Translating an EAD 4.0 archival finding aid is not simply changing English collection descriptions into another language. A successful multilingual finding aid must preserve the hierarchy of fonds, series, files and items, maintain stable archival identifiers, explain access and rights accurately, and keep the record of who created the materials intelligible to readers in different countries.
This guide explains how to localize Encoded Archival Description 4.0 (EAD4) XML finding aids for archives, libraries, museums, universities and public records offices. It examines collection titles, scope and content, dates, name authorities, controlled subjects, multilingual descriptions, digital-object links, repository codes, XML schemas, migration, validation and realistic translation review. The practical aim is accessible discovery without falsifying the archive.
EAD 4.0 was released on 31 July 2026 as part of the revised Encoded Archival Standards suite. Its digital-first direction and closer alignment with EAC-CPF 3.0 and EAC-F 1.0 matter to translation because the description of records, the description of their creators and the description of the functions that produced them are related but distinct. Translation may clarify those relationships; it must not silently rewrite them.
What an archival finding aid actually does
A finding aid is the map by which a reader understands a body of records before requesting or opening it. It explains what survives, how the records are grouped, who held them, and where important subjects may appear. EAD makes that map machine-readable as well as human-readable.
Imagine municipal planning records that contain fifty years of maps, correspondence and committee minutes. A translated heading such as ‘planning appeals’ helps researchers, but only if it still points to the correct series and date range. The descriptive label is a signpost; the identity and arrangement of the records are the terrain.
Before translating, identify which text is explanatory, which is a proper name, which is an identifier and which is XML syntax. Review each category differently. A reader may need a locally familiar term for ‘finding aid’, but a database still needs the original record code and the correct parent-child relationships.
Why EAD 4.0 deserves its own translation policy
EAD 4.0 is not just an alternative spelling of EAD3. The 2026 revision reorients the standard toward contemporary digital interoperability and alignment with related archival standards. Archives should treat the release as a new versioned contract rather than assuming every previous element, attribute and export convention will work unchanged.
A repository may retain EAD3 records for years while publishing new EAD4 descriptions. Translation teams need to know which version each source actually uses, which schema the publication pipeline validates against, and whether the public display comes from transformed XML or from an independent editorial database.
Record the version and schema location in project documentation, not in improvised editorial text. Do not describe an EAD3 export as EAD4 because its prose was refreshed. If migration is planned, perform structural migration and its validation separately from translation so responsibility for an error remains traceable.
The difference between the records and their description
The archive contains documents, recordings and other records. Its finding aid contains assertions about those records. Translation changes the language of some assertions; it does not change when a document was created, which organization produced it, or whether a box exists. This distinction prevents a subtle but common class of catalogue errors.
Suppose a letterhead uses an obsolete institution name. A translator might prefer today’s institutional name for readability, but replacing the historic form inside a title can erase evidence of organizational history. A useful description may show both the historical name and a modern explanatory equivalent, explicitly distinguishing them.
Keep a field-level map showing original text, approved target-language text and contextual notes. When a translation changes interpretation rather than expression, pause for archival review. The record’s identity, dates and provenance must remain supported by source evidence rather than by the translator’s expectations.
Collection hierarchy is more than a visual indentation
Archival description commonly moves from a collection or fonds into series, files and individual items. The levels communicate how records were accumulated and organized. EAD encodes this nesting so a website can create a browsable tree, export an inventory or connect an item to its parent.
Translating a series label as though it were a collection title can make fifty subordinate files appear to be independent deposits. Likewise, promoting an attractive item description into the main collection summary can exaggerate its significance. Preserve the nesting and check whether inherited information still applies to every child.
Validate the number and order of components before and after publication. Open the public viewer at a deep component and walk upward to the parent. The user should see the same archival family even when the displayed words are different. Translation quality here is structural, not merely grammatical.
Call numbers, reference codes and persistent identifiers
A repository’s reference code is a retrieval instrument, not an ordinary phrase. It may include the repository code, fonds number, series code and item number. Repositories can also expose persistent URLs or other identifiers. These values allow requests and citations to land on the intended material.
A user searching for ‘AC/12/4’ may be referring to a folder in a records centre. Replacing the letters with an abbreviation of the translated title would manufacture a new call number. Even adding language-specific punctuation to the identifier could make retrieval fail or create a false duplicate.
Keep exact machine identifiers in protected fields and show translated explanatory labels beside them. Test copied reference codes in the catalogue and request workflow. Where transliteration is needed for a human-readable name, label it as a display variant and do not substitute it for the accession or reference number.
Titles need both fidelity and discoverability
A collection title may be formally assigned, transcribed from a source, supplied by a repository or derived from the predominant creator. Each origin affects what a translator may responsibly change. The best translation is one a reader understands without mistaking an editorial title for wording on a historic document.
Consider ‘Papers of the Harbour Improvement Board, 1927–1964.’ An international reader may need an explanation of the board’s function, but the official corporate name should still be recognizable in the authority record. A translated title and a note identifying its English source can work together.
Ask whether title language, authority names, capitalization conventions and date display are separate fields in the archive system. Prefer controlled bilingual alternatives over overwriting a sole title string. Evaluate search results using original names, translated variants and known repository references, not only the translated main heading.
Language of the materials is not the language of the interface
A finding aid can be written in French while describing letters written in Arabic, English and Tamil. These are different facts. Metadata may indicate the language of description, the language or script of the archival materials, and the language of particular contextual notes. Collapsing them produces misleading filters and access expectations.
A catalogue reader choosing Spanish as the interface language should not be told that the manuscripts themselves are Spanish. Nor should a translated abstract imply the documents have been translated. The interface may be accessible even when the underlying handwritten record remains in its original language.
Document these distinctions explicitly in the target publication system. Inspect language-specific display and indexing rules, especially when several scripts occur in one collection. Verify that source-language codes do not change as a side effect of localizing a summary or a menu label.
Dates describe events, not editorial convenience
Archives often distinguish creation dates, accumulation dates, publication dates, transfer dates and coverage periods. Finding aids also include approximate or uncertain dates. A translator should preserve the event and the degree of certainty before changing how a date is displayed for a locale.
An entry described as ‘circa 1910–1925’ is not equivalent to ‘1910–1925’ without qualification. A collection transferred in 1984 may contain photographs made in the 1930s. Conflating accession date with creation period changes how historians interpret the evidence.
Keep normalized values machine-readable and preserve qualifiers in translated display text. Review calendars, day-month ordering and open-ended spans separately. If the source date appears internally inconsistent, flag it rather than fixing it in translation without confirmation from a records specialist.
Extent and container quantities must remain auditable
Archival extent can combine linear metres, boxes, folders, pages, gigabytes and counts of objects. Those units are not interchangeable. A translation may explain a local measurement convention, but a converted number needs a defensible calculation and a label showing its relationship to the original.
If a finding aid says 18 boxes, the translator cannot assume a particular number of folders or cubic metres. A ‘file’ may mean a physical folder, a unit of intellectual arrangement or a digital object depending on context. A translated word that merges these senses will mislead users.
Separate the measured quantity from the editorial explanation. Retain the exact inventory count and show conversions only when the repository approves them. Compare totals at collection and component levels so a small translation change does not become an inaccurate claim about the size of the holdings.
Scope and content notes carry intellectual boundaries
A scope and content note says what kinds of material are present and what subjects or activities they document. It should not claim to provide an exhaustive account if the archive is selective, incomplete or unevenly catalogued. The translator’s job is to preserve the writer’s measured degree of certainty.
For example, ‘includes correspondence about resettlement’ is narrower than ‘documents the whole resettlement programme.’ A polished translation that uses the broader phrase may sound more confident yet become historically false. Names of communities and technical topics can also be sensitive to local history.
Translate observable content before generalizing its significance. Keep hedges such as ‘primarily’, ‘possibly’ and ‘selected examples’ where they carry evidence limits. Ask an archivist to review terms that might shift the population, date range or institutional function described by a note.
Arrangement statements explain how evidence is ordered
Researchers need to know whether material follows original order, a reconstructed filing system, chronological arrangement or a repository-imposed sequence. These are different archival assertions. A foreign-language finding aid should explain the sequence clearly without pretending the repository’s present order is necessarily the creator’s original order.
A municipal records office might organize maps by project number, while a later archivist arranges stray correspondence by year. If both are described as ‘original order’, readers can infer an administrative history that the records do not support. The translation must preserve who arranged what and when.
Use precise verbs for ‘retained’, ‘rearranged’, ‘reconstituted’ and ‘listed’. Check series navigation against the actual order shown on screen. When a label changes but position does not, verify that its explanatory note still makes sense in the surrounding tree.
Provenance and custodial history should not be smoothed over
Provenance describes origins and relationships of creation, accumulation and custody. Custodial history may include transfers, gaps, reorganizations and changes of ownership. These descriptions sometimes contain unresolved questions. Translating them into a simple uninterrupted chain can erase precisely the uncertainty scholars need to see.
Imagine a collection attributed to a family office but received through a third-party dealer with incomplete acquisition documents. ‘Received from’ is not the same as ‘created by’ or ‘owned continuously by’. Words that look interchangeable in everyday prose have different consequences in archival description.
Build a short terminology table for creator, donor, depositor, custodian, owner and rights holder. Preserve each relationship in the translation and cross-check against acquisition records if available. Unknown links should remain unknown, not be filled by confident prose.
Creators should link to authority descriptions
The creator of a record can be a person, corporate body or family, and the same entity may have several names across time or languages. EAD records can refer to contextual authority descriptions, including EAC-CPF. The link preserves identity when a displayed name is translated, transliterated or modernized.
A museum’s local-language interface might call a historic trading company by a familiar short name while the archival authority record retains a longer official name. Without the authority link, the search engine may treat them as unrelated creators or merge two entities that happen to share a similar translation.
Keep stable authority identifiers and approved name variants attached to the record. Check one-to-many relationships carefully, particularly where a family and a business share a surname. Search and display should help people discover related materials without making unsupported claims that identities are identical.
Functions explain why an institution produced records
EAD 4.0 is part of an archival suite in which EAC-F describes functions and activities. This gives the translator a useful distinction: a government department is an agent, permitting construction may be a function, and permit files are records. Translating all three as though they were organizational names obscures the administrative story.
A department may be renamed but continue the same function, while one historical department may have carried several functions over time. A multilingual catalogue should not infer that a name change automatically created a new function or that similar activity labels prove the same institution created all associated files.
Use a stable function identifier where the repository maintains one. Preserve temporal relationships and any recorded changes in responsibility. Explain the function in ordinary target-language prose but avoid silently changing the official classification or its relation to creators and collections.
Access conditions need operational precision
Access notes can state whether materials are open, restricted for a period, available on request, subject to privacy review, or accessible only through a particular facility. Those are operational rules for readers. A vague translation may cause someone to travel, request sensitive records or rely on information the repository cannot provide.
‘Restricted until 2032’ is different from ‘may be reviewed in 2032’, and ‘consultation by appointment’ is not the same as ‘freely available online’. Distinctions become especially important when the audience includes students, remote researchers and members of communities represented in sensitive records.
Translate the exact condition and its actor, deadline and exception. Check the published wording against current repository policy before release. If a rule has changed, update through the responsible policy process rather than silently treating translation as authorization to revise access terms.
Rights statements and licences are not equivalent
A finding aid can describe copyright, ownership, reuse restrictions, donor conditions and institutional permissions. These categories may overlap but do not collapse into one legal assertion. Localizers should avoid applying familiar licence language to a record that has no documented open licence or permission.
A public archival website may permit online viewing while reproduction requires written approval. Translating ‘available for consultation’ as ‘free to reuse’ changes the reader’s legal expectation. Descriptions of public-domain status may also require careful attention to jurisdiction, creator and publication history.
Keep the rights holder, basis, scope and date of a restriction separate when the data model allows. Consult legal or institutional specialists for disputed cases. The translation should make the recorded position clearer, not provide a new legal opinion in the name of readability.
Sensitive descriptions require accountable editorial choices
Archives may hold material involving colonial administration, persecution, forced migration, disability, medical history or contested identities. Some historical terms are offensive today; others are important evidence of how a document was originally classified. A translation must balance faithful description with transparent, respectful access.
If a historic title contains a harmful term, replacing it without notice may hide evidence, while displaying it without context may harm readers. Repositories may choose a supplied descriptive title, a content note, a warning and a transcription of the original term in separate fields rather than forcing one string to do every job.
Follow a documented descriptive policy and consult relevant communities where appropriate. Distinguish quoted historic wording from modern repository language. Evaluate both full-text search and visual display so warnings and alternate terms do not inadvertently erase discoverability.
Digital objects and their links must survive translation
A finding aid may link to scanned documents, image services, sound recordings, transcripts and external repositories. The links are structural references and often include identifiers, technical paths or parameters. They should not be localized as free prose even when the visible link label is fully translatable.
Consider an image attached to a file-level entry. Replacing the English text ‘View digitized folder’ with a French label is helpful; translating characters in its persistent URL is normally disastrous. A redirect may mask the error temporarily, but broken targets can emerge when the source system changes.
Check every class of link, including downloadable files and third-party digital object viewers. Verify that the object displayed belongs to the correct archival component and that translated captions do not imply the digitized subset equals the entire physical collection.
Metadata about the description itself matters
Archival records often include who prepared the finding aid, when it was revised and which cataloguing practices were followed. These are provenance statements about the description, not the archival material. They let future users understand why language or organization has changed.
A collection may have been described in 1998 and translated in 2026. The translation date must not replace the date of the original archival description or the creation dates of the records. The institutions responsible for description and translation may be different, and that distinction should remain legible.
Record editorial responsibility, translation version and review outcome in appropriate administrative fields or workflow records. Maintain the original description and published language versions in a way the repository can audit. A changed text should be attributable rather than anonymous.
Subject vocabulary and classification are controlled evidence
Subject headings often draw from controlled vocabularies, local taxonomies or international authority files. The terms can have identifiers, broader and narrower relationships, scope notes and history. Localizing a displayed label does not authorize replacing the underlying concept with a nearby but different subject.
An old record may be indexed under ‘harbour works’ while current users search for ‘port infrastructure’. Search expansions can connect the two, but the original subject identifier may encode a specific scope. Translating one term into another without checking that scope can distort both discovery and statistics.
Retain stable concept identifiers and source vocabulary provenance. Add approved preferred labels and synonyms in target languages. Inspect a sample of searches for precision and recall, including cases where everyday words have multiple disciplinary meanings.
Places, jurisdictions and evolving names
Place names may be written in different scripts, governed by different authorities and changed over historical periods. The place named in a nineteenth-century file is not necessarily identical to the current municipality bearing a related name. Translators need to know whether a field records historical wording, a modern geographic entity or an access point.
A repository might preserve a former district name in a record title and add a modern place heading to help current users. Converting every old reference to the modern name will make the record look anachronistic. Doing nothing may make it difficult for international readers to find the material.
Separate the transcribed name, preferred access label and geographic authority identifier. Include historical context where needed. Test that maps, subject indexes and translated descriptions all point to the intended place without claiming that historical boundaries were unchanged.
Names across scripts need a repeatable rule
Personal and corporate names can have original-script forms, conventional English forms, transliterations and translated institutional names. These may be parallel labels for one authority, or they may signal distinct historical entities. The difference should not be decided by a general-purpose machine translation engine.
An archive cataloguing family correspondence in Chinese may show characters, a romanized form and a local-language explanatory note. If the romanization scheme changes from record to record, researchers may think the collection contains different people. If a family name is translated semantically, the error is even more serious.
Agree on source-script retention and romanization policy before editing. Record uncertainties in notes rather than manufacturing a single authoritative spelling. Test name search across scripts, diacritics and alternate forms, especially for materials shared between institutions.
Mixed-language notes need segment-level reasoning
Not every note can be assigned a single language. A German finding aid may quote an English document title, cite a Latin motto and describe correspondence in Polish. The translation process should preserve quotations and titles appropriately while giving readers a clear account of each segment.
A specialist might translate the narrative explanation but keep a letter’s original title in quotation marks. Another might add a bracketed translation for access. These choices are acceptable only if the repository’s conventions are consistent and the reader can tell where original wording ends.
Review short language-bearing spans rather than assigning one global language to the document. Ensure that fonts, bidirectional layout and search indexing work for all scripts. Never treat a change of script as permission to normalize identifiers or remove diacritics from a cited source.
XML markup is a data model, not a sentence
EAD4 is encoded in XML. Its element names, namespaces, required attributes and content model belong to the schema. They are not translated because they happen to resemble English words. Human-readable element values may need translation, but their technical wrappers must continue to validate.
A translator handed a raw XML file might see labels in both element names and text nodes. If they translate the tag names, a parser may no longer recognize the document. If they insert typographic quotation marks into an attribute, they may break its syntax. A translation can be fluent and still fail as XML.
Use extraction rules that expose only approved language fields to linguists, and protect identifiers, technical references and markup. After reintegration, perform XML well-formedness checks, schema validation and a human review in the actual public display rather than trusting the exported text alone.
Validation can reveal structural errors but not historical lies
Schema validation checks whether an XML instance follows technical constraints. It cannot determine whether a scope note is historically accurate, whether a name translates properly or whether a restriction is legally appropriate. Both machine tests and expert judgement are necessary.
A migrated EAD file may parse successfully while assigning a translated title to the wrong component. Conversely, a historically sound prose translation can fail because an exporter omitted a required technical field. Treat the two failures as separate problems and route them to the right owners.
Keep validation reports with versioned inputs. Run schema and Schematron checks where supported, compare hierarchy and reference-code inventories, then conduct bilingual editorial review. Success means the content remains meaningful and the XML remains usable, not simply that an automated tool reported zero errors.
Migration from EAD3 must not be mistaken for translation
Many archives retain EAD3 finding aids and will not replace them at once. A migration project may map old elements, normalize identifiers and change the handling of legacy mixed content. A translation project may add multilingual labels. Running both transformations together without a log makes regressions difficult to diagnose.
Imagine a title that disappears after an EAD3-to-EAD4 conversion. A translator might assume the source omitted it and invent a summary. The real fault may be a migration mapping. Without separate test baselines, the invented summary can conceal data loss from the technical team.
Create a frozen source inventory and test migration independently. Compare content counts and structural relationships before localizing. Use the official EAD4 tag library and best-practice resources for the target model instead of relying on informal substitutions between similarly named elements.
Interoperability requires considering who consumes the feed
Archives publish finding aids to local websites, national catalogues, research portals and data harvesters. Each consumer may expect certain fields, identifiers or display conventions. A multilingual export should be prepared for the real receiving systems rather than only the editorial preview.
A record that displays correctly in one institution’s interface may still lose accents in another search index or drop translated summaries during harvesting. A third-party portal might display only one language when it cannot interpret a custom extension. Readers may then see inconsistent descriptions across platforms.
Test representative records through actual interchange routes. Document supported language tags, fallbacks and mapping limitations. Prefer standards-aligned approaches and agreed extension profiles, and do not promise full multilingual interoperability without inspecting a downstream consumer.
A worked accession example reveals the boundaries
Consider a deposit of school board minute books, planning sketches, attendance registers and correspondence from 1945 to 1978. The finding aid has a collection code, four series, eleven folders, a donor statement and several restricted student records. A target-language edition should translate explanatory notes while preserving the institution’s original name, provenance and access rules.
The first series contains minutes; the second contains building plans. A reader searching for school architecture should find the plans, but the series order must not be rearranged to prioritize the more popular topic. Dates and folder counts remain factual inventory. Restrictions on student records do not vanish because their notice is translated.
An effective review compares the original and translated hierarchy side by side, follows at least one digital object link from each series, checks restrictions at folder level and verifies that the published viewer offers the right title variants. The exercise makes invisible metadata choices visible.
A translation review should include multiple kinds of readers
A linguist can catch false friends, idiomatic errors and script problems. An archivist can identify provenance mistakes and arrangement claims. A digital repository specialist can confirm schema validation and link behavior. A community adviser may flag harmful descriptions or missing cultural context. None of these perspectives substitutes for the others.
A polished sentence about a restricted personnel folder may be understandable yet operationally wrong; an XML validator will not notice. A technical specialist may preserve the folder hierarchy perfectly while allowing an offensive modern description to remain unexplained. Review must be distributed across the kinds of harm and confusion that can occur.
Define clear acceptance criteria for every review role. Capture disagreements rather than burying them, and identify who is empowered to approve descriptive revisions. For large collections, sample by language, period, level, restriction type and object format instead of checking only the most visible records.
A practical prepublication release checklist
Before release, compare source and target record counts, component IDs, reference codes, date intervals, creator links, subject terms and digital-object destinations. Then check translated titles, arrangement notes, access statements, rights descriptions and sensitive terminology. A translated catalogue can fail on either side of this boundary.
Choose representative records: one collection-level overview, a deeply nested item, a restricted file, a collection with mixed languages and a record containing a digital object. Open each in the public interface and confirm that navigation, search and requesting still work. Review with keyboard navigation and accessible text presentation in mind.
Save the source XML, target XML, validation output and approval record as separate artifacts. If the public export differs from the signed-off working file, stop the release and investigate the transformation. The final result should be readable for global audiences and accountable to the original archival evidence.
One final bilingual catalogue stress test
Picture a researcher in another country attempting to trace a displaced village through a multilingual archive. The researcher enters a present-day place name and follows a subject link to a set of historic surveys. A successful target-language finding aid makes the collection understandable without implying that the current municipality existed under the same boundaries when the surveys were created. The original survey titles and dates still matter. The repository’s collection code must survive copying into an email to the reading room, and the digitized plan should open from the correct component. A clear translation does not collapse a geographic authority link, a historic transcription and an editorial explanatory note into one unqualified sentence.
Now repeat the test from a different entry point. Ask a family historian to search an original-script personal name, choose the translated description, and navigate to a folder with an access restriction. The reader should not lose the name variant, be led into a different family’s papers, or see a misleading promise that restricted correspondence is available for immediate download. Repeat the check for a screen reader and for a narrow mobile display, where breadcrumbs, long names and warning text may be truncated. If the archive passes these tests, its localization is serving real research rather than merely producing attractive text.
Common questions
**Should the reference code be translated?** No. Keep reference codes and other machine identities as stored. Translate explanatory labels around them.
**Can translated finding aids change collection titles?** They may provide a properly identified translated or supplied title. A translated label should not erase the transcribed or officially assigned source title.
**Is EAD4 identical to a digitized archive?** No. EAD describes the material and its organization. Digital surrogates may be linked, but the finding aid does not guarantee all holdings are online.
**Will XML schema validation prove a translation is accurate?** No. Schema validation checks structure and constraints. Bilingual archival review is still required for semantics, history and access conditions.
**Can EAD3 holdings remain online during an EAD4 rollout?** Yes, if systems and public communications state their versions accurately and migration claims are not made without evidence.
Further reading
Society of American Archivists / Library of Congress: https://www.loc.gov/ead/v4/EAD4-TL-eng.html
Encoded Archival Standards Suite: https://saa-sdt.github.io/EAS-TagLibraries/
Encoded Archival Standards Best Practices: https://saa-sdt.github.io/EAS-Best-Practices/
Related eduKateSG coverage: https://edukatesg.com/2026/10/30/translate-like-a-pro-3410-mets2-packages-file-inventories-structural-maps-navigation/ and https://edukatesg.com/2026/10/30/translate-like-a-pro-3400-mods-38-library-metadata-titles-names-language-subjects-related-items/
Final principle
Good archival translation expands who can find, interpret and request records. It does not manufacture a cleaner past, a new identifier, a different collection or an access right the repository never granted. EAD4 works best when translated language, machine structure and archival responsibility remain visible as three complementary parts of the same public record.