Oct 2026· Research Journal in Advanced Humanities· 0 citations· 62 references
TL;DR
The study concludes that AI systems require substantially more training, particularly in Arabic-language contexts, and recommended further research on homonym translation, focusing on different contexts as well as the rendition of homonyms outside of pairings.
Abstract
Homonymy is one of the most persistent challenges in translation, particularly for AI-based systems. This study examines the abilities of three prominent AI chatbots, ChatGPT 5, DeepSeek (V3.1), and Gemini (2.5 Flash), in the English translation of Arabic homonyms. The dataset comprised Qur’anic verses and classical poetry. The homonyms were first collected, then analyzed for contextual meaning, and then translated by the chatbots. The data were presented without diacritics to replicate the conditions of authentic Arabic text. After the items were translated, the data were assessed using the TAUS Adequacy/Fluency model to determine the extent to which the systems preserved meaning while maintaining linguistic naturalness. The results showed that while all chatbots often achieved high fluency, their adequacy slightly differed across text types and homonym categories. Gemini outperformed the other models overall, followed by ChatGPT and DeepSeek. The Qur’anic translation performed best due to the high prevalence of such texts in online corpora. However, the performance declined noticeably in the poetic and classical domains. While large language models (LLMs) represent a significant improvement over traditional machine translation (MT) systems in handling lexical ambiguity, they remain unable to reliably disambiguate complex Arabic homonyms without contextual or cultural grounding. The study concludes that AI systems require substantially more training, particularly in Arabic-language contexts. The study therefore recommended further research on homonym translation, focusing on different contexts as well as the rendition of homonyms outside of pairings.
Despite the growing accuracy of machine translation (MT) in general-purpose tasks, its performance on culturally embedded, humor-driven text remains largely unexamined, particularly for the English-Indonesian language pair. This study examines the acceptability of humor translation produced by three MT tools, Google Tr...
The increasing use of large language models (LLMs) in professional translation has reshaped multilingual news production. While prior research has reported gains in fluency and surface-level accuracy, limited attention has been paid to LLMs’ ability to translate figurative language, which plays a key role in meaning co...
Phuong Luong, Huong Thi Thanh Nguyen, V. Tran· Theory and Practice in Langu...· 0 citations
Abstract: This study examines translation fidelity in Pétit Minitaire, the French version of Ken Saro-Wiwa’s Sozaboy as rendered by Samuel Millogo and Amadou Missiri. The mixture of Nigerian pidgin, broken English, standard, and idiomatic English in which the original text is written is the key point that prompted my s...
Cocou André Datondji· Akofena revue scientifique d...· 0 citations
This study investigates stylistic features in English–Italian machine translation by comparing the outputs of Google Translate and ChatGPT-5.2, and shows that the two systems seem to adopt similar translation solutions, but they differ in the quantitative distribution of translation equivalents.
Natália Resende, Sheila Castilho, Maria Clara Menezes et al.· 0 citations
This article examines the core challenges of translating phraseological units in Machine Translation (MT) and Artificial Intelligence (AI), focusing on the Georgian-Russian language pair. Phraseologisms, characterised by cultural specificity and semantic indivisibility, pose a significant hurdle for automated systems t...
Shorena Tsabadze· Online Journal of Humanities· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.