Redefining efficiency in post-editing of LLM-generated translation: why expert translators take more time
Year: 2026
Type: article
Abstract: This study examines how expertise influences the post-editing of translation generated by large language models (LLMs) through the lens of relevant efficiency—the adaptive regulation of effort in response to task demands, which is operationalized here as temporal regulation, intervention calibration, and resource adaptability. We used keystroke logging and mixed-effects modeling to compare student and professional translators who performed full and light post-editing (FPE/LPE) tasks on LLM-generated translations under structured termbase and open-ended Internet-search settings. The findings suggest that post-editing expertise is not necessarily reflected in faster processing, but in how post-editing behavior relates to task requirements. Experts showed a greater increase in time allocation as the density of critical errors increased, indicating more task-sensitive temporal regulation. Expertise-related patterns of error correction varied with post-editing guidelines, while resource conditions were differentially associated with new-error introduction across the two groups. Keystroke-based process indicators also related to editing outcomes, highlighting the importance of evaluating editing activity in relation to its results. These findings support a task-relative understanding of relevant efficiency in the post-editing of LLM-generated translation, suggesting that translator training should prioritize task-sensitive decision-making, temporal regulation, and effective use of external resources rather than processing speed alone.
Source: Frontiers in Psychology
Authors: Longhui Zou(Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft), Michael Carl(Conceptualization, Formal analysis, Methodology, Software, Validation, Writing – review & editing), Jia Feng(Data curation, Methodology, Resources, Validation, Writing – review & editing)
Language: English
Subfield: Language and Linguistics
Field: Arts and Humanities
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7237210548
