La post-édition en bref (en anglais)


Due to machine translation becoming more commonplace, there is a large demand for post-editors to correct computer-translated texts.


In recent years, post-editing has become a full-time job for many translators. Because neural machine translation has become the standard for a lot of translation work, the demand for post-editors has dramatically increased. Many experienced translators have switched from translation work to at least part-time post-editing.

Why post-editing is needed for content translated with NMT?

Texts translated by a neural machine translation (NMT) system are highly accurate and human-like, but some errors can still slip through. Grammatical errors, which are often difficult to spot, are quite common, as well as mistranslations of fixed expressions, metaphors, sarcasm, or humour.

Read more about the implications of NMT in our post on this topic.

Which are the tasks of the post-editor?

A good eye for detail is necessary when evaluating and correcting texts produced with NMT systems. Post-editors are tasked with finding these errors and correcting them each time. The errors are often repetitive and predictable, which makes them difficult to find. So the post-editor refers to the original text, as well as other translation system outputs to ensure they do not miss any mistakes. Checking for spelling is of course part of their task.

However, they do not fix any style or linguistic mistakes, as long as such corrections are secondary to the overall meaning of the text. In addition, post-editors must always edit any culturally offensive term and make the necessary terminology adjustments in their application.

It is thus not as easy as it might seem; which is why even experienced translators need specific training to learn about post-editing.

What are the types of post-editing?

Depending on the task, a translation can be “lightly” post-edited, “fully” post-edited, or “automatically” post-edited.

The first type of post-editing only takes out obvious mistakes, and leaves style and linguistic errors as they are. Full post-editing is more detailed and will correct more mistakes. The NMT system can also automatically self-edit, but this is less reliable. As the same errors are often recurring, the NMT system could easily miss all of them.

A post-editor usually focuses on words and phrases to avoid overlooking mistakes. They generally do not focus much on the sentences or paragraphs as a whole because fixing errors on that level may change the overall meaning, which is not the goal of post-editing.

To ensure that the post-edited translations are up to standard, a sample of the output is usually provided to the client before the entire post-editing job is done. This is to ensure that the amount of post-editing that has been done is sufficient.

However, post-editors may use slightly different guidelines, depending on the type of machine translation system they use. The domain or topic of the text also plays a part in this, as well as the language pair. For this reason, it’s hard to set universal rules for post-editing and to indicate what is the correct amount of “light” or “full” post-editing. The ISO 18587 norm has been set as a rule for post-editing of machine translated texts, but it is not comprehensive for each translation and post-editing job just yet. The field of neural machine translation is constantly changing, which makes it difficult to define general rules for post-editing.

What is pre-editing and what does it involve?

Certain texts in the source language are pre-edited before they are translated by means of NMT systems. During the pre-editing process the texts are slightly adjusted without changing the content, so that the system can understand them more easily and process them better. This reduces the frequency of errors.

Pre-editors would prefer using specific strategies like applying minimum changes and quick fixes over rewriting entire sentences. All in all, the final translation should match the source text as well as possible, without altering too much of the text.

Which are the further advantages of post-editing?

Besides fixing the translations, an added advantage, of course, is that post-editing provides feedback to the developers of neural machine translation tools. Accordingly, these tools can be improved for the specific mistakes that post-editors identify.

Further, even though the process of post-editing is very meticulous and not at all fast-paced, the combination of using neural machine translation with post-editing is still three to four times faster than traditional translating. An average post-editor can process 4,000 to 8,000 words of translated text per day, showing that the production of translation is very fast. Because of these advantages, the demand for professional post-editors has grown exponentially, and continues to do so. It secures new sources of revenues, but it requires additional skills and even experienced translators still have to learn the trade. They have to become familiar with the neural machine translation software, as well as the software for the post-editing itself. It also takes some time to train for finding small, seemingly correct mistakes.

Which are the professional perspectives for translators as regards post-editing?

With the developments in neural machine translation during the past few years, and in the years to come, post-editing will become a more sought-after profession. Even right now, many translators are already shifting their priorities to accommodate this change. The translation standard in the future may soon become a combination of neural machine translation and post-editing.



Pour en savoir plus sur nos services de post-édition, n’hésitez pas à nous contacter.