Translation Quality in Human and Machine Translation: A Small Corpus-Based Study

Translation quality assessment remains a central concern in Translation Studies, particularly as neural machine translation (NMT) becomes increasingly integrated into professional workflows. This study presents a small corpus-based evaluation of English sentences translated into Persian under three conditions: (۱) professional human translation, (۲) neural machine translation, and (۳) machine translation with human post-editing (MT+PE). Drawing on functionalist perspectives on translation quality and established frameworks for equivalence, the study operationalizes quality through three dimensions: adequacy, fluency, and annotated error counts. Adequacy and fluency are scored on a five-point ordinal scale, while errors are categorized into lexical/terminological, syntactic, semantic/pragmatic, and register/fluency-related types. Results indicate that human translation achieves the highest mean adequacy and fluency scores. NMT underperforms primarily in semantic/pragmatic fidelity and targeted lexical precision. Post-editing substantially improves both adequacy and fluency, narrowing the gap to human translation, though not fully eliminating discrepancies—especially in pragmatic positioning and discourse-level cohesion. The findings suggest that post-editing is most effective when human expertise targets systematic NMT weaknesses rather than performing surface-level edits. Overall, the study contributes fine-grained evidence for translation quality assessment in a human-machine continuum.

کلیدواژه ها:

Translation quality Assessment ، Adequacy ، Fluency ، Equivalence ، Functionalism ، Neural Machine Translation ، Post-editing