If you need to translate MODS 3.8 metadata for a digital library, multilingual catalogue, institutional repository or cultural heritage collection, the central problem is not simply choosing words in another language. It is deciding which XML values describe an item, which identify it, which connect it to authority records, and which must remain unchanged. A beautifully translated title becomes a cataloguing error if it is quietly substituted for the title actually printed on a book.
This practical guide explains how to localize Metadata Object Description Schema records without breaking discovery, identity, provenance or machine-readable structure. It covers titleInfo, names and roles, subject headings, origin information, language and script, relatedItem, location, accessCondition, identifiers, recordInfo, XML validation and interoperability with MARC 21 and Dublin Core. Readers will learn how to give people understandable descriptions while preserving the stable evidence that lets libraries know what a record refers to.
MODS is an XML schema maintained through the Library of Congress MODS programme. Version 3.8 provides a comparatively rich descriptive structure, less granular than MARC in some respects and more expressive than a minimal fifteen-element Dublin Core record in others. Translation work can therefore happen at several layers: original transcription, parallel-language description, searchable local display, authority-controlled terminology and an export crosswalk. Those layers should cooperate rather than replace one another.
Begin with the object, not with the translation
Imagine a university library holds a book published in Tokyo. The cover carries a Japanese title, the catalogue has a romanized title, a scholar supplies an English translation of that title, and a discovery interface wants to show the reader a helpful English summary. All four strings can be useful, but they make different claims. The cover title is an observed bibliographic fact. Romanization represents the original script using another writing system. An English supplied title is a translation and may require a source or institutional policy. The summary is newly written explanatory prose. A responsible MODS record and its public interface tell those differences apart instead of compressing them into one field.
The first working question is therefore: what is being described? Is the record for an individual physical volume, a digitized reproduction, a born-digital text, a serial issue, a manuscript, a collection, or a component part? A translation decision suitable for a digital exhibition label may be wrong for a manifestation-level title. Establish the record’s intended unit of description and the responsible cataloguing practice before editing text.
Understand what MODS can and cannot do
MODS carries descriptive metadata using named XML elements. A record may describe titles, creators, publication information, physical characteristics, languages, subjects, classification, related resources, identifiers and administrative details about the catalogue record itself. These elements tell systems where information belongs, but the schema alone cannot guarantee that a translated statement is historically accurate, culturally respectful or authorized by the institution. XML validity is a necessary structural check, not a substitute for editorial judgement.
MODS is not a complete preservation-event ledger, a digitized-file inventory or a page-delivery API. Those are different jobs. METS can assemble file and structural relationships; PREMIS can record preservation actions and agents; IIIF can present pages and media to a viewer. Keeping this division of labor clear is one of the easiest ways to avoid duplicating content across metadata systems. A single object may use all four standards and still have one authoritative original title.
Treat the source record as evidence
Before translating, preserve the original export and record its origin, retrieval date, schema version, character encoding, identifiers and institutional owner. Inspect the record in a structured XML viewer rather than relying on a screenshot of a catalogue page. A display may hide attributes, truncate repeated fields, synthesize punctuation or merge several title elements into one line. You cannot reliably decide what to translate until you see the underlying data.
Create a field-level inventory with three columns: protected values, translatable descriptive values, and values requiring cataloguer review. Typical protected values include stable identifiers, authority URIs and controlled codes. A prose abstract may be translated. A subject label drawn from a named authority may need an approved multilingual label rather than an ad hoc translation. A publisher’s historical name may need to remain as transcribed, with a separate note explaining it to readers. Record each decision so a future update can be repeated.
TitleInfo carries several kinds of title
The titleInfo element is central to discovery, but it is not a licence to overwrite an original title with a familiar English phrase. MODS allows title information to be qualified. Its title and subTitle components, among others, convey parts of a bibliographic title, and appropriate attributes distinguish different uses. A catalogue may contain an original-language form, another-script representation and an explicitly identified translated or alternative title. The exact model depends on the local application profile and the source evidence.
A useful review compares three things: the title as it appears on the item, the title as stored in the source record, and the title shown in the translated interface. If a translator supplies a new English title, readers should not mistake it for the title printed on the document. When a source does not provide enough evidence to distinguish a formal alternative title from an explanatory phrase, keep the translation in an editorial display layer or clearly identified note pending a cataloguer’s decision.
Alternative titles should improve access without inventing editions
A translated title can help a user discover a work, but it does not prove that an English-language edition exists. Consider a manuscript whose original title literally refers to a garden of remedies. A reader may search for “Herbal remedies in the palace garden,” while a specialist needs the exact original wording and its transliteration. Presenting the search phrase as though it were an edition title would mislead both audiences. Describe it as a supplied translation and retain the original evidence.
This distinction also affects deduplication. Automated systems may group records with similar translated titles even when they describe different works. Never let translated text alone determine record identity. Use trustworthy identifiers, cataloguing relationships and source-specific matching rules, and test ambiguous cases manually. Translation expands access; it should not silently merge works or manifestations.
Names are identities before they are words
The MODS name element can describe a personal, corporate, conference or family name and can carry associated roles. Personal names may appear in different scripts, different orders or established authorized forms. Translating a family name by its literal dictionary meaning is generally inappropriate. Romanization is not the same as translation; an authorized heading is not simply whichever spelling sounds natural to the local reader. Historical figures may have multiple established names that need context.
For a creator recorded in Japanese characters, keep the original-script form when available, use a documented transliteration policy, and link a reliable authority identifier when the catalogue provides one. If the public interface wishes to display a familiar English form, establish its relationship to the authorized form without destroying the latter. Evaluate culturally sensitive particles, honorifics, family-name order and diacritics with knowledgeable reviewers. The same principles apply to institutions renamed after mergers or political changes.
Roles must survive the change of language
The person who wrote a text is not necessarily the person who edited it, translated it, illustrated it or wrote its introduction. MODS role information can distinguish these contributions. A localization that renders every role as “author” deprives readers of useful attribution and may affect rights and academic citation. If a role uses a controlled code, preserve the code and translate the display label using the vocabulary’s approved meaning.
Test role mapping with a mixed-contribution volume: an original author, a modern translator, an editor, a photographer and a sponsoring institution. After localization, a reader should still be able to answer who performed each contribution. If a role is uncertain in the source, preserve that uncertainty instead of guessing based on the order of names. Meaningful translation respects attribution, including the distinction between a work’s creator and the translator of a particular version.
TypeOfResource and genre serve different purposes
An institution may describe a resource as text, cartographic material, a sound recording or another broad category, while genre provides a more specific form or intellectual classification. These are not interchangeable. Calling an object a “book” in a user interface may be convenient, but it might be a manuscript, a thesis, a diary, an oral-history transcript or an archival finding aid. Local labels should be understandable without erasing the source categorization.
Where a vocabulary supplies an identifier or code, do not translate the machine value. Translate the human-readable label or select an authorized label in the target language. Check how the destination system filters by resource type and genre. A genre translation that looks fluent but maps to the wrong concept can move an item into the wrong discovery facet. Human language and machine classification need to remain tied to the same concept.
OriginInfo describes publication, not translation history
OriginInfo may capture publication or production details, including dates, publishers and places. Those statements are frequently evidence about an edition. A catalogue field naming a publisher in a particular city should not be rewritten into the institution’s current corporate name simply because users recognize that name. Similarly, a publication date should not become the date of metadata translation, ingestion or scanning. These events belong to different timelines.
Preserve original dates and any relevant encoding attributes. A date expressed in a historical calendar or uncertain chronology may require a carefully documented conversion for users, not replacement of the source evidence. Distinguish publication, copyright, creation, issuance and digitization dates wherever the application profile allows. A good multilingual display can explain complicated dates while leaving machine comparisons, chronological filtering and catalogue history intact.
Language identifies content, metadata and interface separately
Language data can describe the language of the resource being catalogued. The XML attribute xml:lang may identify the language of textual metadata supplied in an element or its context. The language selected in a website interface is yet another variable. A French user reading an English summary of a German book must not cause the record to declare that the book itself is English. Treat these three levels as separate assertions, even when a simple interface shows only one language selector.
Prefer documented language tags and vocabulary codes rather than freehand national labels. Where a resource is multilingual, record that accurately. If a scan contains Arabic commentary around a Persian text, a single convenient language choice may hide the research significance of the combination. Validate actual tags against the conventions used by the system and inspect fallback behavior for fields that have no translation. Never label untranslated original text as if it were a finished target-language translation.
Script and transliteration deserve explicit decisions
Language does not determine script. Serbian material may appear in Cyrillic or Latin; Chinese may be written in different script conventions; romanized forms of names and titles may coexist with original-script forms. A user may request “English letters” when the actual task is transliteration rather than English translation. Clarify the transformation in the record’s editorial documentation even if the public interface uses simple labels.
Choose a consistent romanization authority when one is available and preserve diacritics where supported. Do not remove non-Latin characters because an older downstream system is inconvenient. Modern UTF-8 pipelines can usually preserve them, although clients, sorting rules and search analyzers still need testing. A sound practice compares a native-script search, a transliterated search and a translated-keyword search against the same known object, then verifies that all paths reach the correct record without conflating variants.
Abstracts explain without rewriting evidence
An abstract is often the most natural MODS field for substantive translation. It can tell a reader what a resource covers and why it is relevant. Nevertheless, abstracts can include claims, quotations, chronology, specialized terms, sensitive descriptions and scholarly interpretations. Translation should conserve these distinctions. Avoid improving an ambiguous source sentence by introducing a decisive claim unsupported by the resource.
If the original abstract is outdated or demonstrably wrong, record a correction through the institution’s editorial process instead of hiding it inside the translated text. Use terminology consistently, especially for colonial-era names, scientific classifications, historically contested places and archival descriptions. Readers should be able to tell whether they are seeing a translation of the institution’s abstract, a new summary written for a local audience or an interpretive note supplied by a scholar.
TableOfContents requires structural awareness
A table of contents is not simply one paragraph to run through a translation engine. It may encode chapter order, page references, part titles, nested subsections or references to physical volumes. Translating the words while rearranging the order can misrepresent the work. Replacing page numbers based on a different digital edition can make the catalogue useless for readers consulting the physical copy.
Protect numbers, original sequence and references to source structure. Translate chapter titles when appropriate, but note when a supplied English title is not printed in the source. For heavily structured collections, consider presenting translated navigation separately from the authoritative transcription. Test by choosing a middle chapter and following its reference into the resource. A successful localization should take the reader to the same part of the object as the original metadata.
Subject headings are concepts, not keyword decorations
Subject information may include topical, geographic, temporal, title or name components. A subject string often follows a controlled vocabulary and its rules for combining subdivisions. Word-by-word translation can destroy those relationships. An institutional heading about migration in a specified country during a particular period may be more specific than the literal rendering of its individual terms suggests.
Where possible, resolve the source concept to an authority URI or code and find an approved target-language preferred label. Preserve alternative labels for search if licensing and vocabulary policy allow. Where no accepted equivalent exists, use a documented local label and keep the source concept as the reference. Verify the facet behavior after export: a multilingual reader must reach the same conceptual grouping, not a new unrelated category created by a fluent but uncontrolled phrase.
Geographic names carry boundaries and history
Place names may have multilingual forms, historical names, transliterations, disputed naming conventions and changes in administrative boundaries. A field naming a place of publication in 1910 is not automatically equivalent to a modern geopolitical label. Similarly, a subject heading about a historical region might span several current countries. The translator should not silently choose a contemporary label that changes the scholarly claim.
Keep the original evidence, prefer recognized geographic identifiers when supplied, and add carefully scoped display labels. If the source says “Bombay” in a nineteenth-century imprint, an English display that substitutes only “Mumbai” loses historical context. A reader-friendly display might retain the historical name and explain the present-day place name where editorial policy supports it. Test both forms in search while preserving the temporal meaning of the source.
Classification is not prose to be translated
A classification number is a placement or subject-navigation code according to a classification scheme. Its components should not be translated, rounded, reordered or replaced by an approximate local subject name. The accompanying human-readable caption may be localized if the scheme’s documentation and licensing permit it, but the encoded identifier remains a key to the chosen classification system.
Keep scheme information alongside numbers so a value cannot be mistaken for a code from another system. If the receiving library uses a different classification scheme, treat mapping as a separate expert cataloguing decision. A translation alone cannot reliably convert one classification hierarchy into another. Verify sorting, shelf browsing and category filtering on realistic examples, including leading zeros, letters, punctuation and longer class marks.
RelatedItem is where many subtle errors occur
MODS can express relationships with other resources, such as a host work, an earlier edition, a series or another related item. Those relationships are essential to scholarly interpretation. A journal article belongs to a particular issue; a translated novel may be a related expression or edition; a digitized facsimile may be related to a physical source. Translating the title of a related work must not disconnect the relationship from the correct object.
Follow every related-item link and compare the destination’s identity. Distinguish “part of,” “other version,” “preceding” and other link purposes as modeled by the source record. If a local interface labels every relationship simply “Related,” the user loses important information. Provide comprehensible labels while retaining the exact relationship data underneath. Test nested records where an item is part of a chapter, volume, series or archive. Navigation errors often show up only in these deeper structures.
Identifiers should be copied, not interpreted
An ISBN, ISSN, DOI, local catalogue number, handle or repository accession identifier is a machine-recognizable reference. Translation does not change its characters. A number formatted for human reading may contain spaces or hyphens, but its underlying identifier should remain valid and traceable. When a record contains multiple identifiers, their types matter: an ISBN is not interchangeable with a DOI, and a digital-file checksum is not an item identifier.
If a source record has an incorrect identifier, flag it as a source-data issue. Do not invent a repair because a similar book appears in search results. Validate identifier syntax where the scheme permits and dereference externally resolvable identifiers when feasible. Keep local identifiers scoped to the institution that issued them. A good multilingual catalogue should allow a patron searching in any language to converge on the same item reference.
Locations and holdings need a different layer of caution
The location element can help users find a resource, link to its electronic address or understand its physical holding context. Human-readable branch names may be translated, but call numbers, access URLs and shelf coordinates are operational data. A wrong translation of “reference use only” can send readers toward an item they cannot borrow. A wrong copied URL can make a digitized collection appear to have vanished.
Distinguish a permanent digital identifier from a transient service URL where the platform supplies both. Test links after localization. If a local branch has an official bilingual name, use the institution’s approved form rather than an informal phrase. A reader should be able to act on the record: locate the correct shelf or open the correct digital object. Translation quality becomes observable at this point, not merely stylistic.
AccessCondition is about rights and permission
Access conditions may include rights statements, reproduction restrictions, embargoes or access notes. The translator must preserve the difference between permission to view, permission to reproduce, and ownership of copyright. “Available online” does not necessarily mean “public domain,” and “copyright held by the library” may not imply that users can freely redistribute the work. These differences are consequential.
Keep license identifiers and authoritative rights statement URIs unchanged. Translate their explanation when an official localized wording exists, or provide a carefully scoped plain-language paraphrase that does not override the legal text. Dates and geographic restrictions need exact interpretation. If the institution’s source note is unclear, request policy review rather than improvising legal advice. Test the catalogue’s access buttons and download flows against the translated note to detect contradictions.
Notes can be descriptive, evidential or cautionary
A note may contain a cataloguer’s observation, a source quotation, provenance information, bibliographic history, a content warning or another contextual statement. The correct treatment depends on the note’s purpose. For a quotation engraved on an artifact, translation should be distinguishable from transcription. For a conservation condition note, uncertainty and qualification must remain intact. For an administrative note not meant for public display, localization may be unnecessary.
Build a note-type policy rather than applying one universal instruction. Ask whether the sentence is human-facing, whether it records a verifiable observation, and whether a translated version is authorized for publication. Protect documentary spelling when it is itself the object of study. Some difficult historical terms should be contextualized rather than silently modernized, especially where the record documents discriminatory language or contested institutional practices.
Part and enumeration control retrieval
A serial issue, chapter, volume, fascicle or special part may be identified by numbering or chronological designation. The wording around those values can be localized, but the values themselves must remain attached to the correct entity. “Vol. 2, no. 3” cannot become “Book 3 of 2” just because a display parser misunderstood a local abbreviation. Ordering mistakes lead readers to the wrong source and can break citation.
For each record, identify the structural unit the numbers refer to and compare them with the original publication. Preserve pagination conventions and special-case numbering such as combined issues or supplements. Test sort order across at least three adjacent parts. Do not assume the same punctuation or word order works identically in every language. The data model should drive the display, not the other way around.
RecordInfo records the metadata’s own history
The record and the resource are different objects. A cataloguer may create a description decades after the item was published; another editor may correct it later; a translation may be added much later still. MODS recordInfo is one place where information about the descriptive record can be expressed. Do not replace the object’s publication date with a catalogue revision timestamp or present a translator as the original creator of the resource.
Preserve record origins, responsible agencies and relevant modification information. Where local workflow needs a translation timestamp, manage that information according to the repository’s provenance policy rather than overwriting an unrelated field. Reviewers should be able to reconstruct which version of the source metadata informed a translated edition. That matters when source cataloguing changes: a well-formed translation can become stale even without anyone touching its text.
Extensions are not automatically safe to translate
The extension mechanism can carry local or third-party metadata outside the core MODS model. One institution may store a public label there; another may store a controlled identifier, workflow code or serialized data. Translating an extension by guessing from an element name can damage a system that is otherwise schema-valid. Treat unknown namespaces as protected until their responsible profile and documentation are reviewed.
Inventory extension namespaces, local schemas, allowed vocabularies and downstream consumers. When a team wishes to localize a value, test whether that extension is meant to accept language variants. If not, prefer a separate presentation mapping or an explicitly designed extension. If a destination silently drops unknown extensions, investigate the consequences before declaring conversion successful. A valid MODS export can still be operationally incomplete.
XML namespaces and character encoding are structural contracts
XML element names, namespace URIs, attributes and schema declarations are not sentences to translate. Changing them may prevent a record from being parsed as MODS. The safest workflow keeps the structural markup stable and edits only explicitly authorized textual values. UTF-8 can carry many writing systems, but a valid encoding declaration is not proof that every downstream database and interface handles the characters correctly.
Test non-ASCII text, combining marks, right-to-left scripts, quotation marks and ampersands. Preserve XML escaping rules, including treatment of reserved characters within attribute values and element text. Confirm the resulting document parses and validates against the intended MODS schema. Then inspect rendered output for correct directionality, word order and character display. Machine validity and human readability must both pass before public release.
Repeated fields mean one record may have multiple true descriptions
Some MODS constructs can occur more than once. A catalogue may hold parallel title statements, multiple names with different roles, several subjects, many identifiers and more than one related item. A translation pipeline that assumes one value per field may discard meaningful alternatives. Another pipeline may flatten everything into one unlabeled string that is impossible to search or cite accurately.
Count occurrences before and after localization, and preserve the purpose of each. If a multilingual title is added, it should not consume or erase an original title entry. When the same concept has several language labels, retain their language identity and relationship to the underlying controlled value. Review output on a realistic crowded record rather than testing only a simple single-author monograph. Complex records reveal mistakes that toy examples hide.
Authority records turn translation into accountable mapping
A controlled name or subject may point to an authority system with a stable identifier. That link lets different languages converge on the same concept without pretending all labels are interchangeable. It also allows a catalogue to distinguish people with identical names, regions with changing borders and concepts that are easily confused. Translating a visible heading while retaining the authority identifier can improve accessibility without sacrificing precision.
Do not infer an authority identifier from a surface-language match alone. Confirm the entity through biographical, geographic, bibliographic or scheme-specific evidence. If a cataloguer has intentionally left a name uncontrolled, respect that status unless an authorized remediation project resolves it. Keep the difference between an approved preferred label, an alternative label and a newly supplied local translation visible in editorial records.
Search behavior is part of translation quality
A library user rarely sees the XML. They type a title, author, subject or question into a search box and judge the result by what appears. Therefore localization must be tested with representative user searches, not only a word-for-word review. Run original-script queries, romanized queries, translated-keyword queries and identifier queries. Confirm that search results point to the same appropriate records and that facets do not split a single controlled concept into unrelated categories.
Also inspect snippets, sort order and highlighted terms. A search analyzer may tokenize an East Asian title differently from an English phrase or strip accents from one language but not another. A translated abstract may improve recall while introducing irrelevant matches. Evaluate precision as well as recall, using a small curated set of known records. A successful conversion supports human discovery without inventing relationships between resources.
Mapping MODS to MARC requires more than tag substitution
MODS is often used alongside MARC 21, but a round trip is not automatically lossless. MARC includes specialized tags, indicators and subfields that may not map one-to-one to MODS elements. Conversely, a local MODS extension may have no tidy MARC destination. The goal is to understand the information each side can carry and document transformations that are irreversible, approximate or dependent on local practice.
Use published crosswalk guidance and representative sample records. Test title variants, non-Latin scripts, role codes, publication dates, linked authority names and related-item structures. Compare meaning after conversion rather than relying solely on file validity. A translated display label should not be written into a MARC field in place of the original transcription unless the relevant cataloguing rule explicitly supports it. See the existing MARC 21 translation guide for the field-level distinction.
Mapping MODS to Dublin Core changes granularity
A simple Dublin Core export may expose a title, creator, subject, description and date but omit the finer distinctions represented in MODS. That can be appropriate for basic discovery, yet it creates a risk: multiple semantically different values may end up in one broad property. Translation can amplify the confusion if target-language labels are merged without their type, language or provenance.
Decide what a destination genuinely needs. Keep a rich canonical MODS record and derive an access-oriented export when the receiving system is less expressive. Document which values were flattened, omitted or synthesized. Where language variants are exported, keep their language identity if the destination supports it. Do not treat a successful import as proof that the source could be reconstructed. The user should know whether the delivered record is a preservation-quality description or a simplified discovery representation.
A worked example: a bilingual map collection
Suppose an archive holds a nineteenth-century map of a port city. The title is in French, explanatory notes are in a local language, the map identifies streets using older place names, and the accession record contains modern institutional cataloguing information. An English-speaking visitor needs to understand what the map depicts. The correct localization might present an English supplied title, translate a contemporary curator’s abstract and explain historical names without rewriting labels printed on the image.
Review the MODS titleInfo, language, subject, relatedItem and location data together. Keep the record identifier and image URI fixed. Give historical place names appropriate date context. Verify that an English search for the city reaches the map, while a source-language search remains possible. If the map has multiple digital images, link their delivery through the repository’s structural or presentation system rather than trying to encode page geometry inside a title field.
A worked example: one person, several scripts
A research archive contains a writer’s publications under an original-script name, a transliterated catalogue heading and a familiar English spelling. A careless automated translator creates separate “authors,” dividing the bibliography. A careful localization recognizes that all forms refer to one person, but only after checking authority evidence. The source strings may legitimately have different functions: preferred heading, alternate script, historical pseudonym or reader-friendly form.
Do not remove the variants. Associate them using the authorized identity model, preserve exact strings where they serve bibliographic functions and test searches by each form. If the writer shared a name with another individual, avoid automatic merging. For people whose self-identification has changed, follow current ethical and cataloguing policy while retaining necessary historical evidence. Good internationalization makes the record more findable without manufacturing a different person.
A worked example: a translated edition of a novel
A catalogue holds the original novel and a later translation. A language model might help create summaries for both, but the two entries are not interchangeable. They may have different publishers, publication dates, translators, ISBNs, pagination and rights terms. The work-level relationship can be described without collapsing the edition-level facts. In an access interface, make the distinction visible: an item can be a translation of a work without being the original language of that work.
Keep each manifestation’s own identifiers, originInfo and contributors. Link the related works according to the repository’s bibliographic model. Test the translated-title search against both records and check that holdings, full-text links and circulation options lead to the selected edition. A user choosing a text for close reading needs this distinction as much as a cataloguer does.
A worked example: a collection-level archival record
A record for an oral-history collection may include many interviews in different languages, created over decades with varying consent restrictions. A short English collection abstract can help researchers, but it should not imply that every interview is in English or available online. Subjects and contributor roles may apply to the collection as a whole rather than each individual recording. That scope distinction is part of the meaning.
Translate collection-level descriptions at their proper scope. Preserve interview identifiers, dates and access conditions. Where a child item has its own language, rights or speaker information, do not blindly inherit the parent’s display labels. Test a nested item and ask whether a reader can tell what belongs to the collection, what belongs to the recording and what belongs to a translated interface note. Archive users depend on these boundaries.
Prepare a field-level translation brief
A good brief defines the target audiences, target languages, relevant schema version, source of authority terms, transcription policy, transliteration standard, rights boundaries and the meaning of “translated title.” It identifies which fields may be edited, which may gain parallel labels and which are protected. For each language, record fallback behavior when no equivalent is available. This is more useful than a vague instruction to “translate all XML text.”
Include an escalation route for disputed names, uncertain dates, subject terminology and historically harmful expressions. Choose representative records for review: a single-author book, multilingual manuscript, journal issue, map, digital collection and translated edition. Assign responsible reviewers for cataloguing semantics and language quality. A translator should never have to guess whether a string is a shelf mark, a quoted title or a free-form explanatory note.
Validate the XML, then validate the catalogue
First parse the localized XML with a standards-aware tool. Check well-formedness, namespace names, element placement, required attributes in the selected profile and schema compliance. Then compare record identity, identifiers, counts of repeated fields, authority links and related-item targets before and after localization. Machine checks can detect structural loss but cannot tell whether a historical term has been mistranslated.
Next perform a discovery test. Search in multiple languages, open the item, check citation details, follow related items, examine rights language and confirm access to digital objects or physical holdings. Have a knowledgeable reviewer examine at least one difficult record in each category. Keep the original alongside the localized record for audit. The acceptance question is not only whether the XML works; it is whether the reader reaches the right object and understands the limits of what the catalogue asserts.
Common mistakes and their repairs
One frequent mistake is replacing original titles with translated titles. Repair it by restoring transcription and storing translation in the appropriate qualified or presentation context. Another is translating authority codes or classification numbers. Repair it by protecting coded values and localizing only the corresponding labels. A third is conflating language of item, metadata and interface. Repair it by expressing these as separate assertions and testing mixed-language records.
Other failures include silently renaming historical places, treating transliteration as translation, dropping repeated fields, converting identifiers to local number formats, flattening related-item types and implying broader reuse rights than the license allows. The repair for every case begins with an explicit invariant: which source fact or relationship must survive? Once the invariant is visible, language choices become easier to judge. Never call a record complete while a critical link is unresolved.
Frequently asked questions about translating MODS records
Can I translate every XML text node automatically? No. Text nodes can contain transcribed evidence, controlled labels, identifiers embedded as strings or human-facing prose. Field meaning and local policy determine whether translation is appropriate. Preserve a copy of the original.
Is a romanized title an English translation? No. Romanization changes the writing system, not necessarily the language or meaning. It should be described accordingly, and it may follow a specific recognized standard.
Should an English abstract replace the original-language abstract? Usually it is safer to keep provenance and language clearly expressed. Whether the repository stores parallel abstracts in MODS or in an associated display system depends on its application profile.
Can the translation change subject headings to sound more natural? Human-readable display labels may be improved, but controlled concepts and authority identifiers should be preserved. Mapping a subject to a different concept requires cataloguing review, not stylistic editing.
Will an XML schema catch a wrong relatedItem target? Not reliably. A structurally valid link can still point at the wrong bibliographic object. Verify destinations and relationship semantics separately.
Does MODS provide everything needed for digital preservation? No. MODS chiefly serves descriptive metadata. File inventory, preservation events and user-facing presentation may involve complementary systems such as METS, PREMIS and IIIF.
How should rights language be localized? Preserve authoritative conditions and links, use approved wording where available, and avoid adding permissions or removing restrictions. Escalate ambiguous legal meaning for human review.
What is the best sign that a localization works? Readers using different languages should be able to discover, identify, cite and access the same intended resource, while cataloguers can still trace its original evidence and metadata history.
A final standard of quality
The purpose of MODS localization is not to make every field appear English, French, Arabic or any other language. It is to make a bibliographic object intelligible and discoverable without changing what the underlying record knows. A catalogue succeeds when it can show a welcoming translation while preserving the item’s original identity, relationships and evidence. Protect identifiers and source transcription; distinguish translated descriptions from originals; keep names and subjects tied to authority; test the actual reader journey. These habits apply to libraries of every size, in every country.
Authoritative references and further reading
The Library of Congress MODS 3.8 User Guidelines explain the schema and cataloguing considerations. The MODS standards home provides schemas, conversion resources and implementation information. For neighbouring responsibilities, see the METS specification, the PREMIS 3.0 Data Dictionary and the IIIF Presentation API 3.0. These sources are the proper starting point for version-specific implementation decisions.