AI Translation of Full-Text Arabic Research Articles: The Case of Educational Polysemes

Abstract

Due to the latest technological advancements, AI tools, assistants and chatbots have been used to perform tasks in a variety of domains, including translation. Some researchers and graduate students use ChatGPT, Google Translate, QuillBot, Smartling, and DeepL to translate research articles from Arabic to English for their theses or assignments as AI saves them time and effort. Although AI translation of full texts sounds natural uses good style and sentence structure, there are still contextual and semantic inaccuracies. There is insufficient research on the quality of AI translation of full-text articles from Arabic to English. Therefore, this study explores the problems that AI has in translating polysemes in research articles from Arabic to English. Mistranslated Arabic polysemes in full-text education articles were identified. Data analysis showed that AI has difficulty translating polysemes that have general and specialized meanings and two or more English equivalents (i.e., one to many), such as صدق which has the general equivalent “honesty” and the technical equivalent “validity” used in research;  والمحكمون التحكيم are used in legal, sports and research contexts, but AI gave the equivalent used in legal contexts not the one used in an educational contexts. It gave “arbitration” & “arbitrators” rather than “peer reviewing” & “reviewers”. AI translated المنهج المحوري to “axial” instead of “spiral” curriculum. رسالة has 4 meanings in Arabic with 4 English equivalents (thesis, message, mission & letter) depending on the context. Most occurrences of رسالة were translated into “message”, rather than “thesis”. تصورات was translated into “visions” not “models”, خطة > “plan” not “proposal”, لجنة المناقشة > “discussion committee”  not “defense committee”. Further mistranslations were given to junior and senior high school grades, الدليل الإرشادي, المادة العلمية, العبء التدريسي and others. It was noted that AI tends to give literal, not conceptual, equivalents to Arabic terms and those used in a particular domain. The study recommends that researchers use AI translation with caution, post-editing the translation, and using the technical terms commonly used in education. Results, causes of AI mistranslations and recommendations for improvement are given.