The Ontology Web Language (OWL) for a MultiAgent Understating System Mostafa M. Aref Zhengbo Zhou Department of Computer Science and Engineering University of Bridgeport, Bridgeport, CT 06601 email: aref@bridgeport.edu Abstract— Computer understanding is a challenge problem in Artificial Intelligence. A multi-agent system has been developed to tackle this problem. Among its modules is its knowledge base (vocabulary agents). This paper discusses the use of the Ontology Web Language (OWL) to represent the knowledge base. An example of applying OWL in sentence understanding is given. Followed by an evaluation of OWL. 1. INTRODUCTION One of various definitions for Artificial Intelligence is “The study of how to make computers do things which, at the moment, people do better”[7]. From the definition of AI mentioned above, “Understanding” can be looked as the first step for a system to realize the ability of doing things as well as humans. Natural language processing needs an understanding system to make the machine understand human languages. Understanding is a transformation from one representation to another [1]. To achieve this transformation, the input will be processed through a series of agents. From morphological analysis to pragmatic analysis, the machine can “read” the input and has its own representation. Several applications may be developed based on the understanding system. Some examples of these applications are Machine learning, machine translating, and expert systems with better performance. A multi-agents understanding system accepts a user input in a form of speech (typed or voice). Then, the user may enter several questions concerning the user input. The system should answer these questions that reflects the understanding of the input [1]. The multi-agents understanding system consists of the following agents: a morphological analyzer, a semantic analyzer, a discourse analyzer, a user interface, and a knowledge base. The knowledge base is the main module in the understanding system. It contains the English vocabulary agents and all the linguistic information about the vocabulary using objectoriented technology [1]. OWL is a Web Ontology Language. It is built on top of RDF – Resource Definition Framework and written in XML. It is a part of Semantic Web Vision, and is designed to be interpreted by computers, not for being read by people. OWL became a W3C (World Wide Web Consortium) Recommendation in February 2004 [2]. The OWL is a language for defining and instantiating Web ontologies. OWL ontology may include the descriptions of classes, properties, and their instances [3]. Given such ontology, the OWL formal semantics specifies how to derive its logical consequences, i.e. facts not literally present in the ontology, but entailed by the semantics. Section 2 describes a multi-agents understanding system. Section 3 gives a brief description of a newly standardized technique, Web Ontology Language—OWL. A working example of the OWL applied in knowledge representation is given in section 4. Section 5 evaluates the performance of OWL. Conclusion and directions of the current research are presented in section 6. 2. MULTI-AGENT UNDERSTANDING SYSTEM To understand something is to transform it from input representation into internal representation has been chosen to correspond to a set of available actions that could be performed [1]. The process of natural language understanding is as follows [7], as shown in Figure 1. Morphological Agent: given the input text, morphological analyzer converts the text into group of words in the basic form and their linguistic information. It also separates the affixes from the input tokens [1]. Semantic Agent: structures are created to represent meanings of a group of words (sentence). In other words, a mapping is made between the input sentence and objects in the task domain. Discourse Agent: Given the agent sub-societies of set of sentences, discourse analyzer agent resolves references between these sentences. The user interface is needed to facilitate the communication between the understanding system and the user [1]. For example, a web page containing several text input boxes can get input from a human and then gives another page or dialog box with the answer or some other actions. Morphological Analyzer Agent Knowledge Base Vocabulary Agents Semantic Analyzer Agent Discourse Analyzer Agent Speech-to-text Agent Agent Sub-Societies of the User Input Query Analyzer Agent Text-to-Speech Agent I n t e r n e t User Interface Figure 1 - Multi-agent understanding system 2.2 Knowledge Base A knowledge base is a collection of knowledge expressed using some formal knowledge representation language [7]. In the understanding system, the knowledge base is the main module. It contains the English vocabulary agents and all linguistic information about this vocabulary. Good Knowledge representation is the basis of a good knowledge base. To evaluate one knowledge representation, there are the following four criteria [7]: Representational Adequacy: the ability to represent all kinds of knowledge that are needed in that domain. Inferential Adequacy: the ability to manipulate the representational structures to derive new structures corresponding to new knowledge inferred from old. Inferential Efficiency: the ability to incorporate into the knowledge structure additional information that can be used to focus the attention of the inference mechanisms in the most promising directions. Acquisitional Efficiency: the ability to acquire new information easily. The simplest case involves direct insertion, by a person, of new knowledge into the database. Ideally, the program itself would be able to control knowledge acquisition. The objective of knowledge representation is to organize the information necessary to the application such that it is easily accessed and manipulated. The knowledge content must be sufficient to solve problems in the domain and it must be efficient [1]. There are several knowledge representations such as: predicate logic, procedural, semantic nets, conceptual dependency and object-oriented representation [1] and [7]. Object-oriented knowledge representation organizes knowledge into classes of objects, subclasses and superclasses. That is an important issue in knowledge representation. By this organization, a class may inherit the properties of any of its superclasses and it may pass properties to any one of its subclasses [1]. However, only traditional object-oriented technique is not enough for a good knowledge representation or knowledge base. 3. OWL – ONTOLOGY WEB LANGUAGE Ontology is about the exact description of things and their relationships. It is an old study of philosophy from ancient Greece. As the study of artificial intelligence growing, the concept of ontology have been using more and more in the formalization of knowledge in terms of classes, properties, instances and the relations. So, for knowledge, ontology is about the exact description of the representation of the knowledge itself, as well as the relationships among different categories of the knowledge. Moreover, for the web, ontology is about the exact description of web information and relationships between web information [2]. The OWL Web Ontology Language is designed for use by applications that need to process the content of information instead of just presenting information to humans. Because it provides additional vocabulary along with a formal semantics, OWL facilitates greater machine interpretability of Web content than XML, RDF, and RDF Schema (RDFS). 3.1 OWL Development One of the effective approaches to solve the data exchange among different computers via the computer networks is using XML – eXtensible Markup Language. HTML is to solve how the data appears in front of people. The documents written in XML can describe what data is and be shared and exchanged among different systems [2]. XML provides a syntax for structured documents, however it imposes no semantic constraints on the meaning of the document. Following this, the W3 Consortium introduces the idea of ontology when creating RDF – the Resource Definition Framework. RDF is a data model for objects and relations between them, providing a simple semantic for this data model. RDF uses XML syntax to describe objects and relations in the data model [5]. After that, RDFS is developed to describe properties and classes of RDF resources, with a semantic for creating hierarchies of such objects and classes and thus providing the means for generalization. RDFS is considered to be an ontology language, containing classes and properties and being aware of concepts of range and domain, as well as having the ability describe subclasses and superclasses. However, for implementing the Semantic Web RDFS is not quite optimal as it lacks the features necessary to describe resources in sufficient detail. As Santtu Toivonen concludes in his research [9], RDFS is suitable for providing the means for an ontology that characterizes some environment, no matter how abstract [5]. RDFS alone, however, suffers from its dependence on domain-specific and case-specific details. RDFS suffers from an expressive inadequacy and it lacks a number of important relations between classes such as equivalence and disjointedness, as well as cardinality and characteristics of properties. To solve the RDFS problems, two languages were developed almost concurrently. OIL (Ontology Inference Layer) in Europe and DAML (DARPA Agent Markup Language) in U.S.. Both of them are based on top of RDFS. After submitted the combination of the two—DAML+OIL to W3 Consortium for standardization, OWL –the Ontology Web Language is born as a new W3C standard language. OWL is layered on top of RDFS, using its syntax for Define vocabularies expressing ontological primitives such as Class, Relation, Subclass etc. In addition OWL adds a much richer set of its own primitives, such as transitivity, cardinality, disjunction etc. Also, it adds characteristics of properties like symmetry, richer typing of properties (e.g. nonNegativeInteger), and enumerated classes [5]. As a result, OWL has more facilities for expressing meaning and semantics than XML and RDF (S). Thus OWL goes beyond these languages in its ability to represent machine interpretable content on the Web. Figure 2 shows the development from XML to OWL. The main purposes of OWL can be concluded as below: 1- Formalize a domain by defining classes and properties of those classes, 2- Define individuals and assert properties about them, and 3- Reason about these classes and individuals to the degree permitted by the formal semantics of the OWL language. 3.2 The Species of OWL All of these have led to a set of requirements that may seem incompatible: efficient reasoning support and convenience of expression for a language as powerful as a combination of RDF Schema with a full logic. Indeed, these requirements have prompted W3C's Web Ontology Working Group to define OWL as three different sublanguages [4], each of which is geared towards fulfilling different aspects of this incompatible full set of requirements: OWL Full—The entire language is called OWL Full, and uses all the OWL languages primitives. It also allows combining these primitives in arbitrary ways with RDF and RDF Schema. This includes the possibility (also present in RDF) to change the meaning of the pre-defined (RDF or OWL) primitives, by applying the language primitives to each other. For example, in OWL Full people could impose a cardinality constraint on the class of all classes, essentially limiting the number of classes that can be described in any ontology. The advantage of OWL Full is that it is fully upward compatible with RDF, both syntactically and semantically: any legal RDF document is also a legal OWL OIL Combine Vocabularies Revision XML RDF RDFS Extend Vocabularies DAMLONT Figure 2 - the Development of OWL. DAML (DAML+OIL) OWL Full document, and any valid RDF/RDF Schema conclusion is also a valid OWL Full conclusion. The disadvantage of OWL Full is the language has become so powerful as to be undecidable, dashing any hope of complete (let alone efficient) reasoning support [3]. OWL DL (Description Logic)—In order to regain computational efficiency, OWL DL is a sublanguage of OWL Full which restricts the way in which the constructors from OWL and RDF can be used. Roughly this amounts to disallowing application of OWL's constructor's to each other, and thus ensuring that the language corresponds to wellstudied description logic. The advantage of this is that it permits efficient reasoning support. The disadvantage is the lose of full compatibility with RDF. An RDF document will in general have to be extended in some ways and restricted in others before it is a legal OWL DL document. Conversely, every legal OWL DL document is still a legal RDF document [3]. OWL Lite—An ever further restriction limits OWL DL to a subset of the language constructors. For example, OWL Lite excludes enumerated classes, disjointness statements and arbitrary cardinality (among others). The advantage of this is a language that is both easier to grasp (for users) and easier to implement (for tool builders). The disadvantage is of course a restricted expressivity [3]. Ontology developers adopting OWL should consider which sublanguage best suits their needs. The choice between OWL Lite and OWL DL depends on the extent to which users require the more-expressive constructs provided by OWL DL and OWL Full. The choice between OWL DL and OWL Full mainly depends on the extent to which users require the meta-modeling facilities of RDF Schema (e.g. defining classes of classes, or attaching properties to classes). When using OWL Full as compared to OWL DL, reasoning support is less predictable since complete OWL Full implementations will be impossible. There are strict notions of upward compatibility between these three sublanguages. Every legal OWL Lite ontology is a legal OWL DL ontology. Every legal OWL DL ontology is a legal OWL Full ontology. Every valid OWL Lite conclusion is a valid OWL DL conclusion. Every valid OWL DL conclusion is a valid OWL Full conclusion [3]. 3.3 Structure and Basic Element of OWL OWL as a Web Ontology Language can define classes using XML syntax. Figure 3 shows a simple name classes. Actually, people know almost nothing about these classes other than their existence, despite the use of familiar English terms as labels. Within this document, the Noun class can now be referred to using #Noun, e.g. rdf:resource="#Noun". Another form of reference uses the syntax rdf:about="#Noun" to extend the definition of a resource. It permits the extension of the imported definition from sources in other OWL construct without modifying the original document and supports the incremental construction of a larger ontology [3]. <owl:Class rdf:ID="Verb"/> </owl:Class> <owl:Class rdf:ID="Noun"/> </owl:Class> Figure 3 - Simple Named Classes The fundamental taxonomic constructor for classes is rdfs:subClassOf. It relates a more specific class to a more general class. If X is a subclass of Y, then every instance of X is also an instance of Y. The rdfs:subClassOf relation is transitive. If X is a subclass of Y and Y a subclass of Z then X is a subclass of Z. Apparently, with the definition of a class and a subclass, we can apply “is a” relationship very easily by OWL. Individuals— In addition to classes, it is possible to be able to describe their members. We normally think of these as individuals in our universe of things. An individual is minimally introduced by declaring it to be a member of a class e.g. <Verb rdf:ID="buy1" />. “rdf:type” is an RDF property that ties an individual to a class of which it is a member. There are a couple of points to be made here. First, it has already decided that “buy” (a specific verb) is member of Verb, the class containing all kinds of verbs. Second, there is no requirement in the twopart example that the two elements need to be adjacent to one another, or even in the same file (though the names would need to be extended with a URI in such a case). People design Web ontologies to be distributed. They can be imported and augmented, creating derived ontologies. Figure 4 shows an example of a subclass and an individual. <owl:Class rdf:ID="CountableNoun"> <rdfs:subClassOf rdf:resource="#Noun" /> </owl:Class> <owl:Verb rdf:ID="buy1" /> <rdf:type rdf:resource="#Verb"/> </owl:Thing> Figure 4 - A Subclass and An Individual Simple Properties—This world of classes and individuals would be pretty uninteresting if there is the only definition of taxonomies. Properties let us assert general facts about the members of classes and specific facts about individuals [3]. A property is a binary relation. Two types of properties are distinguished [3]. Datatype properties: relations between instances of classes and RDF literals and XML Schema datatypes. Object properties: relations between instances of two classes. Figure 5 shows an object property example. From the given above, it is not only to know that “raise” is transitive, but also it is able to infer from the domain that “raise” is a verb. Data types [3]— OWL uses most of the built-in XML Schema datatypes. References to these datatypes are by means of the URI reference for the datatypes [2]. Now OWL is still under testing and study, the effectiveness of its expression and communication among machines attracts the experts in related area to use this language and do research with it. Next section is going to present current implementations. <owl:ObjectProperty rdf:ID="isTransitive"> <rdfs:domain rdf:resource="#Verb"/> </owl:ObjectProperty> <owl:Class rdf:ID=”Word”/> </owl:Class> <owl:Class rdf:ID="Verb"> <rdfs:subClassOf rdf:resource="#Word" /> </owl:Class> <owl:Word rdf:ID="Raise"> <isTransitive/> </owl:Word> Figure 5 - Object Property Example 3.4 Linguistic OWL To use OWL in the understanding system knowledge base, it is by first to gather linguistic information. OWL version of Wordnet is a project that translate the Wordnet database into OWL. There is an ongoing project called “Wordnet in RDFS and OWL” [10]. It is one of the Semantic Web Best Practices, and developed by Wordnet Task Force. One of its approaches is integrating existing datamodels in order to provide a unified OWL vocabulary for RDF versions of Wordnet [11]. There is a WordNet.OWL which is an OWL-ontology based on WordNet 1.7.1 [12]. Also, a little bit earlier, there is another project to develop ontologies known as SUMO—Suggested Upper Merged Ontology, and now there is also this version (partial) of Wordnet database [13]. This ontology is being created as part of the IEEE Standard Upper Ontology Working Group. The goal of this Working Group is to develop a standard upper ontology that will promote data interoperability, information search and retrieval, automated inferring, and natural language processing [13]. 4. OWL IN KNOWLEDGE BASE To illustrate the working of OWL in knowledge representation, an example of a text input is given. Chris bought an old car with $5000 last Friday. • • • • • The verb: "buy" Verb’s agent: "Chris" Result of the action: “car” Time of the action: “last Friday” Money in the transaction: “$5000” According to the requirement of knowledge representation of natural language, there is the corresponding OWL code for the content in knowledge base: In the figure 6, “agent”, “transactionAmount”, “time” and “result” are attributes of object “buy1”. <Buying rdf:ID="buy1"> <agent rdf:resource="#Agent" /> <agent rdf:ID="Chris" /> <transactionAmount rdf:resource="#CurrencyMeasure" /> <transactionAmount rdf:ID="$5000" /> <time rdf:resource=”#TimePosition”/> <time rdf:ID=”Friday”/> <result rdf:resource=”#Entity”/> <result rdf:ID=”car1”/> </Buying> <Human rdf:ID="Chris"> <immediateInstance rdf:resource=”#Human”/> </Human> <Vehicle rdf:ID="car1"> <monetaryValue rdf:resource=”#CurrencyMeasure”/> <monetaryValue rdf:ID=”$5000”/> <property rdf:resource=”#Attributes”/> </Vehicle> <UnitedStatesDollar rdf:ID="$5000"> <lessThanOrEqualTo rdf:datatype="&xsd;string”>5000 < /lessThanOrEqualTo> <greaterThanOrEqualTo rdf:datatype=”&xsd;string”>5000 </greaterThanOrEqualTo> </UnitedStatesDollar> <Friday rdf:ID="Friday1"> <PastFn rdf:resource=”#TimePosition” /> </Friday> Figure 6 - An Example in OWL 5. EVALUATION OF OWL The four criteria to evaluate a proper knowledge representation are representational adequacy, inferential adequacy, inferential efficiency and acquisitional efficiency. However, unfortunately, up to now no single system that optimizes all of the capabilities for all kinds of knowledge has yet been found. Following these four principles, we will see how OWL can work for the knowledge representation in a natural language processing system. Since OWL is using XML, it has a strong ability that can be shared and exchanged between different types of computers using different types of operating system and application languages. Representational adequacy—By building up from XML, OWL inherits its main function of describing data. By using ontology in XML syntax and necessary RDF Schema, OWL describes the domain knowledge in ontology primitives (objects, classes, properties etc). Basically, by using this kind of representation, inheritance can be performed efficiently using ontologies. As from examples in section 3.3, the property “subclassof” can explicitly describe the relation. Furthermore, the whole fact set of OWL for classes and objects include the description of an ontology, and the axioms describes the manipulation on each ontology. As a result, it is not only able to describe things in detail, but also able to provide a large base of relationships and the probability of doing reasoning among different ontologies. Inferential adequacy—As mentioned above, axiom can play a role of knowledge operator. Axioms are used to associate class and property identifiers with either partial or complete specifications of their characteristics, and to give other information about classes and properties. Axioms used to be called definitions, but they are not all definitions in the common sense of the term and thus a more neutral name has been chosen [5]. Besides axioms, OWL also has other kind of manipulation on ontologies. Acquisitional efficiency—Compared to some of the knowledge representation method mentioned in [7], ontologies developed by OWL have their own advantages in this criterion. But, as deputed by a lot of researchers, this kind of knowledge representation still lacks of acquisitional efficiency. Most of the ontologies are created by people, not by machine. Disadvantages—So far, it may not be possible to describe OWL semantics with logic programs or rule base directly. There should be some tools, including reasoners, validators or so to do some ancillary work during using OWL. Another inconvenience is, as a new w3c standard, it is not prevailed in a very wide range, and a lot of research work and testing work are undergoing to exploit its usage. Only few organizations, including university laboratories and academic organizations, are studying or implementing OWL. Similar to other knowledge representation technique, it is not easy for OWL to have good performance on knowledge acquisition. However, OWL is still being studied and developed, so, probably it is possible to find out the breakthrough of this principle in OWL in the future. 6. CONCLUSION AND FUTURE TASK Natural language understanding may be considered as being a mapping of a natural language text to an internal representation that capture the meaning of that text. This paper presents an ongoing research project about a multiagents understanding system and the construction of its knowledge base. The knowledge base is the foundation of the work and communication between each of the modules in the system. The new technique of OWL—Web Ontology Language, provides a new tool to implement knowledge representation. Some of its feature fits the requirement of knowledge representation and are proved to be efficient by some tests and implementations. However, as a new standard and language, there is the opportunity to develop its potential ability in knowledge representation and natural language processing. After explored the OWL and evaluated its ability in knowledge representation, the current research focuses on its implementation in this area and discuss its ability in machine learning based on the NLP with OWL. To do this research, WordNet provides a certain kind of platform and foundation. A task force of semantic web is doing the work to convert the WordNet database into OWL. REFERENCE [1] Mostafa Aref, “A Multi-Agent System for Natural Language Understanding”, Proceedings of the International Conference on Integration of Knowledge Intensive Multi-Agent Systems, 1-3, Cambridge MA, PP 36-40, 2003. [2] W3C School, “Tutorial of RDF OWL”, www.w3schools.com, 2004. [3] Michael K. Smith, Chris Welty, Deborah L. McGuinness, “OWL Web Ontology Language Guide”, W3C Recommendation, www.w3.org/TR/owl-guide, 10 February 2004. [4] Deborah L. McGuinness, Frank van Harmelen, “OWL Web Ontology Language Overview”, W3C Recommendation, www.w3.org/TR/2003/PR-owlfeatures-20031215/#ref-rdf-schema, 10 February 2004. [5] Kostyantyn Vovk, Ontology and Object-Oriented Representation, Spring 2004. [6] Sandro Hawke, “OWL Implementation”, http://www.w3.org/2001/sw/WebOnt/impls, 2003. [7] Elaine Rich, Kevin Knight, Artificial Intelligence, Second Edition 1991 McGraw-Hill Inc. [8] Stuart Russell, Peter Norvig Artificial Intelligence: A Modern Approach, 1995 Prentice Hall ; Pearson Education. [9] S. Toivonen. “Using RDF(S) to Provide Multiple Views into a Single Ontology”. SemWeb 2001 Workshop, Hong Kong, 2001. [10] “Wordnet in RDFS and OWL”, http://www.w3.org/2001/sw/BestPractices/WNET/word net-sw-20040713.html W3C 2004. [11] Coordinator Aldo Gangemi, “Semantic Web Best Practices”, WordNet Task Force http://www.w3.org/2001/sw/BestPractices/WNET/tf, Wordnet Task Force, 2004. [12] KID Processing Group, “knOWLer: Ontology-based Information Management System”, http://taurus.unine.ch/knowler/ 2004. [13] Deborah Nichols and Allan Terry, “SUMO Ontology”, http://ontology.teknowledge.com/ Teknowledge Corp.2004.