The Economics That Kept Guides Monolingual
Museums have wanted multilingual audio guides for as long as audio guides have existed. The obstacle was never appetite; it was per-language production cost.
A traditional guide requires a script, a translator, a voice artist, a studio session, an editor, and a quality pass — per language. At that cost structure, an institution rationally supports three or four languages: the local language, English, and whichever two markets dominate its visitor mix. Everything else is unserved.
The result is a familiar scene in almost every major institution. A visitor group arrives from a market that represents a meaningful share of admissions, and the guide is not available in their language. They walk the galleries reading wall labels in a language they half-follow, and leave having experienced a fraction of what the institution spent decades curating.
The cost structure has changed enough that the calculus is genuinely different. An institution can now realistically support fifteen or twenty languages, including the long tail that no per-language production model could ever have justified. What has not changed is that doing it badly is worse than not doing it, because a museum's authority rests on precision.
Curatorial Register Is the Central Challenge
Museum interpretation is written with unusual care. Curators choose words deliberately, hedge attributions precisely, and avoid claims the evidence does not support. A phrase like "attributed to," "workshop of," "circle of," or "probably" carries specific scholarly meaning within a discipline that has spent a long time agreeing what those words mean.
Automated translation flattens exactly this kind of hedging. "Attributed to" becomes "by." "Probably fifteenth century" becomes "fifteenth century." A carefully scoped claim about a maker becomes a confident assertion the institution would never make.
This is not a cosmetic problem. Museums are cited, quoted, and trusted. An audio guide that asserts an attribution the curatorial department deliberately avoided has misrepresented the institution's scholarship.
The controls that address it:
Lock attribution vocabulary as terminology. Every hedging formula used in your interpretation, with the approved rendering in each language, taken from established scholarly usage in that language rather than generated fresh. Art-historical terminology has settled equivalents in the major languages, and using them is both more accurate and more defensible.
Review by someone who knows the field, not only the language. A fluent translator without art-historical or archaeological background will not notice that a hedge has been lost, because the resulting sentence reads perfectly well.
Keep dating and period conventions consistent. Period names, era conventions, and calendar systems differ across languages and cultures, and a mechanical translation of a date range can shift its meaning.
Retain original-language titles alongside translations. A work's title in its original language is part of its identity and is how a visitor will find it in a catalogue or search for it later.
Sensitive Collections and Contested Histories
Many collections contain material whose acquisition, ownership, or representation is genuinely contested. Interpretation of such material is usually written with great care, often after long internal debate and consultation with source communities.
Translation can undo that care silently.
The specific risks:
Community naming. The name a community uses for itself is frequently not the name historically used by the institution, and the correct term differs by language. Automated translation will use whatever form is statistically common, which may be an exonym the community rejects.
Provenance language. Terms describing how an object entered a collection — acquired, collected, removed, taken, donated — carry enormous weight in restitution discussions, and the differences between them are precisely what translation tends to blur.
Sacred and restricted material. Some objects carry cultural protocols about who may view or discuss them. Interpretation written with those protocols in mind must retain them in every language.
Human remains and funerary material. Language here is carefully chosen in most institutions and should be treated as fixed text rather than translatable prose.
Colonial-era terminology. Historical terms that appear in archival records need contextualising rather than reproducing, and the contextualising language must survive translation.
The practical approach: identify sensitive content explicitly, exclude it from automated pipelines by default, and route it through the same consultation process the source interpretation went through. Where source communities were consulted on the original text, they should be consulted on the translation into languages they speak. This is slower and it is the only defensible way to do it.
What to Localize, and In What Order
Institutions have more content than budget, and the sequencing that produces the most visitor value is fairly consistent.
First: orientation and navigation. How the museum works, what is where, ticketing, facilities, accessibility provision, rules about photography and touching. This is the content that determines whether a visitor can use the building at all, and it is almost always neglected in favour of collection interpretation.
Second: the highlights tour. The twenty to forty objects most visitors come to see. Highest leverage per minute of production.
Third: permanent collection interpretation. Deep, evergreen, and worth doing properly since it will serve visitors for years.
Fourth: exhibition-specific content. Temporary by definition, so the economics depend entirely on run length. A six-month exhibition can justify localization; a six-week one usually cannot, unless it is the institution's flagship.
Fifth: educational and family content, which frequently has its own register requirements and benefits from separate treatment rather than a simplified version of the adult guide.
Ongoing: digital collection content. Online catalogue video, curator talks, and recorded lectures reach an audience far larger than the physical visitor base and are usually the least localized material an institution holds.
That last category deserves emphasis. A museum's digital audience is typically many times its physical one and is far more internationally distributed. Localizing recorded curator talks and collection films often reaches more people per unit of effort than anything happening in the building.
Voice and Atmosphere
Audio guides have a particular listening context: a visitor standing in a gallery, looking at an object, often for two to four minutes.
Practical guidance from that context:
Pacing matters more than in most media. The visitor is looking while listening, and narration that races ahead of their looking defeats the purpose. Timing that keeps speech natural rather than compressing it to match a source duration is important here, and target-language text length differences make this a real risk.
A consistent voice per language across the guide builds a sense of a single knowledgeable companion. Switching voices between objects is disorienting.
Curator and expert voices are an asset worth preserving. Where a guide features the curator who researched an object, that authenticity carries value. Voice preservation across languages retains it, with the individual's consent.
Ambient and archival audio should survive. Guides often include music, archival recordings, oral history, and environmental sound. Separating narration from that bed rather than replacing the whole track preserves what makes the guide immersive.
Register should be warm and direct, not lecturing. This is a translation instruction as much as a writing one, and it is worth stating explicitly to reviewers, because institutional register tends to reassert itself in translation.
Accessibility Is Part of the Same Pipeline
Museums carry accessibility obligations that overlap efficiently with translation if planned together.
The same reviewed transcript that feeds translated audio also feeds same-language captions for deaf and hard-of-hearing visitors, published transcripts for visitors who prefer reading, and the text base for audio description.
Audio description for blind and low-vision visitors is a distinct discipline — describing what an object looks like, in words, is a specialist craft — but it is a script that then localizes like any other. Institutions that produce audio description in the source language and never translate it are leaving a significant accessibility gap for non-native-speaking visitors with vision impairment, a group nobody plans for and which certainly exists.
Two practical points: accessibility requirements apply to each language version, so translated content needs translated captions and transcripts; and providing the guide as text as well as audio serves visitors with hearing impairment, visitors in noisy galleries, and visitors who simply read faster than they listen.
Rotating Exhibitions and Maintenance
Temporary exhibitions create a recurring operational problem: content produced under deadline, used for a fixed period, then retired.
What makes this manageable:
Build the terminology asset once, institution-wide. Artist names, period terms, technique vocabulary, material names, and attribution formulas recur across every exhibition. A mature glossary makes each new exhibition dramatically faster than the last.
Template the recurring structures. Introduction, room texts, object entries, and conclusion follow a consistent pattern. Templating the non-object content means each exhibition localizes only its genuinely new material.
Decide language tiers by run length and expected audience, as an explicit policy rather than a per-exhibition negotiation.
Plan localization into the exhibition timeline, not after opening. Content produced two weeks before opening leaves no room for review, which is when quality failures happen.
Archive everything. Exhibition content is frequently reused in touring versions, online exhibitions, and publications. Localized versions retained properly are assets rather than sunk costs.
Measuring Whether It Is Used
Museums have unusually good instrumentation available and rarely apply it to language provision.
Worth tracking: guide activation rate by language against admissions by visitor origin, which reveals whether provision matches demand; completion rate per stop and per tour, which identifies content that is too long or badly paced in a particular language; drop-off points, which frequently cluster around one object where the translation is confusing; and the gap between languages offered and languages requested at the front desk, which is the simplest possible demand signal and almost never recorded systematically.
Two findings recur across institutions that measure this. Uptake in a newly added language is usually far higher than visitor-origin data predicted, because visitors from many countries prefer a language other than their national one, and because residents of the museum's own city are frequently not native speakers of the local language. And completion rates in localized versions are often lower than in the source language, with the cause usually being pacing rather than translation quality — narration generated to match the source duration ends up too fast for comfortable listening while looking at an object.
That second point has a direct fix: let the localized narration run longer than the source rather than compressing it. Audio guides are not synchronised to picture, so there is no technical reason to preserve the original duration, and preserving it degrades exactly the experience the guide exists to provide.
Front-desk staff are also an underused evidence source. They hear every request for an unavailable language, and a simple tally kept over a month is frequently more actionable than any analytics dashboard.
Frequently Asked Questions
Will automated translation preserve careful curatorial hedging?
Not by default, and this is the primary quality risk. Attribution formulas like "attributed to," "workshop of," and "probably" carry precise scholarly meaning, and fluent translation tends to flatten them into confident assertions. Lock these formulas as terminology using established scholarly equivalents in each language, and have output reviewed by someone with subject knowledge rather than language skill alone.
How should sensitive or contested collection material be handled?
Exclude it from automated pipelines by default and route it through the same consultation process the source interpretation received. Community self-designations, provenance verbs, and protocols around sacred material and human remains all carry weight that translation can silently undo. Where source communities were consulted on the original text, consult them on translations into languages they speak.
What should a museum localize first?
Orientation and navigation content — how the building works, ticketing, facilities, accessibility. It determines whether a visitor can use the institution at all and is routinely skipped in favour of collection interpretation. The highlights tour comes second.
Can we use the curator's own voice across languages?
Yes, with their consent, and it preserves genuine authority where the curator researched the object being discussed. Obtain written permission specifying the languages and uses, and settle institutional policy on whether visitors are told the voice is synthesised.
Does translated audio satisfy our accessibility obligations?
No. Captions for deaf and hard-of-hearing visitors, published transcripts, and audio description are separate requirements, and they apply to each language version — so translated content needs translated captions too. Planning translation and accessibility as one pipeline from a single reviewed transcript is substantially more efficient than treating them as separate projects.
Related reading: Audio Description and Translation | Video Accessibility Guide | Subtitle to Audio Conversion
