Copilot’s English Translation of Contrastive Emphatic Negation in Arabic Discourse: An Analytical Study

Abstract

This study explores how AI, vis Microsoft Copilot (MC) translates Arabic contrastive emphatic negation (CEN) expressions, types of translation errors and translation strategies used. Analysis of a sample of 436 Arabic CEN expressions using a variety of negative particles showed that MC gave 65% correct translations of CEN expressions in the sample. MC gave more correct translations of expressions with a single negative particle than correlative conjunctions as لا ... ولا neither …  nor, when the structure has a transparent rather than an underlying meaning, when negation is literal, and structurally simple (الكبير مش الصغير the big stand, not the small one; الكيف وليس الكم quality, not quantity; عاجلا غير آجل sooner not later). On the contrary, MC made semantic and syntactic errors as failure to understand and convey the meaning of idiomatic expressions and equivalent set idioms/phrases, difficulty with polysemous words, rendering faulty lexical choices, faulty structure and faulty wording, faulty choice of negative particles, faulty use of articles, and equivalents with faulty part of speech and derivatives. MC failed when the negation is idiomatic, culturally loaded, polysemous, or pragmatically marked.  Nevertheless, CEN expressions were easier for MC to translate than zero-expressions, Gaza-Israel war terminology, Arabic grammatical terms used metaphorically, expressions of impossibility, Arabic folk medical terms with om and abu, Arabic abu-brand names using different prompts, and metonymic abu and umm animal and plant names. As in prior studies by the author, MC tended to translate word for word, rather than giving the fixed English equivalent or an accurate semantic equivalent (Finally, not lastly for أخيرا وليس آخرا rather than last but not least; دون كلل أو ملل without fatigue or boredom instead of tirelessly. Faulty translations demonstrated that MC’s weaknesses are concentrated in expressions that require recognition of multiword expressions with and idiomatic or metaphorical meaning, accurate particle selection, lexical choice, pragmatic interpretation, cultural competence and semantic inference.  The study recommends that translators and students should not take MC translations for granted but should verify and post-edit its output.