Putting Shared Meanings on the Web

Two catalogues both use the word “mercury.” One means a chemical element. The other means a planet. A third collection includes a journal with that title.

The spelling gives a search system something to match. It does not tell the system whether the records concern the same thing. Library authority work had long dealt with that difficulty. The web made it possible to share the answer across institutional boundaries, provided the answer had an identifier other systems could use.

This is where the history of library vocabularies meets the standards that many ontology practitioners use today.

A name that another system can follow

Tim Berners-Lee’s Linked Data principles describe using web identifiers for things, supplying useful information when those identifiers are looked up, and linking to other identifiers. A shared identifier can connect descriptions without requiring every organization to maintain an identical local label. Berners-Lee, Linked Data

RDF provides a common structure for such statements: a subject, a relationship, and an object. A record might say that a particular report has a particular person as its author. The person and report can have identifiers; other statements can describe them. W3C, RDF 1.1 Concepts

The relationship is part of the statement, not a line whose meaning the reader has to guess. That brings us back to the problem early AI exposed.

Imagine two firms exchanging project information. Each has a project called “Riverside.” The title is a poor joining key. Stable identifiers let the firms keep the projects distinct, while an explicit mapping can record that two particular identifiers do refer to the same project when evidence supports it.

An identifier solves a reference problem. It does not certify every claim attached to the thing it identifies. A mistaken completion date remains mistaken even when its subject has a beautifully maintained web address.

Library vocabularies did not need to become something else

SKOS, the Simple Knowledge Organization System, became a W3C Recommendation in 2009. It provides a web-compatible way to represent concept schemes such as thesauri and classifications. Alistair Miles and Sean Bechhofer edited its reference; Antoine Isaac and Ed Summers edited the accompanying primer.

SKOS distinguishes concepts from the labels used to name them. It supports preferred and alternative labels, broader and narrower relationships, related concepts, and mappings between schemes. That is valuable for existing library and information work without requiring every vocabulary to become a detailed logical theory. W3C, SKOS Reference; SKOS Primer

Suppose a company uses “lift” and a supplier uses “elevator.” A concept can carry the labels people need, with language and scope handled explicitly. But “vertical transport” may be broader than either. A mapping that declares all three equivalent would make some searches easier while quietly losing a distinction.

The ontologist’s work is to inspect that difference, not simply fill the mapping column.

This is also where early classification habits need care. A broader subject relationship is not automatically a formal subclass relationship. A vocabulary may arrange topics for browsing in ways that do not justify saying every instance of one class is an instance of another.

That difference matters when software begins drawing conclusions from the arrangement.

OWL asks for stronger commitments

OWL, the Web Ontology Language, can express logical relationships among classes, properties, and individuals. Its commitments allow reasoners to derive consequences and identify certain inconsistencies. The language also uses an open-world approach: something absent from the recorded knowledge is not automatically false. W3C, OWL 2 Primer

Consider an equipment record without a manufacturer. An application may need to reject the record as incomplete. A logical ontology need not conclude that the equipment has no manufacturer. It may simply lack a statement identifying one.

Those are different judgments.

SHACL provides a way to validate RDF data against specified shapes, including requirements about values and relationships. A shape can say that records submitted to a particular application must include a manufacturer identifier. The resulting failure tells the team that a requirement has not been met; it does not establish that the physical equipment was never manufactured. W3C, SHACL

You do not need to memorize the acronyms to use the distinction. A vocabulary organizes terms. A logical model states relationships from which conclusions may follow. A validation rule checks whether supplied records meet a requirement. A project may need all three, but asking one to stand in for the others creates avoidable confusion.

The ordinary work behind a linked model

A professional vocabulary needs someone to maintain its identifiers, definitions, labels, and history. If a term changes, the team must decide whether it is a new label for the same concept or a different concept altogether.

Suppose “active project” originally meant any project with an open financial account. The organization later decides it means a project with work undertaken in the past ninety days. Renaming the definition in place can change the apparent meaning of old reports.

A versioned definition, a new concept where appropriate, and a recorded mapping give analysts a chance to compare like with like. The technology makes those decisions expressible. It does not make them on the organization’s behalf.

The web expanded the audience for a description. It also expanded the consequences of an ambiguous one. Readers outside the originating team need enough context to use it sensibly.

Libraries were meanwhile examining an ambiguity inside one of their most familiar words. When a reader asks for a book, do they mean the work, a translation, a published edition, or the particular copy sitting on a shelf?