The Semantic Web

OIL: An Infrastructure for the Semantic Web

Dieter Fensel and Frank van Harmelen, Vrije Universiteit, Amsterdam Ian Horrocks, University of Manchester, UK Deborah L. McGuinness, Stanford University Peter F. Patel-Schneider, Bell Laboratories

esearchers in artificial intelligence first developed to facilitate knowl- R edge sharing and reuse. Since the beginning of the 1990s, ontologies have become a popular research topic, and several AI research communities—including Ontologies play a knowledge engineering, natural language processing, and knowledge representation— major role in have investigated them. More recently, the notion of application programs have direct access to data an ontology is becoming widespread in fields such semantics. The resource description framework2 supporting information as intelligent information integration, cooperative has taken an important additional step by defining information systems, information retrieval, elec- a syntactical convention and a simple data model exchange across tronic commerce, and knowledge management. for representing machine-processable data seman- Ontologies are becoming popular largely because of tics. RDF is a standard for the Web metadata the various networks. A what they promise: a shared and common under- World Wide Web Consortium (www.w3c.org/rdf) standing that reaches across people and application develops, and it defines a data model based on prerequisite for such a systems. triples: object, property, and value. The RDF Currently, ontologies applied to the World Wide Schema3 takes a step further into a richer represen- role is the development Web are creating the Semantic Web.1 Originally, the tation formalism and introduces basic ontological Web grew mainly around HTML, which provides a modeling primitives into the Web. With RDFS, we of a joint standard for standard for structuring documents that browsers can can talk about classes, subclasses, subproperties, translate in a canonical way to render those docu- domain and range restrictions of properties, and so specifying and ments. On the one hand, HTML’s simplicity helped forth in a Web-based context. We took RDFS as a spur the Web’s fast growth; on the other, its simplic- starting point and enriched it into a full-fledged exchanging ontologies. ity seriously hampered more advanced Web appli- Web-based ontology language called OIL.4 We cations in many domains and for many tasks. This included these aspects: The authors present led to XML (see Figure 1), which lets developers define arbitrary domain- and task-specific extensions • A more intuitive choice of some of the modeling OIL, a proposal for (even HTML appears as an XML application— primitives and richer ways to define concepts and XHTML). attributes. such a standard. XML is basically a defined way to provide a seri- • The definition of a formal semantics for OIL. alized syntax for tree structures—it is an important • The development of customized editors and infer- first step toward building a Semantic Web, where ence engines to work with OIL.

38 1094-7167/01/$10.00 © 2001 IEEE IEEE INTELLIGENT SYSTEMS Ontologies: A revolution tion spread over different sources. for information access and • Maintaining: Sustaining weakly struc- DAML-O OIL integration tured text sources is difficult and time- Many definitions of ontologies have sur- consuming when such sources become RDFS faced in the last decade, but the one that in large. Keeping such collections consis- our opinion best characterizes an ontology’s tent, correct, and up to date requires a XHTML RDF essence is this: “An ontology is a formal, mechanized representation of semantics HTML XML explicit specification of a shared conceptual- and constraints that can help detect ization.”5 In this context, conceptualization anomalies. refers to an abstract model of some phenom- • Automatic document generation: Adaptive Figure 1. The layer language model for enon in the world that identifies that phe- Web sites that enable dynamic reconfigu- the Web. nomenon’s relevant concepts. Explicit means ration according to user profiles or other that the type of concepts used and the con- relevant aspects could prove very useful. tomer as an instant market overview. Their straints on their use are explicitly defined, and The generation of semistructured infor- functionality is provided through wrappers, formal means that the ontology should be mation presentations from semistructured which use keyword search to find product machine understandable. Different degrees data requires a machine-accessible repre- information together with assumptions on of formality are possible. Large ontologies sentation of the semantics of these infor- regularities in presentation format and text such as WordNet (www.cogsci.princeton. mation sources. extraction heuristics. This technology has edu/~wn) provide a thesaurus for over two severe limitations: 100,000 terms explained in natural language. Using ontologies, semantic annotations On the other end of the spectrum is will allow structural and semantic definitions • Effort: Writing a wrapper for each online (www.cyc.com), which provides formal of documents. These annotations could pro- store is time-consuming, and changes in axiomating theories for many aspects of com- vide completely new possibilities: intelligent store presentation or organization increase monsense knowledge. Shared reflects the search instead of keyword matching, query maintenance. notion that an ontology captures consensual answering instead of information retrieval, • Quality: The extracted product informa- knowledge—that is, it is not restricted to document exchange between departments tion is limited (it contains mostly price some individual but is accepted by a group. through ontology mappings, and definitions information), error-prone, and incomplete. The three main application areas of ontol- of views on documents. For example, a wrapper might extract the ogy technology are knowledge management, product price, but it usually misses indi- Web commerce, and electronic business. Web commerce rect costs such as shipping. E-commerce is an important and growing Knowledge management business area for two reasons. First, e-com- Most product information is provided in KM is concerned with acquiring, maintain- merce extends existing business models—it natural language; automatic text recognition ing, and accessing an organization’s knowl- reduces costs, extends existing distribution is still a research area with significant edge. Its purpose is to exploit an organization’s channels, and might even introduce new dis- unsolved problems. However, the situation intellectual assets for greater productivity, new tribution possibilities. Second, it enables com- will drastically change in the near future value, and increased competitiveness. Owing pletely new business models and gives them when standard representation formalisms for to globalization and the Internet’s impact, a much greater importance than they had data structure and semantics are available. many organizations are increasingly geo- before. What has up to now been a peripheral Software agents will then understand prod- graphically dispersed and organized around aspect of a business field can suddenly receive uct information. Meta online stores will grow virtual teams. With the large number of online its own important revenue flow. with little effort, which will enable complete documents, several document management Examples of business field extensions are market transparency in the various dimen- systems have entered the market. However, online stores; examples of new business sions of the diverse product properties. these systems have weaknesses: fields are shopping agents and online mar- Ontology mappings, which translate different ketplaces and auction houses that turn com- product descriptions, will replace the low- • Searching information: Existing keyword- parison shopping into a business with its level programming of wrappers, which is based searches retrieve irrelevant informa- own significant revenue flow. The advan- based on text extraction and format heuris- tion that uses a certain word in a different tages of online stores and their success sto- tics. An ontology will describe the various context; they might miss information when ries have led to a large number of shopping products and help navigate and search auto- different words about the desired content pages. The new task for customers is to find matically for the required information. are used. a shop that sells the product they’re seek- • Extracting information: Current human ing, getting it in the desired quality, quan- Electronic business browsing and reading requires extracting tity, and time, and paying as little as possi- E-commerce in the business-to-business relevant information from information ble for it. Achieving these goals through field (B2B) is not new—initiatives to sup- sources. Automatic agents lack the com- browsing requires significant time and only port it in business processes between dif- monsense knowledge required to extract covers a small share of the actual offers. ferent companies existed in the 1960s. To such information from textual representa- Shopbots visit several stores, extract prod- exchange business transactions electroni- tions, and they fail to integrate informa- uct information, and present it to the cus- cally, sender and receiver must agree on a

MARCH/APRIL 2001 computer.org/intelligent 39 The Semantic Web

an ontology should have a frame-like look and feel. Car Daimler 230 SE • It must have a well-defined formal seman- Daimler 230 SE 40.000 DM tics with established reasoning properties 23,000 $ to ensure completeness, correctness, and efficiency. Translation • server It must have a proper link with existing Web languages such as XML and RDF to Order Bestell- ensure interoperability. information information Many of the existing languages such as Product Product CycL,6 KIF,7 and Ontolingua8 fail. However, catalogue catalogue OIL9 matches these criteria and unifies the Business Business three important aspects that different com- munities provide: epistemologically rich Figure 2. The translation of structure, semantics, and language. modeling primitives as provided by the frame community, formal semantics and efficient standard (a protocol for transmitting con- need ontologies in two important ways: reasoning support as provided by description tent and a language for describing content). logics, and a standard proposal for syntacti- A number of standards arose for this pur- • Standard ontologies must cover the vari- cal exchange notations as provided by the pose—one of them is the UN initiative, ous business areas. In addition to official Web community. Electronic Data Interchange for Adminis- standards, vertical marketplaces (Internet tration, Commerce, and Transport (Edifact). portals) could generate de facto stan- Frame-based systems In general, the automation of business trans- dards—if they can attract significant The central modeling primitives of pred- actions has not lived up to the propagan- shares of a business field’s online trans- icate logic are predicates. Frame-based and dists’ expectations, partly because of the actions. Examples include Dublin Core, object-oriented approaches take a different serious shortcomings of approaches such as Common Business Library (CBL), Com- viewpoint. Their central modeling primi- Edifact: It is a procedural and cumbersome merce XML (cXML), ecl@ss, Open tives are classes (or frames) with certain standard, making the programming of busi- Applications Group Integration Specifi- properties called attributes. These attributes ness transactions expensive and error-prone, cation (OAGIS), Open Catalog Format do not have a global scope but apply only and it results in large maintenance efforts. (OCF), Open Financial Exchange (OFX), to the classes for which they are defined— Moreover, the exchange of business data Real Estate Transaction Markup Lan- we can associate the “same” attribute (the over extranets is not integrated with other guage (RETML), RosettaNet, UN/SPSC same attribute name) with different range document exchange processes—Edifact is (see www.diffuse.org), and UCEC. and value restrictions when defined for dif- an isolated standard. • Ontology-based translation services must ferent classes. A frame provides a context Using the Internet’s infrastructure for link different data structures in areas for modeling one aspect of a domain. business exchange will significantly im- where standard ontologies do not exist or Researchers have developed many other prove this situation. Standard browsers can where a particular client needs a transla- additional refinements of these modeling render business transactions and transpar- tion from his or her terminology into the constructs, which have led to this modeling ently integrate them into other document standard. This translation service must paradigm’s incredible success. exchange processes in intranet and Internet cover structural and semantic as well as Many frame-based systems and lan- environments. However, the fact that HTML language differences (see Figure 2). guages have emerged, and, renamed as does not provide a means for presenting rich object orientation, they have conquered the syntax and data semantics hampers this. Ontology-based trading will significantly software engineering community. OIL XML, which is designed to close this gap extend the degree to which data exchange is incorporates the essential modeling primi- in current Internet technology, drastically automated and will create completely new busi- tives of frame-based systems—it is based changes the situation. We can model B2B ness models in participating market segments. on the notion of a concept and the defini- communication and data exchange with the tion of its superclasses and attributes. Rela- same means available for other data Why OIL? tions can also be defined not as an attribute exchange processes, we can render transac- Effective, efficient work with ontologies of a class but as an independent entity hav- tion specifications on standard browsers, requires support from advanced tools. We ing a certain domain and range. Like and maintenance is cheap. However, need an advanced ontology language to classes, relations can fall into a hierarchy. although XML provides a standard serial- express and represent ontologies. This lan- ized syntax for defining data structure and guage must meet three requirements: Description logics semantics, it does not provide standard data DL describes knowledge in terms of con- structures and terminologies to describe • It must be highly intuitive to the human cepts and role restrictions that can auto- business processes and exchanged products. user. Given the success of the frame-based matically derive classification taxonomies. Therefore, XML-based e-commerce will and object-oriented modeling paradigm, Knowledge representation research’s main

40 computer.org/intelligent IEEE INTELLIGENT SYSTEMS thrust is to provide theories and systems for term definitions, Instance OIL includes a expressing structured knowledge and for full-fledged database capability. Heavy OIL (possible future extensions) accessing and reasoning with it in a princi- Heavy OIL will include additional repre- pled way. In spite of the discouraging theo- sentational (and reasoning) capabilities. A Instance OIL retical complexity of the results, there are more expressive rule language and metaclass (Standard OIL + RDFS) RDFS now efficient implementations for DL lan- facilities seem highly desirable. We will Standard OIL guages, which we explain later. OIL inher- define these extensions of OIL in coopera- Core OIL its from DL its formal semantics and the tion with the DAML (DARPA Agent Markup (Standard OIL ^ RDFS) efficient reasoning support. Language; www.daml.org) initiative for a rule language for the Web. Web standards: XML and RDF OIL’s layered architecture has three Reification Modeling primitives and their semantics advantages: are one aspect of an ontology language, but Figure 3. OIL’s layered language model. we still have to decide about its syntax. • An application is not forced to work with Given the Web’s current dominance and a language that offers significantly more importance, we must formulate a syntax of expressiveness and complexity than is an ontology exchange language with exist- needed. ing Web standards for information repre- • Applications that can only process a lower sentation. First, OIL has a well-defined syn- level of complexity can still catch some of tax in XML based on a document type an ontology’s aspects. … definition and an XML Schema definition. • An application that is aware of a higher Second, OIL is an extension of RDF and level of complexity can still understand RDFS. With regard to ontologies, RDFS ontologies expressed in a simpler ontol- provides two important contributions: a stan- ogy language. It encounters that herbivore is a subclass of dardized syntax for writing ontologies and animal and a subclass of a second class, which a standard set of modeling primitives such Defining an ontology language as an it cannot understand properly. This seems to as instance-of and subclass-of relationships. extension of RDFS means that every RDFS preserve complicated semantics for simpler ontology is a valid ontology in the new lan- applications. OIL’s layered architecture guage (an OIL processor will also understand A single ontology language is unlikely to RDFS). However, the other direction is also An illustration of the OIL fulfill all the needs of the Semantic Web’s possible: Defining an OIL extension as modeling primitive large range of users and applications. We closely as possible to RDFS allows maximal An OIL ontology is itself annotated with therefore organized OIL as a series of ever- reuse of existing RDFS-based applications metadata, starting with such things as title, increasing layers of sublanguages. Each and tools. However, because the ontology creator, creation date, and so on. OIL follows additional layer adds functionality and com- language usually contains new aspects (and the W3C Dublin Core Standard on biblio- plexity to the previous one. Agents (humans therefore a new vocabulary, which an RDFS graphical meta date for this purpose. or machines) that can only process a lower processor does not know), 100 percent com- Any ontology language’s core is its hier- layer can still partially understand ontolo- patibility is impossible. Let’s look at an archy of class declarations, stating, for exam- gies expressed in any of the higher layers. A example. The following OIL expression ple, that DeskJet printers are a subclass of first and very important application of this defines herbivore as a class, which is a sub- printers. We can declare classes as defined, principle is the relation between OIL and class of animal and disjunct to all carnivores: which indicates that the stated properties are RDFS. As Figure 3 shows, core OIL coin- not only necessary but also sufficient condi- cides largely with RDFS (with the excep- tions for class membership. Instead of using tion of RDFS’s reification features). This classes. ited capabilities. We can declare slots (relations between Standard OIL aims to capture the neces- classes) together with logical axioms, stating sary mainstream modeling primitives that whether they are functional (having at most provide adequate expressive power and are ing which (if any) slots are inverse. We can the semantics and making complete infer- state range restrictions as part of a slot decla- ence viable. ration as well as the number of distinct values Instance OIL includes a thorough individ- that a slot may have. We can further restrict ual integration. Although the previous slots by value-type or has-value restrictions. A layer—Standard OIL—includes modeling An application limited to pure RDFS can value-type restriction demands that every constructs that specify individual fillers in still capture some aspects of this definition: value of the property must be of the stated

MARCH/APRIL 2001 computer.org/intelligent 41 The Semantic Web

formal semantics, but it must exist and be ontoedit). Currently, OntoEdit supports class-def Product consulted whenever necessary to resolve dis- Frame-Logic, OIL, RDFS, and XML. It is slot-def Price putes about the meaning of language con- commercialized from Ontoprise (www. domain Product structions. It is an ultimate reference point ontoprise.de). slot-def ManufacturedBy for OIL applications. • OILed is a freely available and customized domain Product Figure 4 shows a very simple example of editor for OIL implemented by the Uni- class-def PrintingAndDigitalImagingProduct an OIL ontology provided by SemanticEdge versity of Manchester and sponsored by subclass-of Product (www.interprice.com). It illustrates OIL’s the Vrije Universiteit, Amsterdam, and class-def HPProduct most basic constructs. SemanticEdge (see http://img.cs.man.ac. subclass-of Product This defines a number of classes and uk/oil). OILed aims to provide a simple slot-constraint ManufacturedBy organizes them in a class hierarchy (for freeware editor that demonstrates—and has-value “Hewlett Packard” example, HPProduct is a subclass of Product). stimulates interest in—OIL. OILed is not class-def Printer Various properties (or slots) are defined, intended to be a full ontology development subclass-of PrintingAndDigitalImagingProduct together with the classes to which they environment—it will not actively support slot-def PrinterTechnology apply (such as a Price is a property of any the development of large-scale ontologies, domain Printer Product, but a PrintingResolution can only be the migration and integration of ontolo- slot-def Printing Speed stated for a Printer, an indirect subclass of gies, versioning, argumentation, and many domain Printer Product). For certain classes, these proper- other activities that are involved in ontol- slot-def PrintingResolution ties have restricted values (for example, the ogy construction. Rather, it is a NotePad domain Printer price of any HPLaserJet1100se is restricted for ontology editors that offers just enough class-def PrinterForPersonalUse to $479). In OIL, we can also combine functionality to let users build ontologies subclass-of Printer classes by using logical expressions—for and demonstrate how to check them for class-def HPPrinter example, an HPPrinter is both an HPProduct consistency. subclass-of HPProduct and Printer and a Printer (and consequently inherits the • Protégé11 lets domain experts build knowl- class-def LaserJetPrinter properties from both classes). edge-based systems by creating and mod- subclass-of Printer ifying reusable ontologies and problem- slot-constraint PrintingTechnology OIL tools solving methods (see www.smi.stanford. has-value “Laser Jet” OIL has strong tool support in three areas: edu/projects/protege). Protégé generates class-def HPLaserJetPrinter domain-specific knowledge acquisition subclass-of LaserJetPrinter and HPProduct • ontology editors, to build new ontologies; tools and applications from ontologies. class-def HPLaserJet1100Series • ontology-based annotation tools, to link More than 30 countries have used it. It is subclass-of HPLaserJetPrinter and PrinterFor unstructured and semistructured informa- an ontology editor that can define classes PersonalUse tion sources with ontologies; and and class hierarchy, slots and slot-value slot-constraint PrintingSpeed • reasoning with ontologies, which enables restrictions, and relationships between has-value “8 ppm” advanced query-answering services, sup- classes and properties of these relation- slot-constraint PrintingResolution ports ontology creation, and helps map ships. The instances tab is a knowledge has-value “600 dpi” between different ontologies. acquisition tool that can acquire instances class-def HPLaserJet1100se of the classes defined in the ontology. Pro- subclass-of HPLaserJet1100Series Ontology editors tégé, built at Stanford University, currently slot-constraint Price Ontology editors help human knowledge supports RDF—work on extending it to has-value “$479” engineers build ontologies—they support the OIL is starting. class-def HPLaserJet1100xi definition of concept hierarchies, the defini- subclass-of HPLaserJet1100Series tion attributes for concepts, and the defini- Ontology-based annotation tools slot-constraint Price tion of axioms and constraints. They must Ontologies can describe large instance has-value “$399” provide graphical interfaces and conform to populations. In OIL’s case, two tools cur- existing standards in Web-based software rently aid such a process. First, we can derive Figure 4. A small printer ontology in OIL. development. They enable the inspecting, an XML DTD and an XML Schema defini- browsing, codifying, and modifying of tion from an ontology in OIL. Second, we ontologies, and they support ontology devel- can derive an RDF and RDFS definition for type; has-value restrictions require the slot to opment and maintenance tasks. Currently, instances from OIL. Both provide means to have at least values from the stated type. two editors for OIL are available, and a third express large volumes of semistructured A crucial aspect of OIL is its formal is under development: information as instance information in OIL. semantics.10 An OIL ontology is given a for- More details appear elsewhere.4,12,13 mal semantics by mapping each class into a • OntoEdit (see Figure 5) is an ontology- set of objects and each slot into a set of pairs engineering environment developed at the Reasoning with ontologies: Instance of objects. This mapping must obey the con- Knowledge Management Group of the and schema inferences straints specified by the definitions of the University of Karlsruhe, Institute AIFB Inference engines for ontologies can rea- classes and slots. We omit the details of this (http://ontoserver.aifb.uni-karlsruhe.de/ son about an ontology’s instances and

42 computer.org/intelligent IEEE INTELLIGENT SYSTEMS schema definition. For example, they can automatically derive the right position of a new concept in a given concept hierarchy. Such reasoners help build ontologies and use them for advanced information access and navigation. OIL uses the FaCT (Fast Classification of Terminologies, www.cs. man.ac.uk/~horrocks/FaCT) system to pro- vide reasoning support for ontology design, integration, and verification. FaCT is a DL Figure 5. A screen shot of OntoEdit. classifier that can provide consistency checking in modal and other similar logics. FaCT’s most interesting features are its expressive logic, its optimized tableaux Swiss Life: Organizational memory heterogeneous information sources to en- implementation (which has now become the Swiss Life19 (www.swisslife.ch) imple- sure smooth transfer to others. standard for DL systems), and its Corba- ments an intranet-based front end to an based clientÐserver architecture. FaCT’s organizational memory with OIL. The start- EnerSearch: Virtual enterprise optimizations specifically aim to improve ing point is the existing intranet informa- EnerSearch is a virtual organization the system’s performance when classifying tion system, called ZIS, which has consid- researching new IT-based business strate- realistic ontologies. This results in perfor- erable drawbacks. Its great flexibility gies and customer services in deregulated mance improvements of several orders of allows for its evolution with actual needs, energy markets (www.enersearch.se).20 magnitude compared with older DL sys- but this also makes finding certain infor- EnerSearch is a knowledge creation com- tems. This performance improvement is mation difficult. Search engines help only pany—knowledge that must transfer to its often so great that it is impossible to mea- marginally. Clearly, formalized knowledge shareholders and other interested parties. sure precisely because nonoptimized sys- is connected with weakly structured back- Its Web site is one of the mechanisms for tems are virtually nonterminating with ground knowledge here—experience shows this, but finding information on certain top- ontologies that FaCT can easily deal with.14 that this is extremely bothersome and error- ics is difficult—the current search engine For example, for a large medical terminol- prone to maintain. The only way out is to supports free-text search rather than con- ogy ontology developed in the GALEN pro- apply content-based information access so tent-based search. So, EnerSearch applies ject,15 FaCT can check the consistency of that we no longer have a mere collection of the OIL toolkit to enhance knowledge all 2,740 classes and determine the com- Web pages but a full-fledged information transfer to researchers in the virtual orga- plete class hierarchy in approximately 60 system that we can rightly call an organi- nization in different disciplines and coun- seconds of CPU (450-MHz Pentium III) zational memory. tries and specialists from shareholding time. FaCT can be accessed through a Corba companies interested in getting up-to-date interface. British Telecom: Call centers R&D information. Call centers are an increasingly impor- Applications of OIL tant mechanism for customer contact in Earlier, we sketched three application many industries. What will be required in areas for ontologies: knowledge manage- the future is a new philosophy in customer ment, Web commerce, and e-business. Not interaction design. Every transaction should surprisingly, we find applications of OIL in emphasize the uniqueness of both the cus- all three areas. On-To-Knowledge (www. tomer and the customer service person— ontoknowledge.org)16 extends OIL to a full- this requires effective knowledge manage- fledged environment for knowledge man- ment (see www.bt.com/innovations), in- IL has several advantages: it is prop- agement in large intranets and Web sites. cluding knowledge about the customer and Oerly grounded in Web languages such Unstructured and semistructured data is auto- about the customer service person, so that as XML Schemas and RDFS, and it offers matically annotated, and agent-based user customers are directed to the correct person different levels of complexity. Its inner lay- interface techniques and visualization tools in a meaningful and timely way. Some of ers enable efficient reasoning support based help users navigate and query the informa- BT’s call centers are targeted to identify on FaCT, and it has a well-defined formal tion space. Here, On-To-Knowledge contin- opportunities for effective knowledge man- semantics that is a baseline requirement for ues a line of research that began with agement. More specifically, call center the Semantic Web’s languages. Regarding its SHOE17 and Ontobroker:18 using ontologies agents tend to use a variety of electronic modeling primitives, OIL is not just another to model and annotate the semantics of infor- sources for information when interacting new language but reflects certain consensus mation resources in a machine-processable with customers, including their own spe- in areas such as DL and frame-based sys- manner. On-To-Knowledge is carrying out cialized systems, customer databases, the tems. We could only achieve this by includ- three industrial case studies to evaluate the organization’s intranet, and, perhaps most ing a large group of scientists in OIL’s devel- tool environment for ontology-based knowl- important, case bases of best practices. OIL opment. OIL is also a significant source of edge management. provides an intuitive front-end tool to these inspiration for the ontology language

MARCH/APRIL 2001 computer.org/intelligent 43 The Semantic Web

DAML+OIL (www.cs.man.ac.uk/~horrocks/ 6. D.B. Lenat and R.V. Guha, Building Large mann, San Francisco, 1991, pp. 238Ð249; DAML-OIL), developed through the DAML Knowledge-Based Systems: Representation http://logic.stanford.edu/kif/kif.html (current and Inference in the Cyc Project, Addison- 9 Mar. 2001). initiative. The next step is to start on a W3C Wesley, Reading, Mass., 1990. working group on the Semantic Web, taking 8. A. Farquhar, R. Fikes, and J. Rice, “The DAML+OIL as a starting point. 7. M.R. Genesereth, “Knowledge Interchange Ontolingua Server: A Tool for Collaborative Defining a proper language is an impor- Format,” Proc. Second Int’l Conf. Principles Ontology Construction,” Int’l. J. HumanÐ of Knowledge Representation and Reasoning Computer Studies, vol. 46, 1997, pp. 707Ð728. tant step to expanding the Semantic Web. (KR 91), J. Allenet et al., eds., Morgan Kauf- Developing new tools, architectures, and applications is the real challenge that will follow.

The Authors

Dieter Fensel’s biography appears in the Guest Editors’ Introduction on page 25.

Ian Horrocks is a lecturer in computer science with the Information Man- agement Group at the University of Manchester, UK. His research interests include knowledge representation, automated reasoning, optimizing rea- Acknowledgments soning systems, and ontological engineering, with particular emphasis on We thank Hans Akkermans, Sean Bechhofer, the application of these techniques to the World Wide Web. He received a Jeen Broekstra, Stefan Decker,Ying Ding, Michael PhD in computer science from the University of Manchester. He is a mem- Erdmann, Carole Goble, Michel Klein, Alexander ber of the OIL language steering committee and the Joint EU/US Commit- Mädche, Enrico Motta, Borys Omelayenko, Stef- tee on Agent Markup Languages, and is coeditor of the DAML+OIL lan- fen Staab, Guus Schreiber, Lynn Stein, Heiner guage specification. Contact him at the Dept. of Computer Science, Univ. of Stuckenschmidt, and Rudi Studer, all of whom Manchester, Oxford Rd., Manchester, M13 9PL, UK; [email protected]; were involved in OIL’s development. www.cs.man.ac.uk/~horrocks.

Frank van Harmelen is a senior lecturer in the AI Department at Vrije Uni- versiteit in Amsterdam. His research interests include specification languages for knowledge-based systems, using languages for validation and verification of KBS, developing gradual notions of correctness for KBS, and verifying weakly structured data. He received a PhD in artificial intelligence from the University of Edinburgh. Contact him at Dept. of AI, Faculty of Sciences, Vrije Universiteit Amsterdam de Boelelaan 1081a, 1081HV Amsterdam, Netherlands; [email protected]; www.cs.vu.nl/~frankh. References

1. T. Berners-Lee, Weaving the Web, Orion Busi- ness Books, London, 1999. Deborah L. McGuinness is the associate director and senior research sci- entist for the Knowledge Systems Laboratory at Stanford University. Her 2. O. Lassila and R. Swick, Resource Descrip- main research areas include ontologies, description logics, reasoning sys- tion Framework (RDF) Model and Syntax tems, environments for building and maintaining information, knowledge- Specification, W3C Recommendation, World enhanced search, configuration, and intelligent commerce applications. She Wide Web Consortium, Boston, 1999, www. also runs a consulting business dealing with ontologies and artificial intelli- w3.org/TR/REC-rdf-syntax (current 6 Dec. gence for business applications and serves on several technology advisory 2000). boards and academic advisory boards. She received a PhD from Rutgers Uni- versity in reasoning systems. Contact her at the Knowledge Systems Labo- 3. D. Brickley and R.V. Guha, Resource ratory, Stanford Univ., Stanford, CA 94305; [email protected]; Description Framework (RDF) Schema www.ksl.stanford.edu/people/dm. Specification 1.0, W3C Candidate Recom- mendation, World Wide Web Consortium, Boston, 2000, www.w3.org/TR/rdf-schema (current 6 Dec. 2000). Peter F. Patel-Schneider is a member of the technical staff at Bell Labs Research. His research interests center on the properties and use of descrip- 4. J. Broekstra et al., “Enabling Knowledge Rep- tion logics. He is also interested in rule-based systems, including standard sys- resentation on the Web by Extending RDF tems derived from OPS as well as newer formalisms such as R++. He received Schema,” Proc. 10th Int’l World Wide Web his PhD from the University of Toronto. Contact him at Bell Labs Research, Conf., Hong Kong, 2001. 600 Mountain Ave., Murray Hill, NJ 07974; [email protected]; www.bell-labs.com/user/pfps. 5. T.R. Gruber, “A Translation Approach to Portable Ontology Specifications,” Knowl- edge Acquisition, vol. 5, 1993, pp. 199Ð220.

44 computer.org/intelligent IEEE INTELLIGENT SYSTEMS Special Issue, IEEE Intelligent Systems, Sept./Oct. 2001

9. D. Fensel et al., “OIL in a Nutshell,” Proc. European Knowledge Acquisition Confer- ence (EKAW 2000), R. Dieng et al., eds., IN BIOLOGY Lecture Notes in Artificial Intelligence, no. 1937, Springer-Verlag, Berlin, 2000, pp. 1Ð16. MOTIVATION 10. I. Horrocks et al., The Ontology Inference Biology is rapidly becoming a data-rich science owing to recent Layer OIL, tech. report, Vrije Universiteit massive data generation technologies, while our biological colleagues Amsterdam; www.ontoknowledge.org/oil/ are designing cleverer and more informative experiments owing to TR/oil.long.html (current 9 Mar. 2001). recent advances in molecular science. These data and these experi- ments hold the keys to the deepest secrets of biology and medicine, 11. W.E. Grosso et al., “Knowledge Modeling at the Millennium (The Design and Evolution but cannot be analyzed fully by humans because of the wealth and of Protege-2000),” Proc. 12th Workshop complexity of the information available. The result is a great need for Knowledge Acquisition, Modeling, and Man- intelligent systems in biology. agement, 1999. Intelligent systems probably helped design the last drug your doc- tor prescribed, and intelligent computational analysis of the human 12. M. Klein et al., “The Relation between genome will drive medicine for at least the next half-century. Even as Ontologies and Schema-Languages: Trans- you read these words, intelligent systems are working on gene expres- lating OIL Specifications to XML Schema,” Proc. Workshop on Applications of Ontolo- sion data to help understand genetic regulation, and thus ultimately gies and Problem-Solving Methods, 14th the regulated control of all life processes including cancer, regenera- European Conf. on Artificial Intelligence, tion, and aging. Modern intelligent analysis of biological sequences Berlin, 2000. results today in the most accurate picture of evolution ever achieved. Knowledge bases of metabolic pathways and other biological net- 13. M. Erdmann and R. Studer, “How to Struc- works presently make inferences in systems biology that, for example, ture and Access XML Documents with let a pharmaceutical program target a pathogen pathway that does Ontologies,” Data and Knowledge Eng., vol. not exist in humans, resulting in fewer side effects to patients. Intelli- 36, no. 3, 2001. gent literature-access systems exploit a knowledge flow exceeding half a million biomedical articles per year, while machine-learning sys- 14. I. Horrocks and P.F. Patel-Schneider, “Opti- mizing Subsumption,” J. tems exploit heterogeneous online databases whose exponential Logic and Computation, vol. 9, no. 3, June growth mimics Moore’s law. Knowledge-based empirical approaches 1999, pp. 267Ð293. are the most successful method known for general protein structure prediction, a problem that has been called the “Holy Grail of molecu- 15. A.L. Rector, W.A. Nowlan, and A. Glowin- lar biology” and “solving the second half of the genetic code.” ski, “Goals for Concept Representation in the This announcement seeks papers and referees for a special issue GALEN Project,” Proc. 17th Ann. Symp. on Intelligent Systems in Biology. Preferred papers will describe an Computer Applications in Medical Care implemented intelligent system that produces results of significance in (SCAMC 93), McGraw-Hill, New York, 1993, pp. 414Ð418. biology or medicine. Systems that extend or enhance the intelligence of GUEST EDITOR 16. D. Fensel et al., “On-To-Knowledge: Ontol- human biologists are especially wel- ogy-based Tools for Knowledge Manage- come. Referees are solicited from Richard H. Lathrop ment,” Proc. eBusiness and eWork 2000 experts in the field who do not intend Dept. of Information and Conf. (EMMSEC 2000), 2000. to submit a paper. Computer Science Univ. of California, Irvine 17. S. Luke, L. Spector, and D. Rager, “Ontol- Irvine, CA 92697-3425 ogy-Based Knowledge Discovery on the World Wide Web,” Proc. 13th Nat’l Conf. Phone: +1 949 824 4021 Artificial Intelligence (AAAI 96), Ameri- Fax: +1 949 824 4056 can Association for Artificial Intelligence, [email protected] Menlo Park, Calif., 1996. www.ics.uci.edu/~rickl SUBMISSION 18. D. Fensel et al., “Lessons Learned from GUIDELINES Applying AI to the Web,” J. Cooperative Information Systems, vol. 9, no. 4, Dec. 2000, IEEE Intelligent Systems is a scholarly pp. 361Ð382. peer-reviewed publication intended for a broad research and user community. An informal, direct, and 19. U. Reimer et al., eds., Proc. Second Int’l Conf. lively writing style should be adopted. The issue will con- Practical Aspects of Knowledge Management tain a tutorial and an overview of the field, but explicitly biolog- (PAKM 98), 1998. ical terms or concepts should be explained concisely. Manuscripts should be original and should have between 6 and 10 magazine pages (not more 20. F. Ygge and J.M. Akkermans, “Decentralized than 7,500 words) with up to 10 references. Send manuscripts in PDF format to Markets versus Central Control: A Compara- [email protected] by 25 May, 2001. Potential referees and general inquiries should con- tive Study,” J. Artificial Intelligence Research, tact [email protected] directly. vol. 11, JulyÐDec. 1999, pp. 301Ð333.

MARCH/APRIL 2001 computer.org/intelligent 45