Has the age of machine translations arrived?

Machine translations have been widely used for many years. However, they’ve entered into broader consciousness in recent years, especially thanks to the improved quality of the translated text. In fact, they have shifted into the next developmental stage – the earlier statistic machine translations (SMT) models are being replaced by neural machine translations (NMT). Simply put, machine translators have started to learn. And not only from data that we provide them but also from their own mistakes. But these apparent tiny steps mean giant steps for the final quality of the translated text. It still holds true that machine translations will not replace translations done by people. They are not the death knell for translators and translation agencies! Rather working with translations is changing or, more accurately, the services associated with translations.

A variety of machine translators

For clients interested in machine translation, the translation agency Aspena uses two main machine translators – DeepL and Google Cloud Translate. Each is suitable in a different way – DeepL, unlike Google, offers a limited number of languages and is slower, but it often produces a higher-quality final translation. Other translators used by Aspena include KantanMT, Omniscien, Amazon MT and Tilde MT.

Here we arrive at the fundamental question – when and how to use machine translation?

As stated earlier, translations by people cannot be replaced by machine translation. However, there are cases when using a machine translation may be suitable – for example, for internal instruction of texts in another language or for testing the multilingual environment in writing software or apps.

Let’s consider a simple translation. This is defined as the conversion of a text from one natural language into another, while retaining the meaning. After being translated by a person, you will end up with a text that you will comprehend without any knowledge of the source language.

What kind of output does a machine translation produce?

In using machine translation, you will obtain a text that can be simply described as follows:

  • a large section will also be comprehensible to you (how large depends on the translator used, language combination, source text, the translator’s degree of training, etc.),
  • a section of the text will also be comprehensible to those who know the source language (they will guess what the machine “meant” when it translated the text),
  • a section of the text does not have to make any sense.

It wasn’t so long ago that machine translations produced incomprehensible (even nonsensical) sections in the order of tens of per cent of the whole text. This has rapidly changed. Primarily it is due to the use of neural networks and deep learning. However, it is ill-advised to rely on machine-translated documents delivering 100% comprehensibility.


Post-editing of the translation

Machine translations with further editing are not suitable for any publication. This is ensured in the next step involving a human translator, which is called post-editing (MTPE – Machine Translation Post-Editing).

Aspena offers clients two post-editing options:

  • Light Post-Editing – a check of the text by a human translator who aims to make only the most essential corrections to the text so that the result will be comprehensible as a whole for the end user – they are not concerned with style and don’t make changes dependent on taste. The intention is not to deliver a translation comparable to a human translation. This is mostly used for urgent internal translations.
  • Full Post-Editing – a more thorough edit of the text, aiming to attain a level of quality comparable to a human translation. As a result, it may mean that the translator may translate some sections again and disregard the machine translation.

Machine translations will not replace human translations, but the two can lead to a beneficial combination

From Aspena’s perspective, machine translation is only one service in the portfolio – it isn’t a replacement for traditional translation. It is crucial to realise that not all documents are right for machine translation. This especially means documents that contain creative texts, a specific style of expression (irony, jokes, poetic licence, etc.), whereas technical, template-based or standardised texts are suitable. It is worth pointing out that a high-quality translation may be made from a machine translation, even if the original text wasn’t suitable. It depends on the properly selected combination of machine and human translators.

Even machine translators have their limitations

An equally important thing to consider is whether the client intends to produce the machine translation alone or to assign it to a translation agency.

DeepL and Google provide translations by copying texts into the webpage. The resulting translation is then copied into a document. This can be laborious and time-consuming, and the formatting is not preserved. DeepL, unlike Google, offers translations of whole files (currently only MS Word and PowerPoint), but the free version is limited in the number and size of the documents (3 documents per month). Even in the case of the paid versions, the number and size of the documents is limited.

For clarity’s sake, it’s necessary to state that Google offers translation of simple documents incorporated into their web app Google Docs. DeepL also offers a desktop app for Windows and Mac. Its operation is easy, but it still has its limits.

And entrust your machine translation to the experts!

But what to do when you don’t want to be limited by the size and type of machine translation? When I need to translate webpages, lines of software code, database files or complicated graphic print material created in inDesign? And even more specifically – what do I need when I translate, for example, an e-shop that contains several million words?

Here Aspena as a translation agency steps onto the proverbial stage. For customers who require a machine translation, it is merely the first step of the standard workflow in the process in working at Aspena. Aspena isn’t limited by the size of the files, number of documents or source format.

Beside the standard formats such as MS Word, Excel, PowerPoint and the like, Aspena will deliver translations of these formats:

  • xml, resx, po, json, csv, files...
  • html and php files...
  • graphics created in inDesign, Illustrator, FrameMaker, AutoCAD...
  • text files in txt, inf, ini, reg...
  • and many more.

The significant advantage in a translation agency is the interconnection of machine translators and computer assisted translation (CAT) tools . This is a software environment for creating translations with the help of translation memory and terminological dictionaries that are tailored to each client. Be aware of the commonly mistaken belief about what CAT exactly is – it isn’t machine translation. It is an environment in which professional translators work.

Combining a CAT tool and machine translator is always more advantageous for Aspena’s clients than simple translation using a webpage, or DeepL Google or other engines’/translators’ apps. The translation memory and terminological dictionary will ensure improved consistency and more accurate translations.

I am “training” several translators at the translation agency on the basis of client data (mainly by using pre-made translations). DeepL doesn’t currently allow training, but Google and other tools do.

Be well informed of the range of options

It is very likely that any version of machine translation will include cooperation with a human translator – whether it is Light or Full Post-Editing or a combination of language, expert, and preprint proofreading. There are many possibilities and combinations. It always depends on what you expect and need from the translation.

Now all that remains is – to visit any of our branches, send us an email, fill in a request form or call us, and our experts will be happy to advise you on the selection of the most advantageous method to create translations, whether it is using machine translation or traditional translation by a person.


Our clients