A museum, archive or research library that needs to translate an IIIF Presentation API 3.0 Manifest cannot treat the job as replacing words inside a JSON file. A reliable IIIF Manifest translation preserves the digital object’s identity, Canvas ordering, image and media references, Web Annotation targets, reading structure, accessibility and rights while making human-facing descriptions understandable in another language. That is the practical question behind searches such as how to localize IIIF Presentation API 3.0, how to translate multilingual IIIF manifests and how to preserve canvas identifiers when translating digital collections.
A good multilingual exhibition lets a visitor understand what they are seeing without accidentally changing what the institution is presenting. The visitor may select English, French, Arabic or Japanese, but the digitized letter remains the same letter, its second page remains the second page, and a transcription remains distinct from a curator’s interpretation. International Image Interoperability Framework technology makes distributed collections portable; careful language work makes that portability useful to people, not only to software.
This guide explains the architecture of a Presentation API 3.0 resource, the boundaries between descriptive text and structural identifiers, and the decisions that matter when a cultural object travels across languages. It includes worked examples involving manuscripts, photographs, oral histories and time-based media, a validation approach, and questions that institutions can use when reviewing multilingual publication. The focus is the complete viewing model, not a narrow checklist for one label field.
One digital object, several ways of understanding it
Imagine a handwritten diary photographed page by page. Its owners want a visitor in Tokyo to read Japanese labels, a student in Paris to encounter French descriptions, and a family in Nairobi to compare an English transcription with the original script. The digital surrogate needs not split into unrelated objects whenever a new language appears. The Manifest can describe one resource through language-dependent descriptions and related annotations, while keeping its structural identity stable.
This distinction is fundamental. Words may be translated, transliterated, quoted in the original or deliberately left alone. Identifiers should normally remain unchanged. The choice depends on the role each value plays. An identifier locates or names a resource for machines; a label introduces it to people. Confusing the two can make a viewer display the wrong page, lose a canvas annotation or generate duplicate collection entries.
Nor does preserving structure mean refusing to translate anything. A technically valid manifest with only one language can still exclude a large part of its possible audience. The appropriate objective is not immobility. It is a controlled change to the user-facing interpretation, with unchanged references to the underlying assets and honest records of which interpretations are available.
What Presentation API 3.0 actually describes
A Manifest represents an object to be presented, such as a book, single artwork, film, map or recorded performance. A Collection groups resources for discovery and browsing. A Canvas provides the spatial or temporal surface on which content is assembled. Annotation Pages collect annotations; individual Annotations connect content bodies to targets, including the area or time range of a Canvas. Ranges can describe logical sequences such as chapters, acts, movements or an exhibit route.
This is not simply a vocabulary list. Each object has a different relationship to what a visitor can do. A Manifest tells a viewer which item it is opening. A Canvas tells the viewer where an image, sound or video belongs. A Range gives structure across those Canvases. An Annotation can paint a photograph onto a Canvas or supplement it with a transcription. The proper translation strategy must respect those differences.
The official specification separates properties that describe meaning from properties that connect resources. Human-readable labels often use language maps, while technical types such as Manifest, Canvas and Annotation are vocabulary terms with prescribed meanings. Editing a type name to match a translated user interface would not make the data multilingual; it would make the data harder to interpret.
Think in layers rather than in one document
The first layer is the machine-readable graph: resource identifiers, types, references and relationships. The second is the presentation layer: labels, summaries, metadata captions and navigation text. The third is the scholarly layer: transcription, translation, commentary, provenance and alternative readings. A fourth layer governs access and use, including rights information and restrictions.
A language project can touch all four layers, but it should not treat them identically. Translating a summary is different from publishing an alternative transcription. Displaying a rights explanation in another language does not alter the legal status of the media. Localizing a Range label called ‘Chapter Three’ does not justify reordering the Canvases. These examples show why a property-by-property review makes more sense than a global search-and-replace.
The layers also make evaluation easier. When a user reports that ‘page four became page five’ after a language update, the problem is structural. When the wrong script appears in the title, the problem may involve language maps or fonts. When a translated paragraph gives a different date or attribution, the problem is editorial and evidential. A good review asks which layer changed before proposing a fix.
Resource identifiers are not sentences to translate
Every Manifest and Canvas has an identifier, generally expressed as a URI. That URI gives other systems a stable target to cite. A catalog, annotation tool, research notebook or federation may have already stored it. Changing the URI because a page title has been localized can sever those relationships. This is particularly dangerous when multiple applications are consuming the same collection.
Suppose a museum has a Canvas at https://collections.example.org/iiif/diary-17/canvas/p2. A French translation can label that Canvas ‘Page 2’ or ‘Deuxième page’ without rewriting the URL. That illustrative address is not a claim that a real record exists; it merely demonstrates the separation between stable reference and display language. The original asset can continue to be referenced while descriptions change.
Identifiers are not always pretty, and they do not need to be. Their job is dependable reference. Public-facing titles may change with curatorial research. Language can evolve, spelling can be corrected, and access notes can be improved. A stable resource identity allows those revisions to happen without losing the path back to the same cultural object.
Language maps are promises about available text
Presentation API 3.0 uses language maps for many human-readable properties. In everyday terms, the structure connects a language tag to one or more strings. A display system can then choose an appropriate label when the user requests a language. Language tags matter because ‘en’ and ‘fr’ are not decorative labels: they tell clients what linguistic material is present.
Do not invent a translation merely to fill every language slot. If a title has no trustworthy English equivalent, preserving the original script may be better than manufacturing a confident-looking guess. A transliteration may be valuable, but it is not automatically a translation. If an expression is a personal name, administrative title or culturally situated term, the editorial choice should be visible in the record or explanatory note.
A language map can support more than one string for a given language. That possibility should not be treated as permission to throw together unrelated title variants. Decide what the strings mean: variant titles, alternate labels or revisions might be better represented in dedicated descriptive relationships depending on context. A clear editorial rule makes multilingual interfaces predictable and reduces accidental inconsistencies across collections.
Language tags, script direction and multilingual reality
A collection described as ‘Arabic’ may include Arabic-language text in Arabic script, but a researcher could also supply an English explanation, a Romanized form of a personal name or a local dialectal term. Language tags must match the actual content and not merely the institution’s geographic location. A French museum may hold a Persian manuscript; its collection language does not determine the language of the manuscript.
Right-to-left presentation raises separate questions. A manifest can contain the correct Arabic words and still be hard to navigate if a viewer handles direction poorly. Mixed strings containing page numbers, Latin identifiers and Arabic descriptions are especially revealing tests. Reviewers should inspect the rendered interface, not conclude that the job is finished because the JSON is valid.
Fallback behavior requires planning too. When an exact preferred language is not available, the client may choose another language. The institution should test what its chosen viewer actually does rather than assume all viewers behave alike. Readers need an intelligible choice, and curators need a way to recognize when a fallback has shown an untranslated passage.
Titles are not the same as catalogue identities
A Manifest label often gives visitors the quickest recognition of an object. Yet a catalogue title can be supplied by the creator, assigned by a collector, inherited from an exhibition or translated by a later scholar. Those possibilities have different evidential strengths. Translators should ask whether the displayed phrase is an original title, a descriptive title or a modern convenience before deciding how much to adapt it.
Consider a photograph catalogued under a colonial-era place name. A contemporary place name may be useful to visitors, but silently replacing the historical wording can hide the document’s context. A well-designed multilingual description may retain the source title, supply a translated title, and explain historical naming in a note. That answer is less tidy than one universal replacement, but it is more faithful to the evidence.
The same issue arises with objects whose titles are themselves culturally significant works of language. For poetry collections, paintings with inscriptions, ceremonial objects or artist-created naming, a translated title may be an aid rather than a substitute. Treat the visitor’s need for comprehension and the object’s historical record as complementary goals.
Metadata entries need semantic care
A Manifest can present metadata entries as readable labels and values. A translator should look at each entry’s function. ‘Date’ may be a simple field label, while the date value may combine a calendar system, an approximation and a scholarly note. ‘Creator’ may refer to a person, a workshop or an uncertain attribution. ‘Place’ can be a historical jurisdiction rather than a modern city.
A literal translation of a field label is easy; preserving the evidence conveyed by the value is harder. The phrase ‘possibly from northern India, late eighteenth century’ should not quietly become a definite production date and location. A translated value must preserve uncertainty, attribution and the distinction between documented fact and later inference. Otherwise users encounter a stronger claim than the catalogue supports.
Decide which values are textual descriptions, which are coded identifiers and which are links to external authorities. A stable authority URI can remain the same even if the adjacent personal-name display changes. This approach allows multilingual catalogues to reuse the same authority while offering readers the scripts and name forms they understand.
A Canvas is a place where content is assembled
Think of a Canvas as a bounded stage, not as an image file. For a photographed page, the Canvas may have width and height. For audio, it may carry a duration. For audiovisual material it can have spatial and temporal dimensions. Annotations then position content relative to that stage. A page translation should normally preserve the Canvas and its dimensions.
Suppose a museum opens a letter with two photographed sides. The curator adds a Spanish label to the front Canvas and a Spanish transcript annotation to it. The image is still painted on its original Canvas. If the translator creates a duplicate Canvas to hold the Spanish words, visitors may experience phantom extra pages, broken page counts or confused navigation. Parallel textual content is better associated with the correct original target.
Viewer testing should include zooming, moving between pages and selecting an annotation after switching languages. The words can look correct in a preview while a client silently loses the spatial target. An error of that kind is not fixed by polishing the translation. It requires restoring the correct relationship between the annotation and the Canvas.
Painting annotations must keep their task
A painting Annotation associates content that the viewer should render as part of the Canvas. An image painted onto a page should continue to be rendered even when a visitor selects another language. The annotation’s motivation and its target express the job it performs. They are not ordinary caption text.
An editor may see the word ‘painting’ and imagine a painting in the museum sense. Here it names an operation in the data model. Translating that controlled value into a local-language word would undermine its machine-readable meaning. By contrast, the descriptive label beside the image may be freely localized where the underlying evidence permits.
This is a useful general principle for machine-readable cultural data. The same written word can serve as a protocol term in one position and as normal human language in another. A professional translator identifies the role first. A technically literate reviewer protects the resource graph while a linguistic reviewer improves the reader’s encounter.
Supplementing annotations distinguish source text from interpretation
A supplementing Annotation may connect transcribed text, subtitles, commentary or other derived content to a Canvas. This is where translation becomes richer than a title field. A manuscript can carry a diplomatic transcription, an edited reading and a modern-language translation, each with a distinct purpose. They are not interchangeable even if all use readable sentences.
One handwritten line may include an uncertain character. The transcription should signal that uncertainty. The translation should not transform it into an established word without explanation. If two readings are plausible, a curatorial note can make the disagreement intelligible. Suppressing the ambiguity may make the display smoother but impoverishes the historical record.
For audio or video, a time-targeted annotation may connect captions to a section of a recording. Correct language is only one requirement. The target time must remain meaningful, and a visitor should be able to distinguish a translation from a verbatim transcription. The objective is not just to display text; it is to display the right kind of text at the right place and time.
Ranges create the reader’s route through an object
A Range can group and organize parts of a resource, making a table of contents or chapter navigation possible. An archival box may have folders; a manuscript may have sections; a recording may have movements or interviews. Those divisions may come from the source object, a cataloguer or an editor. The distinction should not disappear in localization.
Translate navigation labels so visitors can find their way. Preserve the sequence and references to the relevant Canvases or subordinate Ranges unless the institution is deliberately publishing a different scholarly edition. Calling a section ‘Introduction’ in English and ‘Einleitung’ in German should not route to two different portions of the book. A viewer should arrive at the same section regardless of the menu language.
Ranged navigation is also a strong quality test. Open the same manifest in two languages, select a middle chapter, jump forward, then return to the beginning. Compare the target Canvases, not only the visible menu labels. Problems here often reveal incorrect references that a simple text diff would miss.
Collection navigation is its own multilingual experience
A Collection can group many Manifests. A user searching a museum’s holdings might begin with a collection-level title and summary before ever opening an individual object. That makes Collection localization a distinct editorial responsibility. A clear translated heading can help visitors understand the boundary between a thematic exhibition and a formal administrative collection.
Do not casually alter membership to make a translated exhibition seem more coherent. Membership is a curatorial or repository relationship; language selection is a presentation preference. Some collections may contain records described only in their original language. A multilingual Collection can acknowledge that fact rather than hiding objects whose translations are incomplete.
For multi-institution partnerships, this separation matters even more. One member institution may revise an object’s English description while another offers a Spanish translation. Collection identity, member identifiers and rights remain traceable. Visitors can benefit from a translated overview without losing links to each contributing institution.
Rights and access notes should not make new legal promises
A Manifest may link to rights information and can carry human-readable statements explaining access or use. A translation of ‘may be viewed online’ must not become ‘may be reused without restriction’. A credit line may name a rights holder whose name should remain intact. A rights URI is a specific reference, not prose to adapt into a convenient local address.
Access decisions sometimes depend on a resource’s status, institutional policy or legal conditions that vary across jurisdictions. A translator should not infer permissions from a readable interface or translate a technical access message as a universal licence. When the source statement is ambiguous, the correct action is to clarify it with the responsible institution.
A visitor-friendly explanation can summarize restrictions, but it should be linked to the actual controlled rights statement and reviewed for accuracy. This is particularly important for community-sensitive heritage, unpublished correspondence, private archives and images whose access differs by user group.
Accessibility is part of faithful presentation
A translation is ineffective if the intended audience cannot read or navigate it. Screen-reader labels, heading clarity, image descriptions and keyboard behavior deserve attention alongside visible text. A well-localized Manifest should make the object understandable beyond a single visual interface. Language-specific pronunciation is one reason that accurate tags matter.
Image description is not the same as an objective inventory of pixels. A good description selects what a person needs to know in context. For a statistical graph, the takeaway may be the trend rather than the color of its axes. For a historic letter, the presence of a seal, a handwritten correction or damage to the page may matter more than its background color. Translate these descriptions with their informational purpose intact.
The user interface may be provided by an independent viewer rather than the repository. This means quality assessment should cover a real combination of manifest and viewer. It is possible for data to be conformant while a particular application mishandles a long translated title or right-to-left caption. Both outcomes are worth documenting separately.
Worked example: a handwritten travel diary
A university library owns an illustrated nineteenth-century diary with thirty photographed pages. Its Manifest label identifies the work, and each Canvas corresponds to a photographed page. The university commissions a Japanese description and a modern English transcript. The curator notices that one place name is written differently on two pages, perhaps because the traveller heard different pronunciations.
The wrong approach is to normalize both original occurrences silently and to rename the Canvas links after the Japanese page labels. That would replace evidence and risk breaking references. A better approach retains the Canvas identities, provides language-map descriptions, and attaches the relevant textual annotations to the pages. The transcript can retain the original spelling; a note can explain the uncertainty.
When testing, reviewers should open page one, page fifteen and page thirty in both languages, verify the image targets, inspect the textual alternatives and confirm that the Japanese label did not alter page order. A search result citing a Canvas must still point to that same Canvas. The benefit is continuity: teachers, readers and researchers can discuss one object across languages without arguing about which page is the ‘real’ one.
Worked example: an oral-history interview
A regional archive holds a two-hour interview in which an elder alternates between a heritage language and English. The descriptive Manifest is available in several languages. The archive wants translated subtitles but does not want the local-language speech to disappear from the record.
The key decision is to distinguish the original audio from several forms of derived text. A source-language transcription, an English subtitle translation and a French explanatory summary can each carry separate editorial responsibility. Time-targeted annotations should point to the appropriate parts of the recording. A metadata label describing the interview’s subject should not be mistaken for a transcript.
Reviewers need to play the same segment in each viewing mode and confirm that caption timing and speaker attribution remain trustworthy. They should also consider whether sounds, pauses and expressions of uncertainty have been represented. The value of the multilingual resource lies in enabling understanding while preserving the presence and significance of the speaker’s original words.
Worked example: a photograph with contested naming
A municipal museum has a historic street photograph catalogued under a former district name. A new bilingual exhibition uses a contemporary official name in its introductory text. Both names can be legitimate in their proper contexts, yet they answer different questions: what the historical catalogue says and how people locate the place today.
The translator should not silently rewrite the historic title in every metadata field. The exhibition’s summary can explain the relationship between the names. A label can retain a source title and add an understandable modern-language designation where the cataloguing policy permits it. An authority link may help connect the historic and contemporary place entities without pretending the labels are identical.
The quality check is semantic. Does the French visitor see the distinction? Does the local-language visitor understand that the earlier wording is quoted from an old record? Are geographic names separated from names of institutions that once administered the district? This example shows why good localization can be more informative than forced consistency.
Worked example: a scientific field notebook
A research institute publishes specimen sketches from field notebooks. A Canvas displays an illustration, metadata indicates collection date and site, and a supplementing annotation includes the observer’s handwritten comments. Later experts revise the species identification. A translator faces three different kinds of statement: original observation, present-day identification and explanatory commentary.
A dangerous translation would make the historical observer appear to have used a classification introduced decades later. An accurate record distinguishes the date and authority of each assertion. The original text may be translated with an editorial note about older terminology. The revised identification belongs in a clearly attributed modern description, not inside the historical transcription.
Testing should follow the path a researcher takes: open the original image, inspect the transcript, read the translation and follow any cited authority. These steps should make the evidence trail visible. Translators and reviewers add real value when they help readers understand not just the scientific term but who used it, when and with what level of certainty.
Worked example: a bilingual exhibition pathway
An art institution assembles a temporary exhibition across several collections, each hosted by a different repository. The exhibition uses a Collection for browsing, and each Manifest has its own Canvases, labels and permissions. Visitors may browse through an English overview even when a participating repository has only a source-language object title.
The curator wants a seamless experience, but seamlessness should not erase institutional boundaries. The exhibition may provide its own translated introduction while keeping member Manifest identifiers intact. If a partner withdraws a resource, the collection should respond through proper curatorial maintenance rather than by sending users to a copied object with a false origin.
This situation reveals the practical benefit of interoperable identifiers. One repository can add language-map values, another can maintain a source-language label, and the exhibition can still refer to both. Reviewers should distinguish gaps in translation from broken access. A missing French summary and a missing Manifest are different failures and require different remedies.
Worked example: a manuscript with unusual reading order
A book bound from several earlier manuscripts contains inserted pages, foldouts and a section read from right to left. Its physical pagination differs from the scholar’s preferred reading sequence. The Manifest preserves Canvas order as an explicit decision. Ranges provide structured navigation that reflects the edition’s organization. Translating chapter headings does not give an editor permission to improvise a new page sequence.
The difficult task is explaining the structure to a reader without confusing the record. A clear note might tell visitors that page numbers on the object differ from the viewer’s order. The label of a foldout can be localized, but its image remains attached to the same Canvas. A Range can make a chapter easier to find while its references remain stable.
Review the first, middle and final Canvases and follow links from the table of contents. If a chapter entry moves when the language changes, inspect the Range relationships. Translation is successful when different readers encounter the same scholarly construction, not when the interface makes every cultural complexity disappear.
Worked example: an object with uncertain attribution
An archive describes a drawing as ‘attributed to’ an artist rather than definitively created by that person. One language offers several natural constructions for degrees of uncertainty. A rushed translation chooses a short, confident name line, making the attribution appear settled. The change affects historical meaning even though the JSON structure remains perfect.
The appropriate translation preserves the qualifier. The reviewer might compare ‘by’, ‘attributed to’, ‘workshop of’ and ‘after’ as distinct scholarly claims. A label may need a few more words in the target language to retain those distinctions. A metadata value should not replace a statement of scholarly opinion with a categorical factual assertion.
A test should involve a human reader unfamiliar with the original. Ask what certainty they infer. If the answer is stronger than the institution intended, the localization has not passed. This kind of interpretive check cannot be replaced by checking field names, yet it belongs beside technical validation.
Worked example: a film reel with several subtitles
A moving-image archive publishes footage with a time-based Canvas, one original soundtrack and separate translated text. The multilingual experience depends on more than the file title. Captions must align with dialogue, readable labels must distinguish a subtitle translation from an accessibility caption, and notes may identify uncertain speakers or damaged sound.
The translator should not change Canvas duration to fit a longer target-language sentence. Instead, subtitle presentation and editorial review must accommodate the timing constraints of the source. The underlying temporal positions remain authoritative. Likewise, a list of translated scene descriptions should not be treated as a replacement for the film’s structured time references.
A practical test plays short segments at the beginning, middle and end, checks synchronization and confirms that a reader choosing another language still reaches the same moment in the recording. Technical reliability here is inseparable from comprehension: perfectly translated words shown at the wrong time are not a successful presentation.
A practical sequence for reviewing a multilingual Manifest
Begin by identifying the resource’s role: Collection, Manifest, Canvas, Range or Annotation. Read its original structure and record the identifiers and relationships that constitute its identity. Then inventory the human-facing properties that need linguistic review. Separate original titles, assigned titles, metadata captions, textual annotations, explanations and rights statements; they do not all have the same editorial status.
Next define the target audience and the actual language varieties needed. Decide how unknown terms, personal names, historical place names and uncertain readings will be treated. Where content is untranslated, retain an honest fallback rather than adding a fabricated text. Translate in context: a short field label may need a different word when attached to an artwork than when attached to an archival file.
After editing, compare the manifest’s structural references with the original. Check identifier stability, type values, Canvas order, image targets, dimensions, annotation motivations and Range targets. Use a viewer to test selection, reading direction, labels, captions and navigation in the intended languages. Human reviewers should examine a few difficult objects, not only easy representative ones. Record any unresolved uncertainty explicitly.
What a successful quality check should demonstrate
A clean JSON file is a starting point, not proof of localization quality. A successful check shows that the object is still the same object, the translated statements make defensible claims, the viewer can reach the intended material and readers do not lose important distinctions. This can be established through paired examples, comparing the same user action before and after language selection.
For structure, record whether Canvas and Range references resolve, whether Annotation bodies and targets remain valid, and whether the viewer preserves the object sequence. For language, check readable labels, script direction, fallbacks, typography and domain-specific terminology. For evidence, compare dates, attributions, names, uncertainties and historic quotations. For access, check rights statements without inferring any new permissions.
The strongest validation questions are concrete: Can a visitor switch language on page twelve without jumping elsewhere? Can a scholar cite the same Canvas in either interface? Can a blind reader identify the image’s informational role? Can a curator explain which title is original? A release passes because people can actually use it, not because a spreadsheet contains a line labelled ‘translation complete’.
Frequent mistakes and why they matter
One error is translating a machine term such as Canvas or painting where the specification expects a fixed vocabulary value. Another is altering resource identifiers to look localized. A third is converting uncertain catalogue descriptions into definite facts. A fourth is merging a transcription and a translation so that readers cannot identify what the original object says. A fifth is changing navigation order while updating headings.
Institutions also underestimate font and layout issues. A label that works in English may overflow a narrow viewer panel in German or render poorly in Arabic. A long manuscript title may require a different line-breaking strategy. If a reviewer only looks at the raw data, these problems remain invisible. Rendering tests are not decorative polish; they determine whether the translated resource can be used.
Finally, beware of treating ‘multilingual’ as a binary property. A resource may have bilingual titles but monolingual summaries, untranslated transcripts and a rights page available in only one language. Describe coverage honestly so visitors understand what is translated. The objective is a trustworthy experience that can expand over time, not a claim of complete coverage that the record cannot support.
Questions readers and collection teams often ask
Can an institution create a separate Manifest for every language?
It can publish different representations when there is a legitimate need, but language changes alone do not require inventing unrelated object identities. A common Manifest with language maps can give compatible viewers multiple human-readable options while preserving shared Canvas references. If separate editions are deliberately maintained, document their relationship and avoid implying that they represent different physical objects.
Are labels the only fields that can be translated?
No. Summaries, metadata values, navigation labels, descriptive notes and appropriate textual annotations may all have linguistic content. Each needs review according to its purpose. The crucial distinction is between human-facing meaning and structural vocabulary, identifiers or relationships.
Should every original title be translated?
Not necessarily. Historical titles, artwork titles, names, inscriptions and terms without a good equivalent may call for a source-language form, transliteration, explanatory translation or a combination. The objective is to inform the reader without rewriting the object’s record.
What happens when a language is missing?
A viewer may use a fallback, but fallback behavior is application-dependent. Test the intended viewers and make missing translations clear. Do not claim a language is supported for material that remains inaccessible or untranslated.
Is a transcript the same as a translation?
No. A transcript represents spoken or written source content, perhaps with editorial notation. A translation renders meaning in another language. They can be linked to the same Canvas, but should remain distinguishable so readers can evaluate evidence.
Can translated titles change the order of pages?
Not as a side effect of localization. Canvas order and Range references express structure. Reordering is a deliberate editorial change and should be treated separately from translating headings.
What should a small archive review first?
Start with a few representative and difficult objects: a simple photograph, a multi-page manuscript, an object with uncertain attribution and a resource with alternative text or time-based content. Check identity, navigation, language, rights and real viewer behavior before scaling up.
Does a technically valid Manifest guarantee accessibility?
No. Valid syntax does not guarantee good descriptions, correct pronunciation, usable script direction, sensible fallback or screen-reader-friendly navigation. Accessibility must be evaluated in the published viewing experience.
The principle worth remembering
Translating a digital exhibition does not mean creating a new historical object. It means opening a faithful route into the same evidence. The identifiers tell software which object to retrieve; Canvases and Annotations tell viewers where material belongs; Ranges tell readers how to move through it; language maps help people understand the surrounding words. Each layer deserves its own kind of care.
If a museum can show the same item across languages without losing page identity, provenance, uncertainty, access rules or reading structure, it has accomplished something more useful than a file conversion. It has made cultural knowledge easier to enter while keeping the routes back to the source intact.
Sources and further reading
International Image Interoperability Framework, Presentation API 3.0: https://iiif.io/api/presentation/3.0/
International Image Interoperability Framework, Image API 3.0: https://iiif.io/api/image/3.0/
International Image Interoperability Framework, Content Search API 2.0: https://iiif.io/api/search/2.0/
W3C, Web Annotation Data Model: https://www.w3.org/TR/annotation-model/
Related eduKateSG reading: https://edukatesg.com/2026/10/30/translate-like-a-pro-3480-iiif-image-api-3-info-json-regions-tiles-rights/
Related eduKateSG reading: https://edukatesg.com/2026/10/30/translate-like-a-pro-3410-mets2-packages-file-inventories-structural-maps-navigation/