AI vs human lexicographers

Abstract

This contribution interpretatively assesses the quality of English translation from Arabic into English by the two currently leading free-to-use artificial intelligence (AI) tools, ChatGPT and Google Gemini, against translations listed in Al-Mawrid (the globally leading Arabic-English dictionary) and a comprehensive compilation of Arabic phrases by Abu Ssaydeh (2013). Data comprised the d-30 (d=difficult) Arabic phrases (a mixture of collocations and cultural references) that have persistently posed problems for lexicographers of Arabic-English bilingual dictionaries, who mostly simply exclude them. Deploying Pedersen’s 2011 taxonomy of translation strategies, the analysis indicates some interesting results. Al-Mawrid lists only 5 (c. 17%) of the d-30 phrases, but Abu Ssaydeh does better by including 20 (c. 67%). In complete contrast to the two human translators, both AI tools render all the d-30 (100%). While both AI tools displayed some seriously interesting human-like renderings (adding contextual clues), they nonetheless generated some howlers, usually associated with machine translation. Despite their recent launch (November 2022 for ChatGPT and December 2023 for Google Gemini), their English translations of the d-30 Arabic phrase examined here point to excellence seriously being established. The tedious, time-consuming and often thankless task of dictionary-making may well benefit from a clever combination of both humans and AI tools (H-AI, perhaps)