■ CRITICAL RISK ■ Media & Communication
For the bulk commercial work — manuals, product pages, routine business documents — yes, and it's already happened. Human translators are being squeezed into editors of machine output and specialists in the domains where a wrong word costs real money.
“DeepL doesn't charge per word or argue about idioms.”
Our AI replacement risk score — how we score jobs
Working translators rarely sit down with a blank page anymore. The modern workflow runs through CAT tools like Trados or MemoQ, translation memories that recycle previously translated segments, and increasingly a machine-translation engine that produces the first draft of everything. Agencies now hand out MTPE work — machine translation post-editing — at per-word rates well below traditional translation, on the theory that fixing the machine's draft is faster than translating. Sometimes it is. The rate cut happens either way.
Neural machine translation crossed the good-enough threshold for huge swaths of commercial content years ago, and large language models pushed further: they handle context, register, and even some idiom in ways older systems fumbled. E-commerce listings, support articles, internal corporate documents, and user-generated content are now machine-translated at volumes no human workforce could touch, mostly with light or zero human review. The entry-level generalist translator — the person who used to live on that volume — is the one our 85 score is really about.
The defensible ground is where errors are expensive or meaning is genuinely hard. Literary translation is voice work as much as language work. Legal contracts, patents, and clinical documentation need someone accountable who understands the domain, because 'the model usually gets it right' is not a defense in court. Certified translations, high-stakes marketing transcreation, and low-resource language pairs also resist. But that ground is smaller than the profession that used to stand on it, and the path in — years of bulk work building expertise — is exactly what the machines removed.
Automatability: our editorial assessment of current and near-term AI capability
This one's live now. Machine translation already handles most of the world's translated volume, per-word rates have been sliding for a decade, and MTPE has become the default assignment at many agencies. Through the late 2020s expect generalist translation work to keep evaporating while specialized, certified, and literary niches hold — smaller, better paid, and much harder to break into.
As a generalist path, honestly no — the bulk work that sustained new translators has largely gone to machines, and rates reflect it. As a specialist path it can still work: legal, medical, and certified translation pay well and require credentials machines can't hold. If you pursue it, pair language skills with deep domain expertise from day one, because the language skill alone no longer clears the bar.
For a huge share of commercial content, effectively yes: product listings, support docs, and routine business text are machine-translated at scale with minimal human review. Humans remain essential where stakes are high — contracts, clinical trials, literature, certified documents. The profession hasn't vanished; it has narrowed, with post-editing replacing much of what used to be full translation work.
Pick a defensible niche and get credentialed in it. Court certification, medical translation credentials, or patent expertise turn you from an interchangeable vendor into an accountable professional. Learn the MT tooling well enough to charge for quality evaluation, not just cleanup. And raise your floor: MTPE at collapsing per-word rates is a treadmill, not a strategy — move toward project-based and consulting fees.
Slower. Live interpretation adds real-time pressure, accents, crosstalk, and physical presence — courtrooms, hospitals, and diplomacy still insist on humans, often by regulation. Speech-to-speech AI is improving fast and eating low-stakes use cases like tourist interactions and basic business calls, but high-stakes live interpreting is meaningfully more durable than document translation. It shares the same long-term direction, though.