Machine translation – a trojan horse for our times?

Machine in close-up

Why should I pay a translation agency when I can use Google Translate? Many companies ask themselves this question. Ultimately, machine translations can be obtained free online. Here, we will look at the reasons why it still pays to opt for a real human translator, even in the 21st century.

Most of us have probably used a translation tool at some point. Usually, all you need to do is copy the text into the field provided, select the language and click ‘Translate’. Easy, fast and often free of charge.

What could make more sense then saving money on a translator when translating handbooks, apps or websites? After all, machine translations are significantly better today than they were a few years ago. But there is a catch – or several to be more precise.

What are machine translations? – Overview of different methods

Essentially, a machine translation is when a text is translated by a machine, e.g. a computer, instead of a person. The best known translation services include online translators like Google Translate and DeepL.

Machine translations are older than many people realise, first attracting attention in the late 1940s. However, it’s only recently that they have come to the forefront again. This is mainly due to advancements in technology.

The key methods in the area of machine translation are as follows:

Rule-based translation

For rule-based translation programs, grammatical rules need to be defined for each language and relationships between language pairs need to be established. This requires a lot of effort and the results are generally underwhelming.

Statistical translation

Statistical translation programs require large volumes of text in different languages. When translating, the tool searches for similarities and uses them to generate the target text. Any lack of matches will be reflected in the quality of the final text.

Neural translation

Neural translation programs work using artificial intelligence. This means that they learn to translate on their own using neural networks and can therefore optimise themselves. The popular translation tool DeepL works in this way. The quality of neural translations is significantly higher than the rule-based or statistical machine translations.

Good, but not perfect – typical mistakes in machine translations

At first glance, many neural translations look perfect. Less so upon closer inspection.

This is due to some basic weaknesses of machine translations:

  • Machine translation tools often fail to pick up on negations.

Example (EN>DE)

Original: Do not click on “I run ads for someone else” unless you are an agency and you own the payments profile for that account

Wrong machine translation: Klicken Sie nur dann auf “Ich schalte Anzeigen für eine andere Person”, wenn Sie keine Agentur und Inhaber des Zahlungsprofils für dieses Konto sind.

Correct (eine, not keine): Klicken Sie nur dann auf “Ich schalte Anzeigen für eine andere Person”, wenn Sie eine Agentur und Inhaber des Zahlungsprofils für dieses Konto sind.

  • Online translation tools struggle with humour and metaphors.
  • The same applies for homonyms, i.e. words having several meanings.

Example (EN>DE)

Depending on the context, the English translation for the German term ‘Übersetzung’ could be ‘translation’ (for languages) or ‘transmission’ (for forces, e.g. in a gearbox).

  • Specialist translations require a specific, standardised specialist terminology. The limitations of machine translations are also evident here. All it takes is for a word to be translated a certain way one time and a different way the next time round. Specialised dictionaries can only be integrated into professional systems.
  • Another source of errors for machine translations is when sentences refer to previous sentences. This also highlights that translation machines are still machines.

Important: Many people are unaware that many providers of free translation tools use both your source and target texts to improve the quality of their services. If you’re not careful, this might mean that your new product is suddenly not quite as secret as it was if you have used Google Translate. Only paid pro versions offer genuine confidentiality.

When it makes sense to use machine translation and when it doesn’t

This question can be answered as follows:

  • Machine translations are useful if the result doesn’t need to be error-free and is only intended for personal use. Google translate is fine if you just want to quickly translate an email or an extract from a book for yourself.
  • In a professional context, human translators are a better option. Only they can take into account the relevant specialist terminology, the different styles of text, references to other sections of text and words with multiple meanings – basically everything beyond the meaning of individual words. This is only way to be sure that everything is correct in important contractual texts, on your website or in operating instructions.
Remember: Every word counts, particularly when dealing with safety-relevant or legally binding content. That said, an incorrectly translated formulation can also have serious consequences in a marketing text, not least serious damage to a company’s image.

There is also a third option available: You can use an advanced machine translation tool for the first stage and then have the text revised by a professional translator. This is known as ‘post editing’. Whether or not this method is a good choice and what benefits it can provide depends on the individual case.

Looking for experienced translators for post editing or flawless technical translations?

If so, feel free to contact us now to discuss your requirements.