The assumption nobody tests
Staff shortages push organisations to recruit internationally. Polish installers, Romanian welders, Turkish engineers, Filipino marine crew. And the training runs in English, because everybody speaks that.
That last assumption rarely gets tested. Somebody who copes perfectly well in an intake conversation cannot on that basis work through a procedure in which the difference between shall and should decides whether something is mandatory. That is not a matter of intelligence or effort, it is the difference between speaking a language and thinking in it under pressure.
What exactly gets lost
With safety instruction it goes wrong in three places.
Nuance in obligation. Practically every procedure carries gradations: must, should, may, only if. Somebody reading in a second language flattens those gradations into “this is the idea”, and the exception, precisely the case where it goes wrong, drops out.
Technical terms that sit slightly differently. An engineer knows their trade in their own language. The Dutch or English term is a translation they learned alongside it, and under time pressure they fall back on their first language. If the instruction does not exist there, there is nothing to fall back on.
Daring to ask. This is the biggest and the least visible. Somebody who feels uncertain about their command of the language does not ask the question they should ask, certainly not in a group, certainly not of somebody in a hurry. The quiet nod does not mean understood.
Translating is not enough
The reflex is: have the module translated. That is an improvement, but it solves only half.
A literal translation carries the Dutch sentence structure and the Dutch examples with it. References to national labour law, to a VCA certificate, to how things go here: all of that stays in and means little to somebody from a different background.
What does work is material built visually, with language as support rather than as the carrier. Imagery of the real installation, an operation you see rather than read, a situation you recognise. The text then confirms what the image has already said, and the language barrier matters far less.
How it becomes affordable
The objection is always the cost, and until recently that objection was fair: five languages meant recording five times, editing five times, checking five times.
What has changed is not that translating became free, but that keeping five versions consistent is no longer the bottleneck. The first language stays the most expensive. That is where the design sits, the imagery, the didactic structure. Every language after it is a fraction of that, provided the material was designed for it.
That last point is the heart of it. Build a module with text burned into the imagery and you pay nearly full price again for language two. Keep text and imagery separate and you pay a fraction. That choice is made at the start, and it decides whether multilingual delivery is a project or a setting.
That choice does not stand alone. Which other decisions you take at the start and live with for years is in the guide on setting up or renewing a corporate academy. What it looks like in delivery is under multilingual onboarding.