LANDMARK · 5 MIN · USE
Translation
After this landmark, you can tell when machine translation is good enough and when a document’s stakes call for a human translator.
Machine translation has become genuinely strong for gist and everyday messages — a menu, an email, the sense of an article come through reliably. Where it still fails is the part that isn’t literal: idiom, register and formality, cultural nuance, names, and anything high-stakes where a wrong word does real harm. A model can render the words and miss that a phrase is an insult, that the formality is wrong for the relationship, or that a legal or medical term of art has a precise meaning it flattened. The skill is knowing which situation you’re in: “good enough to understand” is fine for most of daily life, but a contract, a diagnosis, or a safety notice needs a human who knows both languages and the stakes.
This is the canonical concept. It stays the same across learner lenses so personalization never changes the underlying facts.
◌ EVERYDAY LENS
What this looks like for you
For understanding a message, a sign, or a reply, machine translation is genuinely good — use it freely. If you’re sending something where tone matters, like a message to someone’s family, have a fluent speaker glance at it, because register and warmth are what slip.
MAKE A DECISION
You need to translate a customer-facing safety instruction into a language you don’t speak. Machine translation gives you a fluent-looking result. What do you do?
CARRY THISTake a short phrase with idiom or tone, translate it and back-translate it, and note what the round trip lost.