A Comparative Evaluation of Generative AI Efficiency in Reproducing Arabic Metaphors: A Benchmark Analysis of ChatGPT vs. Google Translate

Shadman, Yosra

doi:10.22054/rctall.2026.93728.1856

A Comparative Evaluation of Generative AI Efficiency in Reproducing Arabic Metaphors: A Benchmark Analysis of ChatGPT vs. Google Translate

Articles in Press

Document Type : Research Paper

Author

yosra shadman

associate professor of ilam university

10.22054/rctall.2026.93728.1856

Abstract

The translation of metaphor is one of the most challenging areas of literary and intercultural translation, because metaphors carry not only lexical meaning but also the cognitive, cultural, and emotional systems of the source language. The present study adopts a descriptive–analytical approach and employs both quantitative and qualitative analysis to examine the performance of Google Translate and ChatGPT in translating Arabic metaphors into Persian. The research corpus consists of twenty Arabic metaphorical expressions that were analyzed in terms of metaphor type, translation strategy, and the quality of their Persian renderings. To evaluate translation quality, three criteria—accuracy, acceptability, and readability—were applied based on Nababan et al.’s model. The quantitative findings show that the mean accuracy score of Google Translate is 3.00, while that of ChatGPT is 2.75. However, ChatGPT outperforms Google Translate in the criteria of acceptability and readability, with a mean score of 3.00 compared to Google Translate’s mean scores of 2.35 and 2.35 respectively. Accordingly, in translating literary and culturally embedded Arabic metaphors into Persian, ChatGPT demonstrates a relative advantage in communicative and literary effectiveness, although human supervision and final editing remain necessary.

Keywords

Machine Translation

Translation Evaluation

Metaphor

Google Translate

ChatGPT

Subjects