The ISO 18587:2017 standard outlines the process, output quality expectations, and requirements for machine translation post-editing (MTPE). While the standard itself delves into significant detail, in this blog, we’ll discuss these topics at a higher level to help you understand why MTPE is so beneficial, what the workflow involves, and who is qualified to handle the work.

 

Why is MTPE gaining traction in the translation industry? 

In simple terms, adding human editors to MT or AI output can improve timelines and cost-effectiveness without compromising quality. While it’s widely understood that no MT/AI output can match the quality of human translation, incorporating qualified post-editors into the workflow can bridge that gap.

To illustrate the impact of different levels of machine translation quality, we can look at the Bilingual Evaluation Understudy (BLEU) score, a metric that evaluates the quality of machine-translated text by comparing it to human-translated text.

 

Graph comparing machine translation quality using BLEU scores, ranging from Standard MT to Fully Trained AI and Human Translation, showing improvement in quality from left to right.

 

As shown in the image, the BLEU score improves as the quality progresses from standard MT to semi-custom AI, fully trained AI, and ultimately to human translation. If you want to learn about the different use cases for each translation method and how to implement these strategies effectively to maximize your return on investment, check out our blog on launching a successful MT plan. Now, let’s discuss the general workflow and process of MTPE.

 

The objectives of MT post-editing 

The main objectives of the post-editing process consist of three parts:

  1. Comprehensibility: The final translation should make sense.
  2. Alignment: The source and target content must align, with nothing omitted or added.
  3. Compliance: The translation must meet all client specifications, including style guide, terminology, and formatting.

The third objective, compliance with specifications, is crucial. There should always be a discussion between the client and the Language Service Provider (LSP) during project scoping to ensure both parties understand these details. What good is a translation if it doesn’t meet the client’s needs after all?

 

Standard MTPE workflow steps 

While project specifications may differ, the standard MTPE workflow should include the following steps:

  1. Review of source material: The LSP will review source content to ensure it is suitable for MTPE. For example, scanned documents might not be ideal for MT, and a human-only translation might be more beneficial. Understanding the differences between Full and Light MTPE is crucial to determining the most suitable approach for your project.
  2. Pre-processing of source content: This step will aid in reducing the post-editing workload by fixing formatting and spelling issues in the source content to improve MT output.
  3. Agreement on workflow and scope: There must be a written document outlining the steps, project-specific objectives of MT Post-editing, and any client requirements.
  4. Translation production: Using MT or AI, the source content is translated and then post-edited by a qualified post-editor. The post-editor will read the MT output and decide if the translation is suitable or needs reworking. They may use elements from the MT or rewrite the content entirely. There should always be a feedback loop for post-editors to provide feedback on the MT output as well.
  5. Final QA and delivery: The LSP will ensure all steps and quality objectives are met before delivery. If Full Post-Editing is scoped, the final translation must be free from grammar, spelling, and punctuation errors; free from inappropriate, incorrect, or unclear content; and adhere to client-preferred terminology, style guides, and formatting. In short, the final output should be comparable to human translation.

 

Who is qualified to handle post-editing work? 

The standard is clear about the qualifications and training a post-editor must have to properly post-edit MT output. It’s not enough to merely speak another language; specific training and experience in translation are required. Post-editors must meet at least one of the following qualifications:

  1. A degree in translation, linguistics, or a similar field.
  2. A degree in another field plus 2 years of professional translation experience.
  3. 5 years of professional translation experience.

LSPs should provide in-depth information about the translators and post-editors working on your translation project to maintain full transparency. For example, we offer regulatory compliance platform (RCP) reporting, which includes detailed information about the translation project, the project contributors, and their qualifications.
In addition to having a proper background in translation education and experience, LSPs must offer continual training to their post-editors. Post-editors need to understand how MT/AI works and possess significant knowledge of the tools, processes, and quality objectives to produce output on par with human translation.

 

Final thoughts

By understanding the importance, workflow, and qualifications of MTPE, you can better appreciate the value it brings to translation projects. This process ensures quality, efficiency, and compliance with client needs, making it an essential component of modern translation services.

Interested in learning how an MTPE workflow would look for your specific translation needs? Feel free to reach out to one of our AI translation experts!

Related Blog Articles

Full & Light Machine Translation Post-Editing: What’s the Difference?
Full & Light Machine Translation Post-Editing: What’s the Difference?
Read article ›
How to Improve Efficiency with AI
How to Improve Efficiency with AI
Read article ›
Five Tips for Translating Educational Materials
Five Tips for Translating Educational Materials
Read article ›
DeepL vs Google Translate vs Microsoft Translator Review: Which is Best?
DeepL vs Google Translate vs Microsoft Translator Review: Which is Best?
Read article ›
images