In an index, “García Márquez, Gabriel”, “G. García Márquez” and “Gabriel García Marquez” are three different people. Anyone searching under one of those spellings does not find the books filed under the others. Authority control exists so that the three point to the same place: one form of the name, decided once, for the whole catalog.
When Collect reads a document, the model proposes a form of the name from what it sees. That proposal is not taken as good on its own. It is checked against the lists the institution already uses, and the record states which of those lists the final form came from. That trail is what makes the name trustworthy without typing it again. Why a catalog built with AI can be trusted is in Why trust a catalog made with AI.
The house catalog first, then the world
For a personal name, Collect follows a cascade and stops at the first rung that answers. Before it queries anything, it normalizes the form: it turns the name to “family name, given names” and strips courtesy titles (“Lic.”, “Dr.”, “Mr.”). With that clean form it looks up, in order:
- The institution’s authority file (the ILS: Janium, Alma, Sierra or Koha). If the house already fixed an authorized form for that author, that form wins. It is the highest confidence, because it is the decision the institution already made.
- VIAF, the international file where national libraries record the authorized form of each author. Birth and death dates, when the record has them, help tell homonyms apart. If VIAF resolves the name, the record also carries the link to that entry (the VIAF identifier).
- The name normalizer, if no list had the author. It leaves the name well formed —family name first, without titles— but with no external backing and without that link.
The cascade steps down only when the previous rung finds nothing. The order is also descending confidence, and the record keeps which rung resolved it.
Provenance is written in the record
What distinguishes a verified name from one that is only well written is not a promise: it is a datum in the record itself. A name resolved by VIAF arrives with the link to that entry. One that took the form from the house catalog arrives with that form —the one the institution had already fixed— and without the international identifier. One that only went through the normalizer arrives well written and also without that link. The link marks a match with VIAF or with ISNI; its absence does not mean the name is badly written, it means there is no identifier from those lists.
For subjects the mark is a source code in the same record (the $2 subfield in
MARC): it says against which vocabulary the term was validated. A term without
that code was not checked against any list.
The effect for the cataloger is direct: open the record and see which names come backed by an authority and which ones are worth reviewing, without checking them all. No inferred name is presented as if it were verified.
Organizations: companies, agencies, institutions
Entity names —a publisher, a ministry, a company— follow their own cascade, because the lists that cover them best are not the same as for persons: the local catalog first; ISNI when the institution has it on, because it covers organizations better; VIAF as fallback; and last the entity normalizer.
That normalizer does work company names usually need: it strips the legal form (“S.A. de C.V.”, “S. de R.L.”) and puts the subordinate unit in its place, so that “Editorial Planeta, S.A.” enters the index as it should and not as a literal string with the legal name stuck on.
Subjects: the right list for each region
Subjects are not resolved the way names are. A subject often comes from the Library of Congress vocabulary, in English, and has to be taken to the region’s controlled vocabulary. Collect does that by querying id.loc.gov, which links each subject to its official equivalent in other national lists: EMBNE in Spain, RAMEAU in France, GND in Germany, among others.
When there is no official vocabulary for that region, Collect translates the term using local terminology. “Computadora” in Mexico, “computador” in Chile and “ordenador” in Spain are not the same word: the translation uses the region’s form, not a generic one. For Colombia it uses LEMB terminology. Each subject goes out with the code of the source against which it was resolved.
Art: the Getty vocabularies
An art collection uses specialized lists, the Getty ones: ULAN for artist names, AAT for object types and materials, TGN for places. Collect queries them for work records, and here the rule is stricter than elsewhere: it links only when there is an exact, unambiguous match.
The reason is concrete. Getty text search ranks by relevance, not exactness: “Rivera Diego” returns “Museo Diego Rivera” first, and “Zamora” returns thirteen different places. Taking the first result would link to the wrong concept, and a wrong link is worse than none: it looks precise where it is not. If there is not exactly one concept whose name matches the term, Collect does not link and leaves the term unmarked.
Limits
- Homonyms remain human work. Two “Juan Pérez” born in the same decade, with no dates to tell them apart, are not disambiguated by any list. Collect leaves both in catalog form; deciding whether they are the same person is done by whoever knows the collection.
- Names no list knows —a local official, a company that is not in VIAF or ISNI— still enter well written, family name first and without titles, but without the authority link. They are known to be unbacked precisely because that mark is missing.
- Bringing variants under a single heading for the collection is something Collect does not yet do. Today it leaves each name in its authorized form; unifying one author’s spellings into a single heading for the holdings is a planned extension, and we will say so when it is ready.
- The form is always corrected, even when there is no match. Name normalization happens before any list is queried; what external validation adds is provenance, not good formatting.
None of these lists replaces the cataloger: they take away the repeated typing and the manual matching of spellings, and leave the judgment —homonyms, people no international list records— where that judgment is needed.
If at your institution the work goes into rewriting names and matching spellings of the same author, write to us at info@janium.com; we want to know which authorities you catalog against today.