Translation of Zero-Expressions by Microsoft Copilot and Google Translate

Abstract

A corpus of 318 English and Arabic zero-expressions used in general as well as specialized contexts as math, technology, law, military, economics, finance, and others was collected from Al-Maany Online dictionaries. The expressions were translated by Microsoft Copilot (MC) and Google Translate (GT) to find out the percentage of expressions correctly translated by both, the translation strategies used, and to explore the semantic, lexical, syntactic, and contextual inaccuracies that mistranslations reveal. It was found that 29% of the zero-expressions in the sample were correctly translated by both MC and GT. This percentage represents less than the correct translations of medical and Gaza-Israel War Terminology rendered by MC and GT. In 52% of the translations given by MC and 50% of the translations given by GT, the Arabic equivalent zero expressions consisted of a noun + a derived adjective صفرية الصفرية/ /صفري/ الصفري. In 31% of the data, MC gave definite equivalents (zero rating التصنيف الصفري) compared to 9% by GT. In 11%, GT rendered equivalents with an awkward word order (zero for zero approach صفر لنهج الصفر). In 12%, MC and GT gave similar Arabic equivalents with a reversed word order (zero fraction كسر الصفر (MC), صفر الكسر (GT). In 5%, MC and GT gave faulty Arabic equivalents with different derived forms (output zero إخراج الصفر (MC) & صفر المخرج (GT) instead of صفر مخرجات). The most common translation strategy used was word-for-word translation. Conceptual translation and modulation were not frequently used (zero position وضعية صفرية (MC), موضع الصفر  (GT) instead of وضع الابتداء ; Zero duties واجبات صفرية instead of بدون رسوم).  Zero expressions containing a polyseme were mistranslated (false zero صفر خاطئ (MC), صفر زائف (GT) instead of صفر غير حقيقي). Both MC & GT failed to give the underlying meaning of idiomatic phrases as الشمال صفر على which means has no value. Both gave a word-for-word translation zero on the north (MC) and zero to the north (GT), which are meaningless. Problems that AI has in translating zero-expressions are described and discussed in detail. The article concludes with some recommendations for AI specialists and translation pedagogy.