Are you ready for the translation singularity? Machine translation and the role of the translator in 2030 and beyond

At this point, who hasn’t heard the story of the Tower of Babel? Mankind grows proud and builds a tower to reach the heavens. Divine retribution follows—the tower is destroyed, and one language splits into several thousand. Unable to understand one another, mankind scatters to all edges of the earth.

For us in the language industry, it makes for a compelling origin story, no? Not so much the death of a legendary ”universal language“, of course. But rather, the birth of many languages, and with them a diversity of cultures that expand the richness of human experience. The tower may lie in ruins, but humanity finds a way to transcend difference and continue to cooperate with each other. And that way is translation.

Of course, our origin story has its own hero—the humble translator. Translators connect people across the language divide, and play a critical role in fostering goodwill and meaningful exchange among people of different languages and cultures.

Translators have played this role since antiquity; even the most ancient literature known to us had already been translated over and over in its day. And through the ages, the translator has continued to play the central role in the story.

But for the moment, let’s turn our faces from the past. A new story is brewing over the horizon, one that poses a challenge to that longstanding role. I’m talking, of course, about machine translation. What does its rise mean for the language industry? Will we ever see a future where machine translation has perfected itself, and if so, what place does that leave for translators?

Let’s begin our journey for answers to these questions. This article will be an extended thought exercise on the future of machine translation, the role of the translator in that future, and the inevitability of what we would like to call the translation singularity.

What is the translation singularity?

When we talk about the future of the translation industry, the figure of machine translation inevitably pops up in the discussion. Within the past decade, machine translation has significantly transformed the industry landscape, and while it is far from perfect, research and development into this technology continues to make breakthroughs at an accelerating pace.

Under the singularity, artificial intelligence overtakes our own, and will change the world in ways that are beyond our wildest imagination.

The theme of continued technological improvement is a fundamental theme in science fiction. We build machines that can solve problems that are normally beyond the capability of man, and over time we get closer to building the perfect machine—a machine that not only solves these problems, but can continually improve itself without human input.

The creation of such a machine is a flashpoint in human history that sci-fi writers call the singularity. Vernon Vinge, one of the earliest proponents of the idea, talks about the singularity as “a point where our old models must be discarded and a new reality rules, a point that will loom vaster and vaster over human affairs until the notion becomes commonplace”.

Under the singularity, artificial intelligence overtakes our own, and will change the world in ways that are beyond our wildest imagination. Imagine a world where we no longer have to invent anything, because machines can do that for us. Anything we want is just one verbal command away, and each step of the way is perfectly handled either digitally or by machines.

We are still some time away from that sort of reality, but we are definitely much closer to it than in previous generations (despite all the challenges to their optimism!). The prominent futurist Ray Kurzweil, who serves as Google’s Director of Engineering, predicts that the first major step toward the singularity will happen in 2029.

According to Kurzweil, this is the year that a computer will pass the Turing test and thus be considered as having human-like intelligence. In a 2018 interview, Kurzweil reiterated his position on the accelerating development of AI:  “They can already do things greater than humans in certain areas. But they’ll have (access to) the wide range of human intelligence, and in particular really understand language at human levels by 2029.”

2029 is less than a decade away. If Kurzweil’s prediction turns out to be true, we may see the singularity occur within our lifetime. How likely is this to happen, and what might be in store for translators in the coming decade?

Machine translation: A history of uneven progress

It’s always difficult to grasp the full significance of the present time. We mark history as a series of discrete points which build a bigger picture in retrospect, and through that history build some idea of the future.

As it turns out, the history of machine translation is intimately tied to the history of AI. Telling the whole story would be a whole article in itself, which is already covered from many different angles. But it’s important to mention some major points to help us get an idea of that bigger picture and have a greater appreciation of how these fields are developing, and may continue to develop.

In 1945, the first true computer was built. ENIAC (short for Electronic Numerical Integrator and Computer) was the size of three large SUVs stacked front to back, and weighed four times as much. First built to calculate missile trajectories, ENIAC could solve simple mathematical problems at a rate of 5,000 a second—far beyond the ability of even twenty people working together at the same time!

Within the next year, scholars had already begun to consider the possibility of using this machine to perform translations, and research was soon underway. Warren Weaver, then president of the Rockefeller Foundation, posed the question to Professor Norbert Wiener of the Massachusetts Institute of Technology: “I have wondered if it were unthinkable to design a computer which would translate. . . . and even if it did produce an inelegant (but intelligible) result, it would seem to me worthwhile.”

Weaver’s memorandum

The year 1949 marked the publication of Weaver’s landmark memorandum on machine translation. In it, he suggested four directions that research in the field could take, one of which included intersections with other scientists’ early work on neural networks. This was a remarkable bit of foresight, as the technology of the time was not yet capable of achieving his propositions, and it is only recently that our technology has become effective enough to act on them in a meaningful way.

The next two decades saw increased interest in the possibilities of machine translation, but the results of these early forays did not live up to the initial hype. Things came to a head in 1964 when the US government formed the committee ALPAC (Automatic Language Processing Advisory Committee) to check the state of research and development in the field.

The ALPAC report

In 1966, this committee would go on to conclude that machine translation was a failed enterprise, and that there was “no immediate or predictable prospect of useful machine translation”. The US government withdrew support for research in the field, and private funding also diminished drastically. The bulk of research shifted to other countries. Thus, 1967 – 1976 became known as the Quiet Decade for machine translation.

This event foreshadowed a broader trend of disillusionment in the promises of AI. Like with machine translation, funding was pulled from research in this field, leading to the first AI winter.

After the Quiet Decade, research in machine translation and AI began to trickle in once again. But at no point during this time did machine translation ever reach the level where its quality could be taken seriously.

It was only in the past decade that a major breakthrough would truly revolutionize the field of machine translation in the form of Google Translate.

Google Translate and the deep learning revolution

Originally launched in 2006, Google Translate used a phrase-based approach that suffered from the same problems that other then-state of the art machine translation engines had, such as frequent mistranslations and poor grammatical accuracy. Its results were often a source of humor for its users.

In 2016, Google unveiled a new machine translation engine which resulted in a dramatic improvement in quality for Google Translate’s results practically overnight. This was made possible by a revolution in machine learning that came about only a few years prior, called deep learning.

Deep learning uses a collection of algorithms called a neural network in an attempt to imitate the way the human brain works. It processes a vast amount of data to learn how to perform complex tasks such as identifying objects in images or process human speech. These tasks require a lot of computing power that was simply not available before the turn of the millennium.

Google Translate’s success was the first to break into the mainstream, but the effect was immediate. Other tech companies and language industry players either accelerated their own research or began investing resources into deep-learning based systems.

Neural machine translation became the order of the day, and soon it would become the indispensable feature throughout the language industry that it is today.

Accelerating toward the singularity

So we’ve seen eight decades of development in machine translation. Eight decades matching the history of computing itself. What does this all mean?

What we believe, following the line of thought set by Kurzweil, is that the continued development of machine translation will go much faster than we think. Progress throughout these eight long decades has not happened incrementally, or in a straight line. There have been periods of boom and bust, followed by a veritable explosion of development only recently.

It would seem that the continued progress of machine translation now hinges on the development of deep learning and neural networks. The advantage of neural machine translation is the vast amount of linguistic data it can use to continually hone the accuracy of its translations with much less human supervision than previous models. And this data is ever increasing in volume.

AI is becoming more able to imitate the way the brain works through deep learning neural network technology. This progress owes much to the deep learning revolution which, as mentioned earlier, was in turn made possible by the increased computing power afforded to us by advances in computing hardware.

The phones that fit in the palm of our hand today are hundreds of times more powerful than ENIAC. Computers are becoming faster and costing less to produce. Humanity’s ambitions for technological development have always only been hindered by the capability of hardware, but now we’re seeing hardware improve at an accelerated pace.

There’s much to be optimistic about today. But on the other side of the coin, history has already shown the result of unbridled optimism. Who’s to say that we won’t experience a third AI winter? What if there’s a limit to what deep learning can do for machine translation?

These are questions we should consider soberly as well, but they shouldn’t lead us all the way over to the other side, into the province of complacency and nihilism. It may not be soon, but the singularity is inevitable, and we need to think of it as such. And a world where machine translation can perfect itself is something we need to think about, if only to answer the question—what does such a world mean for translators?

The current landscape for translators

Much has happened since the ALPAC report. In less than a decade since its breakthrough, neural machine translation turned machine translation from a fascinating triviality into a useful, if still limited, tool for professional translators. This has had ramifications not only for their work but also for the structure of the translation industry. 

At Tomedes (our parent company), around 17% of the language professionals we work with have not worked with machine translation, and half say they work with machine translation frequently or every day.

Most of this work involves machine translation post-editing, or MTPE. In MTPE, translators are provided with a document that has been pre-processed through a machine translation engine alongside the original text. The task of the translator is to proofread this text and make sure the translations are correct.

MTPE is a service that only became truly feasible with the rise of neural machine translation. Before Google Translate changed things up, MTPE was dismissed as more of a failure. Today, while still far from perfect, the machine output it produces is often in a condition that is workable enough for common translation projects.

One might expect that the rise of machine translation would be met broadly with suspicion by language professionals. After all, man versus machine is a classic trope of many stories in science fiction, and any form of automation has historically come with the fear that machines are coming to take people’s jobs.

Surprisingly, many of our translators do have a positive view of machine translation. The number of those who like machine translation outweighs those who dislike it, and over half of the translators we surveyed said that machine translation has had a more or less positive effect on the way they work.

“It has improved the speed at which I work and the volume of work I can handle,” says Hermione, who has been in the profession for over 15 years. “I’ve found that editing even bad text is faster than typing.”

Others have said that MTPE jobs are just another form of work, even if it’s paid a lower rate than regular translation. Because it makes translation more affordable for clients, it increases the overall volume of work available on the lower end of the spectrum. Indeed, MTPE is the fastest-growing service segment in the industry, while the demand for human translators also continues to grow.

Understandably, not all language professionals have the same benevolence toward the issue. “My experience so far shows that it worsens the quality of the translated texts,” shares Maria, another critic of machine translation. “Even when doing post-editing, the odd structures of the source language turn into ‘false friends’ in the target language, and the final effect is deterioration of the target language.”

Others criticize how machine translation can lower rates but add more work for translators. “Some clients may ask you to do an MTPE job, but the machine translation is of bad quality and requires a lot of editing, says MG, another translator. “Clients who aren’t familiar with the translation industry may think that machine translation can largely replace manual translation, so they will offer less rates as well.”

What most translators on both sides agree upon is that machine translation will never reach the level of perfection that would allow it to replace human translators. Some language professionals do, however, have thoughts about what it might be like in a world where the singularity has come for machine translation.

“Machine translation will perfect itself to a point where only niche translators and agencies will have anything to do,” says Roy, one of our most veteran translators who we work with on a regular basis. “Even MTPE will disappear, as it would not be needed. “General translation will be done, and is already being done, without our intervention.”

Despite this, Roy also believes translators will still have a role to play after the singularity. “I think that in order to survive, our business will end up returning to its roots,” he says. “Niche and creative translations, transcreation and the such, and high quality services, even though that is a very little business nowadays.”

When talent is gone and replaced by machines, the intermediary also loses its function. This is why translation agencies shouldn’t neglect the needs of translators.

Roy also touches upon an intriguing point that may play into the anxieties for translation agencies—that, in this future, perhaps it is we who may end up becoming irrelevant. 

“Mainstream translation agencies are running the same road as businesses like Tower Records or Blockbuster,” he says. “In the long run, with us translators and agencies enabling it (and we think benefiting from MT), our business will end up in the hands of the IT companies owners of the MT engines.”

That’s one sobering possibility. As things currently stand, translation agencies and translators exist in a mutually beneficial relationship. Clients choose reputable agencies for ease and reliability in managing their projects, while agencies direct appropriate work toward language professionals.

But the role of translation agencies in this relationship remains, in essence, an intermediary one. When talent is gone and replaced by machines, the intermediary also loses its function. This is why translation agencies shouldn’t neglect the needs of translators.

Imagining the translation singularity

The fact of the matter is the translation singularity will affect all players in the language industry. And it may do so in ways that we have not yet imagined. This is why it’s important for us to discuss a future after it, and how we can work together to thrive in the new industry environment that it will bring about.

“The translation singularity is the most important challenge to unlock the next level of human evolution.”

—Marco Trombetti, CEO of Translated

Recently, we had the pleasure of corresponding with Marco Trombetti, CEO of Translated. Marco is also an avid advocate of the translation singularity, and believes the future of the industry lies in symbiosis between human and machine.

“The translation singularity is the most important challenge to unlock the next level of human evolution,” he asserts. “Through language, we have understanding, and thus cooperation. Right now, we humans cooperate on a local level. If we could truly cooperate globally we could solve some of the biggest problems of humanity.”

Language is not just about bringing across the meaning of words and sentences to someone else. Language is a means of sharing the very heart of one’s message, which is something that goes beyond mere logic and sense. It is what the best translators put foremost in their task.

One of the main limitations of machine translation is its inability to transmit culture, which is the one element of language that is peculiarly human. Culture is an expression of shared experience among a group of people that is built through cooperation and cemented through language.

Going back to the myth of Babel, the fragmentation of language is what gave rise to diverse cultures. Translation helps bring us together, but it’s hard to say that there’s one truly “human” culture that allows us full, perfect understanding of each other. But that imperfection, and how we navigate that imperfection, that makes us human, and makes translators such an important agent in the affairs of the world.

Machine translation has access to a vast range of information about language, but not human experience. It is unable to process culture, so its access to the full possibilities of human communication remains limited. As such, this might be the final frontier to be breached before achieving the translation singularity.

A post-language world would, after all, probably need new, post-language solutions. Only it’s hard to imagine what those would be in the present time.

There may very well be a day where there will be no human translators. But before this day, there will also be no human drivers, and no human lawyers, doctors and programmers. Communication between human beings is one of the most challenging tasks to tackle for AI, and other kinds of professionals will become obsolete before translators do. As such, translators will still be needed for quite a long time.

However, it is also true that as technology develops the profession does change, and will change more. We need to continue adapting and looking toward the frontier of our industry, and learning from others in the field.

Marco does also say it could take an infinite amount of time to reach the singularity. “That is, if by singularity we mean a machine is able to translate better than the best translator on earth,” he clarifies. “But that is only because we are trying to simulate human speech communication. There is probably a better way to communicate that machines could develop.”

Now there’s an interesting thought exercise—what if the true singularity takes us beyond the need for human speech as we know it?

Perhaps in the future, we will be able to transmit their very thoughts as images to each other, bypassing the need for language. Perhaps when humans become cyborgs embedded with neural microchips, or have uploaded our consciousness to a machine body.

A different, more efficient way to transmit not only messages, but experience, and hence culture. That future would be the story of Babel finally rebuilt. But what a distant future that must be!

Under the singularity, translation will once again have a new face. One in which human creativity and imperfection must contend with sophisticated technology the likes of which we have never seen before.

It is perhaps selfish to say that we hope so. After all, such a future would leave us with nothing to do. Or would it? A post-language world would, after all, probably need new, post-language solutions. Only it’s hard to imagine what those would be in the present time.

In his article “A Journey Into the Future of the Translation Industry”, Jaap Van der Meer of preeminent language and data think tank TAUS advises translation agencies to follow the outlook of “kill your business before it kills you”.

Resistance is not the way for a business to survive changing tides, especially in an industry whose model is reliant on cutting-edge technology.

As mentioned earlier, it was only with the development of neural machine translation, brought about by the deep learning revolution, that machine translation became a serious figure in the story of translators. Jaap asserts the divide between human and machine is eroding at a swift pace.

“The question for operators in the translation industry, therefore, is whether the two processes continue to exist, or whether MT will completely wash away the old business,” he says in the article. Jaap believes that data and technology are the key to continued growth in the language industry.

But where does that leave translators?

Under the singularity, translation will once again have a new face. One in which human creativity and imperfection must contend with sophisticated technology the likes of which we have never seen before.

In much older times, translators were scribes or monks preserving important religious or literary texts on clay tablets or scrolls. But translation is no longer the task of one lonely linguist with a penchant for converting significant texts from one language to another. Now, there is more involved in the process, and the responsibilities of translators have also continued to grow.

We have a responsibility to think about how the idea of translation itself will change under the new technological reality.

Translators may not yet be cyborgs, but already digital technology has reshaped the way they work. The translator of today is a digital native, translating software and websites, providing subtitles and captions for videos, some even working with graphic design. In addition to specialized kinds of translation, translators are increasingly being called upon to hone their skills in varied fields like marketing, diplomacy, and big data.

They may no longer exist in the form we know now. But rather than disappear, language professionals will evolve, as they always have.

And it’s important not to take the language industry in isolation. The singularity will come not only for translation, but for other aspects of society that require technology as well. It will completely revolutionize the way we live, and the kind of opportunities that become available for us to grasp.

For example, localization isn’t just about cultural context, but also the medium in which it appears. It’s no surprise that the rise of localization coincided with the rise of personal computers, and later the Internet, putting language and technical prowess together.

Today we are even talking about hyper-localization, in which language, culture, marketing, and big data come together to provide precisely targeted digital experiences to consumers.

Transcreation takes things even further. No longer are language professionals bound to the text on the page. What they are given to work with is simply the heart of the message itself, requiring greater creative thought and writing skill on their part.

But it is also technology that drove the growth of transcreation. The increasingly globalized world economy was made possible by the Internet, allowing businesses to reach out to new markets around the world at a much faster pace.

Transcreation’s growth was also moved from inside by the rise of machine translation post-editing, which began to take up the more mundane translation jobs. This could be seen as meaning less jobs for translators, but it could also be seen as giving language professionals more room to specialize.

Translation agencies need to step up to the plate as leaders, helping to shape the new roles for translators in the post-singularity future.

Many language professionals have thrived in these fields because they saw the need and the opportunity to specialize.

This is why it’s unlikely we’ll see translators become endangered after the translation singularity. They may no longer exist in the form we know now. But rather than disappear, language professionals will evolve, as they always have.

Machine translation doesn’t have to be perfected yet to be a catalyst for this evolution, as we have already begun to see. In less than a decade, machine translation has already made waves in the industry, and while players may have differing opinions on the good and the bad of it, the impact of machine translation is definitely impossible to ignore.

It’s our imperfections that make us human. Imperfection is what encourages creativity, because if we were already perfect we wouldn’t need to innovate.

Human imperfection will remain after the singularity. And with it the need for language. But as we move toward greater technological mastery over language, the creative possibilities for managing it will also increase.

We believe the role of translators will become less technical. The creative and communicative aspects of language will become fertile ground for language professionals to grow into new roles. Translators will become trusted advisors and experts in terms of using their mind to facilitate greater communication between parties.

Overall, machine translation is making the industry more mature. It sets a mirror in front of us translation companies, and really challenges us every day to get better, and to be more advanced.

Translation agencies, too, have a role to play. We shouldn’t be content to be intermediaries. Translation agencies need to step up to the plate as leaders, helping to shape the new roles for translators in the post-singularity future.

We see the amazing work big companies are doing with AI and machine translation on a regular basis. We learn the use cases of what works and what doesn’t work, and what are the best options with regard to machine translation. And we are constantly developing the methodology and the principles in using machine translation.

Translation companies are here to set the perfect infrastructure and environment for translators to work better. Translators are part of an ecosystem of software tools that is becoming increasingly complex, and our task is to organize this for translators to be able to work at their best. As such, we should be adopting and shaping practices in ways that bring out the best in our translators.

Overall, machine translation is making the industry more mature. It sets a mirror in front of us translation companies, and really challenges us every day to get better, and to be more advanced. There haven't been many changes in this industry in past decades, and these latest developments have been very significant.

It’s an exciting and very challenging thing to be part of the industry at this disruptive inflection point. Translation might be perceived by a lot of people as a traditional industry, but it is growing every year, and growing rapidly.

Even in a world after the translation singularity, we believe that the story of the translator is not yet over. The current story may end, but new ones will begin, with the translator taking more specialized, more central roles. And machine translation will not just be another tool, but a key element to meeting the challenges that come with those roles.