TRENDS & CONTROVERSIES

The semantic Web and its languages Editor: Dieter Fensel Vrije Universiteit Amsterdam [email protected]

The Web has drastically changed the availability of electronic informa- epistemologically rich modeling primitives, provided by the frame com- tion, but its success and exponential growth have made it increasingly munity; formal semantics and efficient reasoning support, provided by difficult to find, access, present, and maintain such information for a wide description logics; and a standard proposal for syntactical exchange nota- variety of users. In reaction to this bottleneck, many new research initia- tions, provided by the Web community. (See the essay by Frank van tives and commercial enterprises have been set up to enrich available Harmelen and .) information with machine-processable semantics. Such support is essen- Another candidate for such a Web-based ontology modeling lan- tial for bringing the Web to its full potential in areas such as knowledge guage is DAML-ONT. The DARPA Agent Markup Language is a management and electronic commerce. This semantic Web will provide major, well-funded initiative, aimed at joining the many ongoing intelligent access to heterogeneous and distributed information, enabling semantic Web efforts and focused on bringing ontologies to the Web. software products (agents) to mediate between user needs and available The DAML language inherits many aspects from OIL, and the capabili- information sources. Early steps in the direction of a semantic Web were ties of the two languages are relatively similar. Both initiatives cooper- SHOE1 and later Ontobroker,2 but now many more projects exist. ate in a Joint EU/US ad hoc Agent Markup Language Committee to Originally, the Web grew mainly around HTML, which provided a achieve a joined language proposal. The next step will be DAML- standard for structuring documents so that browsers could translate Logic, a language with sufficient means for expressing axioms and them in a canonical way. On the one hand, it was HTML’s simplicity rules. (See the essay by James Hendler and Deborah L. McGuiness.) that enabled the Web’s fast growth, but on the other, its simplicity seri- Defining languages for the semantic Web is just the first step. Devel- ously hampered more advanced Web applications in many domains and oping new tools, architectures, and applications is the real challenge for many tasks. This was the reason for XML (see Figure 1), which lets that will follow. us define arbitrary domain- and task-specific extensions. Even HTML has been redefined as an XML application—XHTML. Consequently, References we define the semantic Web as an XML application. 1. S. Luke, L. Spector, and D. Rager, “Ontology-Based Knowledge The Resource Description Framework took the first step toward Discovery on the World-Wide Web,” Working Notes Workshop defining the Web in XML terms. RDF defines a syntactical convention Internet-Based Information Systems at the 13th Nat’l Conf. Artifi- and a simple data model for representing data’s machine-processable cial Intelligence (AAAI 96), 1996. semantics. It is a standard for Web metadata developed by the World 2. D. Fensel et al., “Ontobroker: The Very High Idea,” Proc. 11th Wide Web Consortium. (See the essay by Ora Lassila.) Int’l Flairs Conf. (FLAIRS 98), 1998, pp. 131Ð135. The RDF Schema candidate recommendation that defines basic onto- logical modeling primitives on top of RDF took the second step, fol- lowed by the Ontology Inference Layer, which uses the RDFS as a start- Dieter Fensel is an associate professor at the Division of Mathematics ing point and extends it to a full-fledged ontology modeling language. and Computer Science, Vrije Universiteit, Amsterdam, and he is a new department editor for Trends & Controversies. After studying mathe- OIL unifies three important aspects provided by different communities: matics, sociology, and computer science in Berlin, he joined the Institute AIFB at the University of Karlsruhe. His major subject was knowledge engineering and his PhD DAML-O OIL thesis was about a formal specification DAML-O DARPA Agent Markup Language- for knowledge-based systems. RDFS OIL Ontology Inference Layer Currently, his focus is on the use of ontolo- gies to mediate access to heterogeneous XHTML RDF RDF Resource Description Framework knowledge sources and to apply them in HTML XML RDFS Resource Description Framework Schema knowledge management and electronic commerce. Contact him at [email protected]. Figure 1. The layer language model for the Web. nl; www.cs.vu.nl/~dieter.

The Resource Description Framework Modeling alog information), RDF is suitable for Ora Lassila, Nokia Research Center describing any Web resource, and as such it All items that RDF expressions describe provides interoperability between applica- are called resources, and, broadly speaking, The Resource Description Framework is tions that exchange machine-understand- anything a Universal Resource Identifier a standard for Web metadata that the World able information on the Web. Its goal is to can name is also a resource.4 Consequently, Wide Web Consortium (W3C) devel- add formal semantics to the Web, thus RDF can describe not just things on the oped.1Ð3 Expanding the traditional notion paving the way for the so-called semantic Web (such as pages, parts of pages, or col- of document metadata (such as library cat- Web. lections of pages) but also things not on the

NOVEMBER/DECEMBER 2000 67 Web—as long as they can be XML.” Sure, you could do that, named using some URI scheme. but essentially you would end up (Hypothetically, we could name or reinventing the wheel by building describe any person with a “per- a similar layer on top of XML that son” URI scheme—for example, RDF already introduces. The XML “person:US:123-45-6789.”) The developer community has focused RDF description model uses on RDF’s use (and abuse) of XML, objectÐattributeÐvalue triples: we sometimes forgetting that RDF is a can view instances of the model as data model whose syntax is largely directed or labeled graphs (which irrelevant. (During development of resemble semantic networks), or the RDF standard, several syntaxes we can take a more object-centric were proposed, some of which view and think of RDF as a frame- were not based on XML.) based representation system. In RDF’s object-oriented extensi- RDF, these triples are known as bility lets developers take pieces of statements. existing RDF schemata and extend Descriptions can be limited to one them as they see necessary. This resource (for example, a library cata- might lead to some type of Dar- log card that names a document’s winian evolution of metadata author and publisher), in which case where the strong solutions will the values of resource attributes (or Illustration by Sally Lee survive and evolve further (as properties, as they are called in RDF) opposed to the all-or-nothing situ- are typically strings. Descriptions can ation that users of XML DTD also span multiple resources: values of proper- RDF model.5 The metaconstructs for the [document type definitions] often face). ties can be other resources—so we can thus type system are terms and concepts that describe arbitrary relationships between multi- URIs name, so RDF effectively represents Applications and future directions ple resources. URIs name properties, which and defines classes and properties. Class In addition to the “syntax wars,” contro- are also resources, and as such they can definitions can be derived from multiple versy surrounds RDF within the knowledge describe a property by asking, “What are a superclasses, and property definitions can representation community. RDF has been particular property’s permitted values, specify domain and range constraints. We criticized for its low expressive power (it does which types of resources can it describe, and can also think of RDFS as a set of ontologi- not have variables, negation, or quantifica- what is its relationship to other properties?” cal modeling primitives on top of RDF. tion—a far cry from first-order predicate Meaning in RDF comes from specific terms calculus, for example). However, RDF is and concepts that URIs define and then Syntax what it is by design. Sometimes you need to name. Because URIs can be unique, two RDF uses XML for the syntactic expres- take baby steps before you can run—Web- systems can define some concept (say, “per- sion of model instances, which is a source of related issues are important (such as simplic- son”) and each use a different URI to name much controversy and confusion. RDF is ity, which enables wide adoption). it to avoid clashes; however, two systems essentially a data model and does not strive Several interesting RDF applications have agreeing on a common concept will use the to replace XML. Instead, it builds a layer on already emerged. Mozilla (also known as same URI and effectively share semantics. top of it, making interoperable exchange of Netscape 6) uses RDF internally as a repre- The RDF model also defines some meta- semantic information possible (for example, sentation format. In 1999, Netscape also level constructs (such as container types for the object-oriented extensibility is intended introduced the RSS formalism (RDF Site describing collections of resources) and to enable a partial understanding of data). Summary, www.egroups.com/group/rss-dev), higher-order statements (statements about RDF lacks primitive data types (such as which has now grown into a broader effort to other statements). Higher-order statements integer, float, and so forth), so strings are build an extensible information description are modeled in RDF and allow the repre- essentially the only literals available; XML and syndication format. sentation of modalities such as beliefs. Fur- atomic datatypes will be used once W3C Dublin Core is—at least initially—a meta- thermore, an extensible, object-oriented completes work on the XML Schema.6 data element set for describing cataloging type system (known as RDF Schema) is We often hear claims such as, “You don’t information, such as that needed by digital introduced as a layer on top of the basic need RDF; you can do everything with libraries. The DC initiative early on embraced

Correction

On page 30 of our September/October issue, the fifth sentence under the heading “IEA responses to some social and philosophical issues” called out the wrong figure. The sentence should read, “Similarly, as shown in Figure 5, on creative narratives, IEA scores agreed with highly trained expert grader scores as well as the latter agreed with each other.” We apologize for this error, which occurred during final production of the issue.

68 IEEE INTELLIGENT SYSTEMS RDF as the framework on which to build tion Framework (RDF) Model and Syntax FAQs on OIL: The Ontology such metadata (www.dublincore.org). Specification, W3C Recommendation, World Wide Web Consortium, 1999; www. Inference Layer RDF is also used as the basic building w3.org/TR/REC-rdf-syntax (current 6 Dec. block for other W3C standards; for exam- 2000). Frank van Harmelen, Vrije Universiteit ple, the Composite Capability/Preference Amsterdam Profile—or CC/PP—uses RDF to express 2. O. Lassila, Introduction to RDF Metadata, Ian Horrocks, University of Manchester profiles of mobile device characteristics to W3C Note, World Wide Web Consortium, 1997; www.w3.org/TR/NOTE-rdf-simple- let servers better tailor content for these intro (current 6 Dec. 2000). The Web is aimed at human readers. often restricted Web clients (www.w3.org/ Machines are oblivious to its informational Mobile/CCPP). 3. O. Lassila, “Web Metadata: A Matter of content: Web browsers, Web servers, and As to what will happen in the future, Semantics,” IEEE Internet Computing, vol. even search engines can’t really distinguish 2, no. 4, July/Aug. 1998, pp. 30Ð37. RDF is playing an important role as a basis weather forecasts from scientific papers, and for the emerging DARPA Agent Markup 4. T. Berners-Lee, R. Fielding, and L. Masin- they can’t tell a personal home page from a Language (see the related essay by Hendler ter, Uniform Resource Identifiers (URI): major corporate Web site. This inability to and McGuiness or visit www.daml.org), a Generic Syntax, Internet Draft Standard process content seriously hampers the Web’s large research program that will build more RFC 2396, Aug. 1998; www.isi.edu/ functionality—computers are limited to in-notes/rfc2396.txt (current 6 Dec. 2000). expressive layers of logic on top of the transmitting and presenting information on basic RDF framework. This is a large-scale 5. D. Brickley and R.V.Guha, Resource Descrip- it, and they can’t really help us process that effort to do real knowledge representation tion Framework (RDF) Schema Specification information. The semantic Web aims to cre- on the Web, and it is seen as a strong push 1.0, W3C Candidate Recommendation, World ate a Web in which both humans and Wide Web Consortium, 2000; www.w3.org/ toward building the semantic Web. TR/rdf-schema (current 6 Dec. 2000). machines can understand the information lurking in cyberspace. This, of course, 6. P.V. Biron and A. Malhotra, XML Schema requires representing information in such a Part 2: Datatypes, W3C Candidate Recom- way that its meaning (or “semantics”) is References mendation, World Wide Web Consortium, 2000; www.w3.org/TR/xmlschema-2 (cur- machine-accessible. The Ontology Infer- 1. O. Lassila and R. Swick, Resource Descrip- rent 6 Dec. 2000). ence Layer is designed to be exactly such a

Ora Lassila is a research fellow at the Nokia Research Center and is a James Hendler is a professor at the University of founder of the center’s Agent Technology group. In his own research, he Maryland, where he has joint appointments in the focuses on knowledge representation for the Web and its applications to Department of Computer Science, the Institute for ubiquitous computing. Previously, he held research positions at MIT, Advanced Computer Studies, and the Institute for CMU, and Helsinki University of Technology. Contact him at ora.las- Systems Research; he is also an affiliate of the [email protected]. Electrical Engineering Department. He was the recipient of a 1995 Fulbright Foundation Fellow- ship, is a member of the US Air Force Science Advisory Board, and is a fellow of the AAAI. Frank van Harmelen is a senior lecturer with the He is on leave from the university, working at AI research group at the Vrije Universiteit, Amster- the Defense Advanced Research Projects Agency dam. His current interests include specification as chief scientist of the Information Systems Office. Contact him at languages for KBS, using these languages for vali- [email protected]. dation and verification of KBS, developing gradual notions of correctness for KBS, and verifying weakly structured data. He studied mathematics Deborah L. McGuinness is the associate director and computer science in Amsterdam and was and senior research scientist of the Knowledge Sys- awarded a PhD from the Department of AI in Edin- tems Laboratory at Stanford University. She has burgh for his research on metalevel reasoning. Con- built and deployed numerous ontology environments tact him at [email protected]. and ontology applications, including some that have been in continuous use for over a decade at AT&T and Lucent. She is coauthor of the current ontology Ian Horrocks is a lecturer in computer science at evolution environment from Stanford University and the University of Manchester, UK. His research of one of the more widely used interests include knowledge representation; auto- systems—CLASSIC from Bell Laboratories. She is mated reasoning (particularly decision procedures also coeditor of the recently released DARPA Agent for description and modal logics); ontological engi- Markup Language. She is on the advisory board for Ontology.Org and neering, and optimizing, testing, and evaluating Powermarket, is on the executive council for the AAAI, and is on the exec- reasoning systems. Contact him at horrocks@cs. utive steering board for the international organization for description log- man.ac.uk. ics, the Ontology Inference Layer effort. Contact her at [email protected].

NOVEMBER/DECEMBER 2000 69 class-def Product slot-def Price domain Product slot-def ManufacturedBy knowledge-representation research, and they domain Product describe knowledge in terms of concepts class-def PrintingAndDigitalImagingProduct (comparable to classes or frames) and roles subclass-of Product class-def HPProduct (comparable to slots in frame systems). DLs subclass-of Product have well-understood theoretical properties. slot-constraint ManufacturedBy In addition, the meaning of any expression in has-value “Hewlett Packard” a DL can be described in a mathematically class-def Printer precise way, which enables reasoning with subclass-of PrintingAndDigitalImagingProduct concept descriptions and the automatic slot-def PrinterTechnology derivation of classification taxonomies. domain Printer There are now efficient implementations of slot-def Printing Speed DL reasoners that can perform these tasks. domain Printer OIL inherits from DLs both their formal slot-def PrintingResolution domain Printer semantics and efficient reasoning support. class-def PrinterForPersonalUse Besides modeling primitives (which subclass-of Printer frame systems provide) and their semantics class-def HPPrinter (which description logics provide), we have subclass-of HPProduct and Printer to decide about the syntax of a markup lan- class-def LaserJetPrinter guage for the semantic Web. Any such syn- subclass-of Printer tax must be formulated using existing W3C slot-constraint PrintingTechnology standards for information representation. As has-value “Laser Jet” a possible candidate, OIL has a well- class-def HPLaserJetPrinter defined syntax in XML based on a docu- subclass-of LaserJetPrinter and HPProduct class-def HPLaserJet1100Series ment type definition (DTD) and an XML subclass-of HPLaserJetPrinter and schema definition. Second, OIL is defined PrinterForPersonalUse as an extension of the RDF and its schema slot-constraint PrintingSpeed definition language RDFS. RDFS provides has-value “8 ppm” two important contributions: a standard set slot-constraint PrintingResolution of modeling primitives such as instance-of has-value “600 dpi” and subclass-of relationships and a stan- class-def HPLaserJet1100se dardized syntax for writing such writing subclass-of HPLaserJet1100Series class hierarchies. OIL extends this approach slot-constraint Price to a full-blown modeling language. has-value “$479” class-def HPLaserJet1100xi subclass-of HPLaserJet1100Series What does OIL look like? slot-constraint Price Figure 2 gives a very simple example of has-value “$399” an OIL ontology (it illustrates only the most basic constructs). Figure 2 also defines a number of classes and organizes them in a class hierarchy (for Figure 2. A very simple example of an Ontology Interface Layer ontology. example, HPProduct is a subclass of Product). Various properties (slots) are representation of machine-accessible one or more other classes—using slot- defined, together with the classes to which semantics of information on the Web. value pairs to specify additional constraints they apply (for example, a Price is a prop- on instances of the new class. Many frame- erty of any Product, but a PrintingRes- How is OIL trying to achieve this? based systems and languages with addi- olution can only be stated for a Printer OIL synthesizes work from three differ- tional refinements of these modeling primi- (an indirect subclass of Product). For cer- ent communities to achieve the ambitious tives have been developed, and, renamed as tain classes, these properties have restricted aim of providing a general-purpose markup object orientation, they have been very values (for example, the Price of any language for the semantic Web. It uses successful in the software engineering HPLaserJet1100se is restricted to $479). frame-based systems, description logics, community. OIL, based on a concept and In OIL, classes can also be combined using and Web standards (XML and RDF). the definition of their superclasses and logical expressions—for example, an Frame-based languages have a long his- slots, incorporates the essential modeling HPPrinter is both an HPProduct and a tory in AI. Their central modeling primi- primitives of frame-based systems. It also Printer (and consequently inherits the tives are classes (known as frames) with treats slots as first-class objects that can properties from both these classes). properties (known as slots). A frame pro- have their own properties (such as domain vides a context for modeling a class— and range) and be arranged in a hierarchy. What does the acronym OIL mean? which is generally defined as a subclass of Description logics have been developed in There are at least two possible meanings

70 IEEE INTELLIGENT SYSTEMS of the acronym—Ontology Inference Layer or Ontology Interchange Language—but all 399 > contain the word ontology. An ontology is a consensual, shared, and formal description of the important concepts in a given domain. Typically, an ontology identifies classes of Figure 3. The last class of the example in Figure 2 in Resourse Description Framework syntax. objects that are important in a domain and organizes these classes in a subclass hierar- chy. Each class is characterized by proper- ties that all elements in that class share. mats, with links to explicit and shared uni-karlsruhe.de/ontoedit), developed at Important relations between classes or ele- ontologies that agents can use to con- the Knowledge Management Group of ments of these classes are also part of an struct mappings between these catalogs. the AIFB Institute at the University of ontology. Ontologies are now an important ¥ Knowledge management. An increasing Karlsruhe; notion in diverse areas such as knowledge number of companies rely on intranet ¥ OILedit (http://img.cs.man.ac.uk/oil), a representation, natural language processing, technology as a knowledge repository freely available and customized editor information retrieval, databases, knowledge for their employees. Traditional docu- for OIL, developed by the University of management, and multiagent systems. They ment-management systems provide Manchester; and are widely considered to be a crucial ingre- insufficient means to structure and ¥ Protégé (www.smi.stanford.edu/ dient for the semantic Web’s infrastructure. access the knowledge in such a reposi- projects/protege), an ontology editor tory. Explicit ontologies are the most built at the University of Stanford. Cur- Which applications will OIL enable? promising technical vehicle for trans- rently it supports only RDF, but work is Machine-processable representations of forming document repositories into starting on extending Protégé to OIL. ontologies will be crucial to many applica- proper knowledge repositories. tions of the semantic Web. We briefly men- Inference engines can be used to reason tion a few: What are the design principles about ontologies, helping build and use behind OIL? them for advanced information access and ¥ Search engines. Current search engines The design of OIL was motivated by a navigation. OIL uses the FaCT system (fast are seriously limited by their reliance desire to classification of terminologies, www.cs. on keyword matching. They can’t find man.ac.uk/fact) to provide reasoning sup- relevant information that is described in ¥ maximize compatibility with existing port for ontology design, integration, and different terms, they often return infor- W3C standards—XML and RDF; verification. FaCT is heavily optimized to mation that uses the same words with a ¥ maximize partial interpretability by less deal with very large ontologies. It can check different meaning, and they can’t com- semantically aware processors; the consistency of thousands of classes and bine information from multiple sources. ¥ provide modeling primitives that have automatically derive their underlying class We can alleviate these problems with proven useful for large user communi- hierarchy in a matter of seconds, running on search engines that search the semantic ties; standard desktop hardware. concepts underlying the information in ¥ maximize expressiveness to model a Web pages rather than searching for wide variety of ontologies; How does OIL relate to RDF and matching keywords. ¥ provide a formal semantics (a mathe- RDFS? ¥ E-commerce. Currently, consumers can matically precise description of the The example in Figure 2 was stated in only compare online shops by visiting meaning of every expression) to facili- OIL’s presentation syntax, which is intended each shop and doing the comparison. tate machine interpretation of that for human readers and writers of OIL ontolo- So-called “shopbots” that try to perform semantics; and gies. For machines, OIL uses RDF as its this task do this by screen scraping: ¥ enable sound, complete, and efficient syntax. OIL exploits as much as possible the retrieving the information by interpret- reasoning services by limiting the modeling primitives of RDFS. This provides ing regularities in the layout of the Web expressiveness of the language. crucial backward compatibility, allowing pages of the various shops. They typi- OIL ontologies to be treated as extensions of cally only retrieve limited information What tools are available? RDF and RDFS ontologies and making OIL from the various shops (such as the Ontology editors help human knowledge ontologies available not only to OIL aware price) and ignore information such as engineers develop and maintain ontologies. applications, but also to applications that are shipping conditions (which are harder to They support the definition and modification only RDF-aware. Such RDF-aware applica- retrieve). In addition, they are cumber- of concepts, slots, axioms, and constraints tions can still process and reason with signif- some to construct and hard to maintain and enable the inspection, browsing, and icant portions of OIL-ontologies. For illus- (they must be updated every time a Web codifying of the resulting ontologies. Cur- tration purposes, the last class of the example shop changes the layout of its pages). rently, two editors for OIL are available and in Figure 2 in RDF syntax would look like Comparison shopping will become a a third is under development: the code presented in Figure 3. reality only when Web shops offer their To a program that is only RDF-aware catalogs in machine-processable for- ¥ OntoEdit (http://ontoserver.aifb. (not OIL-aware), this would still be inter-

NOVEMBER/DECEMBER 2000 71 It is unlikely that a single pretable as saying that the 1100xi printers ontology language can recommendation effort(www.w3.org/TR/ are a special type of the 1100 Series print- fulfill all the needs of the REC-rdf-syntax). ers. The specific restriction that the 1100xi semantic Web’s large range costs $399 would only be available to OIL- of users and applications. Motivation aware programs. The modern information technology world is a dynamically changing environ- How is OIL different from DAML? ment with an exponentially increasing abil- The DAML language inherits many ity to create and publish data that rapidly aspects from OIL, and the capabilities of the exception of RDFS’s reification features). swamps human abilities to process that two languages are relatively similar. Both This means that even simple RDFS agents data into usable information. Agent-based can process the OIL ontologies and pick up computing can potentially help us recog- ¥ support hierarchies of classes and prop- as much of their meaning as possible with nize complex patterns in this widely dis- erties based on subclass and subprop- their limited capabilities. tributed, heterogeneous, uncertain informa- erty relations; tion environment. Unfortunately, this ¥ allow classes to be built from other Where can I find out more about potential is hampered by the difficulty classes using arbitrary combinations of OIL? agents face in understanding and interact- intersection (AND), union (OR), and The European Union IST program for ing with data that is either unprocessed or complement (NOT); Information Society Technologies funds in natural languages. The inability of agents ¥ allow the domain, range, and cardinality the OIL initiative under the On-To-Knowl- to understand the concepts on a Web page, of properties to be restricted; edge project (IST-1999-1013) and Ibrow their difficulty in handling the semantics ¥ support transitive and inverse proper- (IST-1999-19005). OIL’s homepage inherent in a program’s outputs, and the ties; and (www.ontoknowledge.org/oil) provides complexity of fusing information concepts ¥ support concrete data types (integers, definitions of the OIL syntax, papers and from the outputs of sensors—to name but a strings, and so forth). presentations explaining OIL (ranging few problems—truly keep the “agent revo- from the very introductory to the very for- lution” from occurring. However, there are also some important mal), case studies using OIL, and tools that One potential solution is for humans to differences, which we only briefly discuss have been developed for OIL. meet the computer half way. By using tools here. First, OIL achieves a greater back- to provide marked annotations attached to ward compatibility with RDFS than data sources, we can make information DAML. Second, OIL was designed to available to the agents in new and exciting enable reasoning services that are sound ways. Beyond XML, the program’s goal and complete as well as efficient. Some The DARPA Agent Markup is to develop a language aimed at repre- constructions in DAML make similar rea- Language senting semantic relations in machine- soning services impossible. Third, OIL can James Hendler, DARPA/ISO readable ways compatible with current and state either sufficient conditions for a class Deborah L. McGuinness, Stanford University future Internet technologies. Prototype or conditions that are both sufficient and tools are being developed to show the necessary. This last option makes it possi- The DARPA Agent Markup Language potential of such markups to provide revo- ble to perform automatic classification. (DAML) program is a US Government- lutionary capabilities that will change the Given a specific object in a domain, OIL sponsored endeavor aimed at providing the way humans interact with information. can automatically decide to which classes foundation for the next Web evolution—the Deploying such tools to military and intel- this object belongs. In DAML, this distinc- semantic Web. The program is funding criti- ligence users and showing the incredible tion is not as well developed. cal research to develop languages, tools, and dual-use potential of such a technology techniques that will make considerably more caps off the program’s goals. Will OIL be a one-size-fits-all? Web content machine-understandable. This It is unlikely that a single ontology lan- should lead to the next major generation of Description guage can fulfill all the needs of the seman- Web technology and enable considerably To realize this solution, Internet markup tic Web’s large range of users and applica- more machine-to-machine (agent-based) languages must move beyond the implicit tions. We have therefore organized OIL as a communication. Academic researchers, gov- semantic agreements inherent in XML and series of ever-increasing layers of sublan- ernment agencies, software development community-specific controlled languages guages. Each additional layer will add func- companies, and industrial organizations such and move toward making semantic entities tionality and complexity to the previous as the World Wide Web Consortium are par- and markup a primary goal. To this end, layer, so that agents (humans or machines) ticipating in the program. The DAML pro- DARPA is working with numerous partners who can only process a lower layer can still ject is also working closely with other and communities to create an eventual partially understand ontologies expressed in efforts, including European Union-funded Web-standard semantic language and any of the higher layers. A first and very semantic Web projects such as On-To- demonstrate the utility of such a language. important application of this principle is the Knowledge (www.ontoknowledge.org) and We are doing this by developing an exam- relation between OIL and RDFS. Core OIL Ibrow (www.swi.psy.uva.nl/projects/ibrow/ ple language—DAML—and sample tools coincides largely with RDFS (with the home.html) and the ongoing W3C RDF and applications. DAML will tie a page’s

72 IEEE INTELLIGENT SYSTEMS Next in Trends & Controversies information to machine-readable semantics (specifications of term meanings stored in ontologies) and eventually provide a logi- Ontologies and cal language embedded on the Web. It will Electronic Commerce let communities extend simple ontologies for their own use, so that they can use a bottom-up design for meaning while shar- ing higher-level concepts. DAML-ONT specification and supporting nect to deductive engines that look for logi- In addition, DAML will provide mecha- documents at www.daml.org. Also avail- cal implications of the implicit information. nisms for explicitly representing services, able are early versions of DAML support Other times, users will connect to more or processes, and business models, so that tools and pointers to various Web resources less complete inference engines that infer humans and programs running on the Web marked up with DAML. the logical completion of all the statements. can recognize and understand nonexplicit DAML-ONT aims to capture the com- This provides opportunities for a spectrum information (such as that encapsulated in monly used modeling primitives used in of users and does not require users to have programs or sensors). Eventually, it will object-oriented modeling, frame systems, extensive computational resources con- also supply a mechanism by which logical and conceptual schemas and include them nected to their systems. Ultimately, the statements and proofs can become first- in an integrated language for the Web. It language should express the meaning of class Web entities, allowing a new set of attempts to join the ease of modeling in information on Web pages or in applica- capabilities in machine-to-machine com- frame systems, the ubiquity of the Web, tions and lead to a Web-standard language munication. This will enable the develop- and the formal foundations of knowledge for expressing semantic content. ment of a wide new range of software tools representation in description logics to pro- for industrial and government applications, vide a sound language for representing and with diverse uses ranging from business-to- reasoning with term meanings. Currently, it Acknowledgments business e-commerce to government efforts is being used to mark up project pages used in combating the spread of weapons of in the DAML project and to facilitate ser- The US Defense Advanced Research Projects Agency funds the DAML project. The program mass destruction. vices and applications work. In addition, is the result of many people’s work, and the DAML will provide a number of advan- several other communities (including the evolving list of major contributors is maintained tages over current markup approaches. It OIL project described in the previous on www.daml.org. will enable semantic interoperability— essay) are developing tools to translate instead of enabling only syntactic interoper- markup to the DAML-ONT language and ability as is done in XML. We will also be to provide pages in the DAML repository References able to mark objects on the Web (manually (www.daml.org/ontologies). or automatically) to include descriptions of The project’s next goal is to create an 1. O. Lassila and R. Swick, Resource Descrip- tion Framework (RDF) Model and Syntax information they encode, functions they early version of a logic language—DAML- Specification, W3C Recommendation, World provide, and data they can produce. Agents Logic. We expect DAML-Logic to include Wide Web Consortium, 1999; www.w3.org/ that use DAML will be able to link together both a language for expressing constraints in TR/REC-rdf-syntax (current 6 Dec. 2000). Web pages, databases, programs, models, DAML-ONT and for adding inference rules 2. O. Lassila, “Web Metadata: A Matter of and sensors to recognize the concepts they to the language. The first version should Semantics,” IEEE Internet Computing, vol. need. Thus, Web-based information fusion come out in the Spring of 2001. In addition, 2, no. 4, July/Aug. 1998, pp. 30Ð37. from diverse sources will become a reality. work continues on the evolution of DAML- ONT. We’ve published documents showing 3. D. Brickley and R.V. Guha, Resource Status report its intended meaning (http://www.ksl. Description Framework (RDF) Schema Specification 1.0, W3C Candidate Recom- DARPA kicked off its DAML program in stanford.edu/people/dlm/daml-semantics mendation, World Wide Web Consortium, the summer of 2000, and research is ongoing. and http://www.cs.man.ac.uk/~horrocks/ 2000; www.w3.org/TR/rdf-schema (current It’s delivering the language in two portions— DAML-OIL/semantics.html)4 and how it 6 Dec. 2000). an ontology language (DAML-ONT) and maps to current research in semantic Web 5 4. R. Fikes and D.L. McGuinness, “An later DAML-Logic. It released DAML-ONT languages such as OIL, SHOE (www.cs. Axiomatic Semantics for DAML-ONT,” 10 6 version 0.5 on 5 October 2000, and it main- umd.edu/projects/plus/SHOE), and KIF Dec. 2000, www.ksl.stanford.edu/people/ tains an ongoing discussion of the language (http://logic.stanford.edu/kif). We’re also dlm/daml-semantics (current 6 Dec. 2000). and logic issues on the mailing list www- exploring how it maps to emerging Web [email protected]—archived at http://lists. languages (such as XML and RDF) and to 5. S. Bechhofer et al., “An Informal Descrip- tion of OIL-Core and Standard OIL: A Lay- w3.org/Archives/Public/www-rdf-logic, with ongoing efforts in agent-based systems, ered Proposal for DAML-O,” www.onto- the goal of reaching a more stable version. particularly the FIPA standardization effort knowledge.org/oil/downl/dialects.pdf The language extends W3C’s Resource (www.fipa.org). (current 6 Dec. 2000). Description Framework1,2 and its associ- We’re also trying to encourage use from a 3 6. J. Heflin and J. Hendler, “Semantic Interop- ated object-oriented type system. It aims broad spectrum of users. Sometimes, users erability on the Web,” Proc. Extreme Markup to add expressive power suited to agent and will use DAML as a unifying language for Languages 2000, Graphic Communications service interoperation. You can find the stating explicit information and won’t con- Assoc., Montreal, 2000, pp. 111Ð120.

NOVEMBER/DECEMBER 2000 73