Where it genuinely works
For plain, factual, present-tense text with no cultural weight, modern machine translation is good. Opening hours, ticketing, directions, accessibility information, café details, and the rules about photography all translate cleanly, because the sentences are simple and the vocabulary is common. A museum that machine-translates its visitor information is doing something useful, and doing it for free.
It also works as a first pass on longer interpretive text. Machine translation followed by a native speaker editing the result is cheaper than translating from scratch and is now standard practice in the translation industry. The important word is followed: the machine drafts and a person is accountable for what gets published.
Where it fails, and it fails in public
The failures cluster in exactly the content a museum cares most about. Proper nouns get translated when they should be left alone, so a place name becomes a phrase and a person's name becomes a common noun. Historical terminology gets modernised, which can turn a careful description into an offensive one. Register collapses, so text written with dignity comes out casual or clinical.
The dangerous property is confidence. A machine translation is never hesitant and never flags the sentence it got wrong, so a museum with nobody on staff who reads the language has no way of knowing which lines are fine and which are embarrassing. The error is discovered by a visitor who reads that language, standing in the gallery, and it is discovered by the visitor you were most trying to welcome.
Indigenous and minority languages are the sharpest case. Machine translation of te reo Māori and comparable languages is weak, the cultural stakes around getting it wrong are high, and the appropriate people to check it are usually not the people who could check a French translation. That content is not a candidate for an unreviewed machine pass.
Names of people, places, and objects, which should often stay untranslated
Historical and cultural terminology, where the modern word may be wrong or offensive
Anything with tone: memorial text, personal testimony, contested history
Indigenous and minority languages, where quality is weakest and stakes are highest
Anything a visitor might quote, complain about, or photograph
The other problem: asking visitors to do it
Some museums leave the panels in English and rely on visitors pointing their phone's camera translation at the wall. It works, in the narrow sense that the visitor gets words they can read.
What it communicates is different. It tells that visitor the museum did not prepare anything for them, and it makes them do a piece of work every few metres that everyone around them is not doing. It also fails on exactly the panels that matter, because camera translation struggles with the low light, reflective surfaces, and styled typography that exhibition design tends to favour. It is a fallback, not a plan.
The practical middle
The position most museums end up in is straightforward. Machine-translate the operational text and publish it. Machine-translate the interpretation as a first draft, then have a speaker of that language edit it before it goes live. Never machine-translate the emotionally weighted material, the culturally sensitive material, or Indigenous language content without a reviewer who is qualified to sign it off.
The constraint that follows is not budget, it is reviewers. Committing to a language you cannot get reviewed is worse than not publishing it, because an obviously wrong translation damages more trust than an absent one. Publish the languages you can stand behind and add the rest when you have found someone to check them.
Frequently asked questions
Can museums use Google Translate for exhibit text?
For operational text like hours, directions, and rules, yes, and it is a sensible use of a free tool. For interpretation, use it as a first draft that a speaker of the language edits before publication, and keep it away from culturally sensitive or Indigenous language content entirely.
How accurate is machine translation for museum content?
Good on plain factual sentences, unreliable on names, historical terminology, and tone. The problem is not the error rate so much as the confidence: it never flags the line it got wrong, so a museum with no reader of that language cannot tell which lines are safe.
Is it acceptable to expect visitors to translate the signs themselves?
It works technically and reads badly. It tells a visitor the museum prepared nothing for them, and camera translation struggles with the dim lighting, reflective cases, and styled type that exhibitions are full of.
What is machine translation post-editing?
A machine produces the first draft and a human translator edits it to publishable quality. It is standard practice, noticeably cheaper than translating from scratch, and the accountability stays with a person.
Should Indigenous language content be machine translated?
No. Quality is weakest for exactly those languages, the cultural consequences of an error are highest, and the right reviewer is usually a specific person or group rather than any fluent speaker.