Risques associés à la traduction automatique (en anglais)
Not all what glitters is gold. And this holds true for machine translation.
Machine translation (MT) is a great technological advancement, which can make translating much easier for translators and any other people alike. However, it does come with some risks in accuracy and data protection.
As a form of artificial intelligence, Deep Learning (also known as DeepL) and neural machine translation have revolutionised the world of translation. In a matter of seconds, these systems can produce a human-like translation, with just a few errors. After correcting these mistakes, also known as post-editing, the translation is completely fluent. This makes for a fast and efficient way of translating. Even commonly used online applications employ these methods and are available to the public. Google Translate, for example, is used by millions every day. But there are some pitfalls to using this technology as well.
Which are the most common problems occurring with MT?
First of all, the translations that these tools provide often contain mistakes. Small grammar errors may not seem like a big deal, but they can really diminish the fluency of a text. This can come across as unprofessional when these tools are used in a work setting, for example. Neural machine translation tools are also not yet equipped to translate anything that is figurative, or non-literal. Humour, sarcasm, metaphors and some fixed expressions do not literally translate, which can cause misunderstandings. Translating long, complicated sentences with uncommon grammar structures may also result in incorrect translations.
Using these tools can thus be a great help, but the translations should always be checked for mistakes. Hiring a professional post-editor is a great solution for this, as they are trained to spot small mistakes. Another option is to run the translated text through a computer evaluation system, which can accurately grade it. Still, even though this method is more accessible, the system may not catch each mistake.
Which text types are not suitable for MT?
Precision and accuracy are factors that are important in translations of any kind of text, regardless of the technology used (or not used). When it comes to machine translation, however, the type of text should also be taken into account before realising a translation using this technology. Some texts are better suited to machine translation than others.
Especially creative texts are difficult to translate with machine translation technology. Literature and poetry contain creativity and style that is difficult to translate, and machine translation systems do not pick up on this. Even if the machine translation is correct, it often does not contain the same stylistic elements, text register or tonality found in the original texts. Marketing texts, for example, are also a difficult task for machine translation systems. They cannot convey the same message as well as a human translator could. For such texts, human translators are necessary, as machine translations are far from being able to translate them well at the present time. Qualified and skilled human translators are still preferred for translations of any kind of creative text.
Is data confidentiality guaranteed with MT?
Secondly, a big problem with online public translation tools is that they are often not secure when it comes to protecting your texts. When using these tools, it is important to remember not to translate classified or sensitive texts, or any texts containing personal data. The input of this kind of data into an online tool means that it is not protected and could be found by third parties. There is a risk of breach of data confidentiality, not only for the user, but also for whatever (or whoever) is mentioned in the translated texts. On the other hand, public tools can even claim any text that is translated online as their own, making it very unsafe for translating anything but a few words or sentences without confidential information. Texts like official contracts or client information should never be translated with a public tool, though. It is important to keep in mind the security of your own data, as well as anyone else’s with regard to the generally applicable data protection provisions, when working on translations or any task that requires some translation in general.
How can the gaps of MT be closed in the translation process?
When machine translations are not optimal, post-editing is a great solution to correct any mistakes. Implementing MT systems represents a major investment. It is therefore wise to make sure that the text is suited to be processed with machine translation in the first place. Large portions of recurring content and a critical mass of text, as well as short turnaround times are a prerequisite. As mentioned before, simple and objective texts (such as technical texts) are a great fit, whereas creative texts are not ideal.
A different way of using the new technology can be to use translation memories to some extent, in combination with traditional human translating or machine translation. This hybrid solution can provide an easy start to getting acquainted with machine translation, but still gives the translator some oversight in the process.
All things considered, translating a few words or simple sentences that do not contain any sensitive information are likely to be translated accurately by an online translation tool, and do not form a security risk. For larger texts on specific topics, especially those containing classified data, using a professional (offline) tool is a better choice. In combination with post-editing, this yields highly accurate translations. In general, when using neural machine translation tools, one should be aware of the potential risks, as well as of the types of texts suited to the technology.
- FIT. (Mai 2021). FIT Position Paper on Post-Editing. The voice of associations of translators, terminologists and interpreters around the world. In: Féderation Internationale des Traducteurs, https://www.fit-ift.org/publications/papers/#:~:text=FIT%20Position%20Paper%20on%20Post-Editing%20FIT%2C%20as%20the,Paper%20on%20the%20Role%20of%20FIT%20May%202020 [zit. 06.01.2022].
Pour en savoir plus sur nos services de post-édition, n’hésitez pas à nous contacter.