DeepSeek, Google Translate and Copilot’s Translation of Arabic Grammatical Terms Used Metaphorically

Abstract

Arabic has unique linguistic phenomena, one of which is that some Arabic grammatical terms (AGTs) are used as fixed expressions and metaphors in daily speech, with no such phenomenon in English. A sample of 52 AGTs that are used as metaphors and fixed expressions in Standard and Colloquial Arabic was collected and translated by three AI tools: DeepSeek (DS), Google Translate (GT) and Microsoft Copilot (MC). The sample contains conjoined prepositions, particles, pronouns, verb to be, and others. Data analysis showed that the highest correct translations were rendered by MC (43%), followed by DS (29%), and the least by GT )23.5%(.  DS, GT and MC gave identical  (correct and incorrect) translations to 57% of the terms. They tended to translate word for word, which sometimes resulted in weird and funny equivalents as in بين بين  between between, لعل وعسى perhaps and perhaps. Examples of terms translated correctly by MC only are الحكاية فيها ان there is something fishy, من إلى from A to Z; بفعل فاعل by an unknown person, حاشا وكلا absolutely not. An example of a term correctly translated by DS is بين بين  in between. Examples of terms with identical translations by the three AI tools are كائنا من كان whoever it may be, داعش واخواتها ISIS and its sisters, الدول العربية مفعول به the Arab countries are the object.  In الأردن المبتدأ ونحن الخبر Jordan is the subject, we are the predicate, المبتدأ & الخبر were mistranslated as they are polysemous, المبتدأ means subject and starting point, and الخبر means predicate and news. Specifically, the three AI tools have difficulties with polysemous AGTs and those with a cultural content as those used in titles of TV shows, those requiring a historical background (فيها ان) and those used as slogans. The percentage of correct translations rendered by AI herein is the lowest compared to medical and Gaza-Israel War terminology, zero-expressions and educational polysemes. Reasons for AI mistranslations, recommendations for raising students’ awareness of the weaknesses that AI has in translating AGT metaphors and suggestions for improvements are given.