Are We Thinking About Ontologies Wrong?
The article challenges RDF’s focus on global IRIs, showing how conversational and narrative contexts rely on deferred classification, consensus and scoped definitions rather than fixed identifiers. It outlines approaches—contextual definitions, master data management, local IRIs and self-learning AI—to build flexible, locally scoped ontologies.
- Real conversation shows meaning is built incrementally through context, not looked up via fixed global identifiers like RDF's IRIs/URIs
- Understanding of a term (like "bill") solidifies through accumulating clues and social consensus, mirroring how ontologies actually form in practice
- Different organizations naturally build conflicting local schemas for the same concept (e.g. "Customer"), so forcing one universal identifier without first aligning schemas creates clashes
- True interoperability comes from scoped, negotiated consensus between authorities rather than dumping all terms into one master global ontology
RDF’s reliance on global identifiers–URIs or IRIs–breaks down the moment you try to map real conversation into triples. At a cocktail party, someone mentions “Jane Doe’s bill,” sponsors, committees and floors, and you immediately build an understanding without any formal ontology. You don’t pause for a dictionary lookup; you infer meaning from context. A bill here is a proposed piece of legislation, tied to committees and votes, not a restaurant receipt or a bird’s beak. All those alternative senses exist in your head, but you narrow them down on the fly.
As you move around the party, you pick up more snippets–another vote failed and sent back to committee, a separate bill passed and moved on to the Senate. Each new piece nudges your internal definition toward consensus. You drop irrelevant properties (state of origin) and add process details (voting outcomes). In effect, the “class” Bill takes shape through shared experience. That mirrors how real-world ontologies emerge: groups negotiate terms until most members agree, then treat that agreement as a standard.
Apply this to organizations and you see why global IRIs alone can’t enforce uniform meaning. One company’s “Customer” ontology might require address validation, another’s might focus on purchase history. Both models work internally but clash if you demand a single universal identifier without aligning schemas first. In practice, identifiers stay local to each authority, and true interoperability comes from scoped consensus, not from dropping every term into one giant, pre-made ontology.