How do you translate an E-ARK information package for international partners without breaking the archive that receives it? E-ARK CSIP, the Common Specification for Information Packages, defines shared packaging conventions so digital materials, metadata, documentation and representations can move between systems with reliable relationships. The translation challenge begins when names that look descriptive are actually structural signals. If an editor changes a folder label, a METS reference or a controlled value because it sounds awkward in another language, a receiving repository may no longer recognize the package correctly.
People searching for E-ARK CSIP translation, CSIP 2.2.0 localization, archival package METS XML, information package structure, representation folders, xlink:href file references, submission information packages, archival information packages or multilingual preservation metadata need an approach that protects structure first. This guide distinguishes machine-controlled paths from translateable descriptions, explains what validation can and cannot prove, and shows how a package can offer a new language without pretending that its original records were written differently.
The intended reader might work in a national archive, university, museum, government office, commercial records service or cross-border research partnership. You do not need to become a preservation-system engineer before reading it, but you do need to distinguish content, context and responsibility. We will start with a plain-language model, move through package rules and source-to-target decisions, then work through practical handover cases. The purpose is to make records understandable across languages while keeping their chain of custody, technical meaning and internal references testable.
A package is more than a folder of documents
An information package combines digital content with enough description and structure for another institution to understand what it received. The files may be PDFs, images, messages, databases or datasets, but the package also needs identification, relationships, metadata, manifests or equivalent control information. The recipient should be able to distinguish original content from later derivatives and determine whether something is missing. Translation therefore affects several layers: words in the documents, descriptive metadata about the documents, human operating instructions and the technical package that binds everything together.
Imagine a box of records arriving with a multilingual contents sheet. Translating the sheet can help a receiving archivist; renumbering the box labels without updating the authoritative inventory can cause confusion. A digital package is less forgiving, because software can use exact XML values and relative paths to locate each file. A fluent title is not a substitute for a valid reference. Translation should make the structure legible, not rename the underlying evidence merely to make it look tidy.
What the CSIP specification contributes
E-ARK CSIP is a common set of requirements for the structure and metadata of information packages. It builds upon METS conventions and defines how packaging information is represented, including the use of structural maps and links to files and representations. E-ARK SIP, AIP and DIP specifications address more particular stages or purposes, while CSIP supplies common foundations. A local implementation may impose additional obligations; compliance with a shared specification does not mean that every recipient has identical ingest policies.
The DILCIS Board publishes the CSIP materials and maintains their version history. Its release listing identifies version 2.2.0, but repositories can operate different versions or profiles at the same time. Confirm the profile agreed for the transfer instead of assuming that a current website automatically describes every old package. For translation decisions, the significant distinction is between a rule that a recipient must recognize and a readable description that can vary by language.
Four layers of language in one transfer
The first layer is preserved content: the letter, spreadsheet or recording being transferred. The second is descriptive metadata that tells us what an object is and may contain natural-language titles, scopes, notes and names. The third is packaging structure such as file paths, METS element names, identifiers, controlled tokens and technical relationships. The fourth is human handover communication: transfer agreements, instructions, acceptance notices and user-facing catalogue text.
These layers are related but not interchangeable. A translated collection title can improve discovery without changing a package identifier. A translated transfer note can explain an exception without editing an original document. A structural value such as an XML element name must remain the value expected by the schema. Before work begins, mark each text-bearing field with its layer and intended use. Doing this prevents the common error of treating every visible string as if it were ordinary prose.
METS file sections, structural maps and representations
METS can describe files through fileSec and groupings, and express relationships through a structMap. CSIP constrains particular packaging uses, including the structural map labelled CSIP. Representations can encapsulate content arranged in different forms, for example original office documents alongside a normalized access representation. Their identities and relationships matter more than how conveniently an editor can pronounce a folder name. A multilingual public description should refer to representations without changing the pointers that identify them.
Consider a representation containing scanned page images and another containing text extracted from those images. The translated introduction may explain that the first preserves page appearance and the second supports searching. But if the references point to the wrong representation, the recipient can confuse derivative text with the source image. A good translation walks the reader through the relationship while the actual METS mapping remains unchanged and separately validated.
Package paths, filenames and controlled values
A path in a package is an address within a defined structure. It may appear human-readable, yet changing its letters or case can break the address, particularly on case-sensitive systems. CSIP-related conventions can attach specific meanings to folders or labels, so do not translate recognized structural names without a planned migration and a validator-backed mapping. Treat FLocat references and xlink attributes as technical relationships, not captions. A relative path is not a sentence; it has to resolve from a specific location.
Free-text filename components are a harder case. A collection may contain personal names and language-specific titles in filenames created by the donor. Preservation policy may require them to remain unchanged. If a repository publishes a translated display name, record it as metadata and keep it linked to the actual file. Avoid promising that Unicode normalization, case conversion or transliteration is harmless. Different tools and filesystems can interpret those operations differently.
Preservation metadata, technical metadata and rights
Descriptive metadata explains an object to people. Technical metadata may record formats, sizes, checksums or extraction details. Preservation metadata may describe events, agents and outcomes over time. Rights metadata communicates permissions and constraints. Translators have meaningful work in each area, but not the same freedom. Translate explanatory notes carefully; do not rewrite hashes, format identifiers, event times, controlled event types or persistent identifiers as though they were ordinary vocabulary.
Rights require particular restraint. A statement that a file was deposited under certain terms is not proof that public translation or republication is allowed. A multilingual access copy may need a separate rights review. A correct package can contain material that cannot responsibly be placed on an open website. Technical validation and legal authorization answer different questions, and the transfer documentation should not blur them.
How to handle the receiving institution’s expectations
Every transfer needs an agreed interpretation of which packages, profiles and local extensions the receiver accepts. Before translation, ask what counts as the authoritative source, which metadata fields are editable, which validation tool will be used, and which language versions are expected. A donor’s phrase for ‘series’ might not have a one-to-one match with the receiving archive’s arrangement vocabulary. Solve that ambiguity in a shared terminology note rather than changing the structural hierarchy to suit an English label.
Acceptance should be measured through independent tests. A document can be linguistically impeccable but fail because the METS references an absent file. Conversely, a package can validate structurally while giving a reader a misleading title or an inaccurate rights summary. Both tests matter. Keep a record of who approved textual interpretation and who approved package integrity; that separation clarifies accountability when a future audit asks what changed.
Practical translation and validation sequence
Inventory the original package without editing it. Identify the CSIP version and applicable SIP, AIP or DIP profile. Export the file list and relationships, record integrity information, and classify every proposed translated field as source content, descriptive metadata, documentation or machine structure. Translate the human fields in a derivative or permitted metadata area, retaining source language and provenance. Then validate both package structure and linguistic accuracy against the original.
Choose pilot cases that represent complexity rather than convenience: a package with two representations, one with non-Latin filenames, one with rights restrictions and one with a complicated hierarchy. Test file links, METS constraints and counts on a clean copy. Make sure an archivist who does not know the source language can identify the content, understand the limitations, and locate the actual evidence without guessing which of several translated titles is a file path.
Worked translation cases and package handover decisions
A municipal register entering a national archive
A city transfers a register containing handwritten names, digital scans and an index created by the local office. The destination archive uses another language and wants searchable titles for the records. The donor’s existing directory names and METS references are part of the submitted evidence.
Translate series titles, descriptions and explanatory notes into recipient-facing metadata. Keep identifiers and file addresses fixed. The index should state whether a name is transcribed, transliterated or translated; those are different editorial acts. Do not replace a handwritten name with its guessed modern equivalent.
Acceptance should show that each scan still resolves from the package inventory, the original register remains identifiable, and the new-language catalogue description cites its precise source. A useful summary does not require changing the deposited object.
Two versions of a research report
A research project deposits a draft and a signed final report. Both have similar titles, but the final contains approved figures. An editor creating multilingual descriptions prefers a single title because the distinction feels cumbersome in search results.
Keep the version relationship explicit. Translate each title with an unambiguous label for draft or final, preserving the exact version identifiers and file assignments. A reader should not mistake the translated draft for the released document simply because the target language has a shorter title.
Compare the original and translated inventory at the file level. Verify dates, creators and representation references, and ask whether the translated public interface still allows the reader to request the precise version cited in an academic reference.
A relative METS file path breaks after translation
An XML reference points to a file under a known relative directory. Someone renames a folder into the target language and leaves the FLocat value unchanged. The description reads beautifully, but the receiving system reports that the referenced content is missing.
Restore the structural path and put the translated folder explanation in a separate description. If a migration genuinely requires renaming, treat it as an engineered package transformation with a complete mapping and new validation—not as a copy-editing preference.
Run a path-resolution check from the package location, confirm every referenced file exists and compare digests to the source. A zero-error language review cannot rescue a broken path.
A package contains original and access representations
An archive receives source TIFF scans and compressed JPEG access images. A translator labels both representations ‘original pages’ because both visually show the same document. The wording conceals their different preservation roles.
Describe the TIFF representation as the preserved source if that is the institution’s designated role, and the JPEG representation as a derived access set. Avoid assumptions based solely on extensions: confirm the repository’s documented provenance and processing history.
A reviewer should be able to trace each representation to its files and know which copy was transformed, by whom and for what purpose. The translated text must not upgrade a derivative into primary evidence.
A non-Latin filename that looks corrupted
A donor’s folder uses accented characters and a right-to-left script. A recipient’s file browser shows irregular order or replacement symbols. The localization team assumes that transliterating every filename will repair the transfer.
First inspect byte representation, normalization and tool behavior. Preserve the deposited names as required by policy and create display labels or a carefully controlled mapping if access tools need alternatives. Transliteration might be useful to humans but is not a universal fix for path compatibility.
Test extraction on the destination environment, inspect referential integrity and confirm that the package inventory still resolves. An issue in display software should not be disguised by silently editing records.
A rights note is too broad in translation
A donor agreement says files may be used for on-site research under certain restrictions. A hurried translation renders this as open access for research, inviting international users to download and redistribute the materials.
Revisit the exact wording and the applicable agreement. Translate rights conditions with their limits, not with the broadest attractive synonym. Where legal concepts have no clear equivalent, provide a qualified explanatory note rather than a false jurisdictional match.
Legal and archival reviewers should approve the target text independently of package validation. The descriptive field must not promise rights that the repository cannot grant.
A separate translation file is added to the package
A partner wishes to add translated captions directly into an already accepted information package. The additional file appears harmless, yet package inventories, checksums or representation lists may not include it.
Treat the addition as a new documented derivative or package version according to the agreed lifecycle. Put the translation in an appropriate permitted location, update required references and integrity information through authorized package tools, and keep the old package identifiable.
Validate the new package as a whole. The original record should remain retrievable and the new translation should carry language, source and responsibility information rather than appearing as an unexplained original file.
A checksum mismatch after a transfer
A repository verifies a transferred package and finds a digest mismatch in one object. The translator says only human-readable labels were changed, but the file that now fails was also passed through a text editor.
Do not normalize the failure by recalculating the checksum against whatever file is present. Investigate the exact bytes that changed, compare original and received objects, and document whether a transformation was authorized.
Preservation validation should conclude with a clear distinction among unchanged evidence, intentional derivatives and unexplained changes. A signed-off translation is no substitute for a verified handover.
An old package points to a newer CSIP website
An archive receives a package created years earlier. Its profile reference leads users to a current specification page, and an international reviewer assumes the package must pass the latest published version’s exact rules.
Identify the producing software, declared or documented package version, agreement and relevant profile. Validate against the applicable requirements, recording any compatibility judgment. The version history is not a word to translate but a basis for interpreting the handover.
A fair acceptance note distinguishes a valid older package, a migration candidate and a genuinely malformed package. Otherwise a localization project can mistakenly report historic material as defective simply because standards evolved.
A digitized oral-history interview
An archive stores an original audio recording, a source-language transcript and a translated transcript. The translated file uses different sentence boundaries and timestamps. The receiving team wants to place all three under one vaguely named representation.
Identify each item by its actual role and relationship. Preserve the original audio, keep timestamp conventions clear and mark the translated transcript as a derivative. If transcript alignment is approximate, say so rather than implying word-perfect synchronization.
Test that an archivist can locate the source audio from a translated quotation. A good package supports reading and auditability without pretending that text and sound have identical timing or evidential status.
Metadata encoded in a locally defined extension
A local government producer uses an extension schema for offices and administrative functions. It contains labels in one language alongside controlled identifiers. A target repository has no direct equivalent for some office names.
Preserve the extension’s structural identifiers while translating the human descriptions with an explicit crosswalk note. Do not invent replacement element names or unsupported local vocabulary codes to make the XML sound familiar.
Confirm that the schema and any referenced controlled vocabularies are available to the recipient. Validate the extension separately and record which concepts were approximated, translated literally or kept in the source language.
An organizational name changed after deposit
A corporation donated records under an earlier legal name. Years later the company merges and uses a new name. A translator working on the access description wants every reference to use the currently recognizable identity.
Keep the depositor and creator identities at the times evidenced by the package. An explanatory note can connect historical and current names, citing the relevant authority record or verified source. Avoid treating a change in display vocabulary as a change in past custody.
Check whether the package’s creation, transfer and later enrichment events remain separate. Future researchers may need to distinguish the record’s original producer from a modern successor.
A multilingual court file with incomplete pages
A court transfers a scanned case file whose page numbering jumps near the end. An editor decides to smooth the translated summary so it reads as a complete record, assuming the omission is merely clerical.
Record the gap and describe what is verifiably included. Do not reconstruct missing testimony or imply that a ruling is present when it is not. If another collection provides related pages, identify that as a distinct source and document its relation.
Compare item counts and page evidence against descriptive metadata. Package validation can confirm that every submitted file is present, but it cannot by itself establish that the court’s original paper file was complete.
A translated README used as acceptance guidance
A complex package includes human instructions describing expected file order and naming patterns. A translator modernizes the examples and changes an identifier in the prose to make it intuitive for an international partner.
Examples in a handover document can be operational instructions. Translate ordinary language, but verify tokens, paths and commands against the actual package. If a translated example must be illustrative rather than executable, mark it clearly.
Have a recipient follow the translated instructions in a test environment. If the directions lead to nonexistent files or conflict with METS, the documentation has failed even if every sentence is grammatically elegant.
Confusing SIP, AIP and DIP
A partner calls every transferred package an archive package. In a bilingual contract, the translator uses one target-language phrase for submission, long-term archival storage and dissemination copies.
Use terminology that preserves the distinct functions of submission, preservation and dissemination. The same intellectual object can pass through different package roles, but the roles are not synonyms. Name the stage in the workflow rather than assuming a generic word conveys it.
Review contract definitions, package profile references and acceptance criteria. The user should be able to identify which package was received, which was preserved and which was distributed.
A database package with readable table names
A relational database export includes technical table names that resemble English phrases and user-facing column descriptions in another language. The translation team wants to replace all names for clarity.
Do not modify database identifiers unless the approved migration plan includes schema mapping, queries, relationships and validation. Translate explanatory data dictionaries and public descriptions separately. This also applies when a field name is an abbreviation whose meaning was defined by the original application.
Verify that every table, relationship and schema reference still resolves. Where a query example is translated, check whether its identifiers are literally executable or are explicitly labelled as pseudocode.
An image collection with two descriptive languages
A museum submits paintings with source captions in Spanish and English, but some English captions differ in interpretation from the Spanish originals. A third-language translation is being prepared.
Preserve both supplied language versions and their provenance. Do not assume one is a mechanically exact equivalent of the other. The new translation should identify which text it follows and explain any consequential discrepancy.
A review sample should compare author names, dates, materials, uncertainties and rights. The preservation package can hold several descriptions; the public interface should not suggest that they all carry identical editorial authority.
An access copy with redacted personal information
A local archive creates a dissemination version of case files in which personal addresses are hidden. A translator has access to the original unrestricted package and begins writing a detailed summary from it.
Keep translation privileges aligned with the intended audience. The public translation must not reintroduce information deliberately redacted from the access representation. Describe that redactions exist when policy allows, without disclosing their contents.
Test public metadata and search snippets as well as downloadable files. A correctly redacted PDF can still be undermined by an unreviewed translated title or index entry.
A workflow migration to a new preservation system
An institution migrates E-ARK packages between repository platforms. One platform uses a local display label and the next uses a controlled token. Translation appears to be the natural bridge, but the deeper issue is a data mapping.
Record the source and destination models, identify what each value means and build a controlled crosswalk. Keep translations for user-facing labels separate from schema transformations. Test round trips when feasible; information lost in conversion cannot be restored by elegant prose.
Validate sample packages before a mass transfer, including exceptions and local extensions. A successful migration should preserve object identity, relationships and audit evidence, not merely make the new interface readable.
An archival export contains a broken link
A translated summary cites a document stored in a representation folder. The receiving tool loads the summary but returns an error for the file pointer. Staff assume the issue is a temporary hyperlink problem.
Inspect the METS file group, pointer location, filename, path encoding and package content. Repairing a human-facing URL alone may not fix the authoritative structural reference. Document whether the defect existed in the source or arose during transformation.
Acceptance should include link-resolution results, not just a visual inspection of the public catalogue. A translation layer that hides missing content makes the package appear more reliable than it is.
An overly literal accession description
A historical institution used a term whose literal translation sounds like property ownership, even though the original administrative practice referred to temporary custody. The target reader could draw a false legal conclusion.
Translate the institutional function rather than forcing a familiar but inaccurate legal term. Preserve the original label when useful and give a concise note distinguishing custody, ownership and access responsibility.
Ask an archivist familiar with the source jurisdiction to review. A good explanation should let readers understand what was transferred without implying an ownership right that the documents do not establish.
A controlled event type is translated inside XML
A preservation event contains a controlled event-type value and a free-text outcome note. The editor translates both, believing consistency requires that every word appear in the target language.
Retain the controlled value exactly as required by its vocabulary and translate only the relevant descriptive note. If the repository supports language-tagged labels for the controlled concept, supply those through the appropriate semantic mechanism.
Check both schema validity and vocabulary conformance. A syntactically valid element may still contain a token that a receiving application does not recognize.
A transfer notice reports missing documentation
A donor’s handover notes mention that two files could not be recovered. The target-language summary is shortened to ‘transfer completed successfully’ because the central payload passed validation.
Keep the exception visible. Technical success for the submitted set does not prove the original collection was complete. Translate missing-item notices and distinguish integrity of what was received from completeness of what should have been received.
Acceptance minutes should record the exception, responsible party and next action. International collaborators need the limitation clearly explained rather than hidden by an optimistic status label.
A public translation is challenged years later
A local community identifies a misleading term in an archival collection description. The package itself remains unchanged, but the public translation has shaped research citations for several years.
Correct the descriptive layer through the repository’s normal version and audit procedures. Preserve the source-language statement, document why the interpretation changed and identify affected discovery records where appropriate.
Readers should be able to distinguish an updated translation from a newly discovered original document. Translation history is a form of accountable editorial history, not a reason to rewrite package provenance.
An audit across three partner institutions
A university, a municipal archive and a national library each inspect the same package. Their user interfaces use different languages and local terminology, and each reports a different total because one counts representations while another counts files.
Agree on shared units of measurement and identifiers before comparing reports. Translate metric labels but define precisely whether they count packages, intellectual objects, representations or physical files. Do not force counts to agree by adjusting the descriptive records.
Reconcile sample records against the authoritative METS structure and transfer manifest. A multilingual report is reliable when different institutions can reproduce the count from the same defined evidence.
An executable validation report is translated
A validator emits machine codes, test identifiers and free-text error descriptions. The recipient asks for a translated report that can still be correlated with the software logs. Replacing every token with a target-language phrase makes comparison difficult.
Keep the test IDs, paths, severity codes and machine messages traceable. Place translated interpretations beside them, with practical guidance when appropriate. A warning about a missing path is different from a warning about an undocumented local extension.
Have staff locate the same failure from both the translated report and the raw tool output. The workflow is successful when human understanding improves without destroying the identifiers used for remediation.
Review checklist before accepting a multilingual package
Check the exact CSIP release and local profile; inventory original files; verify paths, declared identifiers, METS structure, link targets and applicable integrity evidence; separate original content from translated derivatives; confirm the language and provenance of descriptions; test rights constraints; and use the recipient’s validation process on a clean copy. Keep a record of every exception, including incomplete source records, unavailable schemas and local vocabulary differences.
Then ask a person unfamiliar with the source language to locate a specific file, distinguish the original from a derivative, identify the package stage, understand a rights limitation and retrieve the unchanged source. If they cannot do these tasks, the multilingual description may be too vague. If they can do them only because the package’s technical identifiers were rewritten, the translation has solved the wrong problem.
Frequently asked questions
Can I translate CSIP folder names? Do not change structural names or labels recognized by the applicable profile merely for readability. Put translated explanations in supported descriptive layers or documentation.
Does passing an XML schema test prove the translated package is correct? No. Validation may detect technical defects, but not all rights errors, misleading descriptions, incorrect identities or incomplete source collections.
Is CSIP the same thing as an AIP? No. CSIP sets common packaging requirements; E-ARK specifications for submission, archival and dissemination packages address different lifecycle roles and additional requirements.
Can a translation be added after ingest? Often it can be managed as a derivative or metadata update, subject to policy. The change must be documented, relationships updated appropriately and integrity checked.
Should existing archived documents be rewritten into the target language? Preserve source evidence. Offer a separately identified translation or access representation instead of silently replacing the historical object.
What is the first question for a new partner? Ask which CSIP release and specific package profile their system accepts, and which metadata fields may contain new-language human descriptions.
References and neighboring guides
Consult the E-ARK Common Specification for Information Packages and DILCIS Board release history for formal definitions and release context. The connected eduKateSG articles on METS package maps, BagIt packaging and OCFL preservation storage deal with neighboring systems. The distinctive job here is managing multilingual language within the E-ARK CSIP transfer contract and its structural rules.
Archive transfer succeeds when the receiving institution can understand the material and verify the same identities, relationships and responsibilities that the originating institution intended. Translation should widen understanding, not silently redesign the package. A clear descriptive note, a maintained source record and a reproducible validation result are more valuable than a polished label that no longer points to the right thing.