You’d like to share an interesting article on your page to increase impressions and engagement, but unfortunately, it’s in French and most of your followers are English speakers?
Or perhaps your boss has asked you to write a document in French for a presentation that will take place the next day with important colleagues, but unfortunately, your skills in the language of Molière are, to say the least… improvable? What should you do?! No problem, let’s ask ChatGPT, Claude, or Gemini, you’ll probably say.
Let’s take a closer look at the blind spots of AI in translation.
A Brief History of AI and the Rise of Machine Translation
Change Over Time
As we know, the somewhat stereotypical image of the translator that we used to have—buried under piles of dictionaries and scouring libraries in search of the “right word,”—has faded somewhat over time. Today, with the rise of generative AI robots such as ChatGPT and its counterparts, one might imagine that a single click is all it takes to obtain an almost perfect result, regardless of the type of text and language.
AI in Translation: A Reality That Isn’t So New
Artificial intelligence (commonly known as AI) has been used for a long time in the field of translation, particularly through machine translation tools. Over the past 10 years or so, with the development of neural machine translation (NMT) engines, we have seen the task of translation gradually shift to post-editing, i.e., the refinement or revision of a translation performed by a “machine.”
Technology is evolving at breakneck speed, and the tools we used 10 or even 5 years ago are no longer the same. It is also true that those available today can produce particularly impressive results, especially with tools like ChatGPT. In general, the translated sentences will be fairly well constructed and vivid, at least at first glance, with no apparent spelling or grammar errors (unless there is a bug).
Nevertheless, in most cases, it does not take long to notice that the translations contain errors, ranging from minor to substantial.
AI’s Blind Spots in Translation and the Importance of Human Intelligence
Tools such as ChatGPT, Gemini, and Claude were not designed specifically for translation, but rather to understand and generate text. They use probabilities to predict the logical sequence and thus suggest a result that gives the impression of a harmonious statement. They offer no guarantee as to the reliability or consistency of the translations or texts provided.
Grammatical Agreement and Context: When Their Relevance Takes on Its Full Meaning
Among the most common mistakes are grammatical agreement. The famous “they,” which refers to a neutral or unknown gender in English, will often be translated as “ils” in French, even though it is clear in the preceding sentences, for example, that we are talking about a single person or that the referent is feminine. The same applies to figures of speech or expressions specific to a particular field that AI does not understand: it then becomes easy to end up with an incomprehensible or awkward text.
These small details may seem trivial for an everyday text or one that has little visibility. However, in a professional, artistic, or technical context, such errors could have serious consequences for the accuracy, clarity, or credibility of the message.
Calques, Anglicisms, and Other False Friends
Let us also remember the rather clumsy translations of a series of Telus Fund advertisements a few years ago, one of which read in French: “Prenez une profonde respiration, broyez-vous. Va le tuer.” (Literal translation: Take a deep breath, put yourself in a blender. Go kill him.) The translation (and proofreading!) was probably not done by professional translators. The original text stated the following: “Take a deep breath, ground yourself. Go kill it.” Such an abomination not only compromises the clarity of the message, but also significantly damages the company’s brand image and reputation.
Unsupervised machine translation also comes with its share of calques, false friends and anglicisms. For example, AI tools will sometimes translate “actually” by “actuellement” (which means “right now”) instead of “en fait.” You might also notice a mismatch in tone, like using “vous” (formal) instead of “tu” (informal), or vice versa.
Ultimately, accepting such discrepancies, however small, will end up normalizing the dumbing down of digital writing. You just have to browse the unintelligible websites of several organizations and companies—some of which are even Canadian—to see the extent of the damage being done.
For more examples, see our post on web expressions [a].
Consistency: More Than Just a Matter of Style
There is also the issue of consistency. In a long report where the same key terms appear several times, or in a legal text where terminology is crucial, for example, we still see many errors generated by AI. This reality likely stems from the probabilistic model with which artificial intelligence works (we may return to this in a future post).
Thus, without human intervention, the English term “agreement” could be translated as “accord,” “entente,” “convention,” or “contract.” And all this wonderful mix of terminology can be found in a single text!
That is why it will be a long time before we can do without the human eye in professional and wide-ranging writing. At least, that is our hope for companies that do business in French; otherwise their brand image could be seriously damaged.
The Future of Translation
Partnership or Rivalry?
In professional translation, the idea is not to completely reject tools or be resistant to them. Such an unwelcoming attitude can only harm the profession, isolate us from the rest of the world, and weaken our position in a reality where content creation is booming.
Generative AI is useful because it can provide valuable assistance in organizing our ideas effectively and suggesting more concise formulations when dealing with character or space limitations, for example.
Technologies like this can also help us navigate specialized topics and complex references, thus serving as a guide. Let’s not forget: the translation profession often involves becoming an expert in a wide range of subjects in a very short time!
We always have to work faster and faster, so in order to maintain quality, it is essential to focus on efficiency to increase productivity.
The Transformation of the Profession and our Place in the Digital Landscape
With the tools available to us today, the challenge for translators (and editors) is no longer detecting errors in agreement or syntax, but ensuring accuracy, consistency, and uniformity. In addition to machine translation engines, professionals work with structured translation memories and terminology databases (or glossaries) customized for each client.
But beyond the tools, these language professionals also draw on their sense of taste, their culture, their sensitivity, and their human ability to detect the subtleties of speech. This makes them, and will always make them, valuable allies.
Continue exploring the topic by consulting our guide to choosing a translation and editing agency [b].
Who or What Should You Turn to?
So, what about “asking ChatGPT” for our everyday translation needs? As mentioned above, tools like this are very useful for occasional troubleshooting or assisting with professional translation or revision work. Obviously, using an agency where the quality of work is guaranteed by true professionals with experience and qualifications remains the safest solution.
The team at Maxime Collins Inc. is at your disposal for all your language needs, whether it be translation, revision, or proofreading. We provide a personalized, turnkey service; and, above all, a human one. Unlike machines, we don’t forget you once the task is complete. We will always be available to answer your questions and provide thorough follow-up.
Conclusion
Artificial intelligence has profoundly transformed the world of translation by offering speed and accessibility. Despite impressive progress, AI remains limited when it comes to grasping cultural nuances, tone, intent, or subtleties specific to a particular context. The blind spots of AI in translation (linguistic ambiguities, cultural references, puns, or implied meanings) remind us of an essential truth: language is alive, human, and deeply contextual.
That is why the human eye remains, and will always remain, indispensable. Translators do more than just transpose words; they interpret a message, adapt it to a target audience, and validate its impact. The future of translation therefore does not lie in opposition between humans and machines, but in intelligent collaboration, where technology serves human work without ever replacing it.
N.B. This post was NOT generated by artificial intelligence! 😉
FAQ
Can AI completely replace human translators?
No. AI excels at simple, repetitive translations, but it struggles to grasp nuance, communicative intent, and cultural subtleties. Human judgment remains crucial to ensuring accurate and natural translations.
What are the main blind spots of AI in translation?
Among the most common blind spots of AI in translation are interpreting context, humour, irony, idiomatic expressions, cultural references, and variations in tone depending on the target audience.
In what cases is human intervention essential?
High-reach marketing content, as well as legal, literary, medical, or institutional materials—where accuracy, cultural sensitivity, and stakes are high—must be reviewed by a human professional.
Is AI useful in professional translation?
Yes. Artificial intelligence is an excellent translation tool, as it can speed up processes, suggest initial or more concise versions, provide information on certain specialized topics, and improve productivity, provided it is used under the supervision of a professional.
What is the best model for high-quality translation?
To achieve maximum efficiency, the best bet is the hybrid model: a combination of the power of AI and human expertise. This strategy allows us to take advantage of rapid technological advances while ensuring optimal linguistic and cultural quality.