A museum's English catalog says a print was made in Athens. The label on the object gives a different spelling in Greek script. A bibliography uses Athína. An editor suggests that the collection use one standardized title so researchers don't have to put up with three forms.
That suggestion solves a display problem while creating a discovery problem. A reader who copies a label from a photograph should be able to find the record. So should a reader who searches an older catalog's Latin-script form. A reader coming from an English travel book should not be offered a choice among the wrong places merely because several places share the same English spelling. A bilingual catalog works best when it preserves the source and builds several roads to the same entry, rather than deciding one spelling can carry the whole story.
The method in this article is about cataloging public geographic references, not determining where people live or who owns a place name. It treats editorial naming and evidence about actual language use as decisions with sources. A short notebook can use the same approach as a large museum database.
Separate three kinds of variation
One difference is between scripts. A catalog may represent a Greek name using Latin script according to a documented scheme. That is normally called romanization in this context. It changes the written form and does not automatically create a translation. More than one scheme can exist; even a beautifully consistent catalog entry only proves the catalog follows a method, not that every publisher chose the same spelling.
A second difference is between names used in different languages. The United Nations Group of Experts on Geographical Names provides a clean illustration: it distinguishes the Greek name represented as Athína in Latin script from Athens, its English exonym. The script conversion and the different-language name are not the same operation. Write down which one you applied. This is useful even if every reader knows which city the example refers to.
A third difference is temporal. A town, district or feature may carry different names in different periods. A historical name is not necessarily a current translation. A name can also apply at one date or scope to an administrative unit, and at another to a broader informal region. Ask whether the old and new forms refer to the same feature before entering them as interchangeable. Sometimes both are useful to include but need separate relationships and dates.
There are other distinctions: spelling variants, accents, abbreviations, editorial updates, or variants preferred by particular publishers or communities. Don't bury them all under “alternate spelling.” A search interface can group them for convenience while the record labels the relationship. That distinction helps a reader tell whether a result matches the meaning of a search or merely contains a similar string.
Do not let the normalization step consume the source
Start with the exact wording as far as an editor can responsibly transcribe it. Keep the source script or note that the script was not confidently readable. Mark a damaged or ambiguous character. Record where the label appears and how it relates to the object. A printed place beneath a picture could be the subject, issuer, printer's location or part of a series title. You won't solve the role just by correctly translating the word.
Once you have the literal reading, create a separate editorial field for how it should appear to the reader. List the manuscript's dated usage in the record rather than silently replacing it. If a historic caption includes a harmful or obsolete term, your public title may use respectful present-day language without pretending the source always did. Explain the distinction where it affects interpretation, and use institutional guidance for handling sensitive language. Preserve what is necessary for research without deliberately surfacing harmful terms as recommendations.
For romanized search forms, check an institutional table appropriate to the language and workflow. The Library of Congress ALA-LC tables are an example of documented cataloging schemes; existing catalog records may use rules or practices that differ in detail. Adding a romanized search string is not a reason to discard the script material. In an archive that does not have staff who read the script, use an explicit review status rather than authoritatively assigning a language or transliteration by resemblance. If a model supplies a plausible candidate, have a qualified reader or an independent trustworthy reference confirm it before placing it in the public authority list.
Normalize for discovery, not for evidence. A search engine might match diacritics loosely so someone who can't easily type them can still find a record. The stored transcription should retain the characters you actually observed. If you automatically create an unaccented display, someone else may later copy that shortened form back into an exhibit and think it came from the original. Give transformations their own place and refresh them from the stored evidence when your rules improve.
Validate the place ID independently of the string
A string can point to more than one feature. A city can share a name with another city, a region or an ancient site. An archive may use only a name in its original accession description because catalogers believed their audience would know the context. Before consolidating references, compare broad geographic context, feature type, date and any provenance supplied. Add an identifier or authority citation only when the match is justified.
The Getty Thesaurus of Geographic Names FAQ explains why names alone don't identify places: entries may share a name, and broader context and place type are core fields. Getty's editorial guidelines on names show how vernacular, historical and other forms can be labeled within a record. The scope note is also part of the method. The vocabulary is tailored to cultural-heritage work, and is not a promise that every local or current name can be found there.
For U.S. features inside its remit, the USGS Geographic Names Information System overview describes official and variant names, feature attributes and identifiers. Its domestic names guidance distinguishes federally recognized names from variants. Neither is an assertion that the same official name applies around the world, or that a federally recognized designation tells you every way a community names a site. The system's named-feature scope does not cover roads and highways; for a street sign or historical street name, seek local street index, maps and institutional evidence rather than forcing a match in an unsuitable file.
A place authority can corroborate an entity match while having no information about your document's exact phrase. Save the source page and role of both pieces: “object label shows this spelling” versus “authority record identifies the candidate named feature.” A third source may establish association between the two. If the association is only likely, state that without minting a universal equivalence.
Worked example: one Greek caption and one English import
The United Nations example of Athens and Athína is real. The museum, records, acquisition and review steps below are fictional. They show a small editorial workflow and are not descriptions of objects held by an actual collection.
An exhibition database imports two records in a batch. Record A describes a public-square print and gives its city in Greek script as Αθήνα. A 1970s English card for the same print says “Athens, Greece.” A later exhibition list calls the city “Athína.” Record B, a donated print from an unrelated U.S. collection, also uses “Athens” in a caption but has no state given. The automated import suggests merging their place fields because the strings look similar. An editor wants to keep A findable in all three forms without misleading readers of B.
First treat the Greek caption as the observed source, not as proof of where the print was made. The word could identify the pictured city or refer to the subject of the series. Inspect its position and wording and, if possible, find an independent inventory or caption that describes its role. In this demonstration the existing exhibition notes and subject description support the Greek city as the place depicted, but do not establish the workshop location. Put that distinction in the place relationship.
Next create documented retrieval forms for that one resolved feature. Save the Greek string, the romanized version with its source and language, and the English exonym as separate fields. If a local search system accepts accents poorly, consider an unaccented index key as a system-derived form. Label it as search normalization rather than an additional historical spelling. Connect all three to the same authority match only after checking their place role and broader context.
| Value | Role in the fictional record | How to use it |
|---|---|---|
| Αθήνα | Source-script transcription | Preserve and display as recorded, with a review note if needed |
| Athína | Latin-script representation of the Greek name in the UN example | Make available for documented discovery |
| Athens | English-language name for the Greek city; elsewhere also an ambiguous string | Index and disambiguate with geography |
| Greece, city | Broader geography and entity type | Keep with the relevant authority match, not inferred for every occurrence |
| Unqualified Athens in B | Unresolved place | Keep searchable as text but do not attach it to A's feature |
The database can now show a useful title on an English-facing page: “Public square in Athens, Greece,” with the source form presented as an artifact label. A Greek-script search can return A without forcing the entire site to use one language. A bare Athens search might also return B with an unresolved badge; it should not silently promote B to Greek cultural history. If a donor later supplies the U.S. city and state with a reliable record, revise B's match and keep its original text unchanged.
A useful caption for a volunteer editor is “The print identifies the depicted city in Greek; the catalog includes Greek-script, romanized and English retrieval forms.” It doesn't imply the print originated in a workshop at that location. An exhibition label can acknowledge the actual linguistic evidence without requiring viewers to know the technical classification of every field.
Teach the search box to fail usefully
Build three small tests before launch. Search for the exact original text; its item ought to appear. Search for a documented alternate form; the same item should be discoverable without relabeling the source. Finally, search an ambiguous spelling; items that share that string should remain distinguishable by context and unknown matches should not be guessed. Run all three from a reader's view rather than only checking that values exist in a management page. The outcome is useful even on a modest local site.
A table might store item ID, raw place string, script, language when known, source citation, role in the item, authority ID when verified, variant relationship, temporal usage and review state. Have a separate field for display preference. Let query normalization change what a reader can match, but not which relationship the record claims. For a folder or spreadsheet archive, keep a simple controlled list of variant forms and link item rows through local identifiers instead of copying the display title into every sheet.
Time and unresolved scope need their own behavior. If a 1900 label names a district whose extent changed, don't treat all objects labeled with today's district name as identical. Attach the dates and note which named feature the source describes. If an authority lists a historical name as highly likely but not certain, preserve the qualification. Your search interface can still suggest related records without merging their historical descriptions.
For a migration, record whether fields are bilingual at the item level, at the place-authority level, or merely in a view template. This prevents the next volunteer from clearing an old place form after noticing the screen doesn't use it. Validate imported encoding with visible sample names and test the search path of at least one text string in each script used in that sample. An archive can contain the right bytes and still fail its readers if the query processing loses them.
Keep a corrections trail when the keyboard changes
Multilingual records can break in subtle ways during conversion from an older system. One form might have been stored in a font that draws Greek characters but actually contains Latin ones. Accent marks may have been entered in more than one technically valid way. An export might change spacing, punctuation or some characters to placeholders. Before you normalize, take a readable sample from the predecessor and compare it with the migrated result in a modern browser, then try copying text from that result into a query. If the character sequence is not trustworthy, preserve the source screenshot or image locator and mark the transcription for review rather than guessing.
There is a similar trap in translations of evidence. A draft label may have been translated by one person, romanized by another and proofed by a third. Unless the archive stores which version each action applied to, a correction at one step can make the others silently obsolete. Write “translation of transcription v2” rather than putting a revised English sentence into a field called Original. Preserve enough context to reproduce both the editorial display and the source reading if the same image must be recataloged later.
Review one historical mismatch in a maintenance session. Find two catalog entries using an apparently equivalent name and check whether they include different feature kinds or date context. If the match is unsubstantiated, downgrade it to a possible relation and record why. This action need not hide any old description. It simply prevents a machine from converting two interesting local records into one misleading assertion.
Date a name without inventing an official name change
A caption has at least two relevant dates: the date of the described view and the date the caption itself was written or printed. A postcard could carry a new name beneath an older photograph. A family inscription could be decades later than the scene. A modern collection title can identify a place using its current name without showing how people identified it at the time. Record where each date comes from and let the dates remain distinct. A photographed sign, a subsequent annotation on the reverse, and an institution's current preferred label are separate observations.
The start and end of a name's documented use need similar care. Finding a historical label in a 1924 advertisement establishes one source attesting to it then. It doesn't establish a first use date or show that a second label fell out of common use in 1923. A presently official form may coexist with familiar speech in other languages and local descriptions. Authority files, official lists and photographs have different jobs. Where a source documents an official adoption date, keep that date linked to the specific authority and decision rather than advertising that the earlier language disappeared the same day.
A useful data record can document three strengths of time evidence without converting them to one false interval. “Seen on the 1924 caption” is attestation. “Identified by this institution as a historical description” is a classification you can attribute. “Official under this authority from the stated date” is an assertion with a particular source and scope. A local nickname can have many dates of attestation and no legal beginning. A retired cataloging form may have an administrative effective date that says nothing about public speech. If an interface offers only from and to fields, put the fact type somewhere visible or avoid filling unavailable endpoints.
When building a historical timeline, make sure that the date belongs to the use you are documenting. A 1970 title already held on a transparency can be legitimate evidence about the collection practices of 1970 while being a later and weaker witness to a scene from 1910. A modern preferred form can help someone locate the place now. Keep both. An audit can ask “why is this preferred in this view?” without expecting the team to choose a winner across every year and every language. That question is better suited for an archive than deciding a name's entire social history on the basis of two strings.
Use evidence relations more specific than synonym
In general search, all alternate strings may appear to do the same job. In an archival graph, a spelling, nickname and historical place relation should not be automatically declared the same. A foreign-language equivalent can be attached as a name of the same feature after identification. A transliteration can be attached as a representation of a particular local form with recorded scheme and version. A term can be geographically related without naming the same entity at all. “West Basin” could name a neighborhood inside the municipality, a water feature close to it, or the outer part of a harbor authority area. No amount of shared character similarity resolves which kind was used.
Give a name link a direction and source. A synthetic record could say a label in a caption probably refers to a marked riverside settlement based on a local bridge name and transport route; that is different from saying another office directly documented a Latin-script form of the same place name. Preserve the candidate classification, and distinguish what you imported from an institution from what your team inferred. An external subject identifier is worth keeping when it is a good match, but different authorities need not cover exactly the same scope or feature. Connect the records only after checking what each describes. Do not assume same coordinates at map scale are the same identity where sites or named regions overlap.
The UNGEGN modeling discussion offers a distinction between geographical names and different written forms of a name. The Getty name guidelines provide an institutional example of attaching editorial distinctions to recorded terms. Neither turns a local uncertainty into a mandatory synonym. Use source-specific labels when helpful and keep local cataloging decisions about meaning visible enough to review.
Store relationships that could make a shape of several places rather than reaching for a single grouping. A named hill can lend its name to a station built at the base, and a district can borrow that station name. Selecting the hill in an index should not automatically place every station postcard at its summit. Mark the related place without moving the asserted location of a record. For interfaces that offer only one place field, a structured research note with primary place, contextual places and matching caveat protects the underlying evidence from its delivery format.
A second worked example: two Alder Harbors and a spelling that moved
Alder Harbor, the collection, scripts displayed on the teaching card and all observations in this case are synthetic. This example assumes a fictional catalog contains two candidates whose English display strings normalize to Alder Harbor. One is a small waterfront settlement with a tram stop and an east dock. The other is an administrative harbor district whose published boundary extends over two shore settlements. A donated caption contains a local form, which the archive transcribes exactly, followed by an English note reading “Alder Harbour, tram to east.” The collection manager wants to replace all spelling variants with the current regional label.
An earlier ticket brochure links a locally printed equivalent only to the tram terminus. A later public tariff lists the administrative district with a different jurisdictional context. A contact sheet places the donated image among other pictures of the tram but doesn't include a feature confirming which stop it shows. The team logs where the English and local strings were actually printed, and when each source was created. They do not assume the two sources used the same local form with the same scope. A note about the tram increases the plausibility of the settlement candidate without deciding exactly which point is depicted.
The cataloger can record the donor's exact wording, a display transcription with clearly noted corrections if needed, two searchable English spellings and a flagged geographic candidate. The place link remains to an unresolved choice rather than silently assigning the broader district as the only location. The website can display “Alder Harbour (as captioned); likely tram area, exact location not confirmed.” In a general browse query, both place candidates can appear with feature type and a context line. This makes the result understandable to a reader who has never read the cataloger's notes.
Now imagine the team finds a contemporary transit diagram with a label on a stop and another at the mouth of the harbor. The mouth label matches the words at a different scale. The stop label matches a number visible on the car in another frame and a dated route listing. This does not give the photographer's position with pinpoint accuracy, but it can raise confidence that the label under the picture referred to the settlement. A later cataloging change records the new relationship and reasons, preserving the prior uncertainty instead of changing the original caption. If another contributor produces convincing evidence that the car is shown on a branch line, the conclusion is still revisable.
Scope matters if audience queries combine terms. Searching Alder Harbor by itself should probably return both results. Filtering for the administrative district as a political entity shouldn't silently acquire pictures assigned only to the settlement unless defined broader geographic discovery intentionally includes them. A website can offer that broader discovery, but label it as nearby, contained or related rather than “same name therefore same place.” Otherwise a researcher counting historic views of the administration can get a collection of trams merely because the string was convenient.
Test script rendering, indexing and destination separately
Names can arrive correctly in storage and be damaged later. A browser may display accents and direction differently from a print export. A data exchange can silently fail to encode a script or strip a mark on import. A search engine can normalize matching successfully while the view replaces the verified display term with the lossy search key. Make different checks of data fidelity, rendering and retrieval. Passing one isn't evidence about the others.
A compact set of cases could cover exact native-script search, source-faithful display, use of a documented romanized variant, a broad accented or unaccented search, and a deliberately ambiguous place. For each test, write the expected outcome before running it. A normalized search may return several candidates but shouldn't assign one silently. An exact place ID should point to the item described by the scope note rather than the first result matching a name. Do not treat appearance on the result page as confirmation that an item has the correct place link. Open it and check the underlying evidence indication and context.
If the project handles right-to-left writing, verify text direction and surrounding punctuation in the actual publishing contexts. Separators and mixed script can render differently across platforms even where the machine representation is intact. Test the browser and the printable view independently before telling a reader the archival transcription was preserved. A styling repair should not replace the stored text with a manually reordered string that will cause retrieval problems elsewhere. Where support is incomplete, preserve the Unicode source and record the display limitation rather than deleting the source form to make a preview look cleaner.
Romanization choices also deserve separate roles. The Library of Congress's romanization tables document a cataloging system; another institution, local naming authority or community usage may supply equally relevant forms for another purpose. The team should scope each form to why it is present: source written form, recognized searchable romanization, common public English name, alternate recorded wording, or approximate normalized search aid. Mark guesses and machine-suggested forms as candidates needing review rather than assigning them the authority of an institutional table.
Keep reader-facing context proportionate
Not every exhibit placard needs a long note about the history of a label. A short title can use the verified place form appropriate to that audience, while notes give exact source text and other forms. If a historically significant older form is the object of the story, foreground it accurately and explain its source. If the older term is incidental and would mislead modern map users, identify it as source wording rather than silently rewriting the caption. This carries more context with less noise than listing all indexed variants on every public label.
There are also reasons to avoid displaying some retrieved terms with no context. A historical insult, deprecated category or politically charged label may need to remain retrievable for responsible research without presenting it as the chosen public name. Document the editorial and community consultation policies that govern the use in that project. Avoid making a particular unfamiliar language or spelling sound illegitimate just because it differs from a system's default display language. Restore a person's ability to find a record without pretending the retrieval helper has to control what the public title says.
Compare two small public views. A researcher mode can keep source fidelity, authority, term classification and uncertainty visible. A visitor mode can show a concise place identification and a meaningful note when historical wording differs. Both can link to the same underlying record so a future correction doesn't need two different databases. A rendered export can preserve a common citation key even if it doesn't show the entire collection schema. If readers move between views, the main place should stay anchored to the same documented feature, not to whichever equivalent label sorts first.
Check the authority import without replacing local evidence
Authority data can be updated while local records have not changed. The Getty vocabularies FAQ describes identifiers and revision history as tools supporting change tracking. Log the import time and relevant external data version or other locator. When a default form changes, review the kind of change before applying it. A corrected typo in a public label is not a renamed municipality, a source transcription correction or a determination that two sites are identical. Some updates can be automatic while identity changes need review by the people responsible for the collection.
A safe handoff includes original local record ID, external authority and identifier, scope match, confidence and checked date. Keep the exact imported text separate from the editor's note about it. If the external system changes its display label, the archive can retest linking without invalidating the older citation. If the team discovers an old mismatch, correct the connection with supporting evidence and maintain a note for people following it from older research. Avoid redirecting one local record into another solely because they shared a weak candidate link. Need for corrections should not be treated as loss of confidence in the whole collection.
Use a small exception report instead of a global hunt for every variant. Flag unreviewed machine forms, exact retrieval failures, overbroad matches, unresolved scope changes and any cases where chosen titles changed unintentionally. Prioritize test cases that would actively mislocate a reader, such as two identical names in the same search language with distinct feature types. One resolved collision can help define how the team handles a class of records without assuming every unknown term should be folded into the same label. The goal is not a perfect universal name. It is preserving enough evidence to tell someone which form they found and which place, if any, your archive can establish.
Handle identical names differently from changing boundaries
The name of a place can remain recognizable while the geographic feature described by the archive changes. A name printed on a label for a parish may be applied to a municipality formed later. A neighborhood within a harbor district can use a label that also appears in the modern municipal address. Treat the question as one of historical coverage, not merely names: which type of entity did this source describe, and what larger area or relations does it place around that entity in that period? Even a strong name candidate can retain uncertain spatial scope.
Make at least two records when clear evidence shows two identifications with different feature classes. Give each a concise public qualifier, such as “parking area” versus “municipality,” or “river name” versus “historic settlement.” Do not solve duplicate display strings by storing editorial qualifiers inside the original transcription. Put them beside it. This lets anyone using the catalog distinguish an address cue from a source's exact wording. If scope is still unresolved, show the disambiguation question where the result matters; returning two records is sometimes evidence the search works as designed, not a bug.
What if an archival project only knows approximate location near a landmark? A reference to the broader area can be useful as explicit context even when the specific candidate remains uncertain. A map pin, however, can make that temporary context look like an established photography location. Use a non-point summary, a broad geographic range where defensible, or a place-context tag that is clearly not the source subject. Add the exact source annotation to the research record. A display intended for neighborhood exploration should not imply that a reader can use an unverified location to find the structure today.
Ask two team members with different language familiarity to try the same restricted set of test items using the names they would naturally type. Check whether their results point to the same documented object when appropriate, and whether the view distinguishes near but distinct items. You don't obtain a universal truth about local usage from two tests; you identify a concrete problem with the current archive. If their search behavior reveals an unexpected term, save its context and seek documentation or appropriate consultation before making it a definitive name. Most indexing improvements can preserve legibility without rewriting the source text at all.
Make the editorial choice visible
A public exhibit can use the current local name followed by a familiar English form, a historically appropriate name with modern orientation in parentheses, or the source wording plus a brief note. Neither this article nor a database automatically decides which is best. Choose for the audience and historical context, and explain the choice when it could be mistaken for a source quotation. Ask appropriate local or language expertise when the consequences matter. Don't treat a source printed decades ago as the sole voice for a community today.
A good maintenance checklist asks: Can a reader copy text from the image and find the record? Does the screen distinguish translated captions from literal transcriptions? Are two names being matched only because they look alike? Can the original evidence survive a change to our current editorial label? Keep the answer about identity separate from whether a search result looks tidy.
Naming work succeeds when a new writer can extend the archive without flattening it. Let the old label remain legible, keep relevant forms and their sources, and give each piece of evidence a route to the right object. A place can have several names on a page without forcing visitors to choose which one is real.
AI disclosure: Drafted with AI assistance for Apiary. The museum records and decisions are fictional; the linguistic example and institutional data practices are cited. The example does not declare a preferred name for unrelated records.
Source notes
- UN Group of Experts on Geographical Names: Data modeling examples, including Athens and Athína.
- Library of Congress: ALA-LC Romanization Tables.
- Getty Vocabulary Program: Vocabularies FAQ, TGN names guidelines, and TGN scope.
- USGS: GNIS primer and Domestic Names.