The Role of Post Editing in Machine Translation

Post-editing is the process of correcting machine translation output so that it meets objective quality criteria. As an early adopter of machine translation, LexWorks has extensive experience in using post-editing in combination with machine translations to meet our client’s needs.

Post-editing is the process of correcting machine translation output so that it meets quality criteria. As an early adopter of machine translation, Lexcelera has extensive experience in post-editing machine translations.

Depending on the quality needed, post-editing may be light or full. Light post-editing aims at making information understandable, if not perfectly fluent, while full post-editing provides publication quality. Full post-editing ensures that the document reads like a human translation, and cannot be identified as a machine translation.

Not all machine translation output requires a post-edit

With proper training and full customization, some machine-translation (MT) engines are capable of delivering understandable quality for functional content without any post-editing whatsoever. Either raw or lightly post-edited MT output may be appropriate for frequently updated content such as customer support knowledge bases, user-generated content (UGC) and legal eDiscovery.

Full post-editing delivers publication quality. It can actually provide higher quality than a traditional process because:

  • A well-trained rules-based engine will ensure complete terminological consistency
  • Working in a bilingual environment, post-editors have been shown to pay closer attention when correcting MT matches than when correcting TM, (translation memory) matches, and thus they obtain higher quality

  • Quality is ‘publication quality’
  • The goal is a message that is accurate and complete but indistinguishable from a fully human translation
  • The editing effort aims to make more fluent sentences
  • Terminology is generally managed by the trained MT engine
  • The quality must be equal to that of a traditional translation!

  • Quality is ‘understandable’
  • Corrections are made only when necessary for comprehension; subjective changes are avoided
  • The goal is a message that is accurate and complete
  • The text may contain stylistic errors and awkward sentences
  • Terminology is generally managed by the MT engine
  • The information must be able to be understood without consulting the original text