DeepL is the most accurate translator. That’s its tagline and promise. We could even say that Deepl stole the limelight from us human translators.
Many companies wanted to try DeepL, Chat GPT, etc. to translate their content. Some of them just do with it as they only need to make the gross meaning understandable to people speaking a different language. But other companies who really have a clear selling strategy and understand the power of a compelling message decided to come back to human translation.
In this article, I’ll tell you about my clients, of tests they made with our help, freelance translators. We’ll study the different problems and challenges so as the results and reasons of coming back to human translation.
DeepL translations are inefficient for specialized content
One of my clients who sends me digital marketing content to translate into French, announced that they wanted to test pretranslation of blogs with DeepL. My role as a linguist would evolve and I would have to correct and edit the pretranslation. Then, the proofreader would take his/her role and check the text afterwards.
Since this client is a regular one, I accept the mission.
Unfortunately, results are bad, and in the end, I spend more time editing the content than if I had translated it myself. Here is why.
This client operates in a niche, hospitality tech, and especially hotel tech. Consequently, the vocabulary is specific. DeepL does not know this specialism. So it does translate « guests » by « people invited », invited suggesting that these people did not pay for their stay, which actually is a mistranslation.. DeepL also translates PMS by « premenstrual syndrome » although in hotel tech, a PMS is an operating system, a property management system. Translating PMS into premenstrual syndrome is a huge error as it’s totally irrelevant to the topic. Such errors could impact brand image, reputation and trust.
As a professional and skilled translator, I have to be very careful and correct these occurrences and not to forget any of them. This work is additional to a typical translation. Indeed, as I know my client context and specialism, I already know the accurate translations for the above terminology. If I had translated directly myself, I would not have to chase and track these repeated errors. This work is more exhausting on the cognitive plan. If I translate directly, it’s more efficient and my effort makes less effort. It’s a real time saver, opposite to what you would think at first, that machine translation will be faster.
Marketing and idioms
“Répandre le bouche-à-oreille” is another example of a translation suggested by DeepL
or « Sweetheart special » in the context of a special offer for a CRM platform translated by « spécial cœur de jeune fille » “dedicated to young girls heart” in an emailing campaign. L’IA would then be sexist ? Or Black Friday literally translated « vendredi noir ».
On the opposite, these errors have a huge cost and they don’t inspire trust.
Here again, machine translation is generic and does not take the context into account. In specialism, AI is inefficient and synonym of time loss and consequently loss of money.
A specialized translator knows the terminology associated to his/her field of expertise. The linguist won’t then do the error and will immediately write the adequate term. It is consequently useless to use pretranslation step which would be counterproductive. Entrusting your projects to a professional translator represents on many aspects a smart investment.
DeepL and Machine translation biases
Back to the tests done for the hotel tech company, but it could be applied to machine translation in general.
Machine translation creates an availability and anchoring bias. What are these biases ? Availability bias refers to preferring and overestimating information immediately available to our memory. It’s the same for a pretranslated text displayed to you. It’s more difficult for the human brain to detach, to take some distance with what it’s right under your eyes. It demands extra effort to the brain. On the opposite, if there is no suggestion right from the start, the brain is not biased et can immediately find a better language solution.
Anchoring bias
Anchoring bias refers to the same concept as a whole. MailChimp provides a definition.
Anchoring bias, sometimes called departure point bias describes our tendency to rely too much on the first information we receive. When we take a decision, our first point of reference often acts as an anchor. It’s more or less the same as what happens during a negotiation. The rate given at first is an anchor.
Proofreaders report unpredictability, unreliability and increased attention when they proofread a post-edited text (correction of a machine translation), even if a translator already worked on the text. Anchoring and availability bias really influence the translation and have an impact on the cost of a post-edition service as it increases the time spent on the project.
Entrust your translations to skilled and experienced professionals and not to DeepL
As we don’t just take your words to transpose them into another language. We take the time to listen to and to understand your message and intent before creating the adequate message in our native language.
You now know three good reasons to be skeptical about DeepL and machine translation in general for your marketing content. After these tests, and in most experiences, linguists and clients realize that in niche or specialised fields, this technology produces unsatisfying output. Worse, it generates more work and consequently more costs. In marketing, translation suggestions of machine translation are too generic. It does not convey the adequate tone nor is able to respect a style guide or a tone.
Using it is tempting, but to correct a machine translation output, ie to post-edit, is an exhausting task, counterproductive because of the mentioned bias. Even proofreaders and reviewers report it. A specialised translator will help you save time as she/he will be able to find the adequate terminology immediately, and will be more meticulous in producing a excellent result.
That’s why many companies come back to human adaptation, which performs better when you do B2B and talk to skilled professionals.
What about you? What will you choose? After reading this article, does it confirm your trust in human linguists ? Have you decided to reconsider collaborating with a human translator rather than to use AI? I’m all ears and available for a chat with you to talk about it.
Français