The 2nd VietCALL International Convention (VietCALL 2026)

Idioms resist translation because their meaning and rhetorical effect arise from context rather than from their component words. This study compares how a published literary translation and a generative AI system render English idioms into Vietnamese. Seventy-six idiom occurrences from Harry Potter and the Sorcerer’s Stone, the first novel in the series, were aligned with Lý Lan’s published Vietnamese edition and with ChatGPT output elicited through one standardised prompt, then coded using Baker’s strategy categories and assessed for meaning, tone and naturalness. Across 76 comparable items the two versions reach idiomatic equivalence at similar rates, but they diverge systematically: the published translation omits an idiom outright three times, borrows twice and compensates once, whereas ChatGPT does none of these and converts such cases into paraphrase (36 instances versus 25). The study gives learners a concrete profile of how machine translation handles idioms and supports critical evaluation of AI output.