Comparing human and AI translation: Lexical diversity, structural patterns, and semantic fidelity in L2 learners’ texts

This study investigates the differences between translations produced by novice L2 writers and AI systems in terms of lexical diversity, structural patterns, and semantic fidelity. Drawing on a corpus of 100 translations by Korean university students and outputs from three AI platforms, the research analyzes a Korean newspaper article on the dual nature of parody using a mixed-methods approach. The findings reveal that while student translations exhibit emotional depth and interpretive creativity, they often struggle with semantic precision and cohesion. In contrast, AI-generated translations are more concise, structurally coherent, and semantically faithful—but frequently lack cultural sensitivity and nuance. These contrasts highlight the potential of AI to support language learning while reinforcing the irreplaceable value of human insight in culturally rich translation tasks. The study advocates for integrative translation pedagogy, where human and AI strengths complement each other in academic settings.