Translating classical Arabic verse: human translation vs. AI large language models (Gemini and ChatGPT)

Abstract

Classical Arabic poetry is a sophisticated and complex art form, combining intricate themes with refined stylistic forms. This paper examines the translation of 15 individual Classical Arabic verses by comparing the English renditions provided by a human translator and two prominent large language models (LLMs), namely Google’s Gemini (GEM) and OpenAI’s ChatGPT (GPT). The human and LLMs are evaluated in relation to three variables: thematic clarity, creativity, and prosody by 54 Arab professors in Literature, Linguistics, or Translation who use English as a medium of instruction at their universities and are considered qualified as assessors. The translations are also evaluated by the two LLMs. The findings show that participants’ assessment of human and GPT translations are clearly positive on the three variables, while GEM significantly lags behind in terms of prosody (M 2.83) and even more so in its own assessment (M 1). As for AI technology assessment, GPT is clearly consistent with participants’ evaluation, in contrast to GEM assessment, which is markedly different in terms of prosody and creativity. The study concludes that AI technology emerges as a true competitor with human translators, and their interactive potentials need to be further explored in poetry translation.

Keywords

• Arabic classical poetry
• translation
• cognition
• prosody
• ChatGPT
• Gemini