The Industry That Trained Its Replacement

Almost nobody outside it can name a translation company. And almost nothing in world trade moves without one.

The language industry is worth around sixty to seventy billion dollars a year. It is the plumbing of globalisation: every medicine label in twenty-four European languages, every aircraft manual, every subtitle, every patent, every contract that crosses a border.

A few big firms organise the work. Hundreds of thousands of freelancers actually do it, paid by the word, invisible to the clients whose voice they become.

One corner of this industry is more dramatic than the rest, and you should meet it properly.

Telephone and video interpreting. Hospitals, courts, police forces and emergency lines cannot work in only one language. In the United States, having an interpreter in healthcare is a legal right. So specialist companies keep thousands of interpreters available around the clock, reachable in seconds, in more than two hundred languages.

An interpreter on one of those lines might take a heart-attack call at nine, an asylum interview at ten, and a child-custody hearing before lunch. Pure performance under pressure, where one wrong word can cost a case, a diagnosis or a life.

Remember this corner. It behaves differently in every part of this story.

And for decades the whole trade had one quiet habit — the habit this case turns on.

It kept its work. Every translated sentence was saved next to its original, so nobody paid twice for the same text. Public institutions did the same enormously: the European Parliament alone published decades of proceedings, professionally translated, in more than twenty languages.

By the 2010s, humanity had built the largest collection of carefully matched sentences ever created — hundreds of millions of examples of exactly how an expert turns one language into another.

Nobody thought of it as anything except an archive.

For fifty years, machine translation was the future that never arrived. Early systems produced nonsense. The systems of the 2000s were better than nothing and worse than embarrassing — good enough to guess the meaning, and a standing joke among professionals.

Then in late 2016, Google changed its translation service to a neural network trained on matched text. The jump in quality was so sudden that it made headlines around the world. Overnight, in the major languages, the output went from broken to often fluent.

And the data that made it possible was, to a very large extent, that archive. The saved memories. The public collections. Decades of human sentences.

The industry had spent thirty years, without knowing it, writing its replacement's textbook.

Two ironies followed, and both are worth enjoying.

First: in 2017, Google researchers looking for a better translation system published a paper called "Attention Is All You Need". It introduced the transformer — the invention that later powered ChatGPT and the whole modern AI wave. The technology that would unsettle every knowledge profession was built, first and specifically, to translate.

Second: that same summer, a small company in Cologne called DeepL launched a neural translator that independent tests repeatedly ranked above the American giants in its main languages. One of the best translation machines on earth came from a medium-sized German firm.

Inside the profession, the ground moved within a few years. Agencies changed their prices: why pay the full rate when the machine's draft is nearly right? The new offer was post-editing — clean up the machine, at half the rate or less.

The price per word for ordinary commercial translation started a fall that never reversed. When large language models arrived after 2022, the pressure jumped again.

And yet the industry did not shrink. That is the second half of this case.

Total revenue kept growing through the disruption, because demand exploded. At machine prices, the world began translating things that would never have been worth a human's rate: every product page, every support message, every video's subtitles in thirty languages.

The pie grew while each slice got thinner.

The companies that did well changed what they sold. Not words — assurance. Machines make the volume; humans carry the judgement. And the price is attached to the risk of the content, not to its length.

A tourist menu goes through the machine raw. A pacemaker manual, a court file, a drug label goes through people — because when those are wrong, somebody must answer for it. And a model cannot be sued, sworn in, or struck off a professional register.

Which is exactly why the spoken corner became the fortress.

Demand for medical and legal interpreting kept rising all through the AI decade. Migration did not pause. Regulations still required qualified humans. And no hospital's lawyers would put an unaccountable system alone between a doctor and a consent form.

The interpreter's product was never really words per minute. It was a qualified human being, present in the moment, responsible for what both sides understood — closer to a nurse or a court officer than to a document.

Machines came in as assistants: transcribing, suggesting, covering languages nobody could staff. The accountable human stayed on the line.

The pattern generalises, and every profession is now studying it: automation moves fastest where mistakes are cheap, and slowest where somebody must answer for them.

An honest telling needs the open questions too. The machines keep improving, so the line marked "mistakes are cheap" keeps moving. Some routine interpreting is already crossing over.

And the profession has a quiet crack in it. If the easy work that used to train beginners is automated, where does the next generation of experts come from?

The question this industry answered first, every profession now faces. When the machine can produce your output, what exactly are you selling?

The translators still standing have an answer, and it fits on an invoice. Not the words. The responsibility for them.