Verb Concepts: W3C/RDF vs. SBVR

The semantic web has stalled for LLMs. The future for AI lies with SBVR-style concept models.

If you look at the W3C stack (RDF/OWL), a relationship such as "Alice knows Bob" is broken into modular pieces. The property "knows" exists entirely in the wild, decoupled from the entities it connects. Neither RDF or OWL has any name, notion, or structural container for the unified concept: [Person] knows [Person].

In contrast, SBVR (Semantics of Business Vocabulary and Business Rules) and treat this as an atomic, linguistic building block – a "verb concept."

This architectural divide reveals three critical insights for data architects today:

1. RDF allows one system to use "knows" for Person-to-Person friendships, while another uses it for Organization-to-Organization partnerships. They integrate perfectly at a technical level, but any shared operational meaning is completely lost. Without strict structural ‘glue,’ automated integration is a technical illusion that leads straight to semantic chaos.
2. The distinction becomes critical as Large Language Models take center stage. LLMs natively process knowledge through natural language linguistics and grammar – they inherently think in verb concepts (predicates). Forcing an LLM to map business knowledge into pure, deconstructed labels and properties, or OWL axioms, creates significant structural friction. (Read ‘churn and misinterpretation.’)
3. While SHACL (Shapes Constraint Language) was introduced to bring back rigid structural checking to knowledge graphs, it acts as a validation engine for data and code. It strips away the human-readable, business-facing vocabulary that SBVR preserves. It is not predicate-and-sentence friendly.

To build scalable, AI-ready knowledge architectures, strings like “knows” cannot be treated as standalone, ungrounded predicates. True interoperability comes from knowing which things can – and more importantly cannot – relate.