Using Grammatical and Semantic Correction Model to Improve Chinese-to-Taiwanese Machine Translation Fluency

Using Grammatical and Semantic Correction Model to Improve Chinese-to-Taiwanese Machine Translation Fluency Yuan-Han Li ncku10509@gmail.com Chung-Ping Young dryncku@gmail.com Wen-Hsiang Lu whlu@mail.ncku.edu.tw National Cheng Kung University Department of Computer Science and Engineering Abstract Currently, there are three major issues to tackle in Chinese-to-Taiwanese ma- chine translation: multi-pronunciation Taiwanese words, unknown words, and Chinese-to-Taiwanese grammatical and se- mantic transformation. Recent studies have mostly focused on the issues of multi- pronunciation Taiwanese words and un- known words, while very few research pa- pers focus on grammatical and seman- tic transformation. However, there exist grammatical rules exclusive to Taiwanese that, if not translated properly, would cause the result to feel unnatural to na- tive speakers and potentially twist the orig- inal meaning of the sentence, even with the right words and pronunciations. Therefore, this study collects and organizes a few com- mon Taiwanese sentence structures and grammar rules, then creates a grammar and semantic correction model for Chinese- to-Taiwanese machine translation, which would detect and correct grammatical and semantic discrepancies between the two languages, thus improving translation flu- ency. Keywords: Machine translation, Tai- wanese grammatical rules, Lexical transfor- mation, Syntactic transformation, Chinese-to- Taiwanese 1 Introduction Machine translation systems are being increas- ingly used across multiple fields, such as busi- nesses, tourism and medical industries. These systems can translate multiple languages like English, Chinese, Japanese and even obscure traditional languages such as Swahili and Croatian. Taiwanese machine translation is likewise gaining importance as the govern- ment becomes more aware of its historical and cultural importance, thus setting out to digitally preserve the language. There has been multiple research papers on Chinese-to- Taiwanese (henceforth referred to as C2T) machine translation, usually focusing on the issues of multi-pronunciation words and un- known words translation. However, despite the importance of grammatical and seman- tic differences between languages in machine translation, it has been observed that papers that integrate them into C2T translation sys- tems are comparatively rare, which results in the output of C2T systems losing fluency and potentially the original meanings of the orig- inal Chinese input. As such, this paper pro- poses a grammatical and semantic error de- tection and correction model, which can im- prove C2T translation fluency by correcting the discrepancies between Chinese and Tai- wanese grammar. 2 Related Work Considering the importance of machine trans- lation (Raad, 2020; Panayiotou et al., 2020; Kapoor et al., 2019), researchers begin to ap- ply rule-based (Hurskainen and Tiedemann, 2017), statistical (Och et al., 1999; Koehn et al., 2003), and neural machine transla- tion (Sutskever et al., 2014; Vaswani et al., 2017) technology to various languages, includ- ing C2T translation models (Lin and Chen, 1999), in order to tackle various recurring is- sues in this field, such as choosing the correct pronunciation for multi-pronunciation words and unknown words. For multi-pronunciation words, (Wu, 2015) extracts the features of each word in the input sentence, such as part-of-speech (POS) and se- mantic meaning, then employ feature models The 34th Conference on Computational Linguistics and Speech Processing (ROCLING 2022) Taipei, …

Content truncated to 500 words