While, naturally, it's possible to encode Web links in XML (e.g., with <a href="..."/>), note how little in common a Web link has with the XML data model: Web links use URIs instead of tags or QNames; unlike XML, Web links impose no inherent hierarchy, no notion of containment, and no sequencing of the things to which they relate. In fact, a set of Web links doesn't look much like a DOM tree, but does look an awful lot like an RDF (Resource Description Framework) graph, where each link corresponds to an RDF triple. After all, each part of an RDF triple can be, and most often is, identified by a URI, and the structure of a triple is obviously tripartite.

The parts of a triple, by design, correspond to the parts of a Web link:

• The subject of a triple is where you start.
• The predicate connects the subject and the object.
• The object corresponds to the target of a Web link.

RDFS is a simple language for creating Web-based "controlled vocabularies" and taxonomies. The language defines several RDF predicates for making links between concepts. Most notably, RDFS defines class and property relations and provides a mechanism for restricting the domains and ranges of the properties. In RDFS, for example, we can express site-specific Yahoo-like Web site categories as a hierarchy of classes with sets of named (sometimes inherited) properties. This allows other sites to connect their own information to these terms, providing better interoperability for use in B2C, B2B, or other Web transactions.

OWL extends RDFS into a more capable language, usable for thesauruses and domain models. OWL is based on DAML+OIL, a Web language jointly developed by the U.S. Defense Advanced Research Projects Agency (DARPA) and the European Union's Information Science and Technology (IST) program. DAML+OIL has begun to get heavy use in the government, and in November 2001 the W3C chartered a Web Ontology Working Group to refine the DAML+OIL standard into a W3C recommendation language, now called OWL.

OWL extends RDFS with many more constructs for defining the relationships between classes and, more important, placing restrictions on how properties (i.e., predicates) can be used when linking entities. OWL thus allows users to define simple models of their domains using these properties and their constraints. A full discussion of the language is beyond this article (see www.w3.org/2001/sw/WebOnt for more details), but revisiting the "Address" example should make a few things clearer.

XML and OWL
In XML we were able to say that there was a document field called a <US-address> that had subfields of a street address, a city name, a state name, and a zip code. In OWL we can explicitly name these objects as classes and properties, and place constraints on how to legally relate these entities to each other or to entities defined in other documents. Thus, for example, we could mention that cities are in states, and that each city is in one, and only one, state. We could know that a U.S. address is a type of international address where the state field is restricted to be one of Alabama, Maine, New York, and so on, and that these addresses have zip codes that consist of either five or nine numbers. We could also add the information that, in general, international addresses have country codes, and that the country code for U.S. addresses always has the value "USA," and many other such facts. (This sort of specification of these relationships in a formal language is called ontology, thus the term ontology language for OWL.)

Ontologies let us more precisely link to other documents and resources based on shared use of conceptual terms, even where there is only a partial match (a key difference from current XML-based approaches). Our addresses could thus be linked in turn to other vocabularies - for example, knowing that an address names a location, we could link to other location-based Web resources. These could be databases or Web services that would compute the location of the nearest airport (another kind of location) to a given address, the weather forecast for the city the address is located in, or other such location-specific data. Metadata can also be used to link nontext media to ontologies, expressing, for example, that the photo in a particular picture is of a house at a particular address or that the place to complain about the contents of a particular streaming video is in a particular state (allowing you to compare its location to yours and see if local content restrictions might apply).

Figure 1 provides an example of the linking of a Web site to ontological information. In this case, from a presentation on OWL given at the W3C session of the World Wide Web Conference in May 2002, information about the keynote speaker is linked to information about events, photos, and people.

The example of addresses is an extremely basic one, yet we already see a tremendous number of possible uses. By mapping the implicit semantics inherent in XML DTDs and schemas into the explicit relationships expressible in RDFS and OWL, a whole range of new applications, largely created by the linking of existing Web resources, become easily implementable. In the business world this kind of linking to models could be done for SEC filings, supply-chain databases, business services posting WSDL descriptions, and a virtually infinite range of others, allowing enterprise integration on a Web-wide scale. Current Semantic Web-related research is also exploring the use and extension of these RDF-based languages to express trust and authorization relationships, to do the automated discovery and composition of Web services, and to design new languages to continue to enhance the potentially revolutionary capabilities of the Semantic Web.

The Semantic Web is being built on models based on the RDF representation of Web links. To achieve their full impact, however, the enhanced models enabled by the Semantic Web crucially need to be tied to the document-processing and data-exchange capabilities enabled by the spread of XML technologies. If XML- and RDF-based technologies were incompatible, as some people seem to think they are, it would be a true shame. But, in fact, they aren't. While the underlying models are somewhat different, the normative document exchange format for RDF, RDFS, and OWL is XML. Thus, to those preferring to think of the whole world as XML based, RDF, RDFS, and OWL may simply be thought of as yet another XML language to be managed and manipulated using the standard toolkit. To the RDF purist, the documents and datasets being expressed in XML and XML Schema can anchor their models with interoperable data. To those focused on the world of Web services, SOAP and WSDL can carry, in their XML content, RDF models expressing information that can be easily found, linked, and discovered.

Of course, as is the case with any groups doing overlapping tasks, there is friction between some in the RDF community and some in the XML world. RDF folks often complain about the (for them) superfluous intricacies of XML. XML experts shake their heads at the way the RDF/XML serialization abuses QNames and XML Namespaces and treats certain attributes and child elements as equivalent. However, these kinds of complaints are nothing new. In fact, they're common in the XML community itself: witness the fury that some XML people express over XSLT's use of QNames as attribute content (to pick one example). Similarly, the RDF world has plenty of dark and overcomplicated corners. Both sets of languages are also continuing to evolve, and each is also exploring new non-XML syntaxes (consider Relax-NG, XQuery, and XPath).

Best of the Best
In short, the Semantic Web offers powerful new possibilities and a revolution in function. These capabilities will arrive sooner if we stop squabbling and realize that the rift between XML- and the RDF-based languages is now down to the minor sorts of technical differences easily ironed out in the standards process or kludged by designing interoperable tools. Combining the best of all these languages, and their variants, is easily enabled by the combination of the "documents" of the XML Web with the "links" expressed in RDF. Throw interoperable Web services into the mix and the vision is compelling. The future of the Web can be even more exciting than its past, and pulling all these threads together will get us there.

Useful Links