Copilot vs DeepSeek’s Translation of Denotative and Metonymic Abu- and Umm- Animal and Plant Folk Names in Arabic

Abstract

This study investigates the translation of denotative and metonymic Abu- and Umm-animal and plant folk names by Copilot (MC) and DeepSeek (DS) using three prompts (no-domain, with domain and metonymic), the strategies they use, causes of errors, and whether students and specialists can depend on AI in translating Abu-names. Results showed that in the denotative Abu-names, DS gave higher correct equivalents in response to the no-domain prompt (51% by DS vs 46% by MC) and the domain prompt (51% by DS vs 44% by MC). The equivalent animal’s name was directly given without any translation, transliteration or annotation (أبو مَركوب Shoebill). Both gave identical responses to 40% of the denotative items. In the metonymic name list, both MC and DS failed to identify the exact animal or plant type to which each Abu-metonyms in response to all 3 prompts. Both gave fewer than 3% correct responses to all 3 prompts. Similar equivalents were given to each Umm-name in response to the three prompts, of which MC gave 30% correct and 70% faulty equivalents with different wording. By contrast DS failed to give correct responses to all items in the no-domain prompt, 97%-99% faulty responses to the domain prompt & metonymic prompt, respectively. Regarding faulty strategies, MC translated Abu to “father” (46%); translated Abu + Noun semantically without “father” (أبو الشّبت *Dill beetle) (32%); made faulty guesses (أبو حُدَيج Possibly a local fish or bird name) (17%), transliterating the noun following Abu in 57% (البحتري أبو Father of Al Buhturi) and translating it (43%) (إبن جلي Son of Clarity). Both MC and DS considered metonymic names as personal names (55% by MC and 95% by DS). DS translated أبو Abu to “father” in 27%. In the annotation, DS gave the genus, not the specific animal implied (أبو الشّبت *Dill Father (a type of beetle). DS rendered “lizard” as the referent animal/plant in all items in response to the metonymic prompt. The study gives causes of the AI errors and recommendations for improving AI performance in translating Abu- and Umm-animal and plant names to English.