Genre-Based Variation in AI-Assisted Students’ English-Indonesian Translation Quality: A Comparative Assessment across Academic Texts, News Texts, and Poetry

Abstract

Artificial Intelligence (AI) performance in translation varies across genres due to the unique features of each genre. This study aims to explore the quality of students’ AI-aided translations of English academic texts, news articles, and poetry, and to determine whether translation quality varies by genre. A descriptive comparative research design was used in the study, involving a sample of 35 students from the English Departments of four universities in Indonesia. The participants translated an English academic text, a news article, and a poem using an Artificial Intelligence-assisted translation tool. The total number of translations was 105. Six translation experts evaluated all the translation samples using Nababan and Nuraeni’s model. They assessed translation quality in terms of accuracy, acceptability, readability, and a new dimension proposed by the researchers, i.e., poeticness, specifically for the translation of poetry. The study found that translations produced by students using AI ranged from moderate to high in quality. For the dimension of translation quality, the highest number of maximal scores (54.55%) was awarded for readability, followed by accuracy (51.56%) and acceptability (50.31%). Poeticness had the lowest score of 17.95%. It proved that AI’s ability to translate poetry is limited, as AI-generated translation cannot preserve literary devices such as figurative language, imagery, and rhyme, as well as the emotional and aesthetic elements of poetry.