Communications - March 2009

letters to the editor
DOI: 10.1145/1467247.1467250
children’s magic Won’t Deliver the semantic Web

To explain the nature of “Ontologies and the Semantic Web” in his contributed article (Dec. 2008), Ian Horrocks, a leading figure behind the theory and practice of Description Logics (DLs), employed analogous characters and language of the fictional Harry Potter children’s novels. Notwithstanding the fact this did not help readers not already familiar with Potter or even those, as there may exist a few, who find the novels utterly boring and repetitive, hearing the same story over again in a new guise prompts me to ask: When will such presentations evolve from toy examples into more realistic accounts of larger, complex ontologies? That is, when will the important issue of scalability in the storage, retrieval, and use of large ontologies (millions of concepts, hundreds of millions of roles/attributes, nontrivial reasoning) be addressed?

Horrocks wrote, “A key feature of OWL is its basis in Description Logics, a family of logic-based knowledge-representation formalisms that are descendants of Semantic Networks and KL-ONE but that have a formal semantics based on first-order logic.” While this may be true, it could also mislead a neophyte to conclude that DL is somehow the only formalism for representing and using ontologies. This is far from true. There is at least one alternative formalism, also a direct descendant of KL-ONE—Order-Sorted Feature (OSF) constraint logic —that a

lends itself quite well to the task. Elsewhere, I also covered how various DLs and OSF constraint logics formally relate to one another.^b

The trouble I see in such publications by influential members of the

a Ait-Kaci, H. Data models as constraint systems: A key to the semantic Web. Constraint Programming Letters 1 (Nov. 2007), 33–88; www.cs.brown. edu/people/pvh/CPL/Papers/v1/hak.pdf.

b Ait-Kaci, H. Description logic vs. order-sorted feature logic. In Proceedings of the 20th International Workshop on Description Logics. Lecture Notes in Computer Science. Springer-Verlag, 2007; sunsite.informatik.rwth-aachen.de/Pub- lications/CEUR-WS/Vol-250/paper_ 2.pdf.

World Wide Web Consortium (W3C) is that one particular formalism—DL— is being confused with the general issue of formal representation and use of ontologies. It would be like saying Prolog and SLD-Resolution is the only way to do Logic Programming. To some extent, the LP community’s insistence on clinging to this “exclusive method” has contributed to the relative disinterest in LP following its development in the 1980s and 1990s. Similarly, DL formalists have built a de facto exclusive reasoning method—Analytic Tableaux—into their formalism so the same causes always result in the same consequences.

Whether the various languages proposed by the W3C are able to fly beyond toy applications has yet to be proved, especially in light of the huge financial investment being poured into the semantic Web. To realize this promise, we must not mistake the tools for the goal. Indeed, while DLs are admittedly one tool among several for representing and using ontologies, the goal is still to make semantic Web ontology languages work, no matter which method is used, as long as it is formal, effective, and efficient on real data. Otherwise, the semantic Web might well end up being built on nothing more than children’s magic.

hassan aït-Kaci, Vancouver, Canada

author’s Response:

The Harry Potter example was not intended to be representative of realistic application ontologies. As I discussed in the article, such ontologies are often large and complex, making them unsuitable for didactic purposes.

I certainly didn’t mean to suggest that DL is the only possible formal basis for an ontology language. However, it is important to agree on the use of some formalism in order to facilitate the exchange and reuse of ontologies and encourage the development of the tools and infrastructure needed for large-scale ontology development and deployment. This is a major success of RDF and OWL; users now have access to a previously undreamt of range and

quality of tools and is a major factor in their popularity.

Finally, the W3C standards relate only to the languages themselves, leaving the design and implementation of tools to developers. The OWL standard does not specify any particular reasoning algorithm, and existing OWL/DL reasoners are based variously on (at least) analytic tableau, resolution, hyper-resolution, query rewriting, saturation, and rule-extended triple stores.

ian horrocks, Oxford, U.K.

Give me the science of
Virtualization, not Buzzwords

The “CTO Roundtable on Virtualization, Parts I and II” moderated by Mache Creeger (Nov. and Dec. 2008) was a rambling discussion filled with vague assertions, buzzwords, and brand names but few clear concepts. The anecdotal discussion touched on cloud computing, late binding, even the terror attacks of 9/11, without clear logical sequence or relationship with deeper (unstated) definitions or principles.

As far as I know, VM is an operating system concept first implemented by IBM 40 years ago on its punched-card-era mainframes (360– 67) and commercially available on PCs for at least the past 10 years. VM was invented for essentially the same reasons it is used today: run multiple operating systems on one machine in fully isolated ways. Some of these operating systems may be less reliable than others or may still be under test but are unable to interfere with one another. Even if we are talking about the same thing, the roundtable highlighted none of these basic concepts. VM was widely used within a few years of its earliest implementation. One roundtable participant (in Part I, Nov. 2008) said: “I support virtualization.” OK, so I support transistor radios.