plications the ontology could play several roles: provide a formally defined and extensible vocabulary for semantic annotations; describe the structure of existing sources and the information they store; and provide a detailed model of the domain against which queries are formulated. Such queries can be answered by using semantic annotations and structural knowledge to retrieve and combine information from multiple sources. ²² It should be noted that the use of ontologies in information integration is far from new and the subject of extensive research within the database community. ²

With large ontologies, answering conceptual and data-retrieval queries may be a very complex task, and DL

 

reasoners allow OWL ontology applications to answer complex queries and provide guarantees about the correctness of the result. Reliability and correctness are particularly important when ontology-based systems are used in safety-critical applications; in those involving medicine, for example, incorrect reasoning could adversely affect patient care.

However, RDF and OWL are also used in a range of applications where reasoning plays only a relatively minor role in, for example, the Friend of a Friend, or FOAF, project ( www.foaf- project.org) and the Dublin Core Metadata Initiative, ( dublincore.org) and when carrying annotations in Adobe’s Extensible Metadata Platform (www. adobe.com/products/xmp/). In them, RDF is typically used to provide a flexible and extensible data structure for annotations, with the added advantage that IRIs can be used to refer directly to Web resources.

In FOAF, for example, a simple RDF/ OWL ontology provides a vocabulary of terms for describing and linking people and their interests and activities; terms include the foaf:Person class and properties, including foaf:name, foaf:homepage, and foaf:knows. OWL is used to declare that some properties (such as foaf:homepage) are InverseFunctional; that is, they can be used as a key to identify the subject of the property, often a person. However, the semantics of the vocabulary is mainly captured informally in textual descriptions of each term and procedurally interpreted by applications. This informality reduces the need for reasoning systems but limits the ability of applications to share and understand vocabulary extensions.

future Directions

The success of OWL also involves many challenges for the future development of both the OWL language and OWL tool support. Central to them is the familiar tension between requirements for advanced features, particularly increased expressive power, and raw performance, particularly the ability to deal with large ontologies and data sets.

Researchers have addressed them by investigating more expressive DLs, developing new and more highly optimized DL reasoning systems and identifying smaller logics that combine still-useful expressive power with better worst-case complexity or other desirable computational properties. Results from these efforts are being exploited by the W3C in order to refine and extend OWL, forming in October 2007 a new W3C Working Group for this purpose ( www.w3.org/2007/OWL/). The resulting language is called OWL 2 (initially called OWL 1. 1) based on a more expressive DL called SROIQ. ¹⁰ OWL 2 extends OWL with the ability to “qualify” cardinality restrictions to, say, describe the hand as having four parts that are fingers and one part that is a thumb; assert that properties are reflexive, irreflexive, asymmetric, and disjoint (such as to describe hasParent as an irreflexive property); and compose properties into property chains (such as to capture the fact that a disease affecting a part of an organ affects the organ as a whole). OWL 2 also provides extended support for datatypes

and for annotations.

Besides increasing the expressive power of the complete language, OWL 2 also defines three so-called profiles, in effect language fragments with desirable computational properties (www. w3.org/TR/ owl2-profileswww.w3.org/ TR/opw12-profiles/). One is based on DL Lite, a logic for which standard reasoning problems can be reduced to standard query language (SQL) query answering; another is based on EL++, a logic for which standard reasoning problems can be performed in polynomial time; and the third is based on DLP, a logic for which query answering can be implemented using rule-based techniques that have been shown to scale well in practice.

In some cases, even the increased expressive power of OWL 2 may not meet application requirements. One way to further increase the expressive power of the language would be to extend it with Horn-like rules; that is, implications like parent (x, y) ∧ bother (y, z) ⇒ uncle (x, z) stating that if y is a parent of x and z is a brother of y (the antecedent), then z is an uncle of x (the consequent). A notable proposal along these lines is the Semantic Web Rules Language ( www.w3.org/Submission/ SWRL/). If the semantics of Horn-like rules is restricted so it applies only to named individuals, then its addition does not disturb the decidability of the underlying DL; this restricted form of rules is known as “DL-safe” rules. ¹⁷ A W3C working group was established in 2005 to produce a W3C language standard that will “allow rules to be translated between rule languages and thus transferred between rule systems” ( www.w3.org/2005/rules/).

As I discussed earlier, reasoning-enabled tools provide vital support for ontology engineering. Recent work has shown how this support can be extended to modular design and module extraction, important techniques for working with large ontologies. When a team of ontology engineers is developing a large ontology, they should divide it into modules in order to make it easier to understand and facilitate parallel work. Similarly, it may be desirable to extract from a large ontology a module containing all the information relevant to some subset of the domain; the resulting small(er) ontology is easier for