Comparing the Efficiency of Source Text Pre-editing vs. Machine Translation Post-editing

Abstract

As machine translation (MT) becomes increasingly embedded in professional workflows, researchers explore ways to improve quality and efficiency. Although neural MT systems like DeepL and Google Translate improve fluency, they still require human intervention. Two key strategies are pre-editing (PrE), which modifies the source text before MT to reduce errors, and post-editing (PoE), which refines MT output to meet quality standards.
This study compares PrE and PoE in MT workflows through a controlled experiment involving 20 translation students. One group used PoE alone, while the other combined PrE and PoE. Translation quality was assessed using the TAUS Dynamic Quality Framework, with time efficiency also analyzed.
Findings show PoE alone accelerates the process but increases error rates, particularly in accuracy and fluency. PrE enhances translation quality by reducing errors and cognitive load during PoE, though it requires more time upfront. The combination of PrE and PoE produced the highest-quality translations, suggesting that integrating PrE improves accuracy and consistency. These results highlight the importance of combining human expertise with MT to improve workflows, balancing speed and quality in professional translation.