EAMT 2020: MT Post-Editing Levels – Breaking Away from the Tradition & Delivering a Tailored Service

welocalize August 18, 2020

As content volumes continue to rise, global consumers demand local, relatable content and user experiences. Many companies look to language technology solutions as part of their localization program to process more multilingual content.

Post-edited machine translation (PEMT) is a process that involves the reviewing, editing, and correcting of machine translation (MT) output to deliver quality content. Post-editing is completed by linguists and can facilitate the translation of high volumes of content at speed, without the time and costs associated with full human translation.

The level and extent to which post-editing takes place can vary. For example, light editing of knowledge-base articles typically requires correction of major accuracy issues only, whereas full post-editing of branded marketing materials involves a more in-depth approach, often involving adjusting style and tone of voice. There are varying expectations of final quality depending on content type and use, and this impacts how much attention is needed by the post-editor.

For the upcoming European Association of Machine Translation (EAMT) conference, 3-6 November, Mara Nunziatini and Lena Marg, both MT program managers at Welocalize, have provided unique insights in how MT post-editing levels vary and are defined. They will present their paper, MT Post-Editing Levels – Breaking Away from the Tradition & Delivering a Tailored Service at the upcoming virtual conference.

“We have seen that in the language industry and academic literature, there are no shared definitions of the different levels of light, medium, and full MT post-editing,” states Mara. “At Welocalize, as we onboard advanced MT programs, we’re providing more post-editing services on the MT output. We ensure the quality levels and expectations of the post editors are managed so they know what they do and don’t have to fix, from basic accuracy through to style and fluency.”

The paper explains and summarizes the definitions of light, medium, and full post-editing service levels, drawing information from leading industry organizations such as TAUS, GALA, and ISO standards. There are no industry standard guidelines for providers or clients which can result in inconsistencies across providers, post editors, and end users.

Mara continues, “Another interesting point is how many linguists should be involved in the post-editing process. For full MT post-editing, you often have two linguists involved – one for the post-edit, one for the review stage. In a light post-editing situation, you would typically have one post editor only, because they are supposed to introduce a smaller amount of edits. Besides, you need to align post editor expectations with language quality assurance (LQA) measures.”


Some of the key areas the paper focuses on include:
  • Existing industry definitions and standards of post-editing
  • Challenges with existing definitions
  • Challenges with error typologies
  • Use case examples
  • How to measure adequate editing effort
  • How to provide guidance and set expectations

To read Mara and Lena’s EAMT 2020 paper in full, click here to download the PDF.

You can view all papers due to be presented at the EAMT Conference 2020 here: Proceedings of the 22nd Annual Conference of the European Association for Machine Translation.

Welocalize works with many global brands, managing high volume multilingual content strategies that integrate MT and  PEMT into the overall localization program. To learn more about leveraging language automation and striking a balance between MT and human translation to achieve international goals, connect with us here.