Three decades ago, when I started working as a professional translator, I sent documents to clients by post . . . and then waited days or weeks to find out their approval, questions, or additional needs. Now, email makes sending my work a question of seconds at most, and if a client doesn’t get back quickly, I know more is going on than something being lost in the mail.

This is just one of a myriad of ways translation has dramatically changed over the past few decades.

We use technology not just to communicate on projects but also to assist in the work, and it’s become our competition. But while the benefits of technology in the translation world are undeniable in certain ways, I’m not worried that “computers are going to take our jobs” any time soon.

One of the most noticeable ways technology has become a strong presence in translation is the way computer-assisted translation (CAT) tools have become an increasingly common part of most translators’ work routine.

In a 2019 ProZ.com survey of professional translators around the world, 88% reported using at least one CAT tool for some, if not all, of their translation work. At 83%, most use a CAT tool for all of their translation work, and 76% use more than one CAT tool.

Nothing is perfect, not even CAT tools; still, the same survey shows that an impressive 93% of translators feel that CAT tools make translation more efficient.

Not only are a majority of translators using CAT tools; more than 50% of clients require them.

Other types of translation tech — neural machine translation NMT and AI — have become de rigueur elements of online and in-person communication. Likewise, translation bots like Google Translate are readily available to the general public and have been used for everything from on-the-spot translations of a foreign-language website to helping with on-site communication when people travel abroad.

AI and neural machine translation (NMT) work by essentially learning from whatever resources are put into them. This could be anything from dictionaries, to entire bodies of literature, to online interactions and conversations, allowing for insta-translations based on algorithms and averages.

Their convenience and low price (or even no price — after all, apps and software like Google Translate are free) have made machine translators wildly successful. Sources vary on the current worth of the machine translation (MT) market, with figures and estimates starting around $200 million and soaring to over $500 million. But one thing every source will tell you is that the market is continuing to grow at an impressive rate.

Unfortunately, that doesn’t mean that NMT and AI provide accurate translations. Most consumers are probably even aware of Google Translate “fail” memes and the like. But in terms of convenience, machine translators are pretty much unbeatable.

Similarly, even clients for whom accurate translation is a “must” are tempted by automation. Human translators have to take time off, could call in sick, and just take longer to get the job done — all of which means time and money spent that wouldn’t be an issue with machines.

When you look at all of this evidence, it seems like translation might be headed to total automation territory. Some tech might become obsolete, but so will the entire concept of human translators.

Fortunately for us human translators, this worst-case scenario is highly unlikely to come true.

Translation bots may be quicker, cheaper, and more easily available to the general public and clients alike. But as anyone who’s worked closely with AI and translation tools knows, they’re far from error-proof. In addition to meme-able mistakes, several issues make machines unable to accurately translate certain material without human help or verification. And in some cases, machines can’t be relied on to provide a translation at all.

For instance, it’s a well-known fact in the translation community that MT can’t be used for literary translation. While not every translation is literary, the idea behind why machines can’t translate creative writing offers important insights into why they aren’t 100% accurate in general.

Literary language isn’t straightforward. Authors often use figurative language, idioms, ambiguity, sarcasm, humor, and expressions specific to their local community. These are exactly the sorts of things that machines just can’t handle. It may be one thing to translate a relatively straightforward document, such as an invoice, statistics, or automatically generated reports full of standard phrases. But understanding nuanced language is beyond a machine’s capability.

This idea has been proven again and again. For instance, a 2017 joint project between Sejong Cyber University and the International Interpretation and Translation Association of Korea pitted translation bots against human translators. When it came to speed, the machines were the winners, completing their work in a few minutes, while human translators clocked in at nearly an hour. But when the translations were checked for accuracy, the machines scored far lower than their human rivals: just 15-28 points out of 60. The humans, by comparison, averaged 49 points out of 60.

Texts full of complex language aren’t the only thing that causes problems for translation bots. In addition to difficulties understanding the subtleties of language, machines can also have difficulty with grammatical structures that have unclear rules or an ambiguous meaning.

For instance, in an article on Naver Labs Europe’s blog, scientist Marc Dymetman explains that AI had trouble determining which gendered pronoun belonged to which subject in a statement in French. While a reader would understand, for example, that sa condemnation referred to a criminal’s conviction, AI might think the pronoun, which is determined by the gender of condemnation, referred to the other person in the text, the police officer.

Machines also have a limited knowledge of the world’s languages, although you could say the same for human translators. After all, each of us is limited by the amount of languages we can learn and become fluent in. But it turns out that while you could find a human speaker of any living language on earth, there are many languages that machines lack sufficient knowledge of. In fact, make that “most” languages.

In a 2021 BBC article entitled The languages that defy auto-translate, journalist Sophie Hardach reports that only about 100 of the world’s 7,000 living languages can be machine translated.

If there’s little or no printed material in a language, AI can’t produce algorithms that would allow translation. Since 4,000 of the world’s languages aren’t written at all, this may very well mean that there will never be machine translation for them. As for the other 2,900 or so, it will depend on whether enough documentation exists and/or can be uploaded for MT technology to analyze.

So then, maybe we could say that there’s a risk that the most commonly spoken and written languages in the world could be entirely machine translated, even if low-resource languages would not. But even the most common languages, like English, Spanish, and Mandarin, would still be missing something if they were entirely machine translated: transcreation.

As this portmanteau word suggests, transcreation is a combination of translation and creation — that is, the creation of content that can be effectively understood by a different culture, as well as specific local expressions and references. For instance, if characters on a German television show use an idiomatic expression, the subtitles translator should find an equivalent expression in the target language, instead of just translating it word-for-word. That’s what a machine would naturally do.

Although the word “transcreation” isn’t familiar to the general public, it’s extremely important. A number of studies and surveys conducted over the past decade reveal that consumers connect better to advertisements, websites, and apps that are translated and localized into their native language. One of the most recent of these, a survey conducted by CSA Research from November 2019-January 2020, was even entitled Can’t Read, Won’t Buy. The survey found, among other things, that close to half of internet users won’t buy a product or stay on a webpage if it’s not in their native language.

In some fields, money isn’t the only concern, and certainly not the most important. My company, aiaTranslations, specializes in healthcare and pharma translation. Without proper transcreation, documents, information, and instructions related to vital issues like patient consent and education can be misunderstood or disregarded.

Not all human translators think of transcreation, unfortunately, but many do. On the other hand, no machine is capable of knowing a culture and the subtleties of a language on that level.

These are all reasons why, no matter how advanced translation technology gets, human translators will always be needed.

Just like one of my inner-office mottos, translation itself is “humans first.”

Tech like CAT tools, AI, and NMT, not to mention good old email, has made translation easier in many ways. And that’s what it should continue to do. Translators can rely on tech to help with easily automated tasks and quick references to a translation memory, while they focus on the more important aspect of translation: the subtleties that abound in the human experience.