How People Translate Quickly | MARC 21 Field 880 Translation QA: Preserve Alternate-Script Links and Catalogue Identity

MARC 21 field 880 translation QA means checking whether a multilingual catalogue record still connects its alternate-script text to the correct bibliographic field. The quickest reliable method is to freeze the source record, identify every 880 and its partner through subfield $6, edit only authorised descriptive text, and test that each pair still points both ways. Translating a title is not the same task as translating a field tag, changing an access point, or inventing a new work identity.

A librarian may see a Latin-script title in field 245 and a Chinese, Arabic, Cyrillic, or Hebrew rendering in field 880. These are not simply two paragraphs sitting next to each other. MARC uses field identifiers, indicators, subfield codes, and explicit linking data to tell catalogues that the different scripts represent related information. A reader should find the same book after changing interface language, not two apparently unrelated books with conflicting titles.

Begin with a protected copy of the entire record and a small pairing table. For each eligible text element, note whether the work supplies an original-script transcription, a transliteration, or an actual translation. Those three operations can produce very different strings. Only then prepare the target wording, preserve the $6 occurrence and associated tags, run structural checks, and ask a cataloguer to compare search, sorting, and public display using both scripts.

What field 880 actually does

The Library of Congress defines bibliographic field 880 as an alternate graphic representation of another field. Its associated regular field links through $6 to the 880, and the 880 links back through its own $6. The system is particularly useful when a record contains both Latin and non-Latin scripts, but the machinery is about script representation rather than a promise that every alternate field contains a translation. A record could have Japanese characters and romanisation of the same Japanese title without either string being translated into English.

The linked field retains the semantic roles of the ordinary field. An 880 paired with 245 supplies an alternate-script representation of a title statement; an 880 paired with 100 concerns a personal-name main entry; one paired with 264 concerns production, publication, distribution, manufacture, or copyright statement data according to its indicators. A translation programme that sees every 880 as interchangeable prose will confuse bibliographic functions, even when the visible characters appear correctly converted.

Subfield $6 acts as a relationship instruction. In a regular field, a value such as 880-01 identifies the alternate field and an occurrence number; in the alternate field, 245-01 points back to the particular original field. The occurrence number is not a language code, an item identifier, or a chronological sequence. It is an internal pairing number. If there are several possible 880s, matching solely on the nearest neighbouring line is unsafe.

The precise formatting matters. The Library of Congress describes a linkage structure containing an associated tag, a two-digit occurrence number, optionally script identification, and optionally field orientation. Subfield $6 is the first subfield in its field. A code denoting Cyrillic script or right-to-left orientation describes how to understand and display that record content. It does not authorise a translator to rewrite the cataloguing field’s meaning or reorder stored characters to imitate screen direction.

Field 880 can sometimes appear without an associated regular field, using occurrence number 00 for a particular unpaired representation. That is not always a defect requiring a fabricated partner. The catalogue reviewer must decide whether the unpaired case is deliberate and permitted by the record’s cataloguing policy. A structural checker should therefore distinguish a genuinely broken link from the reserved unpaired case instead of demanding that every 880 resolve to exactly one visible regular field.

Decide whether the requested change is translation, transliteration or transcription

Translation replaces meaning in one language with meaning expressed in another. Transliteration changes writing systems according to a rule or convention, often without changing the language. Transcription can capture how a word is pronounced or how a source inscription reads, depending on the cataloguing practice. A romanised Japanese title is not automatically an English title. Confusing these operations can make a catalogue look impressively multilingual while giving a reader false information about what language a book actually uses.

Ask which source is authoritative for each field: a title page in the physical item, an established name authority, the cataloguer’s verified transcription, or a supplied explanatory translation. The title found on a cover may disagree with a publisher’s marketing translation. Catalogues ordinarily distinguish a title proper from a variant or supplied translation rather than silently overwriting the title that identifies the manifestation. Preservation of the evidence is a reader service, not an obstinate refusal to make a display welcoming.

Consider a Japanese-language science workbook whose original title is represented in Japanese characters, whose Latin form is a romanisation, and whose English explanatory label is “Experiments for Young Readers.” Each has a different provenance. If an English interface presents the explanatory label as though it were the transcribed title proper, teachers may order the wrong edition. A safer display says that the English phrase is a supplied translation and keeps the original-script and romanised forms accessible.

Names are more delicate than slogans. A Chinese author’s name may be represented by Chinese characters in an 880 linked with a 100 or 700 field while a romanised authorized form is used for retrieval. Translating the literal meaning of the surname or reversing name components for an unfamiliar display would create a false personal identity. Preserve the established access point and identifiers, then address alternate scripts and display labels using the authority policy chosen by the institution.

The translator should make a compact decision record with five columns: source phrase, field and subfield, linguistic operation, provenance, and destination. That note helps reviewers separate a legitimate translation of an explanatory note from an illicit replacement of a cataloguer-transcribed title. It also prevents reuse of a previously approved rendering when the new edition, script, or source differs from the older item’s evidence.

Read the linkage before reading the words

Start at the bibliographic record level, not an exported spreadsheet of isolated strings. Identify the leader and the character encoding declaration or profile used by the system, locate relevant variable fields, and enumerate all associated 880 pairs. A parser should preserve the record’s order and subfield boundaries. If a library export strips dollar-sign markers or folds two records into one text blob, recover the actual MARC structure before attempting translation.

Build a map keyed by associated tag and occurrence. A 245 carrying 880-02 should be matched to an 880 carrying 245-02, not to an 880 carrying 100-02. Identical occurrence digits may appear with distinct source tags in surprising vendor outputs, so a robust validator examines both halves of the relationship, not just the numeral. The check should also verify that any optional script notation remains attached to the correct $6 and that the normal field’s own indicators are not treated as words.

MARC field indicators are small but consequential. The first and second indicator in a 245 can affect whether a title is treated as having an added entry and how initial articles are skipped for filing. A translator who changes the title language may be tempted to update these indicator values automatically. That is a cataloguing decision, not a search-and-replace operation. Keep the original indicators until an authorised cataloguer has reviewed their relevance to the intended representation and local indexing.

Subfield codes also carry meaning. A title field may use $a for the title proper, $b for remainder of title, and $c for the statement of responsibility. A copied translation that moves a responsibility statement into $a changes the structured statement. A catalogue display may still look acceptable to a person but index the wrong text as the title. Keep distinct subfields distinct and run a display test that exposes, rather than hides, each field boundary.

A paired 880 preserves subfield meanings from its associated field, except for the special linkage. Do not assume there is one plain string to localise. A publisher’s place, name and date may occupy separate subfields in a linked 264 field; a variant script can require particular punctuation, but identifiers and dates remain evidence. If punctuation changes are truly needed for target-script presentation, document why the change follows local cataloguing conventions and confirm that extraction rules still find the same components.

A small, auditable linkage fixture

The following text is a simplified, human-readable field display for teaching. It is not a byte-level MARC21 transmission record and does not replace a MARC parser. It demonstrates two linked field representations with matching 02 occurrence numbers; the second 880 contains a Japanese-script counterpart. A production record requires proper encoding, delimiters and the rest of its bibliographic data before it can be exchanged with another system.

245 10$6880-02$aToshokan no kagaku :$bkodomo no tame no jikken.
880 10$6245-02/$1$a図書館の科学 :$b子どものための実験.

The Japanese text is a controlled illustrative title, not a claim that a real publication with this bibliographic identity exists. Its purpose is to let reviewers see that the 245 and 880 are linked despite different visible scripts. The $1 is part of a MARC script identification notation, not a dollar amount and not a target-language request. The ordinary text changes between records; the associated field tag and occurrence number form the invariant relationship.

A safe validation exercise first reads both lines into structured records. The program checks that 245 has a first subfield $6 pointing to 880-02 and 880 has a first subfield $6 pointing to 245-02, then checks the remaining subfield roles. It must not simply compare the strings after $a, because those are expected to differ. A human reviewer then verifies that the romanisation and original-script text actually correspond, which structural validation alone cannot establish.

Now consider a candidate export in which the 880 begins with 245-03. The characters still look Japanese and the English-language catalogue may still show a title. Yet the internal relation has changed. If an unrelated 245 or another 880 uses occurrence 03, the alternate script could attach to the wrong display or disappear. The repair is to restore the verified source pairing, not to change the name of the book until the pair accidentally matches something else.

Some libraries preserve a translation in an additional note, variant title, or separate multilingual discovery layer rather than reusing an existing 880. This is often the right architectural choice, especially when an original-script form and romanisation already occupy the available linked roles. The exact field selection is a cataloguing decision that depends on the manifestation, authority file, MARC profile and institutional policy. The translator should propose the wording and its provenance, not invent an unreviewed bibliographic encoding.

Control Unicode, direction and search separately

Unicode text is a sequence of characters, not an image of the page. Hebrew and Arabic fields normally store logical character order even though a rendering engine may display their letters right to left. Reversing the code-point order to make an editor preview appear attractive can corrupt search and copying. A test should compare the intended logical text with the stored text, then inspect the display using a bidirectional-aware renderer rather than using the screen as a proxy for stored truth.

Numbers, embedded Latin abbreviations and punctuation complicate bidirectional titles. A book title might combine Arabic words, a Western year, and a series number. A visual line break can make the number look attached to the wrong phrase even when the underlying bytes are correct. Test the exact title in a catalogue results list, a detailed record view, a citation export and a copied plain-text snippet. These contexts have different rendering constraints, so one attractive screenshot is not a sufficient pass.

Normalization deserves controlled handling. Two visually similar sequences may use composed versus decomposed Unicode forms, with effects on search, collation or record equality. It can be reasonable for a search index to normalize query text, but changing authoritative cataloguing data requires a separate policy. Record the original field bytes or a stable digest before any conversion, and test both composed and decomposed input queries. Avoid declaring a failed lookup an error in the translation when the actual defect is a tokenizer or search-normalization mismatch.

Romanisation has its own search behaviour. Readers may type long vowels without diacritics, omit apostrophes, or use different transliteration conventions. An index can offer aliases or query expansion while preserving the catalogued forms. That is better than rewriting the title to satisfy the most common keyboard. Keep the editorial record faithful and treat forgiving search as an additional service whose rules can be tested and documented.

Authority identifiers connect names across forms more reliably than strings alone. If an author is represented by several scripts, a stable authority record or recognised identifier lets search reconcile variants. However, identifiers are useful only when the correct authority has been selected. A translated alias that attaches to a similarly named but different person makes an apparently helpful multilingual index actively misleading. QA should validate the authority match through external evidence and local policy, not by a superficial character resemblance.

A workflow that is fast because it has boundaries

A practical translation queue begins with a record snapshot containing the full bibliographic context, including any paired 880s and the source of the cataloguing statement. The requester should specify the desired reader outcome: improve native-script retrieval, show a supplied English title, correct a romanisation, or explain publication data. Those outcomes require different interventions. Do not accept an instruction such as “translate the whole MARC record” without splitting it into eligible text and non-translatable structure.

Create a protected inventory of tags, indicators, subfield codes, $6 values, control numbers, identifiers and authority references. Compare the inventory before and after localisation using a structural parser. An eligible descriptive value may change, but a protected structural element should not. Where a change to an indicator or authority field is genuinely required, remove it from the translation batch and route it to a separate cataloguing change request with explicit evidence.

Prepare a terminology ledger for names, genres, series titles and institutional language conventions. It should distinguish original language, script, approved romanisation scheme and supplied translations. One term might have an established translated institutional name, while another is a personal name that must be carried rather than translated. Review decisions in the context of the actual record; a generic bilingual dictionary cannot know which meaning a title uses or whether a phrase is deliberately metaphorical.

Check a small batch before converting hundreds of records. Choose records with uncomplicated Latin script, one non-Latin script, mixed-script punctuation, right-to-left content, several alternate fields, and an intentional unpaired 880 if the system supports one. Record the expected matching pairs and public-display outcome. A test fixture with only a single easy Japanese title will not expose Arabic directionality, collisions between occurrence numbers or lost associated fields.

Release only after two independent gates. The machine gate verifies record parsing, encoding, field positions, linkage and agreed invariants. The human gate verifies title identity, correct script, semantic fidelity and retrieval in both languages. A passed machine gate cannot certify that a translation is idiomatic; a passed human reading cannot certify that the 880 relationship has survived an export. Each gate catches a different family of failure.

Failure diagnosis: where apparently correct text goes wrong

If the native-script title disappears from a detailed record, inspect whether the associated 245 still points at the correct 880. The catalogue theme may suppress alternate fields, but start by checking the data. Compare the $6 pair in the saved source and the imported record. If the pair is intact, inspect discovery system field mappings and display settings. Repair the earliest verified failure rather than adding a duplicate title to compensate for an invisible original field.

If an alternate-script title appears under the wrong book, check whether a batch import reused occurrence identifiers across records or misplaced field boundaries. An occurrence number is local to the field relationship in a record; it is not a cross-record identity key. The right fix may be a record-splitting bug in the importer rather than an individual translation. Inspect one affected item alongside its neighbours in the import file to identify a repeatable boundary error.

If a Latin-script search succeeds but the Chinese-script search fails, first verify that the Chinese characters actually survived encoding. A visual square or replacement character suggests data loss; correctly stored characters with no hits suggest indexing or tokenization. Search should be tested with both full phrases and characteristic word fragments using a documented tokenizer. Do not romanise the underlying field merely to make an inadequate search index pass an English-only test.

If a copied Arabic title appears to have its punctuation reversed, check logical order and the renderer. A left-to-right debugging console can display bidirectional data differently from a browser field with appropriate direction metadata. Review the stored code points and test with mixed Arabic, digits and parentheses. The fix should live in display handling where possible, not in artificially reordered bibliographic text that becomes difficult to reuse.

If the catalogue shows two nearly identical entries for one book after a multilingual import, investigate whether the pipeline has mistaken language variants for separate records. A different script does not automatically imply a different manifestation. Compare identifiers, edition statements, publisher, date and authoritative record numbers before deciding whether to merge. Conversely, never collapse genuinely different translations or editions into a single item solely because their titles share a meaning.

If an author search returns the wrong person, inspect the authority linkage and role of the field. A transliterated personal name might match multiple people; changing letter spacing or diacritics may worsen ambiguity. Review dates, works, institutional affiliations and controlled identifiers when available. Maintain a human-review exception for uncertain identity rather than manufacturing a deterministic answer from a name that happens to look familiar.

If a language label promises a translation but the field contains a romanisation, fix the display terminology and provenance before editing record values. Readers need to know whether they are seeing the title written in another script or its meaning expressed in another language. In a school library, this distinction helps a student choose a text they can actually read rather than assuming an English display title means the underlying book is written in English.

If an export changes a book’s series numbering, ISBN, date or source title while translating surrounding descriptions, the scope was too broad. Review the transformation that touched structured identifiers, then regenerate from the protected source. Do not manually retype identifiers from a screenshot; a visually plausible ISBN can still be wrong. A robust release check compares protected identifiers byte for byte and reports differences separately from approved text changes.

Case study: a bilingual school-library search result

Suppose a school’s library holds a Japanese astronomy reader. The native-script title is important to students who read Japanese, and a romanised title is useful to a student whose keyboard is configured for Latin characters. The school also wants an English summary to help parents understand the subject. These are three legitimate reader jobs. A single indiscriminate “English title” field cannot satisfy all three without concealing the difference among transcription, romanisation and explanatory translation.

The cataloguer retains the transcribed title and its linked alternate-script representation according to local practice. The translation editor writes a short English subject summary and labels it as such. The discovery interface allows either script to retrieve the same record, then displays the edition and language of the item. A parent reading the English summary will not mistake it for evidence that the book’s lessons are in English, while a student can still search by the title printed on the cover.

The diagnostic test uses three queries: a distinctive Japanese phrase, a romanised fragment, and an English subject phrase from the summary. The first two should resolve to the intended item, while the third may lead through the searchable summary according to local index rules. The test includes a second book with a similar English subject to prevent the system from “passing” merely by returning any astronomy book. Identity must be tested against the correct catalogue record, not a plausible category of results.

During an import rehearsal, the Japanese title appears on a different edition. Comparing $6 values reveals that the 880 was linked to the wrong 245 occurrence within one merged record. The remedy is not to alter the Japanese title text; it is to restore the field relationship and re-check edition identifiers. This illustrates an important boundary: a fully accurate translation can still be attached to the wrong bibliographic object when record structure is mishandled.

Case study: right-to-left display in a city archive

A municipal archive receives a Hebrew-script publication and a Latin-script cataloguing representation. The public catalogue supports both, but its mobile results card wraps a serial number and a date differently from its desktop view. Staff initially suspect that the Hebrew translator inserted punctuation incorrectly. A comparison of the stored sequence shows the translation is in logical order and the problem lies in the web component’s direction handling.

The team isolates a fixture containing Hebrew words, a Latin abbreviation and a four-digit publication year. It tests selection, copying, screen-reader reading order and keyboard navigation as separate outcomes. One repair adds appropriate directional containment in the result component without modifying the MARC text. After that change, the mobile display becomes legible while record exchange and index matching remain stable.

The team also checks that punctuation was not copied from the rendered screen into the underlying record during a later manual correction. This is a common second-order problem: the first defect is display-related, but an attempt to “fix” what staff see causes a durable data error. Editorial records should therefore include a warning that visual order is not always storage order for bidirectional text, and display changes should be regression-tested before anyone adjusts catalogue fields.

Case study: a name authority with competing romanisations

A researcher finds an author’s name represented in Chinese characters and by two Latin spellings. One form reflects a cataloguing rule, while another appears on a publisher’s English-language website. A translation vendor proposes replacing the authority form with the website spelling to make discovery “more international.” That is a potentially damaging shortcut because it confuses a display alias with the authoritative identity established by the catalogue’s community.

The cataloguer compares identifiers, works and available authority evidence. If both spellings refer to the same person, the catalogue can support the publisher form as a variant access point without removing the established form. The linked native-script representation remains connected to the proper name field. The change is documented as an additional discovery route, not a retranslation of the author’s identity or a silent edit to the authority record.

A useful test searches both romanisations and the Chinese characters, verifies that the same person is returned, and uses a second person with a similar Latin spelling as a negative control. Without that negative case, query expansion can look successful while collapsing two identities. The distinction matters in bibliographies, student citations and historical records, where assigning one person’s work to another may persist for years through copied cataloguing data.

Case study: correcting a mixed-script series statement

An imported textbook series combines English volume numbering, Japanese characters and a publisher abbreviation. A reviewer wants to translate the series title for a bilingual reading list. The safe scope is to prepare a supplied explanatory rendering, preserving the original series numbering and source statement. The list can use the translation for orientation while the catalogue continues to record what the item actually says.

The team checks whether the series has a controlled access point and whether the publisher has authorised or established an English series title. If not, the English rendering should be marked as supplied. The number “12” must not become “21” because a right-to-left preview or a spreadsheet sort has reversed visual order. The abbreviation may identify the publisher rather than form a common word, so translating it literally could change the bibliographic meaning.

In the finished test, users can open the school reading list, see the explanatory English label and navigate to the precise catalogued series volume. The exported MARC record retains the source statement and $6 links. This result is more useful than a universal translation of every string because it improves comprehension without sacrificing the evidence that distinguishes this volume from an adjacent one.

An efficient diagnostic matrix for cataloguers and translators

Begin triage by asking what failed: linkage, encoding, retrieval, display, meaning, or authority. A linkage problem produces missing or misassociated alternate fields. An encoding problem produces replacement characters or broken exchange. A retrieval problem produces correct visible text that the index cannot find. A display problem changes apparent order or punctuation on screen. A meaning problem produces a fluent but incorrect explanation. An authority problem joins the wrong person, work or edition. Each category requires a different owner and test.

For linkage problems, compare the source and candidate’s tag, two-digit occurrence, script notation and first-subfield placement. For encoding, examine actual stored characters, not the editor’s font choice. For retrieval, inspect tokenizer and index mappings while preserving source forms. For display, compare raw logical order with browser rendering and screen-reader output. For meaning, request a bilingual reviewer. For authority, consult identity evidence. A “translation QA passed” label should never conceal an unresolved failure in another category.

A useful change log records the source record ID, field path, approved text, reviewer, local policy, fixture, parser result, search result and final disposition. When a fix is rejected, record the reason. This allows the next librarian to distinguish an intentional unpaired 880 from an accidental orphan, or a documented romanisation variant from an error introduced by a previous vendor. Good logs make the second revision safer and faster than the first.

Tests should also consider downstream exports. A catalogue may display a linked field correctly, but a reading-list generator might flatten both scripts into one string without labelling either. A citation exporter may omit an alternate title. An interlibrary sharing interface may fail to preserve script identification. These do not always require changing the record; often the receiving adapter must be configured to carry the distinctions already present in the authoritative metadata.

Finally, sample the exceptions rather than only routine successes. Include an intentionally unpaired 880 using the reserved occurrence convention, two pairs in one record, a mixed-script title and a right-to-left field. Include a source record with no supplied translation so the system must abstain from inventing one. Reviewers who never test abstention are likely to mistake automatic completion for accuracy.

Changed-context transfer exercise

A community music archive acquires a booklet with a Cyrillic title, a Latin-script transliteration, a translated English description, and two records for different editions. The first record contains a 245 linked to an 880 using occurrence 04; the second contains a similar title but a different publication date and identifier. A contractor suggests copying the English description into the second record’s 880 so both display similarly. Decide which values can be carried into the bilingual discovery interface and identify at least three checks required before anything is changed.

The correct first step is to keep the two edition identities separate. Inspect each record’s title page, linked fields, edition identifiers and publication dates. Preserve the Cyrillic and transliterated source-title representations where authorised. The English description belongs in an appropriately labelled presentation or descriptive field, not automatically in a linked alternate-script field. Copying it across editions risks claiming that both items have the same contents or title, and it does nothing to prove that their $6 relationships are correct.

Next, verify each pairing by tag and occurrence, inspect script and logical character order, and search for both Cyrillic and Latin forms in a test index. Only reuse the English description after checking that its scope fits the second edition and that its provenance is recorded. A bilingual reviewer should validate its meaning, while a cataloguer decides placement. The exercise transfers the principle from school books to music archives: improve discovery, but do not convert a useful explanatory translation into false catalogue evidence.

Practical review checklist

Confirm the reader task: translation, transcription, transliteration, or bilingual discovery.
Preserve an immutable copy of each full source record and its identifiers.
Map each 880 to its associated tag and matching two-digit occurrence in $6.
Treat 00 unpaired cases according to source evidence rather than automatically repairing them.
Confirm indicators, subfield roles, authority references and protected dates have not changed.
Test Unicode logical order, bidirectional display, search and copy/paste independently.
Use qualified reviewers for name identity, original titles and supplied translated descriptions.
Confirm exact record and edition retrieval with positive and negative test cases.
Verify exchange files parse and linked forms survive catalogue and citation exports.
Record every approval, exception and rollback path before releasing a batch.

Limits and editorial responsibilities

Field 880 linkage QA cannot tell a cataloguer what an item’s authentic title should be, which romanisation authority to follow, or whether a supplied translation is suitable for a particular readership. It also cannot guarantee that every catalogue vendor implements right-to-left rendering or alternative-script indexing in the same way. This article explains a safe method, not a universal rule for every bibliographic scenario. Institution-specific cataloguing policy and authoritative source evidence take precedence over attractive automated output.

Search-engine or discovery visibility should not be used as a reason to falsify the record. Public pages may use descriptive titles and accessible summaries, but the bibliographic source remains an evidence-bearing object. When no defensible translation exists, provide the original title, a reliable transliteration if authorised, and an explicit statement of the limitation. A careful refusal to invent a title is a better reader experience than a fluent misidentification.

Summary

MARC 21 field 880 is a relationship between script representations, not a general-purpose translation box. Preserve $6 links, associated tags, indicators, identifiers and authority boundaries. Translate eligible explanations with documented provenance, then test both semantic meaning and machine-readable pairing. The fastest robust work is the work that never has to repair a library catalogue after a seemingly harmless string replacement.

Frequently asked questions

Is field 880 always an English translation?

No. It can hold an alternate-script representation of another field, such as original Japanese characters paired with a Latin-script romanisation. Language, script and translation are different concepts.

What does the number in `$6 245-02` mean?

It identifies the occurrence used to match an alternate 880 with its associated 245 field within a record. It is not a language, edition or catalogue-record identifier.

Can an 880 exist without a regular partner?

Yes. The Library of Congress describes a reserved occurrence number 00 for specified unpaired cases. A validator should not automatically label every such field as damaged.

Should I translate the `$6` script notation?

No. It is structural notation. Use the applicable MARC convention, including documented script codes and optional orientation information, rather than substituting translated words.

Can the public catalogue show a supplied English title?

It can, provided that the label is identified as supplied or translated, placed according to local cataloguing policy and not mistaken for the authentic transcribed title proper.

Does a valid $6 link prove a good translation?

No. A structural validator can confirm pairing but cannot establish that the two language forms express the correct title or person. Human linguistic and bibliographic review remains necessary.

Why can right-to-left titles look broken despite correct storage?

A display may fail to handle bidirectional text, Latin numbers or punctuation. Inspect the stored logical sequence and fix rendering rather than reversing underlying characters blindly.

Should authority identifiers be localized?

No. Stable authority references and identifiers connect representations across scripts. Localize display information without changing which person or work the identifier denotes.

How can a library avoid creating duplicate books after translation?

Compare authoritative identifiers, edition details and cataloguing evidence before linking or merging records. Two languages or scripts for one title do not automatically mean two different books.

What is the minimum regression test before releasing a batch?

Parse source and candidate records, compare protected fields, validate $6 pairs, test retrieval in both scripts, inspect display and copy behavior, and obtain linguistic and cataloguing approval.