What impact has AI had on translators?

The ability to speak multiple languages has long been a gateway to opportunity. Indeed, research from Cambridge a few years ago suggested that improving Britain's language abilities would provide a multi-billion-pound boost to its economy.

Of course, translation has also long been in the crosshairs of AI developers. Initial language models developed in the 1970s evolved to form a key component of Google Translate when it was launched in 2007.

A recent paper from Oxford University's Carl Frey highlights the impact of the latest generation of AI translation tools on the demand for translation services. The study finds that the rise of AI translation has had a clear and noticeable impact on the demand for human translators. What's more, it also reduces the demand for language skills in general.

“Our baseline estimate indicates that a one percentage point increase in the use of Google Translate reduced translator employment growth by 0.71 percentage points," the researchers explain.

"A counterfactual scenario suggests that approximately 28,000 fewer translator positions were created as a consequence of machine translation technology relative to the broader growth trajectory of the profession.”

This had an inevitable impact on wages, with translator income dropping after the release of Google Translate, before rebounding after 2016. The researchers believe this is likely to be because AI took over the simpler translation tasks, with translators then moving on to more complex tasks that were beyond the capabilities of the technology.

Despite this, however, the profession has still been in decline over this period.

Demand for languages

What's more, this has also significantly impacted the general demand for language skills in industry. The study found that, whereas historically, there was a strong demand for bilingual candidates, companies are now using AI to bridge any linguistic gaps. Indeed, demand for Spanish, Chinese, Japanese, German, and French skills has all declined.

“Based on the empirical approach outlined above, we find that the rise in internet searches for Google Translate corresponded with a decrease in demand for all languages spoken by the United States’ top five trading partners," the researchers explain.

“Based on the empirical approach outlined above, we find that the rise in internet searches for Google Translate corresponded with a decrease in demand for all languages spoken by the United States’ top five trading partners."

"Our baseline IV estimates suggest Spanish-speaking workers were the most affected by machine translation, with a -1.4 percentage point reduction in job posting growth for the analyzed period, while demand for Chinese and German speakers declined by -1.3 and -0.8 percentage points, respectively.”

This is despite the continued growth in globalization and the demands that it places on organizations' abilities to communicate across borders. In a world in which language skills "should" be more important, the data points in the opposite direction.

A bitter lesson

The findings have foreboding lessons for other disciplines. In 2019, Rich Sutton penned The Bitter Lesson, an essay outlining the march of AI. The neural network-based approaches he describes have powered the latest wave of translation technologies.

This has allowed translation technologies to move beyond predicting the probability of words based on the frequency patterns to neural networks that are better able to capture the linguistic context of words and sentences.

Of course, that doesn't mean that machine translation is perfect. After all, while Google Translate supports nearly 250 languages, this is still just 4% of the 7,000 or so languages used around the world. Even when translating major languages, the technology can struggle with idioms and legal and technical terms. It's perhaps no surprise that a study last year found that machine translation hasn't been fully cracked yet.

Over-reliance

As with other disciplines, there is a real risk of becoming overly reliant on machine translation. A lack of our own knowledge makes us less equipped to check that what we're given is accurate. This can be hugely problematic in high-risk settings, such as healthcare or the law.

This goes some way to explain why human translators are still important in these disciplines, but even then, their role has changed to more of a quality control role rather than producing the original translation. This, understandably, diminishes the enjoyment people get from that role and blunts their own skills.

Looking to the future

So what does this mean for the future? At the moment, around one in five American schoolchildren is studying a foreign language. How useful will these skills be if AI translation continues to improve? Will these skills continue to be a priority?

“Although the observed effects have been moderate thus far, they are likely to intensify as machine translation technologies continue to advance, particularly in the realm of simultaneous speech interpretation," Frey explains.

"These trends carry significant implications for education policy, especially considering that nearly 20 percent of students in American schools are currently enrolled in one or more foreign language courses.”

Obviously, the benefits of knowing multiple languages extend beyond mere economic value. Language can help to provide a level of cultural understanding that is crucial in our globalized world. More importantly, soft skills such as negotiation, diplomacy, and relationship-building often depend on the ability to communicate in a counterpart’s native tongue. The risk is that overreliance on machine translation fosters a generation of monolingual professionals ill-equipped for the deeper cultural dimensions of global engagement.

To date, the impact has mainly been on written translation, but there are perhaps even bigger risks with real-time speech translation. OpenAI’s recent demonstration of its Sky model, which translated speech between Italian and English with minimal latency, hints at the technology’s trajectory. If such systems become widely adopted, the role of human interpreters may also diminish.

Machine translation is but one of the many professions that will be shaped by AI. As in the past, the key will be ensuring that workers can transition to new roles that require distinctly human capabilities.

For now, the outlook for human translators and foreign language professionals remains uncertain. As AI-powered translation continues to improve, the economic rationale for language learning will likely weaken further.

However, whether AI can ever truly replace the human nuances of language remains an open question.