机器翻译字幕质量评估研究——以“ArcTime”字幕汉译为例

Assessing Machine Translation Quality in Interlingual Subtitling—A Case Study of ArcTime

摘要: AI时代迅猛发展的翻译技术对文本内容的翻译产生了重大影响。然而作为翻译技术中最新的发展之一，机器翻译技术(MT)更常应用于文本翻译之中，而在视听翻译中的应用仍处于初期。本文选取电影《疯狂动物城》机器翻译英译中字幕文本为研究文本，结合彼得森提出的FAR模型采用量化统计与质性分析相结合的方法，评估了ArcTime自动生成的英译中字幕文本。以探索机器翻译字幕质量评估以及制约机器翻译质量的影响因素，总结机器翻译质量以及未来机器翻译发展侧重点。本次机器翻译字幕质量评估研究表明，机器翻译(MT)在功能对等(functional equivalence)方面，尤其是语义选择与字幕切分等方面错误较多，从而导致机器翻译整体上质量偏低。此外，扣分仅次于语义错误的是接受程度(acceptability)方面，相对而言，机器翻译字幕在阅读体验(readability)方面整体上出现的错误相对较少。机器翻译字幕整体质量偏低的原因在于，一方面在于机器翻译无法获取源语言所处的视觉语境，另一方面是影视作品语言整体上口语体居多，而机器翻译其训练语料主要源于书面语，因此以上因素很大程度上制约了机器翻译字幕的质量。

Abstract: Since AI enabled translation technology provides basic and overall support for subtitle translation, it’s crucial to enhance subtitle translation technology research. However, machine translation (MT), as one of the most innovative technologies to be deployed in the field of translation, is currently more often used in (non-AV) text translation than in Audiovisual Translation (AVT), where its ap-plication is rare. In this article, the author intends to assess the machine translation quality pro-duced by ArcTime in translating the American movie entitled Zootopia, and the problems that would have an impact on the quality of the machine translation. For this purpose, the data was col-lected from the movie and qualitatively analyzed using Pedersen’s (2017) FAR model, which is a quality assessment model. The results of the study showed that most subtitles are of a relatively good quality in terms of readability, and the less errors or problems were found in terms of accept-ability. However, most of the errors are found in terms of functional equivalence, particularly in terms of semantic errors. In addition, the major problems that impact the machine translation of subtitles are lack of visual context, in which the source text is embedded and to which the MT has no access, and the generally informal language used in the audiovisual works.