Human vs AI translation of Common Names of Chemical Compounds: A comparative Study

Abstract

In a study by Al-Jarf (2022), undergraduate student-translators had difficulty translating English common names of chemical compounds (CNCC) to Arabic. They gave correct responses to fewer than 20% of the test items and left 55% blank. Their most common translation strategy was literal translation and transliteration of English names. The students reported that they are not familiar with most of the English and Arabic chemical compound names. Based on these findings, the current study aims to find out if student-translators can use Artificial Intelligence (AI) to translate CNCC from English to Arabic and vice versa, % of chemical compound names correctly translated by AI and the translation strategies used by AI. Two samples of English and Arabic CNCC were translated by Microsoft Copilot (MC). Data analysis showed that MC translated 72% of the chemical compounds in the sample correctly. It gave more Arabic-English than English-Arabic correct equivalents (40% & 32% respectively). It gave more correct equivalents when I specified the domain and when I asked for all the equivalents that MC knows. MC gave literal word-for-word translation to terms as (Lunar caustic > الكاوي القمري(نترات الفضة); transliteration (Stearic acid >حمض الستيريك instead of حمض الشمع; a faulty derivative (Chlorinating powder > مسحوق التكلور instead of مسحوق الكلورة) and explanation (Ammonia liquor سائل الأمونيا). Although some English chemical compounds have multiple Arabic equivalents, MC gave only one unless prompted to give all. Even though MC is characterized by breadth of knowledge of English and Arabic chemical terminology and quick access to each term and in which context(s) it is used, in few cases, it failed to match the English chemical compound name with its Arabic equivalent and vice versa. Since translation-students’ knowledge of CNCC is limited, they are advised to specify the kind of terms to be translated and to ask AI to give as many equivalents as possible. They should give AI the context, nuance, and direction so that it matches the term with its correct equivalent. Results and recommendations are given in detail.