Will machines replace language translators? Blueprint's Dutch translator, Yasmine Van Pee, ponders the effects of artificial intelligence and automated tools on her field.
As a budding Trekkie, I was absolutely mesmerized by Lt. Uhura, Star Trek’s iconic exolinguist. The epitome of intergalactic glamour, she was like catnip to my pre-teen sensibilities: zipping from adventure to adventure on an impossibly cool vessel surrounded by all kinds of dashing brainiacs. And, of course, saving the galaxy with a few choice phrases of quickdraw Klingon.
Uhura has been on my mind lately, though this time for somewhat less leisurely reasons. Since I now work as a Dutch translator and given the huge strides made in natural language understanding (NLU) in the past decade, I wondered: What would be the job prospects for an aspiring linguist in the year 2266? Will the galaxy still need human translators, or will machines soon replace us? And what would happen to language in the age of the Universal Translator?
While not yet as advanced as the Universal Translator from Trek lore, a neat device that could translate languages by decoding brainwaves, it’s undeniable that machine translation has made immense progress over the last decade. It has also changed what it means to work as a translator. Modern translation is now inherently machine-aided ― our daily practice would be unthinkable without dynamic translation memories and other forms of automation.
For most translators today, this simply means liberation from the more repetitive aspects of translation, so we can devote more time to solving the daily conundrums that go hand in hand with producing accurate, culturally sensitive and readable prose. For example, how to cut the clutter out of bloated sentences, how to render the cultural richness of a word like “cakewalk” for your average Dutch speaker or how to translate one of those mind-bending German neologisms like Verschlimmbessern, which means to make things worse by trying very very hard to make them better.
For language professionals working in machine translation post-editing (MTPE), the age of artificial intelligence increasingly means acting as an editor rather than a translator. (The mere existence of the acronym should point to the ubiquity of the practice). There, translators are the human-in-the-loop in a largely machine-driven system, correcting and polishing AI-generated translations.
It remains to be seen what broader effects this increased automation will have on language diversity and perhaps even on the very structure of our various mother tongues. For non-English speakers, more and more content we read online is in fact translated content. In this digital age, with its proliferation of online text, translations make up a much more significant portion of the content we consume every day than they ever have before. Take for instance social media platforms, streaming services, online retailers and tech giants like Google, Apple, Netflix and Amazon. They are an integral part of our daily life, the interfaces of their products as familiar to us as a slice of bread or the voice of an old friend.
However, when we as non-English speakers use them in our own language, most of the text we see in these products is not written for us de novo (as original content) in our mother tongue, but rather translated from an English-language source. Since languages are constantly evolving, will the structure of these translations have an impact on real living languages and how they are used and spoken? And more importantly, could the everyday machine-aided tools we use as translators in tech perhaps even push our mother tongues in new directions?
It’s not inconceivable, for instance, that CAT tools ― computer-assisted translation software indispensable for translators today ― could easily lead to translations more tightly source-oriented than before, by which we mean translations that copy the structure of the original overly literally. In these tools, texts are segmented and subsequently translated segment by segment, leaving translators less room to maneuver and restructure sentences and passages as needed. If we return for a moment to those ubiquitous tech products we interact with every day it’s noteworthy that this source language is overwhelmingly English.
It leaves one to wonder: Is real-world Dutch or Thai or French ever so slowly inching toward English? Not so much by adopting the odd English phrase here and there ― borrowed words are inevitable and as old as language itself ― but by more closely mirroring English in their very structure?
Even a tool as pedestrian as a translation memory could potentially lead to a shift in the vocabulary of a language. If not used with skill and sensitivity, these dynamic databases of existing translations made to keep new translations consistent both within one text and across texts, can “disappear” less commonly used words. Not so much intentionally but purely by rote. If there are, for instance, four different ways to translate the word “amazing” in a certain language, using a translation memory tends to make translations curdle around just one of those four, which will be used consistently while the others get filtered out and benched.
A recent very visible linguistic error in the Dutch version of MS Word ― no doubt now populating the Dutch-language translation memory over at Microsoft, hence prone to remain in use ― left me pondering whether it would, by the sheer force of being endlessly reproduced in one of the most-used products on the planet, be folded into living Dutch and end up replacing the original correct idiomatic expression. Will “opslaan in een pc” (to save something in a PC) soon supplant the correct “opslaan op een pc” (to save something on a PC)? And would that be bad or just different? More alarmingly, would I soon sound as antique as a 1930s talkie if I continue using the correct phrase?
Kidding aside, all the automated tools we routinely use as translators can lead to unintended side effects with the potential to inflect the very languages we speak. They can cause unnoticed errors to multiply and become runaway errors with a life entirely of their own. They can increase the gravitational pull of a source language and lead to translations that follow the structure and wording of the original text too closely, like planets tidally locked to their sun. On Blueprint’s localization team, we are keenly aware of the potential pitfalls of our translation tools, and we take pains to produce texts that read as if they were originally dreamt up in their target language. The gold standard for any translation, after all, is one that effaces itself ― one that reads not as translated text, but simply as beautifully crafted text.
To return to our opening question: Will AI soon replace us? For the moment it seems unlikely, but maybe one day it will. In that light, perhaps a more interesting question might be what happens to language in translation now, in the interim, in this moment of machine-aided translation. And how we as translators can take stock of the potential side effects of the indispensable tools we use. Because you never know… maybe, just maybe, in a distant future and lightyears away, you might find yourself having to tinker with a Universal Translator, zipping along at warp speed to your next great adventure, and you better be prepared.
At Blueprint, our in-house localization team is made up of tech-savvy translators who go above and beyond to deliver fresh and vibrant translations. Do you too love thinking about language? Are you as excited as we are about the intersections of language and technology? The Blueprint localization team is hiring. Consider joining our ever-growing team of (inter)stellar linguists and check out the open positions on our Careers page.