Post-editing: What is it used for beyond verifying text accuracy and reliability?

We analyse in which cases machine translation requires post editing and quality evaluation by professional translators and what is the use of it.


GearTranslations Internationalization Content marketing Translation SMBs


In a hyper-connected world, where electronic devices can easily connect with any network or application, the English language has been considered “the business language” for decades. Moreover, it is the official language in more than 20 countries.

However, limiting company’s products or service offering to one single language, in order to reduce translation and post-editing costs in an specific domain (technical vocabulary), may hinder company’s opportunities in new markets.

One option commonly used by corporations to minimise costs is machine translation, which drastically reduces the investment needed for this service. This would be further more efficient if human post-edition wasn’t necessary.

Next, we analyse where the industry is headed in this regard.

Current model: Neither human nor robot

The milestone of fully automated translation was expected to be achieved years ago. We can certainly say there has been significant progress.

Automated translation tools known to us nowadays have evolved from word by word translation, to concept translation. The process of getting one concept translated into the same one in another language can be executed in an automated fashion, but sometimes results do not fully consider context, specialty area, style or other language elements used by people and companies, which are an essential part of communication.

Technological development –such as CAT tools, or the first versions of machine translation technology– has largely facilitated the translation process. However, in specialty areas it’s still necessary that professional translators evaluate and adjust the accuracy of machine outputs.

These types of tools can fairly well translate vocabulary and phrases associated with general topics. But there’s still no reliable machine translation system that doesn’t require some kind of editing to be fully accurate. The need for post-editing is stronger when dealing with domain specific translations, that is, when the content to be translated requires the use of specific vocabulary from a specific sector. The terms and phrases needed are very infrequent outside that particular field, which might result in the machine not knowing what meaning and translation is the most adequate.

This happens not only to translation engines, but also with humans, since a native speaker without knowledge of the specific area wouldn’t know its specific vocabulary. However, it would be possible for the person to learn it, studying the area, and relying on the help of more experienced professionals. But, how does the machine learn what it ignores?

Feedback is always key

Current translation tools are mostly based on hybrid tools, meaning that on top of the automation there is a layer of human post-edition which guarantees translations accuracy and reliability.

Research in this field continues, and work is already being done on the next generation of automated translation tools that will train machines to translate the vocabulary of a specific area (domain specific translations).

In order for a machine to learn and execute an automated activity, large amounts of information and feedback are needed. Simply put, it needs to study and learn what it’s doing wrong to be able to improve it.

Significant breakthroughs have recently taken place both in natural language processing and sentiment analysis. The new generation of neural networks requires multilingual data sets of considerable size to notably improve the accuracy of translations.

This multilingual data consists in former accurate translations and terminology bases used to better train the decision trees that determine all the possible translations that a word or phrase might have.

However, it is also possible to work with assisted machine learning models, where the machine makes decisions that are later evaluated and graded by a person on how suitable the outcome is. In this way, the machine is capable of learning and adjusting its results. It is here that human post-editors play a key role: Registering their corrections enables the machine to learn what it shouldn’t do.

To benefit from the use of this technology, it is key to structure content in a way that’s understandable by a machine. Corporate information has to be machine-readable.

Gear Translations platform can help you structure your multilingual assets to build an automated translation model that achieves savings and improved operational speed.

Would you like to know more about how to turn your translations into valuable corporate data? Contact us