Deviasi Sintaksis dan Kalimat Mubazir pada Terjemahan Akademik Berbasis Generative AI: Studi Komparatif ChatGPT dan Gemini

Abstract

This study analyzes the academic translation performance of two generative artificial intelligence platforms, ChatGPT and Gemini, focusing on identifying syntactic deviations and redundant sentence constructions. While prior research has documented syntactic weaknesses in generative AI translation in general, systematic platform-to-platform comparisons on comparable academic corpora remain scarce. Employing a qualitative approach with a descriptive-comparative content analysis method, this research examines a corpus of 10 English international journal abstracts translated into Indonesian by both AIs. The results indicate that both platforms frequently produce translations with syntactic deviations, such as rigid sentence structures and literal translations that mimic source language patterns, as well as redundant sentence constructions in the form of pleonasm and semantic repetition, though with distinct dominant patterns, namely ChatGPT more frequently produces rigid, literal structures, while Gemini more frequently produces convoluted sentences and semantic redundancy. These limitations suggest that current generative AI is not yet fully capable of internalizing the stylistic nuances and pragmatics of Indonesian in an academic context. The pedagogical implications of these findings highlight the importance of developing platform-aware critical post-editing skills for students and adapting language learning curricula to train them in recognizing and correcting AI-generated errors. This study contributes to understanding the platform-specific strengths and weaknesses of generative AI in academic translation and suggests future directions for AI development.