ASSESSMENT OF HUMAN TRANSLATION VS. AI TRANSLATION IN A LITERARY WORK

Abstract

This study investigates the effectiveness of Gemini 2.0 Flash, a large language model developed by Google, in translating humor and stylistic features from David Walliams’ Awful Auntie into Vietnamese. The AI-generated translation is compared with a human translation by Pham Quoc Hung, using one of Yen Fu’s (1854-1921) classical translation principles, stylistic fidelity, as the evaluation framework. Findings indicate that while Gemini 2.0 Flash achieves a degree of literal accuracy, it struggles to convey the original text’s humorous tone and stylistic creativity. The AI consistently fails to preserve sound-based rhetorical devices such as assonance and consonance, resulting in a loss of auditory playfulness and comedic effect. It also shows significant limitations in rendering invented lexical items (neologisms), often leaving them untranslated or converting them in rigid, context-insensitive ways. Furthermore, the AI’s handling of idiomatic expressions reveals a tendency toward literalism, producing translations that are stylistically unnatural and pragmatically inappropriate. In contrast, the human translator demonstrates flexibility, creativity, and cultural fluency, adapting idiomatic language, wordplay, and stylistic nuance to better suit Vietnamese linguistic and cultural norms. These findings suggest that while AI tools can serve as efficient translation aids, they lack the interpretive depth and creative capacity required for high-quality literary translation, especially in the genre of humorous children’s literature. Human translators remain essential in preserving the artistic integrity and expressive impact of such works.