Thirty pupils, eleven languages, one letter home.

Machine translation is now reliable enough for routine school communication, and using it is usually better than the alternative, which is a family receiving nothing they can read. The risks worth managing are not mistranslated words. They are register, liability and data, in that order of frequency and roughly the reverse order of how often they are discussed.

Can AI translate school communications reliably?

For a trip letter, an absence notice or a newsletter, yes, comfortably. These are short, factual and conventional, which is the easiest case there is. A parent who currently receives an English letter they cannot read is not being protected by the school's caution. They are being excluded by it.

For anything with legal or emotional weight, the picture changes. A safeguarding communication, an exclusion letter, a SEND document or anything a family might later rely on should have a human translator in the loop, because the cost of being subtly wrong is not symmetrical with the cost of taking longer.

What are the real risks of machine translation in a school?

• Register. Many languages encode formality grammatically, and a translation that lands as too casual or too cold changes the relationship the letter was meant to maintain.
• False fluency. A translation that reads smoothly is trusted more than one that reads awkwardly, regardless of which is more accurate. Smooth and wrong is the dangerous combination.
• Untranslatable specifics. Terms like SENCO, EHCP, key stage and pupil premium have no equivalent elsewhere, and a literal rendering can mislead badly. These usually need a gloss, not a translation.
• Direction of trust. Translating a school's letter outwards is lower risk than translating a parent's reply inwards, where a misreading changes what the school believes about a family.
• Who signs it. A translated letter is still the school's letter. Nobody has translated the accountability.

How should AI support EAL pupils in lessons?

As a bridge into English rather than a substitute for it. First-language explanation of a concept, subject vocabulary with cognates identified, and instructions given in both languages until they are not needed all support the pupil towards the curriculum. Translating everything permanently supports them away from it.

The pragmatic version is that a pupil should be able to reach the same learning objective as the class while their English catches up, which is the same principle as differentiation. The language is the barrier. The objective does not move.

Why translation is a data question too

Because translating a document means sending it somewhere. This is obvious when stated and almost never stated. A SENCO pasting a support plan into a translation service has just sent a named child's needs to a third party, usually without a thought, because translation does not feel like the sort of act that has a data protection dimension.

It has exactly the same one as marking. The answer is the same too: if the translation happens on the school's own hardware, the document never leaves, and the question never arises. Aevum's multimedia and language work runs on the same local-first basis as the rest of the family. In development.