Alwazna Rafat Y
Department of Modern Languages and Literature, Faculty of Arts and Humanities, King Abdulaziz University, Jeddah, Saudi Arabia.
Front Artif Intell. 2024 Jan 12;6:1282020. doi: 10.3389/frai.2023.1282020. eCollection 2023.
Efforts to automate translation were made in the 1950s and 1960s, albeit with limited resources compared to current advanced standards. Machine translation is categorised under computational linguistics that examines employing computer software in the rendition of text from one language into another. The present paper seeks to compare five different machine translation systems for the sake of assessing the quality of their outputs in rendering certain articles of the Saudi Commercial Law into English through post-editing based on Human Translation Edit Rate. Each machine translation output is assessed against the same post-edited version, and the closest output to the post-edited version with regard to the use of the same lexicon and word order will achieve the lowest score. The lower the score of the machine translation output is, the higher quality it has. The paper then analyses the results of the Human Translation Edit Rate metric evaluation to ascertain as to whether or not high-quality machine translation outputs always produce acceptable Arabic-English legal translation. The present paper argues that the use of Human Translation Edit Rate metric is a useful tool for the sake of undertaking post-editing procedures as it is a combination of both human evaluation as well as automatic evaluation. It is also advantageous as it takes account of both the use of lexicon and word order. However, such metric cannot be sufficiently depended on as one term substitution, which will be counted according to this metric as a single error, may render the whole sentence invalid, particularly in legal translation. This paper offers a baseline for the quality assessment of machine translation output through post-editing based on Human Translation Edit Rate metric and how its results should be analysed within Arabic-English legal translation context, which may have implications for similar machine translation output quality assessment contexts.
20世纪50年代和60年代就有人努力实现翻译自动化，尽管与当前的先进标准相比资源有限。机器翻译属于计算语言学范畴，该领域研究如何使用计算机软件将文本从一种语言翻译成另一种语言。本文旨在比较五种不同的机器翻译系统，以便通过基于人工翻译编辑率的后期编辑来评估它们将沙特商业法的某些条款翻译成英语的输出质量。每个机器翻译输出都与相同的后期编辑版本进行比较，在使用相同词汇和词序方面最接近后期编辑版本的输出得分最低。机器翻译输出的分数越低，其质量越高。然后，本文分析了人工翻译编辑率指标评估的结果，以确定高质量的机器翻译输出是否总能产生可接受的阿拉伯语-英语法律翻译。本文认为，使用人工翻译编辑率指标是进行后期编辑程序的一个有用工具，因为它是人工评估和自动评估的结合。它还有一个优点，即考虑了词汇的使用和词序。然而，这种指标不能完全依赖，因为一个词的替换，根据这个指标会被算作一个错误，可能会使整个句子无效，特别是在法律翻译中。本文为基于人工翻译编辑率指标的后期编辑机器翻译输出质量评估提供了一个基线，以及在阿拉伯语-英语法律翻译背景下应如何分析其结果，这可能对类似的机器翻译输出质量评估背景有启示。
