xml-dev - Re: Services-based automation (WAS RE: Realistic proposals to the W3C?)

[Martin Bryan <mtbryan@sgml.u-net.com>]

> 
> A basic edge-labeled graph defines its structure mainly by
> edges, keeping its values or data in the nodes.  The node
> model for xml uses nodes to delineate the structure as well
> as to hold data.  The edge-labeled model has a cleaner
> separation of structure and data.  I happen to like working
> with a edge-labeled model better, the DOM uses a node model.

While the distinction between edge-labelling with RDF and node-labelling with XML/DOM is nice it misses the real point. Both are required, neither is fixed, and, in most documents, neither is done correctly. This is why Topic Maps has both (Topics for node labelling, Associations for edge labelling) and why Topics assign names to occurrences within other resources rather than to trying to apply the labelling to the objects themselves.

Think of your average HTML document. How much (semantic) meaning is there in any node label? In 99% of tags, none. Are Word documents (even those stored as XML objects) any better? Or an XML-encoded Star document? No. To apply meaning to such documents you have to associate a meaningful term with the node. There may be more than one relevant term to describe the meaning. Different communities (linguistic, cultural or  commercial) use different terms to identify the same meaning. You therefore need a mechanism for assoication multiple terms to a single node. This is what Topic Maps do. (Unfortunately, despite many years of screaming on my part, Topic Maps fail to require you to record the meaning of the term!!)

Until we learn to assign culturally acceptable terms to both nodes and edges (and to use edges to identify process boundaries!!) we cannot hope to assign meaning to the resources of the web (which is how I interpret the otherwise meaningless phrase "semantic web".)

Martin Bryan
CEN/ISSS working group on Defining and Maintaining Semantics and Datatypes