track, pages 1–11 September 4–8, 2023, Macau SAR, China. ©2023 The authors. This article is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). 1 Multiloop Incremental Bootstrapping for Low- Resource Machine Translation Wuying Liu 1,2 wyliu@ldu.edu.cn 1 Shandong Key Laboratory of Language Resources Development and Applica- tion, Ludong University, Yantai, 264025, China 2 Center for Linguistics and Applied Linguistics, Guangdong University of For- eign Studies, Guangzhou, 510420, China Wei Li 676471162@qq.com School of Computer and Big Data Science, Jiujiang University, Jiujiang, 332005, China Lin Wang * lwang@xdsisu.edu.cn Xianda College of Economics and Humanities, Shanghai International Studies University, Shanghai, 200083, China * Corresponding Author Abstract Due to the scarcity of high-quality bilingual sentence pairs, some deep-learning-based machine translation algorithms cannot achieve better performance in low-resource machine translation. On this basis, we are committed to integrating the ideas of machine learning algorithm improve- ment and data augmentation, propose a novel multiloop incremental bootstrapping framework, and design the corresponding semi-supervised learning algorithm. This framework is a meta- frame independent of specific machine translation algorithms. This algorithm makes full use of bilingual seed data of appropriate scale and super-large-scale monolingual data to expand bi- lingual sentence pair data incrementally, and trains machine translation models step by step to improve the translation quality. The experimental results of neural machine translation on mul- tiple language pairs prove that our proposed framework can make use of continuous monolin- gual data to raise itself. Its effectiveness is not only reflected in the easy implementation of state-of-the-art low-resource machine translation, but also in the practical option to quickly es- tablish precise domain machine translation systems. 1. Introduction Machine Translation (MT) is an algorithmic computing process that uses a target natural lan- guage form to paraphrase the semantics of a source natural language. After the Bronze Age marked by Rule-based MT (RBMT) and the Silver Age marked by Statistical MT (SMT), the Golden Age marked by deep-learning-based Neural MT (NMT) has begun. After more than 70 years of unremitting exploration around the three generations of MT, many excellent algo- rithms and practical products have been produced (Garg and Agarwal, 2018). If the formal language theory and context-free grammar derived from the development of compilers have achieved MT based on transformation generation rules, then language data has become the backbone of MT in the post-rule era. The Bayes conditional probability formula explicitly quantifies the language model and translation model contained in large-scale lan- guage data, which makes the noise channel model to decrypt an encrypted message become a 2 statistical MT paradigm. The deep neural network performs fine-grained characterization of super-large-scale language data, and uses many parameters to simulate the end-to-end NMT model that can generate fluent target language (Tan, Wang, Yang, Chen, Huang, Sun and Liu, 2020). RBMT is time-consuming and labor-intensive, and it is not easy to guarantee the self- consistency among many rules, so it is difficult to popularize into practical applications. SMT often needs …

Content truncated to 500 words