In the age of neural translation technologies and the ever-increasing volume of translations and narrow time frames, machine translation is a must.
In many cases neural machine translation systems are used, but are the translators ready for the new tasks?
This development has led to the creation of a new profession: The post editor.
The work of the post editor is completely different. The post-editor no longer translates from scratch, but already has a more or less good pre-translation.
But what must he do? What does he actually have to revise?
Together with the raw machine translation, the post-editor receives a guideline detailing what he needs to pay attention to. Obviously, the guidelines depend on the working methods:
- Light post-editing
- Full post-editing
- Customized Post-Editing
This type of post-editing is normally used for internal documentation. The final text need not be fluid. Even if the language style varies, this is not important. But the content of the source
text MUST be clear. Names, numbers and facts must be correct, of course.
This type of post-editing is normally used for external documentation. The final result is almost like a traditional translation and in some cases the quality may even exceed the quality of a
In this case we pay attention to brand identity. The terminology must be correct. The text must be consistent, and the context must be fluent and easy to read.
Customers may have specific requirements. It could also be a mix of light and full post-editing.
What should the post-editor not do in any case?
Translators who take up the profession of post-editor often make a big mistake. They tend to rewrite everything.
It is not supposed to be like this. Even if the machine translation does not necessarily correspond to our style, this does not mean that you have to rewrite the translation completely. This leads to too much time and frustration.
However, if the text cannot be translated automatically, the post-editor must inform the customer! There are also clear rules as to what the customer should pay attention to when creating the document.
Let's do a test!
To test how traditional translators react to post-editing, I have done a test on for German and English.
I selected a text that could be easily translated by a neural machine translation system.
1. First, I filtered out the "company terminology".
2. Then the post-editors have defined the terminology.
3. As a supposed customer, I have standardized and approved the terminology.
4. Then I put them into the NMT system.
5. Additionally, I built a termbase with the same terms.
6. The document was machine translated with the set NMT profile.
7. Together with the Full Post-Editing guide and the defined terminology, I sent the package to the post-editors.
8. They then revised the package.
9. They also wrote a report and indicated how long it took them to post-edit the text.
Result of the experiment
- The document was 900 words long.
- The duration of the post-editing varied from 45 minutes to 4 hours.
- In total 6 post-editors were involved.
All post-editors found that the system had adopted the terminology, except for some words written in capital letters. However, some technical terms were missing from the glossary, which meant
that they were not translated consistently by the NMT system.
Translators who had no experience with post-editing often tended to rewrite the sentences completely. This then led to frustration and errors in the target text.
Translators who already had experience with NMT systems did not tend to rewrite the documents and got along very well with the guideline. This saved them a lot of time.
Post-editing is a different way of working and means rethinking. However, you can learn the profession of post-editing. It is important, not to reject machine translations or to falter with
unnecessary translation proposals.
If you see post-editing as an opportunity, you can get a lot of work in the future. The systems do not necessarily take away the translator's work. But the translator's work has evolved: Translators are evolving into post-editors.