Everything is connected ... really ... Putting meaning to work

At its heart, the Semantic Web is a standardized model for distributed creation and referencing of “linked data”—assertions about concepts and their relationships. Concepts are uniquely identified by uniform resource identifiers (URIs). Berners-Lee envisioned this vast web of Semantic metadata making documents more accessible to search engines, but the model can be used—is being used—for a variety of requirements, including development of standard vocabularies within domains and enterprises.

Hundreds of commercial tools—thousands if you include open source and academic products—have sprung up in this simple but fertile ground. Try Michael Bergman’s “Sweet Tools” resource at: http://www.mkbergman.com/new-version-sweet-tools-sem-web, which contains over 800 tools at this writing. You can also find any number of forums on the topic online, and you can get up close and personal with the SemWeb experts at Semantic Web Gatherings across the country (see meetup.com).

If interest in the Semantic Web is strong in the United States, it might be even stronger in Europe, where projects spanning countries and private/public boundaries are well funded.

You can find my own brief take on the Semantic Web at: semanticadvantage.com/pdfs/Sarkozy%20bites%20obama%20child.pdf.

We will have to continue to live in a world of fragments and miscellany, but a meaning-based approach provides the connections that enable substantive, measurable and persistent improvements to knowledge work.

The terminology of semantic applicationsYou can use words to describe corporate policies, disaster recovery procedures, the functions that a person or application performs and similar aspects of the realities of the enterprise. But you end up with ... just words. You aren’t one step closer to turning that information into value. You can’t find or manage that information predictably. You can’t build more complex and useful tools from words themselves. You cannot use that information to detect duplication of effort or anomalous activities.

To understand semantic applications like semantic networks or metadata systems based on Semantic Web principles, you have to start by adopting a very simple notion—that you must deal with concepts rather than words. A concept is a discrete thing (tangible or abstract), often correlated with a single meaning of a word or phrase. (Conversely, several different words or phrases might describe that same concept.)

It’s easier to put that distinction to work if you think of a concept as something that exists across the boundaries of natural languages. Unfortunately, that’s not absolutely true, because language influences our understanding of concepts. Most of us understand that even the names of physical things might have different connotations in other languages.

In semantic networks and other knowledge representation (KR) formalisms, a concept—often presented as a node in a graphic network of symbols (like circles) and arrows—may be described by several different words or phrases. The label is necessary, but the meaning of the concept is embodied in its relationships to other concepts. In most KR formalisms, relationships are labeled to indicate the meaning of the relationship. Relationships are also referred to as links, arcs or edges. (Unlabeled relationships among nodes are more characteristic of “mind mapping.”)

In the Semantic Web, whose foundational model is the Resource Description Framework (RDF) triple, relationships are often referred to as predicates. But Chris Menzel (now of Texas A&M University) noted in a posting to the Ontology forum that “ ... there are no predicates per se in RDF, there are only names [more specifically, uniform resource identifiers (URIs) and literals] that denote resources. Predicate itself just indicates a role that a name can play in an RDF triple—it is the name that ‘connects’ the other two names in the triple.”

In any case, such a relationship is an assertion that concept A is connected to concept B in a specific way. Central to any understanding of the Semantic Web and computer ontologies in general is that formalized relationships among concepts typically support inferencing by software applications—for example, that anything asserted as a characteristic of primates is by inference also characteristic of all humans, a subtype (or subclass) of primates. Human implies primate, but not the other way around. (Thanks, Samir Batla of EMC, for the example.)

I also propose that we make an explicit distinction between concepts and ideas. I define an idea as an expressed observation about reality, the equivalent to a simple sentence in natural language. You might think of ideas as concepts connected by syntax—but in a way that is more rigorous than in natural language. That distinction is very useful because in argumentation/discussion support systems (as well as in most mind-mapping applications), the nodes are usually ideas, not concepts. Labeled relationships between those ideas often reflect influence, causality, constraint (conditions) or opinion.