XML Declarative Description: a language for the Semantic Web

54 1094-7167/01/$10.00 © 2001 IEEE IEEE INTELLIGENT SYSTEMS

T h e S e m a n t i c W e b

XML DeclarativeDescription: A Language for theSemantic WebVilas Wuwongse and Chutiporn Anutariya, Asian Institute of TechnologyKiyoshi Akama, Hokkaido UniversityEkawit Nantajeewarawat, Sirindhorn International Institute of Technology

Today’s Web serves primarily as a global space to present information for human

consumption. However, it lacks a mechanism for providing information that

machines can comprehend or process, allowing them to communicate and interoperate.

So, software engineers must work hard to develop intelligent services and automated

software agents because they must first agree on thedata’s syntax and semantics before hard-coding theminto their applications. In addition, changes to syn-tax and semantics necessitate expensive applicationmodifications.

We need a Semantic Web1 with a well-establishedmechanism to express information that is machine-interpretable and allows syntactic and semantic inter-operability among Web applications. Although XML(eXtensible Markup Language) and RDF (ResourceDescription Framework) offer foundations for,respectively, syntactic and semantic interoperability,their mechanisms cannot accomplish this goal. XMLby itself will let the same semantic unit be expressedin more than one syntactic structure. XML, RDF, andRDF Schema combinations might solve this multi-ple-structures problem, but they would still lackexpressive power. For example, axioms, conditions,and constraints could not be specified.2

OIL3 (Ontology Inference Layer) andDAML+OIL4 (DARPA Agent Markup Language +OIL) are two recent improved frame-based lan-guages. They extend RDF Schema by richer sets ofmodeling primitives for representation of Booleanexpressions, axioms, and property restrictions. Thesemantics of these languages involves mapping

such extended RDF statements on correspondingrepresentations in a particular logical theory—forexample, on first-order logic sentences—followedby corresponding determination of their semantics.OIL’s semantics relies on a translation into thedescription logic SHIQ,5 whereas DAML+OIL’sis based on KIF (Knowledge Interchange For-mat)6—a language designed for knowledge inter-change and based on first-order predicate logic.Their main inference services are class consistencyand subsumption checking with other reasoningservices, for example, query subsumption andquery answering over classes and instances, refor-mulated in terms of subsumption checking.

All these extensions require additional formalismsfor XML and RDF to define their semantics or spec-ify axioms and constraints. We need an XML-basedlanguage with a single formalism that meets all therequirements of the Semantic Web. Its intendedsemantics should be precisely and formally defin-able under its single formalism, and it should sup-port general inference mechanisms. XML Declara-tive Description7 aims to fill this need.

An informal introductionXDD7 is a language that enables representation of

Although XML and

RDF are widely used

languages for Web

applications, they

have insufficient

mechanisms to fulfill

the requirements of

a Semantic Web. The

authors show how

XML Declarative

Description expands

the capabilities of

XML and RDF to meet

these requirements.

Copyright 2001 Instituteof Electrical and ElectronicsEngineers. Reprinted, with permission, from IEEEIntelligent Systems and theIEEE. Such permission of theIEEE does not in any wayimply IEEE endorsement.Internal or personal use ofthis material is permitted.However, permission toreprint/republish this materialfor advertising or promotionalpurposes or for creating newcollective works for resale orredistribution must beobtained from the IEEE bysending a blank email messageto [email protected]. By choosing to view thisdocument, you agree to allprovisions of the copyrightlaws protecting it.

a Web resource’s semantics. It employs XMLas its bare syntax and enhances XML expres-sive power by employing DeclarativeDescription theory.8 A description in XDDis a set of ordinary XML elements, extendedXML elements with variables, and the XMLelements’ relationships in terms of XMLclauses. An ordinary XML element denotesa semantic unit and is a surrogate of an infor-mation item in the real application domain.An extended XML element representsimplicit information or a set of semanticunits. Clauses express rules, conditional rela-tionships, integrity constraints, and ontolog-ical axioms. We define the precise and for-mal semantics of an XDD description as aset of ordinary XML elements, withoutemploying other formalisms.

Important axioms that are missing in XMLand RDF but expressible in XDD includesymmetry, composition-of, and inverserelations. As an example of the inverse-rela-tion axiom, consider the XML clauses A andB in Figure 1a. They model the Creatorand Publication properties’ inversesassumed by some particular domain. Figure1b then gives an example of representing anRDF statement C, “A creator of a documententitled ‘XDD language’ is John.” The

semantics of an XDD description, whichcomprises the clauses A, B, and C, will alsocontain an RDF statement, “A publication ofJohn is ‘XDD language’” (Figure 1c), henceallowing inverse inference of such implicitinformation. Obviously, this axiom cannotbe represented in RDF.

XDD can directly represent all XML-based application markup languages. It alsocan simply represent XML applications thatprovide common conventions of semantics,syntax, and structures for certain specificdomains. In addition to RDF, these domainsinclude the following:

• MathML (Mathematical Markup Lan-guage);

• XMI (XML Metadata Interchange For-mat—the Object Management Group’srecommended technology for represent-ing Unified Modeling Language and MetaObject Facility diagrams in XML); and

• WML (Wireless Markup Language).

Encoded in XDD, these languages canhave their intended semantics formallydefined. XDD removes the boundary andbrings about the convergence of these lan-guages’ syntax and semantics. This readily

enables interoperability of independentlydeveloped Web applications.

XDD applications include

• e-commerce: flexible communication,interaction, and collaboration schemes fornegotiation agents;

• resource discovery and digital libraries:an intelligent search engine that under-stands and catalogs Web contents andincorporates knowledge into the search toimprove precision and recall; and

• software engineering: automatic configu-ration of a new software component basedon existing ones and certain specific con-figuration rules.9

Figure 2 depicts XDD’s role in modelingthe Semantic Web. The Unicode layer merelyviews exchanged data as a stream of Unicodecharacters. Next, the XML layer creates anXML document from the stream. Typically,the obtained document can be merely adocument, an XML Schema, an RDF docu-ment, an RDF Schema, or a combination ofthese, because they are similarly encoded inXML syntax. The document’s semantics isformally determined in the XDD layer, yield-ing a set of data objects that are encoded in

MAY/JUNE 2001 computer.org/intelligent 55

<rdf:Description about=$S:personP> <rdf:type resource="#Person"/> <Publication resource=$S:documentD/></rdf:Description> <rdf:Description about=$S:documentD> <rdf:type resource="#E-Document"/> <Creator resource=$S:personP/> $E:D_properties

</rdf:Description>.

<rdf:Description about=$S:documentD> <rdf:type resource="#E-Document"/> <Creator resource=$S:personP/></rdf:Description> <rdf:Description about=$S:personP> <rdf:type resource="#Person"/> <Publication resource=$S:documentD/> $E:P_properties </rdf:Description>.

<rdf:Description about="http://xdd.org"> <rdf:type resource="#E-Document"/> <Title>XDD Language</Title> <Creator resource="http://smith.com/john"/></rdf:Description>

<rdf:Description about="http://xdd.org"> <rdf:Description about="http://smith.com/john"> <rdf:type resource="#E-Document"/> <rdf:type resource="#Person"/> <Title>XDD Language</Title> <Publication resource="http://xdd.org"/> <Creator resource="http://smith.com/john"/> </rdf:Description></rdf:Description>

% If a creator of an e-document D is % a resource P, it is known that such % a resource P must be an instance % of a class Person and one of P ’s % publications is that e-document D.

% If a publication of a person P is a % resource D, then we can infer % that D is an e-document, the % creator of which is the person P.

% An RDF statement describing a % resource http://xdd.org.

A:

B :

C:

(b)

(c)

(a)

Figure 1. An XDD description example and its semantics: (a) XML clauses A and B model an inverse-relation axiom; (b) XML clause C models an RDF statement; (c) the semantics of an XDD description, which comprises the clauses A, B, and C, contains two RDF statements.

their respective XML applications in the dataobject layer. These objects can be explicitlydescribed by the document or derived fromthe relationships, rules, or axioms in the doc-ument. In the application layer, such objectsare surrogates of real-world objects in a par-ticular application domain.

A formal introductionXDD’s words and sentences are XML

expressions and XML clauses, respectively.XML expressions represent explicit andimplicit as well as simple and complex facts.XML clauses represent ontology, implicitand conditional relationships, constraints,and axioms.

XML expressions and XML elements havea similar form. However, XML expressionscan carry variables to represent implicit infor-mation and to enhance XML elements’expressive power. Every component of anXML expression—the expression itself, itstag name, attribute names and values,attribute–value pairs, contents, subexpres-sions, and some partial structures—can con-tain variables. XML expressions without vari-ables are called ground XML expressions orsimply XML elements. Those with variablesare called nonground XML expressions. Table1 defines all variable types and their usages.

Variable instantiation is defined by basicspecializations, each of which has the form (v, w) where v specifies the name of the variableto be specialized and w the specializing value.For example, ($N:name1, $N:name2),($S:url, “http://smith.com”) and($E:properties, ($E:p1, $E:p2))are basic specializations that rename the N-variable $N:name1 as $N:name2, instan-tiate the S-variable $S:url into the string“http://smith.com/”, and expandthe E-variable $E:properties into thesequence of the E-variables $E:p1 and$E:p2. Basic specializations come in fourtypes:

• rename variables,• expand a P- or an E-variable into a sequence

of variables of their respective types,

56 computer.org/intelligent IEEE INTELLIGENT SYSTEMS


Application layer

XDD layer

XML layer

Unicode layer

Data object (resource) layer

Applications in the Semantic Web

Expressions of explicit facts about,constraints on, and relationships among data objects

XML documents, XML schemas, RDF/XML documents, and RDF schemas

Serialization of exchanged data (XML documents) as Unicode character streams

Data objects in particular applicationdomains that are surrogates of real-world objects and are encoded intheir respective XML applications

MathMLobjects

XMIobjects

WMLobjects

. . . RDFobjects

Figure 2. XDD’s role in the Semantic Web architecture.

Table 1. Variable types.

Variable type Variable names beginning with Instantiation to

N-variables (Name variables) $N Element types or attribute namesS-variables (String variables) $S StringsP-variables (Attribute-value-pair variables) $P Sequences of zero or more attribute-value pairsE-variables (XML expression variables) $E Sequences of zero or more XML expressionsI-variables: (Intermediate expression variables) $I Parts of XML expressions

Let’s look at the definition of specialization systems, defined in the DeclarativeDescription theory.1

Let A, G, and S be sets of objects, ground objects, and specializations, respec-tively, and µ be a mapping from S to partial_map(A) (the set of all partialmappings on A). The quadruple ⟨A, G, S, µ⟩ is a specialization system under theconditions

1. ∀s1, s2 ∈S, ∃s ∈S : µ(s) = µ(s1) ° µ(s2);2. ∃s ∈S, ∀a ∈A : µ(s)(a) = a; and3. G ⊂ A,

where µ(s1) ° µ(s2) is the composite mapping of the partial mappings µ(s1) and µ(s2). Intuitively, Conditions 1 to 3 mean

1. For all specializations s1 and s2, there exists a specialization s such that the cor-responding partial mapping of s is the composition of the two mappings corre-sponding to s1 and s2,

2. There exists a specialization that does not change any objects (identity special-ization), and

3. Ground objects are objects.

When µ is clear from the context, for θ ∈S, we write µ(θ)(a) simply as aθ. If bexists such that aθ = b, θ is said to be applicable to a, and a is specialized to b by θ.

Reference

1. K. Akama, “Declarative Semantics of Logic Programs on Parameterized RepresentationSystems,” Advances in Software Science and Technology, Iwanami Shoten, Publishers andAcadamic Press, Tokyo, vol. 5, 1993, pp. 45–63.

Specialization Systems

• remove P-, E-, or I-variables, or• instantiate variables to XML expressions

or components of XML expressions thatcorrespond to the variables’ types.

A sequence of basic specializations is aspecialization. The data structure of XMLexpressions is characterized by a mathemat-ical abstraction ΓX = ⟨AX, GX, SX, µX⟩, calledthe XML Specialization System (see therelated sidebar), where

• AX is the set of all XML expressions,• GX is the subset of AX that comprises all

ground XML expressions in AX,• SX is the set of all specializations that

reflect the data structure of the XMLexpressions in AX, and

• µX is the specialization operator, whichdetermines for each specialization s in SX

the change of each XML expression in AX

caused by s.

Figure 3a illustrates an example of a non-ground XML expression a in AX. Figure 3billustrates a specialization θ in SX, and the appli-cation of θ to a by the operator µX to obtain aground XML expression g in GX (Figure 3c).That is g = µX(θ)(a) or, by postfix notation, g =aθ. Specialization θ changes the nongroundexpression a to the ground expression g by

• instantiation of the N-variable $N:titleinto the tag name dc:Title,

• instantiation of the S-variable $S:urlinto the string “http://smith.com”,

• expansion of the E-variable $E:prop-erties to the sequence of the E-vari-ables $E:p1 and $E:p2,

• instantiation of the E-variable $E:p1 intothe XML expression <dc:Creatorresource=”http://smith.com/john”/>, and

• instantiation of the E-variable $E:p2into the XML expression <dc:Lan-guage>English</dc:Language>.

Constraints are useful for defining restric-tions on XML expressions or components ofXML expressions. A constraint is a formulaq(a1, …, an), where n > 0, q is a constraintpredicate, and ai is an XML expression. Theapplication of θ ∈SX to a constraint q(a1, …,an) yields q(a1θ, …, anθ). A ground con-straint, which takes the form q(g1, …, gn),gi ∈ GX, has a predetermined truth or falsity.For instance, given two XML expressions a1

and a2, define GT(a1, a2) as a constraint that

will be true if and only if a1 and a2 are XMLelements of the forms<Num>v1</Num> and<Num>v2</Num>, respectively, where v1

and v2 are numbers and v1 > v2.Based on the XML specialization system

ΓX and the concept of constraints, an XMLdeclarative description on ΓX, simply calledan XDD description, is a set of XML clauses.Each clause has the form

H ← B1, B2, …, Bn,

where

• n ≥ 0,• H is an XML expression in AX,• Bi is an XML expression in AX or a con-

straint on ΓX, and

• the order of the Bi is immaterial.

H is the head and (B1, B2, …, Bn) is the bodyof the clause.

For a unit clause, n = 0; for a non-unitclause, n > 0. Normally, we use (H ←) to indi-cate a unit clause. However, when the contextclearly implies a unit clause, we will write itsimply as H. With this representation, everyXML document becomes immediately anXDD description without a non-unit clause.

The meaning of a given XDD descriptionP, denoted by M(P) (see the “Semantics ofDeclarative Description” sidebar), is the setof all XML elements that are directlydescribed by and are derivable from the unit


Here we formally define a given declarative description’s semantics on a particu-lar specialization system.

Let Γ = ⟨A, G, S, µ⟩ be a specialization system and C be a clause (H ← B1, B2, …,Bn) on Γ. The head of C will be denoted by head(C) and the set of all objects and con-straints in the body of C by object(C) and con(C), respectively. For a specialization θ ∈ S, application of θ to the clause C is Cθ = (Hθ ← B1θ, B2θ, …, Bnθ). A clause C is aground clause if and only if C comprises only ground objects and ground constraints.

Let Tcon denote the set of all true ground constraints and P be a declarativedescription on Γ.

Associated with P is the mapping TP on 2G , which we define as this:

For each X ⊂ G, a ground object g is contained in TP(X) if and only if a clause C ∈ P and a specialization θ ∈ S exist such that Cθ is a ground clause with thehead g, and all the objects and constraints in the body of Cθ belong to X andTcon, respectively; that is, TP(X) = {head(Cθ) | C ∈ P, θ ∈ S, Cθ is a ground clause,object(Cθ ) ⊂ X, con(Cθ) ⊂ Tcon}.

Based on TP, the meaning of P, denoted by M(P), is defined as

,

where ∅ is the empty set, and [TP]1(∅) = TP(∅) and [TP]n(∅) = TP([TP]n−1(∅)) for each n > 1.

M (P ) = [Tp]n(∅)∞

n=1

The Semantics of Declarative Description

(c)

(b)

(a)

<rdf:Description about=$S:url> <$N:title>Smith Company’s Page</$N:title> $E:properties</rdf:Description>

<rdf:Description about="http://smith.com"> <dc:Title>Smith Company’s Page</dc:Title> <dc:Creator resource="http://smith.com/john"/> <dc:Language>English</dc:Language></rdf:Description>

($N:title, dc:Title)($S:url, "http://smith.com")($E:properties, ($E:p1, $E:p2))($E:p1, <dc:Creator resource="http://smith.com/john"/>)($E:p2, <dc:Language>English</dc:Language>)

Specializedinto

by x( )θµ

Figure 3. Specialization of a nonground XML expression into a ground XML expression:(a) nonground XML expression a; (b) specialization θ; (c) ground XML expression g = aθ.

and the non-unit clauses in P, respectively—that is,

• Given a unit clause (H ←) in P, for θ ∈SX, Hθ ∈M(P) if Hθ is a ground XMLexpression.

• Given a non-unit clause (H ← B1, ..., Bi,Bi+1, …, Bn) in P, assuming without lossof generality that B1, ..., Bi are XMLexpressions and Bi+1, …, Bn are con-straints, for θ ∈SX, Hθ ∈M(P) if Hθ is aground XML expression; if B1θ, ...,Biθ, ∈M(P); and if Bi+1θ, …, Bnθ are trueconstraints.

Figure 4 illustrates an XDD approach tomodeling objects (Figure 4a) and the rela-tionships between XDD language and a real-world domain (Figure 4b).

XDD and the Semantic WebInteroperation among various applications

in the Semantic Web demands common rep-resentation and interpretation of exchangeddata. Each application domain requires an

appropriate definition of standard documentsyntax together with an agreement about, ora common understanding of, the employedontology.10 In general, an ontology is a spec-ification of concepts, their hierarchical rela-tionships and axioms in a particular applica-tion domain. Objects in the domain areinstances of one or more concepts. Each con-cept also contains a set of properties.

From this point of view, the SemanticWeb’s components are constraints, ontolo-gies, and contents, all of which can be mod-eled and manipulated by XDD:

• Constraints or restrictions on the infor-mation exchange format can be repre-sented and imposed by a correspondingset of XML non-unit clauses.

• Concepts and their properties in an ontol-ogy are described as XML unit clauses.Their hierarchical relationships and onto-logical axioms, such as symmetry andinverse, are modeled as XML non-unitclauses. As an example of modeling con-cept hierarchies, consider the expressionthat the concept “Web-Page” is a special-ization of the concept “E-Document,”which can be represented as a clause,assuming that concepts are described astag names in XML (see Figure 5).

• Contents, describing certain objectsand their relationships, are also mod-eled as XML unit and non-unit clauses,respectively.

Table 2 describes the Semantic Web com-


<Description about=$S:personP> <Description about=$S:docD> <type resource="#Person"/> <type resource="#E-Document"/> <Publication resource=$S:docD/> <Creator resource=$S:personP/> </Description> $E:D_properties </Description>.

<Description about=$S:docD> <Description about=$S:personP> <type resource="#E-Document"/> <type resource="#Person"/> <Creator resource=$S:personP/> <Publication resource=$S:docD/></Description> $E:P_properties

</Description>.

<Description about="http://xdd.org"> <type resource="#E-Document"/> <Title>XDD Language</Title> .

<Creator resource="http://john"/></rdf:Description>

XML expressionNonground XML expressionSet of all XML expressionsSet of all ground XML expressionsXDD description—modeling of objects and their relationships in a particular domain

The semantics of P— that is, a set of ground XML expressions.

M(P)GXAX

M(P)

P

GX

AX

Application domain

An XDD description P

E-Document:http://xdd.org

Person: John

is a surrogate of

(b)(a)

Figure 4. The relationships between XDD and a real-world domain.

... ...<E-Document> $E:X </E-Document> <Web-Page> $E:X </Web-Page> ... ...

% This clause specifies that % every Web page object is% an e-document object.

Figure 5. An example of modeling a concept hierarchy.


ponents and XDD’s role in modeling eachcomponent. On the basis of this modelingtechnique, a Web resource, modeled as theXDD description P, will uniquely convey itsmeaning, represented as a set of groundXML expressions in terms of M(P).

Modeling the Semantic WebWe now show how to employ XDD to

model the Semantic Web. This exampleassumes that RDF syntax, RDF Schemas,and RDF statements describe the constraints,ontologies, and contents, respectively. More-over, we show how to apply XDD to axiommodeling—an important notion missingfrom the RDF framework. We also give thesemantics.11

Constraint modeling: RDF syntaxFor simplicity, consider the partial RDF

serialization syntax in Figure 6a, which wecan represent simply as the XDD descriptionP1 in Figure 6b. Based on P1, we formulatethe clause V to determine whether the RDFstatement of Figure 1b conforms to the givenRDF Syntax or not (see Figure 6c). If thestatement is valid, M(P1 ∪ {V}) will includethe XML expression <xdd:ValidDe-scription about=“http://xdd.org”/>.

We can similarly define XML clauses,which can parse and validate completeRDF serializations as well as abbreviatedsyntax.

Ontology modeling: RDF Schemas and axioms

RDF Schema is a language that providesa simple ontology definition facility. Twoessential constructs, subClassOf andsubPropertyOf, let you specify hierar-chical relationships among a set of classesand properties, respectively. The constructsrange and domain let you impose con-straints on a property’s value and on the types(classes) of objects to which a property canbe applied, respectively.

Figure 7 shows an RDF Schema graph andits corresponding RDF Schema document,which defines a simple ontology for describ-ing electronic resources. That is,

• Person, E-Document, E-Article,and Web-Page are rdfs:Class,

• Person and E-Document are sub-classes of rdfs:Resource,

• E-Article and Web-Page are sub-classes of E-Document (Figure 7a),

• Publication, Name, Title, Cre-ator, Author, and Illustrator arerdf:Property, and

• Author and Illustrator are sub-properties of Creator (Figure 7b).

Because RDF Schema documents areXML documents, they correspond directlyto XDD descriptions containing only unitclauses. So, the example document in Figure7c directly becomes the XDD description P2.P2 comprises 10 XML unit clauses, the headsof which are the RDF statements in theexample document.

To model the meanings of subClassOfand subPropertyOf, which are transitiveand include some notion of implication, weformulated the XML non-unit clauses C6

through C9 in Figure 8. These clauses con-stitute XDD description P3; that is, P3 = {C6,C7, C8, C9}. The semantics of P2 ∪ P3 explic-itly yields the following implicit informationand relations:

• E-Article and Web-Page are sub-classes of rdfs:Resource (through thesubClassOf transitivity property).

• Every instance of E-Article or Web-Page is also an instance of the class E-Document.

• Every instance of Person, E-Docu-ment, E-Article, or Web-Page isalso an instance of rdfs:Resource.

• Every resource having an Author or anIllustrator property also has a Cre-ator property with a value similar to thatof Author or Illustrator.

Additional relations among classes or prop-erties, as well as the meanings ofrdfs:range and rdfs:domain con-straints, are also expressible in XDD. More-over, by facilitating specifications of rules, onto-

logical axioms, and conditions and constraintson classes and properties, XDD overrides RDFSchema, which permits only a simple ontologi-cal modeling mechanism. As we mentionedbefore, Figure 1a shows the modeling of theinverse of the Creator–Publicationaxiom. XDD description P4 denotes the set ofclauses A and B in the figure; that is, P4 = {A,B}. Clause A describes that the inverse of Creator is the Publication property;clause B represents the reverse. Other kinds ofaxioms can be modeled similarly.

Content modeling:RDF statements

RDF statements, which model given con-tents of the Semantic Web and are encoded inXML syntax, are directly mapped onto XMLelements or XML unit clauses in XDD. Fig-ure 9 shows four RDF statements that de-scribe certain Web resources; we call thesestatements XDD description P5.

The semanticsThe union of the XDD descriptions P1

through P5 becomes a simple example of theSemantic Web modeling—that is,

P = P1 Constraints on the data-exchange format (RDF syntax).

∪ P2 Concept and relation descrip-tions (RDF Schema).

∪ P3 Concept and relation hierar-chies (modeling the mean-ings of the RDF Schema con-structs subClassOf andsubPropertyOf).

∪ P4 Axioms (inverses of theCreator and Publica-tion properties).

∪ P5 Resources of the SemanticWeb (RDF statements).


Table 2. Modeling the Semantic Web.

Semantic Web component Expressed as

Constraints on the information exchange format XML non-unit clausesOntologies

Concept and property descriptions XML unit clausesHierarchies of concepts and properties XML non-unit clausesAxioms XML non-unit clauses

ContentsObjects XML unit clausesRelationships among objects XML non-unit clauses

A resource of the Semantic Web Modeled as An XDD description P on ΓX comprising(contents + ontology + constraints) ⇒ XML unit clauses + XML non-unit clauses

The semantics of the resource Is M(P)

The semantics of P, M(P) includes thisinformation, which is implicit in the Seman-tic Web’s content:

• All RDF statements in P2 (see Figure 7c)

and P5 (see Figure 9) conform to the RDFsyntax.

• The resources “http://smith.com”and “http://xdd.org” are instancesof E-Document.

• The resources “http://smith.com”,“ h t t p : / / x d d . o r g ” ,“http://smith.com/john”, and“http://smith.com/joe” areinstances of rdfs:Resource.



% This clause together with the clause C2 specifies that % every rdf:Description element must contain exactly % one about or ID attribute and may contain zero or % more property elements. The xdd:PropertyCheck % element contained in the body restricts that such % property elements will be further restricted by the % clauses whose heads are xdd:PropertyCheck—that is,% C3, C4, and C5.

% This clause and the clause C4 impose restrictions on % the syntax of every property element, contained in % rdf:Description. That is, % • Its tag name could be any valid property name % (represented by the name variable $N:propName), % • If its content is any textual string (represented by % the String variable $S:value), it must contain no % attribute-value pair; and% • If its content is empty, it must contain exactly % one resource attribute.

% This clause together with the clauses C3 and C4 restricts% that rdf:Description may contain zero or more % property elements.

C1: <xdd:DescriptionCheck> <rdf:Description about=$S:uri> $E:propertyElt </rdf:Description> </xdd:DescriptionCheck> <xdd:PropertyCheck> $E:propertyElt </xdd:PropertyCheck>.

C2: <xdd:DescriptionCheck> <rdf:Description ID=$S:uri> $E:propertyElt </rdf:Description> </xdd:DescriptionCheck> <xdd:PropertyCheck> $E:propertyElt

</xdd:PropertyCheck>.

C3: <xdd:PropertyCheck> <$N:propName> $S:value </$N:propName> $E:propertyElt </xdd:PropertyCheck> <xdd:PropertyCheck>

$E:propertyElt </xdd:PropertyCheck>.

C4: <xdd:PropertyCheck> <$N:propName resource=$S:uri/> $E:propertyElt </xdd:PropertyCheck> <xdd:PropertyCheck>

$E:propertyElt </xdd:PropertyCheck>.

C5: <xdd:PropertyCheck> </xdd:PropertyCheck> .

(b)

(c)

(a)

rdf-statement ::= <rdf:Description about=string> propertyElt* </rdf:Description> propertElt ::= <propName> string </propName> | <propName resource=string> string ::= (any XML text) propName ::= (any valid property name)

V: <xdd:ValidDescription about=" http://xdd.org"/> <xdd:DescriptionCheck>

<rdf:Description about="http://xdd.org"> <rdf:type resource="#E-Document"/> <Title>XDD Language</Title> <Creator resource="http://smith.com/john"/> </rdf:Description> </xdd:DescriptionCheck>.

% This clause specifies that if the RDF % statement contained in its body % conforms to the syntax described by % the clauses C1 through C5, the XML % expression specified in its head will % be obtained.

Figure 6. Constraint modeling: (a) a partial RDF serialization syntax; (b) its representation as the XDD description P1 = {C1, …, C5}; (c) an XML clause for checking the conformance of the RDF statement of Figure 1b.

• The person referred to by “http://smith.com/john” is a Creator of“http://smith.com”and “http://xdd.org”.

• The person referred to by “http://smith.com/joe” is a Creator of“http://xdd.org”.

• The resources “http://smith.com”and “http://xdd.org” are Publi-cations of the person referred to by“http://smith.com/john”.

• The resource “http://xdd.org” is aPublication of the person referred toby “http://smith.com/joe”.

So, a search of all E-Documentresources of which John Smith is a Cre-ator, for instance, will return also theresources “http://smith.com” and“http://xdd.org”, although suchresources have not been declared as E-

Document and John Smith has not beendeclared explicitly as their Creator butonly as Author and Illustrator.This implicit information is uncoveredthrough the predefined hierarchicalrelationships among E-Document, E-Article, and Web-Page and amongCreator, Author, and Illustrator.

Requirements of a SemanticWeb language

Tim Berners-Lee pointed out that a goodlanguage for the Semantic Web should have

• compact syntax;• well-defined semantics;• sufficient expressive power to represent

human knowledge;• an efficient, powerful, and understandable

reasoning mechanism; and• the potential for building large

knowledge bases.12

However, the third and fourthproperties conflict. In addition,these five properties focusmerely on the information rep-resentation and computationaspects; information presenta-tion is missing. To emphasizethat the Semantic Web should bea means for not only machine-to-machine communication butalso machine-to-human com-munication, we should add asixth property: it includes a pre-sentation or rendering scheme.

Because XDD concentrateson the information represen-tation aspect with an attemptto provide a concise, expres-sive language with precise,well-defined semantics, it hasall but the fourth and sixthproperties. These limitationscan be resolved in two ways.

The first solution is to pro-vide efficient computation.We employed ET (EquivalentTransformation),13 a new com-putational paradigm, to allowefficient manipulation of andreasoning with XDD. (Anequivalent transformation is atransformation that preservesthe equivalence of the trans-formed descriptions—a seman-tic-preserving transformation.)

We carried out computation through ET bysuccessive transformation of a given XDDdescription R1 into R2, R3, …, until weobtained a desirable XDD description Rn.During the transformation, we must preserveeach XDD description’s semantics—that is,M(R1) = M(R2) = M(R3) = ... = M(Rn). Toguarantee the computation’s correctness, weapplied only semantic-preserving transfor-mations or equivalent transformations atevery step. The unfolding transformation, awidely used program transformation in con-ventional logic programming, is a kind of ET.We can also devise other kinds of ET, espe-cially to improve computation efficiency. ETthus provides a more flexible, efficient com-putational framework.

The second solution is to provide variousflexible presentation forms. To employ avail-able XML applications and products—XMLeditors, parsers, and rendering tools—we


sc

rdfs:Resource

<rdf:Description ID="Person"> <rdf:type resource="rdfs:Class"/> <rdfs:subClassOf rdf:resource="rdfs:Resource"/> </rdf:Description>

<rdf:Description ID="E-Document"> <rdf:type resource="rdfs:Class"/> <rdfs:subClassOf rdf:resource="rdfs:Resource"/> </rdf:Description>

<rdf:Description ID="E-Article"> <rdf:type resource="rdfs:Class"/> <rdfs:subClassOf rdf:resource="#E-Document"/> </rdf:Description>

<rdf:Description ID="Web-Page"> <rdf:type resource="rdfs:Class"/> <rdfs:subClassOf rdf:resource="#E-Document"/> </rdf:Description>

<rdf:Description ID="Publication"> <rdf:type resource="rdf:Property"/> </rdf:Description> <rdf:Description ID="Name"> <rdf:type resource="rdf:Property"/> </rdf:Description> <rdf:Description ID="Title"> <rdf:type resource="rdf:Property"/> </rdf:Description> <rdf:Description ID="Creator"> <rdf:type resource="rdf:Property"/> </rdf:Description> <rdf:Description ID="Author"> <rdf:type resource="rdf:Property"/> <rdfs:subPropertyOf rdf:resource="#Creator"/> </rdf:Description> <rdf:Description ID="Illustrator"> <rdf:type resource="rdf:Property"/> <rdfs:subPropertyOf rdf:resource="#Creator"/> </rdf:Description>

rdfs:Class

E-Article

Person

Web-Page

scsc

sc

t

ttt

E-Document

(a)

tt

Publication

rdf:Property

Author

Name

Title

Illustrator

spsp

t

ttt

Creator

(b)

(c)

===

tscsp

rdf:typerdfs:subClassOfrdfs:subPropertyOf

Figure 7. An RDF Schema graph and its corresponding RDF-Schema document: (a) a class model; (b) a property model; (c) an RDF Schema document denoted by XDD description P2.

should translate XDD descriptions into con-ventional XML documents. For example, byintroducing XML syntax for the encoding of

XML non-unit clauses, the clause C1 in Fig-ure 6b becomes the XML example shown inFigure 10. So, we can represent any given

XDD description equivalently as a corre-sponding XML document, which can thenbe transmitted, exchanged, and handled sim-



C6: <rdf:Description about=$S:resourceR> <rdf:type resource=$S:classB/> $E:R_properties </rdf:Description> <rdf:Description ID=$S:classA> <rdf:type resource="rdfs:Class"/> <rdfs:subClassOf rdf:resource=$S:classB/> $E:A_properties </rdf:Description>,

<rdf:Description about=$S:resourceR> <rdf:type resource=$S:classA/> $E:R_properties </rdf:Description>. C7: <rdf:Description ID=$S:classA> <rdf:type resource="rdfs:Class"/> <rdfs:subClassOf rdf:resource=$S:classC/> $E:A_properties </rdf:Description> <rdf:Description ID=$S:classA> <rdf:type resource="rdfs:Class"/> <rdfs:subClassOf rdf:resource=$S:classB/> $E:A_properties

</rdf:Description>, <rdf:Description ID=$S:classB> <rdf:type resource="rdfs:Class"/> <rdfs:subClassOf rdf:resource=$S:classC/> $E:B_properties

</rdf:Description>.

C8: <rdf:Description about=$S:resourceR> <$S:propertyP2 resource=$S:valueV/> $E:R_properties </rdf:Description> <rdf:Description ID=$S:propertyP1> <rdf:type resource="rdf:Property"/> <rdfs:subPropertyOf rdf:resource=$S:propertyP2/> $E:P1_properties

</rdf:Description>,<rdf:Description about=$S:resourceR>

<$S:propertyP1 resource=$S:valueV/> $E:R_properties

</rdf:Description>.

C9: <rdf:Description ID=$S:propertyP1> <rdf:type resource="rdf:Property"/> <rdfs:subPropertyOf rdf:resource=$S:propertyP3/> $E:P1_properties </rdf:Description> <rdf:Description ID=$S:propertyP1> <rdf:type resource="rdf:Property"/> <rdfs:subPropertyOf rdf:resource=$S:propertyP2/> $E:P1_properties

</rdf:Description>,<rdf:Description ID=$S:propertyP2>

<rdf:type resource="rdf:Property"/> <rdfs:subPropertyOf rdf:resource=$S:propertyP3/> $E:P2_properties

</rdf:Description>.

% If a class A is a subclass of% another class B, every% resource that is an instance of% class A will also be an instance% of B.

% SubClassOf transitivity property:% If A is a subclass of B and B is a % subclass of some broader class C, % this implies that A is also % a subclass of C.

% If a property P1 is a subproperty % of another more general property % P2, and if a resource R has a P1 % property with a value V, it is % implied that such a resource R has % also a property P2 with the value V.

% SubPropertyOf transitivity property:% If P1 is a subproperty of P2 and % P2 is a subproperty of P3, % P1 is also a subproperty of P3.

Figure 8. The XDD description P3 = {C6, …,C9} models subClassOf and subPropertyOf.

ilarly to ordinary XML documents. More-over, we can use XSL (eXtensible StylesheetLanguage) or XSLT (XSL Transformations)to flexibly display an XDD description invarious forms. We can also define an RDFSchema for encoding XDD descriptions.

Instead of developing a single languagethat satisfies the three important but con-flicting aspects of a language for the Seman-tic Web, we propose a new approach thatseparately manipulates each aspect byemploying XDD, ET, and XSLT (see Figure11). (See the “Related Works” sidebar.)

On the basis of the XDD and ET par-adigms, we have implemented the

XDD System—a Web-based XML proces-sor available at http://kr.cs.ait.ac.th/xdd.Preliminary tests on several XML and RDFapplications, including software configu-ration management, human resource man-agement, and agent-based systems, revealits feasibility and potential in real applica-tions. We are conducting a more thoroughevaluation with a large collection of XMLdocuments.

AcknowledgmentsThe Thailand Research Fund partially funded

this work. We’d like to thank the anonymousreviewers for their insightful comments that helpedimprove this article.

References

1. T. Berners-Lee, Weaving the Web, Harper, SanFrancisco, 1999.

2. S. Staab et al., “An Extensible Approach forModeling Ontologies in RDF(S),” Proc. 1stWorkshop Semantic Web at the 4th EuropeanConf. Digital Library (ECDL 2000), Lisbon,Portugal, Sept. 2000, http://www.ics.forth.gr/proj/isst/SemWeb/proceedings/session2-1/paper.pdf (current 2 July 2001).

3. F.V. Harmelen, and I. Harrocks, “FAQs onOIL: The Ontology Inference Layer,” IEEEIntelligent Systems, vol. 15, no. 6, Nov./Dec.2000, pp. 69–72.

4. J. Hendler and D.L. McGuinness, “TheDARPA Agent Markup Language,” IEEEIntelligent Systems, vol. 15, no. 6, Jan./Feb.2000, pp. 72–73.

5. I. Horrocks, U. Sattler, and S. Tobies, “Prac-tical Reasoning for Expressive DescriptionLogics,” Proc. 6th Int’l Conf. Logic for Pro-gramming and Automated Reasoning (LPAR99), Lecture Notes in Computer Science, no.1705, Springer Verlag, Heidelberg, 1999, pp.161–180.

6. M.R. Genesereth, “Knowledge InterchangeFormat,” http://logic.stanford.edu/kif (current19 June 2001).

7. C. Anutariya et al., “Towards a Foundationfor XML Document Databases,” Proc. 1stInt’l Conf. Electronic Commerce and WebTechnologies (EC-Web 2000), Lecture Notesin Computer Science, no. 1875, Springer-Verlag, Heidelberg, 2000, pp. 324–333.


<xdd:Clause> <xdd:Head> <xdd:DescriptionCheck> <rdf:Description about=$S:uri>$E:propertyElt </rdf:Description> </xdd:DescriptionCheck> </xdd:Head> <xdd:Body> <xdd:PropertyCheck>$E:propertyElt</xdd:PropertyCheck> </xdd:Body></xdd:Clause>.

Figure 10. The XML representation of the non-unit clause C1 in Figure 6b.

Informationpresentation

(XSLT)

Informationmanipulation

(ET)

Informationrepresentation

XDD

Renderingtools

Computationalmodels

Figure 11. The three aspects of alanguage for the Semantic Web and theirrelationship.

<rdf:Description about="http://smith.com"> <rdf:type resource="http://schema.org/appl#Web-Page"/> <Title>Smith Company’s Page</Title> <Author resource="http://smith.com/john"/> </rdf:Description>

<rdf:Description about="http://xdd.org"> <rdf:type resource="http://schema.org/appl#E-Article"/> <Title>XDD Language</Title> <Author resource="http://smith.com/john"/> <Illustrator resource="http://smith.com/joe"/> </rdf:Description>

<rdf:Description about="http://smith.com/john"> <rdf:type resource="http://schema.org/appl#Person"/> <Name>John Smith</Name></rdf:Description>

<rdf:Description about="http://smith.com/joe"> <rdf:type resource="http://schema.org/appl#Person"/> <Name>Joe Smith</Name></rdf:Description>

Figure 9. XDD description P5 consists of four RDF statements that describe certain Webresources.

8. K. Akama, “Declarative Semantics of LogicPrograms on Parameterized RepresentationSystems,” Advances in Software Science andTechnology, Iwanami Shoten, Publishers andAcademic Press, Tokyo, vol. 5, 1993, pp.45–63.

9. S. Kitcharoensakkul and V. Wuwongse,“Unified Versioning Using ResourceDescription Framework,” to be published inAnnals of Software Eng., Kluwer AcademicPublishers, Dordrecht, The Netherlands, vol.11, 2001.

10. M. Uschold and M. Grüninger, “Ontologies:Principles, Methods and Applications,”Knowledge Eng. Rev., vol. 11, no. 2, June1996, pp. 93–136.

11. C. Anutariya et al., “Towards Computation


XML Declarative Description is not a logic-programming lan-guage, although its clauses and descriptions are similar to Datalog clauses and Datalog programs, respectively. UnlikeDatalog,1 XDD has been formally defined without such compli-cated concepts as interpretation and model (see the “Specializa-tion Systems” and “The Semantics of Declarative Description”sidebars). Moreover, it has a higher-order syntax because itallows complex, nesting structured objects. For example, XML(eXtensible Markup Language) expressions can be directly rep-resented and manipulated without decomposition or transla-tion into sets of flat structured objects.

The various XML-based rule markup languages in theRuleML Initiative2 are a class of representation schemes thatwe can use for the Semantic Web. Examples include BRML3

(Business Rules Markup Language) and RFML4 (Relational-Functional Markup Language), which merely encode CLP(Courteous Logic Programs) and Relfun-style declarative pro-gramming and knowledge representation in XML syntax,respectively. The defined language’s semantics relies on translation into corresponding sets of rules in its originalframework. Although such languages can be employed formodeling the Semantic Web, providing syntactic and semanticinterchangeability and interoperability among Web applica-tions appears to be unnatural and difficult.

OIL (Ontology Inference Layer) is an ontology-based lan-guage that extends RDF (Resource Description Framework)Schema with more expressive modeling primitives, includingontology metadata, class, and slot (binary relation) definitions.The DAML+OIL (DARPA Agent Markup Language + OIL) ontol-ogy markup language has been defined on the basis of RDF,RDF Schema, and OIL. OIL and DAML+OIL share importantcharacteristics and are equivalent in their expressive powers.However, their current versions do not support a mechanismfor a description of arbitrary rules and axioms.5 Such a mecha-nism is an essential feature in many application domains,because it enables definition of additional relationshipsamong classes and relations other than the generalization–specialization relationship.

Such a limitation would seem to demand major exten-sions and refinement of the languages. Alternatively, wecan employ XDD to serve as their foundation, which not only helps enhance their expressiveness but also lets theirintended meanings be determined directly. Their modelingprimitives—for example, subClassOf, subPropertyOf,inverseOf, and TransitiveProperty—can be modeledby appropriate XML non-unit clauses. Their schemas andinstances, which are encoded in RDF or XML serialization,immediately become XML unit clauses. In addition, definitionsof arbitrary rules and axioms become possible. For example,the assertion “If a person P has authored e-document D inlanguage L, that person must be able to speak that language,”which is inexpressible in OIL or DAML+OIL, can be representedin XDD as a clause (see Figure A).

With the support of XDD—a well-established, generic toolfor the Semantic Web—those languages can have their seman-tics formally defined together with full-fledged ontologicalmodeling and reasoning services.

References

1. M. Liu, “Deductive Database Languages: Problems and Solutions,”ACM Computing Surveys, vol. 31, no. 1, Mar. 1999, pp. 27–62.

2. H. Boley, “Rule Markup Language (RuleML),” 2001, www.dfki.uni-kl.de/ruleml (current 19 June 2001).

3. B.N. Grosof, Y. Labrou, and H.Y. Chan, “A Declarative Approach toBusiness Rules in Contracts: Courteous Logic Programs in XML,”Proc. 1st ACM Conf. Electronic Commerce (EC 99), ACM Press, NewYork, 1999, pp. 68–77.

4. H. Boley, “RFML: Relational-Functional Markup Language,” 2000,www.relfun.org/rfml (current 19 June 2001).

5. S. Bechhofer et al., “An Informal Description of Standard OIL andInstance OIL,” Nov. 2000, www.ontoknowledge.org/oil/downl/oil-whitepaper.pdf (current 19 June 2001).

<rdf:Description about=$S:personP> <rdf:type resource="#Person"/> <CanSpeak>$S:languageL</CanSpeak> </rdf:Description> <rdf:Description about=$S:documentD> <rdf:type resource="#E-Document"/> <Author resource=$S:personP/> <Language>$S:languageL</Language> $E:D_properties

</rdf:Description>.

% If a person P has authored an % e-document D in a language L, % we can infer that such a person % P must be able to speak that % language.

Figure A. An example of modeling an axiom.


Related Work

with RDF Elements,” Proc. Int‘lSymp. Digital Library 1999 (ISDL99), Univ. of Library and InformationScience, Tsukuba, Japan, 1999, pp.112–119.

12. T. Berners-Lee, “The Semantic Webas a Language of Logic,” Sept. 2000,www.w3.org/DesignIssues/Logic.html (current 19 June 2001).

13. K. Akama, “ET Computational Para-digm,” 2001, http://kr.cs.ait.ac.th/et(current 19 June 2001).

For further information on this or any other computing topic, please visit our Digital Library at http://computer.org/publications/dlib.


T h e A u t h o r sVilas Wuwongse is a professor in the Computer Science and Information Management Program at theAsian Institute of Technology, Thailand. He is also an adviser to the Revenue Department of the Thai Min-istry of Finance and the Bank of Thailand, where he leads teams to develop large-scale Web-based softwaresystems and data warehouses. His research interests include knowledge representation, data modeling, andthe Semantic Web. He serves as editor of the Journal of Natural Language Processing and the Computer Pro-cessing of Oriental Languages. He has a BEng, MEng, and DEng from the Tokyo Institute of Technology.He is a member of the IEEE Computer Society, ACM, Information Processing Society of Japan, and Japan-ese Society for Artificial Intelligence. Contact him at AIT/CSIM, PO Box 4, Klong Luang, Pathumthani12120, Thailand; [email protected].

Chutiporn Anutariya is a doctoral student in the Computer Science and Information Management Programat the Asian Institute of Technology. Her research interests include the Semantic Web, knowledge repre-sentation, and software engineering. She is a member of the Knowledge Representation Laboratory. She hasa BS in statistics from Chulalongkorn University, Thailand, and an MS in computer science from the AsianInstitute of Technology, Thailand. Contact her at AIT/CSIM, PO Box 4, Klong Luang, Pathumthani 12120,Thailand; [email protected].

Kiyoshi Akama is a professor at the Center for Information and Multimedia Studies, Hokkaido University.His research interests include artificial intelligence, knowledge processing, and programming languages.He has a BEng, MEng, and DEng from the Tokyo Institute of Technology. He is a member of the JapaneseSociety for Artificial Intelligence, Japan Society for Software Science and Technology, and the InformationProcessing Society of Japan. Contact him at the Center for Information and Multimedia Studies, HokkaidoUniv., Kita 11, Nishi 5, Kita-ku, Sapporo, 060-0811, Japan; [email protected].

Ekawit Nantajeewarawat is an assistant professor of information technology at Sirindhorn InternationalInstitute of Technology, Thailand. His research interests include deductive object-oriented systems, knowl-edge representation, and object-oriented software engineering. He has a BEng in computer engineering fromChulalongkorn University, Thailand, and an MEng and DEng in computer science from the Asian Instituteof Technology, Thailand. Contact him at Information Technology Program, Sirindhorn Int’l Inst. of Tech-nology, PO Box 22,Thammasat Rangsit Post Office, Pathumthani 12121,Thailand; [email protected].

c o m p u t e r . o r g / d s o n l i n e

Distributed Systems Online

cluster computing

distributed agents

distributed databases

distributed multimediagrid computing

middlewaremobile & wireless systems

operating systemsreal-time systems

security IEEE

dependable systemscollaborative computing

XML Declarative Description: a language for the Semantic Web

Documents