From process to product: links between post-editing effort and post-edited quality

Research output: Chapter in Book/Report/Conference proceedingChapter in a book

430 Downloads (Pure)


Post-editing of machine translation (MT) is now increasingly implemented in the human translation workflow after studies in both industry and academia have demonstrated the efficacy of this practice. Post-editing still involves open questions, however, such as how best to train post-editors and how to estimate the effort required by post-editing tasks. In attempting to address some of these questions, many previous studies investigate the post-editing process, but less research has focused on the post-edited product. This chapter examines the link between the process and product of post-editing by checking to see how post-editing effort data related to the quality of post-edited texts, assessed in terms of fluency (linguistic quality) and adequacy (translation accuracy). A statistical analysis indicated that the association between editing operations and the fluency of post-edited texts is dependent on the quality of the raw MT output. Interestingly, a negative association was observed between the number of eye fixations on the text and the quality of the post-edited translations. The chapter shows empirical evidence supporting the distinction between the concepts of translation fluency and adequacy, and postulates that automatic processes play a central role in post-editing performance.
Original languageEnglish
Title of host publicationTranslation in Transition
Subtitle of host publicationBetween Cognition, Computing and Technology
EditorsArnt Lykke Jakobsen, Bartolomé Mesa-Lao
Place of PublicationAmsterdam
PublisherJohn Benjamins Publishing Company
ISBN (Electronic)9789027265371
ISBN (Print)9789027258809
Publication statusPublished - 2 Oct 2017

Publication series

NameBenjamins Translation Library
PublisherJohn Benjamins
ISSN (Print)0929-7316


Dive into the research topics of 'From process to product: links between post-editing effort and post-edited quality'. Together they form a unique fingerprint.

Cite this