6.Semantic Web

8/6/2019 6.Semantic Web

1/6

Semantic Web: The Story So Far

Ian HorrocksUniversity of Manchester

Manchester, UK

[email protected]

ABSTRACT

The goal of Semantic Web research is to transform the Webfrom a linked document repository into a distributed knowl-edge base and application platform, thus allowing the vastrange of available information and services to be more effec-tively exploited. As a first step in this transformation, lan-guages such as OWL have been developed; these languagesare designed to capture the knowledge that will enable ap-

plications to better understand Web accessible resources,and to use them more intelligently. Although fully realis-ing the Semantic Web still seems some way off, OWL hasalready been very successful, and has rapidly become a defacto standard for ontology development in fields as diverseas geography, geology, astronomy, agriculture, defence andthe life sciences. An important factor in this success hasbeen the availability of sophisticated tools with built in rea-soning support. The use of OWL in large scale applicationshas brought with it new challenges, both with respect to ex-pressive power and scalability, but recent research has alsoshown how the OWL language and OWL tools can be ex-tended and adapted to meet these challenges.

1. INTRODUCTIONWhile phenomenally successful in terms of size and num-

ber of users, todays World Wide Web is fundamentally arelatively simple artefact. Web content consists mainly ofdistributed hypertext, and is accessed via a combination ofkeyword based search and link navigation. This simplicityhas been one of the great strengths of the Web, and has beenan important factor in its popularity and growth: naive usersare able to use it, and can even create their own content.

The explosion in both the range and quantity of Web con-tent has, however, highlighted some serious shortcomings inthe hypertext paradigm. In the first place, the requiredcontent becomes increasingly difficult to locate using thesearch and browse paradigm. Finding information aboutpeople with very common names (or with famous name-

sakes) can, for example, be a frustrating experience. More

Permission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.W4A2007 Keynote, May 0708, 2007, Banff, Canada. Co-Located with the16th International World Wide Web Conference.Copyright 2007 ACM 1-59593-590-8/06/0010 ...$5.00.

complex queries can be even more problematical: a query foranimals that use sonar but are neither bats nor dolphins1

may either return many irrelevant results related to bats anddolphins (because the search engine failed to understand thenegation), or may fail to return many relevant results (be-cause most relevant Web pages also mention bats and/ordolphins). More complex tasks may be extremely difficult,or even impossible. Examples of such tasks include locatinginformation in data repositories that are not directly acces-

sible to search engines [34], or finding and using so-calledweb services [24].

If human users have difficulty accessing web content, theproblem is even more severe for automated processes. Thisis because web content is primarily intended for presenta-tion to and consumption by human users: HTML markupis mainly concerned with layout, size, colour and other pre-sentational issues. Moreover, web pages increasingly useimages, often including active links, to present information.Human users are able to interpret the significance of suchfeatures, and thus understand the information being pre-sented, but this may not be so easy for an automated pro-cess or software agent. It is easy to imagine that similardifficulties might be experienced by users with cognitive or

sensory impairments.

The Semantic Web aims to overcome some of the abovementioned problems by making web content more accessibleto automated processes; the ultimate goal is to transformthe existing web into . . . a set of connected applications. . . forming a consistent logical web of data . . . [3]. Thisis to be achieved by adding semantic annotations to Webcontent, i.e., annotations that describe the meaning of thecontent.

In the following we will examine in a little more detailwhat semantic annotations will look like, how they describemeaning, and how automated processes can exploit such de-scriptions.

2. BACKGROUNDAs we mentioned above, the key idea behind the semantic

web is to explicate the meaning of web content by addingsemantic annotations. If we assume for the sake of simplicitythat such annotations take the form of XML style tags, wecould imagine a fragment of a web page being annotated asfollows:

1Thanks to Uli Sattler for this example.

120


2/6

Figure 1: Tree of Porphyry.

WizardHarry Potter/Wizard has a pet calledSnowyOwlHedwig/SnowyOwl.

Taken in isolation, however, such annotations are of onlylimited value: the problem of understanding the terms used

in the text has simply been transformed into the problem ofunderstanding the terms used in the labels. A query for in-formation about raptors, for example, may not retrieve thistext, even though owls are raptors. This is where ontologiescome into play: they provide a mechanism for introducinga vocabulary and giving precise meanings to the terms inthe vocabulary. A suitable ontology might, for example, in-troduce the term SnowyOwl, and include the informationthat SnowyOwls are kinds Owl, and that Owls are kinds ofRaptor. Moreover, if this information is represented in away that is accessible to our query engine, then it would beable to recognise that the above text is relevant to our queryabout raptors.

Ontology, in its original pholosophical sense, is a funda-mental branch of metaphysics focussing on the study of ex-istence; its objective is to determine what entities and typesof entities actually exist, and thus to study the structure ofthe world. The study of ontology can be traced back to thework of Plato and Aristotle, and from the very beginningincluded the development of hierarchical categorisations ofdifferent kinds of entity and the features that distinguishthem: the well known tree of Porphyry, for example, iden-tifies animals and plants as sub-categories of living thingsdistinguished by animals being sensitive, and plants beinginsensitive (see Figure 1).

In computer science, an ontology is usually taken to bea model of (some aspect of) the world; it introduces vo-cabulary describing various aspects of the domain beingmodelled, and provides an explicit specification of the in-

tended meaning of the vocabulary. This specification oftenincludes classification based information not unlike that inPorphyrys famous tree. For example, Figure 2 shows ascreenshot of a Pizza ontology as displayed by the Protegeontology design tool [22]. The ontology introduces variouspizza related vocabulary (some of which can be seen in theleft hand panel), such as NamedPizza and RealItalian-Pizza, and arranges it hierarchically: RealItalianPizza is,for example, a sub-category of NamedPizza. The other pan-els display information about the currently selected vocabu-lary term, RealItalianPizza in this case, describing its mean-

ing: a RealItalianPizza is a Pizza whose country of origin isItaly; moreover, a RealItalianPizza always has a ThinAnd-CrispyBase. Ontologies can be used to annotate and to or-ganise data from the domain: if our data includes instancesof RealItalianPizza, then we can return them in response toa query for instances of NamedPizza.

3. THE WEB ONTOLOGY LANGUAGE OWL

The architecture of the Web depends on agreed standardssuch as HTTP that allow information to be shared and ex-changed. A standard ontology language is, therefore, a pre-requisite if ontologies are to be used in order to share andexchange meaning. Recognising this fact, the World WideWeb Consortium (W3C) set up a standardisation workinggroup to develop such a langauge. The result of this activ-ity was the OWL ontology language standard [25]. OWLexloited existing work on langauges such as OIL [8] andDAML+OIL [14] and, like them, was based on a DescriptionLogic (DL). In the following we will briefly introduce DLsand OWL. For more complete information the reader shouldconsult The Description Logic Handbook [1], and the OWLspecification [25].

3.1 Description LogicDescription logics (DLs) are a family of logic-based knowl-

edge representation formalisms; they are descendants of Se-mantic Networks [35] and KL-ONE [4]. These formalismsall adopt an object-oriented model, similar to the one usedby Plato and Aristotle, in which the domain is described interms of individuals, concepts (usually called classes in on-tology languages), and roles (usually called relationships orproperties in ontology languages). Individuals, e.g., Socratesare the basic elements of the domain; concepts, e.g., Hu-man, describe sets of individuals having similar charac-teristics; and roles, e.g., hasPupil describe relationshipsbetween pairs of individuals, such as Socrates hasPupilPlato.

As well as atomic concept names such as Human, DLs alsoallow for concept descriptions to be composed from atomicconcepts and roles. Moreover, it is possible to assert thatone concept (or concept description) is subsumed by (is asub-concept of ), or is exactly equivalent to, another. Thisallows for easy extension of the vocabulary by introducingnew names as abreviations for descriptions. For example,using standard DL notation, we might write:

HappyParent Parent hasChild.(Intelligent Athletic)

This introduces the concept name HappyParent, and as-serts that its instances are just those individuals that areinstances of Parent, and all of whose children are instancesof either intelligent or athletic.

Another distinguishing feature of DLs is that they arelogics, and so have a formal semantics. DLs can, in fact, beseen as decidable subsets of first-order predicate logic, withindividuals being equivalent to constants, concepts to unarypredicates and roles to binary predicates. As well as givinga precise and unambiguous meaning to descriptions of thedomain, this also allows for the development of reasoningalgorithms that can be used to answer complex questionsabout the domain. An important aspect of DL research hasbeen the design of such algorithms, and their implementa-tion in (highly optimised) reasoning systems that can be

121


3/6

Figure 2: Example pizza ontology.

used by applications to help them understand the knowl-edge captured in a DL based ontology. We will return tothis point in Section 4.

A given DL is characterised by the set of constructorsprovided for building concept descriptions. These typicallyinclude at least intersection (), union () and complement(), as well as restricted forms of existential () and univer-

sal () quantification, which in OWL are called, respectively,someValuesFrom and allValuesFrom restrictions. OWL isbased on a very expressive DL called SHOIN that also pro-vides cardinality restrictions (, ) and enumerated classes(called oneOf in OWL) [15, 17]. Cardinality restrictionsallow, e.g., for the description of a concept such as peoplewho have at least two children, while enumerated classes al-low for classes to be described by simply enumerating theirinstance, e.g.,:

EUcountries {Austria, . . . ,UK}

SHOIN also provides for transitive roles, allowing us tostate, e.g., that ifx has an ancestor y and y has an ancestor z,then z is also an ancestor ofx, and for inverse roles, allowing

us to state, e.g., that if z is an ancestor of x, then x is alsoan descendent ofz. The constructors provided by OWL, andthe equivalent DL syntax, are summarised in Figure 3.

In DLs it is usual to separtate the set of statements thatestablish the vocabulary to be used in describing the do-main (what we might think of as the schema) from the setof statements that describe some particular situation thatinstantiates the schema (what we might think of as data);the former is called the TBox (Terminology Box), and thelatter the ABox (Assertion Box). An OWL ontology is sim-ply equivalent to a set of SHOIN TBox and ABox state-

Constructor DL Syntax ExampleintersectionOf C1 . . . Cn Human MaleunionOf C1 . . . Cn Doctor LawyercomplementOf C MaleoneOf {x1 . . . xn} {john,mary}allValuesFrom P.C hasChild.DoctorsomeValuesFrom r.C hasChild.LawyerhasValue r.{x} citizenOf.{USA}minCardinality ( n r) ( 2 hasChild)maxCardinality ( n r) ( 1 hasChild)inverseOf r hasChild

Figure 3: OWL constructors

ments. This mixing of schema and data is quite unusual (infact ontologies are usually thought of as consisting only ofthe schema part), but does not affect the meaningfrom alogical perspective, SHOIN KBs and OWL ontologies arejust sets of axioms.

The main difference between OWL and SHOIN is thatOWL ontologies use an RDF based syntax intended to fa-cilitate their use in the context of the Semantic Web. Thissyntax is rather verbose, and not well suited for presentationto human beings. E.g., the description ofHappyParent givenabove would be written in OWLs RDF syntax as follows:

122


4/6

4. ONTOLOGY REASONINGWe mentioned in Section 3.1 that the design and imple-

mentation of reasoning systems is an important aspect ofDL research. The availability of such reasoning systems wasone of the motivations for basing OWL on a DL. This isbecause reasoning is essential in supporting both the designof high quality ontologies, and the deployment of ontologiesin applications.

4.1 Reasoning at design timeOntologies may be very large and complex: the well known

Snomed clinical terms ontology includes, for example, morethan 200,000 class names [31]. Building and maintainingsuch ontologies is very costly and time consuming, and pro-

viding tools and services to support this ontology engineer-ing process is of crucial importance to both the cost and thequality of the resulting ontology. State of the art ontologydevelopment tools, such as SWOOP [19], Protege [22], andTopBraid Composer (see http://www.topbraidcomposer.com/), therefore use a DL reasoner, such as FaCT++ [33],Racer [11] or Pellet [29], to provide feedback to the userabout the logical implications of their design. This typi-cally includes (at least) warnings about inconsistencies andredundancies.

An inconsistent (sometimes called unsatisfiable) class isone whose description is over-constrained, with the resultthat it can never have any instances. This is typically anunintended feature of the designwhy introduce a name fora class that can never have any instancesand may be dueto subtle interactions between descriptions. The ability todetect such classes and bring them to the attention of theontology engineer is, therefore, a very useful feature.

It is also possible that the descriptions in the ontologymean that two classes necessarily have exactly the same setof instances, i.e., that they are alternative names for thesame class. This may be desirable in some situations, e.g.,to capture the fact that Myocardial infarction and Heartattack mean the same thing. It could, however, also be theinadvertent result of interactions between descriptions, andso it is also useful to be able to alert users to the presenceof such synonyms.

In addition to checking for inconsistencies and synonyms,ontology development tools usually also check for implicit

subsumption relationships, and update the class hierarchyaccordingly. This is also a very useful design aid: it allowsthe ontology developer to focus on class descriptions, leavingthe computation of the class hierarchy to the reasoner, and itcan also be used by the developer to check if the hierarchyinduced by the class descriptions is consistent with theirintuition.

Recent work has also shown how reasoning can be used tosupport modular design [6] and module extraction [5], im-portant techniques for working with large ontologies. Whendeveloping a large ontology such as SNOMED, it is useful

if not essential to divide the ontology into modules, e.g., tofacilitate parallel work by a team of ontology developers.Reasoning techniques can be used to alert the developers tounanticipated and/or undesirable interactions between thevarious modules. Similarly, it may be desirable to extractfrom a large ontology a smaller module containing all the in-formation relevant to some subset of the domain, e.g., heartdiseasethe resulting small(er) ontology will be easier forhumans to understand and easier for applications to use.

Reasoning can be used to compute a module that is as smallas possible while still containing all the necessary informa-tion.

Finally, in order to maximise the benefit of all these ser-vices, a modern system should also be able to explain itsinferences: without this facility, users may find it difficult torepair errors in the ontology and may even start to doubtthe correctness of the reasoning system. Explanation typ-ically involves computing a (hopefully small) subset of theontology that still entails the inference in question, and ifnecessary presenting the user with a chain of reasoning steps[20].

4.2 Reasoning in deployment

Reasoning is also important when ontologies are deployedin applicationsit is needed, e.g., in order to answer struc-tural queries about the domain and to retrieve data. If weassume, for example, an ontology that includes the descrip-tion ofHappyParent given in Section 3.1, and we know thatJohn is a HappyParent, that John has a child Mary (i.e.,John hasChild Mary), and that Mary is not Athletic, thenwe would like to be able to infer that Mary is Intelligent.

The above example may seem quite trivial, but it is easyto imagine that, with large ontologies, query answering maybe a very complex task. The use of DL reasoners allowsOWL ontology applications to answer complex queries, andto provide guarantees about the correctness of the result.This is particularly important if ontology based systems areto be used as components in larger applications, such as the

Semantic Web, where the correct functioning of automatedprocesses may depend on their b eing able to (correctly) an-swer such queries.

5. ONTOLOGY APPLICATIONSThe availability of tools and reasoning systems such as

those mentioned in Section 4 has contributed to the increas-ingly widespread use of OWL, not only in the Semantic Webper se, but as a popular language for ontology developmentin fields as diverse as biology [28], medicine [9], geography[10], geology [32], astronomy [7], agriculture [30] and defence[23]. Applications of OWL are particularly prevalent in thelife sciences where it has been used by the developers ofseveral large biomedical ontologies, including the Biological

Pathways Exchange (BioPAX) ontology [27], the GALENontology [26], the Foundational Model of Anatomy (FMA)[9], and the National Cancer Institute thesaurus [12].

The importance of reasoning support in such applicationswas highlighted in [21], which describes a project in whichthe Medical Entities Dictionary (MED), a large ontology(100,210 classes and 261 properties) that is used at theColumbia Presbyterian Medical Center, was converted intoOWL, and checked using an OWL reasoner. This check re-vealed systematic modelling errors, and a significant num-ber of missed subClass relationships which, if not corrected,

123


5/6

could have cost the hospital many missing results in variousdecision support and infection control systems that routinelyuse MED to screen patients.

6. FUTURE DIRECTIONSAs we have seen in Section 5, OWL is already being suc-

cessfully used in many applications. This success brings withit, however, many challenges for the future development of

both the OWL language and OWL tool support. Centralto these is the familiar tension between the requirements foradvanced features, in particular increased expressive power,and raw performance, in particular the ability to deal withvery large ontologies and data sets.

Use of OWL in the life sciences domain has brought tothe fore examples of both of the above mentioned require-ments. On the one hand, ontologies describing complex sys-tems in medicine and biology often require expressive powerbeyond what is currently supported in OWL. Two particu-lar features that are very often requested are the ability toqualify cardinality constraints, e.g., to describe the handas having four parts that are fingers and one part that is athumb, and the ability to have some characteristics be trans-ferred across transitive part-whole relations, e.g., to capture

the fact that a disease affecting a part of an organ affectsthe organ as a whole. The former feature (so called qualifiedcardinality restrictions) has long been well understood, andhas been available for some time in DL reasoners; the latterfeature is now also well understood, thanks to recent theo-retical work in the DL community [16, 13], and has recentlybeen implemented in DL reasoners.

This happy coincidence of user requirements and exten-sions in the underlying DLs and reasoning systems has ledto a proposal to extend OWL with these and other use-ful features that have been requested by users, for whicheffective reasoning algorithms are now available, and thatOWL tool developers are willing to support. In addition tothose mentioned above, the new features include extra syn-tactic sugar, extended datatype support, simple metamod-elling, and extended annotations. The extended language,called OWL 1.1, is now a W3C member submission (seehttp://www.w3.org/Submission/2006/10/), and is alreadysupported by tools such as Swoop, Protege and TopBraidComposer.

As well as increased expressive power, applications mayalso bring with them requirements for scalability that area challenge to current systems. This may include the abil-ity to reason with very large ontologies, perhaps containing10s or even 100s of thousands of classes, and the abilityto use an ontology with very large data sets, perhaps con-taining 10s or even 100s of millions of individualsin factdata sets much larger than this will certainly be a require-ment in some applications. Researchers are rising to these

challenges by developing new reasoning systems such as theOWL Instance Store [2], that uses a combination of DL rea-soning and relational database systems to deal with largevolumes of instance data, HermiT (see http://www.cs.man.ac.uk/bmotik/HermiT/), that uses a hypertableau basedtechnique to deal more effectively with large and complexontologies, and Kaon2 [18], that reduces OWL ontologies todisjunctive datalog programs, and uses deductive databasetechniques to enable it to deal with very large data sets.

7. CONCLUSIONS

As we have seen, the goal of Semantic Web research is totransform the Web from a linked document repository intoa distributed knowledge base and application platform, thusallowing the vast range of available information and servicesto be more effectively exploited. As a first step in this trans-formation, languages such as OWL have been developed;these languages are designed to capture the knowledge thatwill enable applications to better understand Web accessible

resources, and to use them more intelligently. It is easy toimagine that these developments might also be of benefit tousers with cognitive or sensory impairments.

Although realising the Semantic Web still seems some wayoff, OWL has already been very successful, and has rapidlybecome a de facto standard for ontology development infields as diverse as geography, geology, astronomy, agricul-ture, defence and the life sciences. An important factor inthis success has been the availability of sophisticated toolswith built in reasoning support.

The use of OWL in large scale applications has broughtwith it new challenges, both with respect to expressive powerand scalability, but recent research has also shown how theOWL language and OWL tools can be extended and adapted

to meet these challenges.

8. REFERENCES

[1] F. Baader, D. Calvanese, D. McGuinness, D. Nardi,and P. F. Patel-Schneider, editors. The DescriptionLogic Handbook: Theory, Implementation andApplications. Cambridge University Press, 2003.

[2] S. Bechhofer, I. Horrocks, and D. Turi. The OWLinstance store: System description. In Proc. of the20th Int. Conf. on Automated Deduction (CADE-20),Lecture Notes in Artificial Intelligence, pages 177181.Springer, 2005.

[3] T. Berners-Lee. Semantic web road map, Sept. 1998.Available at

http://www.w3.org/DesignIssues/Semantic.html.[4] R. J. Brachman and J. G. Schmolze. An overview of

the Kl-One knowledge representation system.Cognitive Science, 9(2):171216, Apr. 1985.

[5] B. Cuenca Grau, I. Horrocks, Y. Kazakov, andU. Sattler. Just the right amount: Extracting modulesfrom ontologies. In Proc. of the Sixteenth InternationalWorld Wide Web Conference (WWW 2007), 2007.

[6] B. Cuenca Grau, Y. Kazakov, I. Horrocks, andU. Sattler. A logical framework for modularintegration of ontologies. In Proc. of the 20th Int.Joint Conf. on Artificial Intelligence (IJCAI 2007),2007.

[7] S. Derriere, A. Richard, and A. Preite-Martinez. An

ontology of astronomical object types for the virtualobservatory. Proc. of Special Session 3 of the 26thmeeting of the IAU: Virtual Observatory in Action:New Science, New Technology, and Next GenerationFacilities, 2006.

[8] D. Fensel, F. van Harmelen, I. Horrocks,D. McGuinness, and P. F. Patel-Schneider. OIL: Anontology infrastructure for the semantic web. IEEEIntelligent Systems, 16(2):3845, 2001.

[9] C. Golbreich, S. Zhang, and O. Bodenreider. Thefoundational model of anatomy in OWL: Experience

124


6/6

and perspectives. J. of Web Semantics, 4(3), 2006.

[10] J. Goodwin. Experiences of using OWL at theordnance survey. In Proc. of the First OWLExperiences and Directions Workshop, volume 188 ofCEUR Workshop Proceedings. CEUR(http://ceur-ws.org/), 2005.

[11] V. Haarslev and R. Moller. RACER systemdescription. In Proc. of the Int. Joint Conf. onAutomated Reasoning (IJCAR 2001), volume 2083 ofLecture Notes in Artificial Intelligence, pages 701705.Springer, 2001.

[12] F. W. Hartel, S. de Coronado, R. Dionne, G. Fragoso,and J. Golbeck. Modeling a description logicvocabulary for cancer research. Journal of BiomedicalInformatics, 38(2):114129, 2005.

[13] I. Horrocks, O. Kutz, and U. Sattler. The even moreirresistible SROIQ. In Proc. of the 10th Int. Conf. onPrinciples of Knowledge Representation andReasoning (KR 2006), pages 5767. AAAI Press, 2006.

[14] I. Horrocks, P. F. Patel-Schneider, and F. vanHarmelen. Reviewing the design of DAML+OIL: Anontology language for the semantic web. In Proc. ofthe 18th Nat. Conf. on Artificial Intelligence

(AAAI 2002), pages 792797. AAAI Press, 2002.[15] I. Horrocks, P. F. Patel-Schneider, and F. van

Harmelen. From SHIQ and RDF to OWL: Themaking of a web ontology language. J. of WebSemantics, 1(1):726, 2003.

[16] I. Horrocks and U. Sattler. Decidability ofSHIQ withcomplex role inclusion axioms. Artificial Intelligence,160(12):79104, Dec. 2004.

[17] I. Horrocks and U. Sattler. A tableaux decisionprocedure for SHOIQ. In Proc. of the 19th Int. JointConf. on Artificial Intelligence (IJCAI 2005), pages448453, 2005.

[18] U. Hustadt, B. Motik, and U. Sattler. ReducingSHIQ-description logic to disjunctive datalog

programs. In Proc. of the 9th Int. Conf. on Principlesof Knowledge Representation and Reasoning(KR 2004), pages 152162, 2004.

[19] A. Kalyanpur, B. Parsia, E. Sirin, B. Cuenca-Grau,and J. Hendler. SWOOP: a web ontology editingbrowser. J. of Web Semantics, 4(2), 2005.

[20] A. Kalyanpur, B. Parsia, E. Sirin, and J. Hendler.Debugging unsatisfiable classes in owl ontologies. J. ofWeb Semantics, 3(4):243366, 2005.

[21] A. Kershenbaum, A. Fokoue, C. Patel, C. Welty,E. Schonberg, J. Cimino, L. Ma, K. Srinivas,R. Schloss, and J. W. Murdock. A view of OWL fromthe field: Use cases and experiences. In Proc. of theSecond OWL Experiences and Directions Workshop,

volume 216 of CEUR (http://ceur-ws. org/

), 2006.[22] H. Knublauch, R. Fergerson, N. Noy, and M. Musen.The Protege OWL Plugin: An open developmentenvironment for semantic web applications. In S. A.McIlraith, D. Plexousakis, and F. van Harmelen,editors, Proc. of the 2004 International Semantic WebConference (ISWC 2004), number 3298 in LectureNotes in Computer Science, pages 229243. Springer,2004.

[23] L. Lacy, G. Aviles, K. Fraser, W. Gerber,A. Mulvehill, and R. Gaskill. Experiences using OWL

in military applications. In Proc. of the First OWLExperiences and Directions Workshop, volume 188 ofCEUR Workshop Proceedings. CEUR(http://ceur-ws.org/), 2005.

[24] S. McIlraith, T. C. Son, and H. Zeng. Semantic webservices. IEEE Intelligent Systems, 16:4653, 2001.

[25] P. F. Patel-Schneider, P. Hayes, and I. Horrocks.OWL Web Ontology Language semantics and abstractsyntax. W3C Recommendation, 10 February 2004.Available at http://www.w3.org/TR/owl-semantics/.

[26] A. Rector and J. Rogers. Ontological and practicalissues in using a description logic to represent medicalconcept systems: Experience from GALEN. InReasoning Web, Second International Summer School,Tutorial Lectures, volume 4126 of LNCS, pages197231. SV, 2006.

[27] A. Ruttenberg, J. Rees, and J. Luciano. Experienceusing OWL DL for the exchange of biological pathwayinformation. In Proc. of the First OWL Experiencesand Directions Workshop, volume 188 of CEURWorkshop Proceedings. CEUR(http://ceur-ws.org/), 2005.

[28] A. Sidhu, T. Dillon, E. Chang, and B. S. Sidhu.

Protein ontology development using OWL. In Proc. ofthe First OWL Experiences and Directions Workshop,volume 188 of CEUR Workshop Proceedings. CEUR(http://ceur-ws.org/), 2005.

[29] E. Sirin, B. Parsia, B. Cuenca Grau, A. Kalyanpur,and Y. Katz. Pellet: A practical owl-dl reasoner. Toappear, 2005.

[30] D. Soergel, B. Lauser, A. Liang, F. Fisseha, J. Keizer,and S. Katz. Reengineering thesauri for newapplications: The AGROVOC example. J. of DigitalInformation, 4(4), 2004.

[31] K. Spackman. Managing clinical terminologyhierarchies using algorithmic calculation ofsubsumption: Experience with SNOMED-RT. J. of

the Amer. Med. Informatics Ass., 2000. FallSymposium Special Issue.

[32] Semantic web for earth and environmentalterminology (SWEET). Jet Propulsion Laboratory,California Institute of Technology, 2006.http://sweet.jpl.nasa.gov/.

[33] D. Tsarkov and I. Horrocks. FaCT++ descriptionlogic reasoner: System description. In Proc. of the Int.Joint Conf. on Automated Reasoning (IJCAR 2006),volume 4130 of Lecture Notes in Artificial Intelligence,pages 292297. Springer, 2006.

[34] R. Volz, S. Handschuh, S. Staab, L. Stojanovic, andN. Stojanovic. Unveiling the hidden bride: DeepAnnotation for Mapping and Migrating Legacy Datato the Semantic Web. Journal of Web Semantics,2004.

[35] W. A. Woods. Whats in a link: Foundations forsemantic networks. In R. J. Brachman and H. J.Levesque, editors, Readings in KnowledgeRepresentation, pages 217241. Morgan KaufmannPublishers, San Francisco, California, 1985.

125

Documents

6.Semantic Web