How to Master Post-Editing Machine Translation

How to Master Post-Editing Machine Translation


Technology in the language industry is evolving at an incredibly fast pace. With new tools and solutions being constantly released, it is important for linguists to know how they can make the most of the technology available to them.

Machine translation is currently a major point of focus in the industry and post-editing machine translation is quickly gaining popularity as a service among language service providers’ clients. With the landscape changing fast, linguists face the need to adapt accordingly and learn how to become skilful post-editors.

These days, machine translation is integrated into majority of CAT tools, which means that linguists can easily use it as an additional productivity booster whenever they can, either globally on entire files or at segment level only.

Opting for post-editing machine translation rather than translation from scratch is intended to reduce the time it takes to localize a text. This means linguists might want to choose this option to handle more repetitive, less creative texts, leaving themselves with extra time to take on more jobs or to allocate more time to creative tasks such as literary translation, transcreation or copywriting.

Translation and post-editing are two very different linguistic tasks. Translation involves the linguist reading the source content, formulating the translation in their mind, which might require a certain degree of research, and then finally writing it down.

With post-editing machine translation, the linguist reads through the source text carefully, reads the machine translation output and compares it with the source text to identify and correct any errors. Linguists have to pay close attention to grammar, punctuation, spelling, word order, style, non-translated words, any potential mistranslations to arrive at a final translated text.

Computer showing a brain on the screen

Depending on the setup of an MT system, by doing post-editing linguists are also automatically feeding the MT engine with corrections, which helps it continuously improve over time as it learns from the edits.

The key to successful post-editing is quick decision making

After reading the machine translation output and comparing it to the source to understand how accurate it is, the linguist needs to be prompt in deciding whether it is more efficient to post-edit the machine translation suggestion or to delete it and translate from scratch.

If the MT output is of good quality and only needs some tweaks, then it should be post-edited. However, if the MT output is of poor quality and would take the more time to post-edit than to re-translate, the linguist should opt for translation from scratch.

Another key moment for decision making is establishing whether a machine translation segment needs editing or not. Some language service providers, in an effort to make sure the process is efficient, ask their linguists to move on to the next line of text if they cannot find anything wrong with a machine translated segment within as little as 3 seconds.

The process of editing machine translation will only prove productive and time-saving if the linguist uses as much of the MT output as possible.

Over-editing goes against that principle, which means that making amendments which are purely preferential or not completely necessary should be avoided. If the post-editor is tempted to replace a word with a synonym whilst both variants are viable options, they should refrain from doing so. The same applies to word re-ordering in languages with no strict sentence word order.

On the other hand, under-editing should also be avoided at all costs. Under-editing means leaving errors in the target copy. For instance, failing to correct mistranslations, spotting punctuation errors, leaving the translation sounding not fluent and robotic, failing to make sure the approved terminology is used.

Therefore, finding the right balance between editing too much and too little is a fine art that can only be perfected by practicing.

It is also useful to always bear in mind the following guidelines:

  • The final post-edited machine translation should always be an accurate representation of the source text that is semantically, syntactically and grammatically correct.
  • No information included in the source should be missing from the target copy.
  • Equally, the target copy should not contain any information that did not feature in the source.
  • Any inappropriate or offensive content should be edited.
  • Ensure that the terminology is consistently used and in line with client’s approved glossaries. Any terms that should not be translated need to be left in the source language.

As machine translation engines improve over time due to advances in technology and linguists become more and more proficient in post-editing, we will start seeing a considerable boost in the way translations are provided.

Written by Kasia Kosmaczewska
Kasia Kosmaczewska
Kasia Kosmaczewska is Machine Translation Programme Manager at TranslateMedia. She has extensive linguistic experience and a keen interest in machine learning. She spends her free time reading about socio-politics, practicing pilates and travelling.

Related posts