Wednesday, May 16, 2007

imho Ocean Semantic....


Semantic interoperability requires shared understanding by all parties of the full context, content, and consequences of their actions within a given application domain.

The focus of the W3C's Semantic Web initiative has been on shared understanding as crystallized and governed by domain-specific "ontologies." An ontology is a grammar and vocabulary for representing entities and relationships, constraints and containments, applications and manifestations within a domain. RDF, with its subject-predicate-object "triples," is one such grammar, and OWL expands RDF's expressive power for making meaningful, machine-readable, URI/XML-based statements about any entity and any relationship in any domain that chooses to represent itself online.

But ontologies can't be shared unless two or more parties choose to agree on them, or if they are mandated by whoever controls the domains within which two or more parties agree to interact. And they can't serve as a stable, broadly understood and accepted common denominator unless they are under governance by some recognized authority within the domain they purport to support. Which leads to some key questions, where semantic inteoperability is concerned. What is a semantic domains? What authorities control the domains? From what sources do these domain authorities, controllers, governors, or stewards derive their authority? What procedures do they use to define, disseminate, administer, and revise ontologies for their domains? How do they square, coordinate, reconcile, and federate their domain-specific ontologies with ontologies in other related domains, or with foundation ontologies that apply more horizontally?

Semantic domains are springing up all over, often as B2B data-standardization initiatives within particular industries. Look at the RDF and OWL specs for a glimpse at domain authorities in the birthing. Under RDF spec, section 6, "Some RDF Applications: RDF in the Field," note the intiatives and their scope/applicability/institutional backing: "Dublin Core Metadata Initiative" (ECM vendors, digital library operators, electronic publishers), "Publishing Requirements for Industry Standard Metadata (magazine publishers), "RDF Site Summary" (online content syndicators and aggregators), "Common Information Model" (electric power research institute), "Gene Ontology Consortium" (biomedical researchers), and "Describing Device Capabilities and User Preferences" (mobile device, infrastructure, carrier). And that document's a couple of years old...there are many fresher examples of horizontal, vertical, regional, B2B, intra-corporate, and other ontology-making bodies and initiatives.

Of course, an ontology is only one component in a "domain model," which should also define the governance, roles, workflows, data matching/cleansing, exception-handling, and master-data-management rules, and other policies and best practices that prevail in that (real or virtual) (B2B or intra-corporate) slice of the universe. Taken all together, the domain model is the practical/institutional context within which all participants adopt a common, controlled vocabulary (i.e., ontology) to catalyze shared understandings.

Notice the word "federation," earlier in this post, in the same sentence as the word "ontologies." Shared understandings are hard to catalyze when the world is fragmented into fiefdoms. But that's the world we will always live in....the SemanticWeb governance structure is and will remain indelibly federated.

More to come.