The Semantic Web envisions an infrastructure in which humans and machines seamlessly

Transcription

1 A r g u m e n t a t i o n T e c h n o l o g y Argumentation-Based Ontology Engineering Christoph Tempich and Rudi Studer, University of Karlsruhe Elena Simperl and Markus Luczak, Free University of Berlin H. Sofia Pinto, Technical University of Lisbon The Semantic Web envisions an infrastructure in which humans and machines seamlessly exchange information on the Web. For this to succeed, we need shared ontologies that enable information exchange between different parties. Engineering a shared The DILIGENT argumentation framework helps capture design deliberations in ontology-engineering discussions. It makes consensus-building tasks more efficient and provides detailed guidance for nonexperts. ontology in this kind of scenario is a social, evolving process, involving a geographically dispersed community with different knowledge and expertise ontology engineers, domain experts, and ontology users. Depending on the methodology used, the ontology s life cycle might consist of a series of stages, in which the engineering team decides how to model a domain to best suit the ontology users needs and the application requirements. 1 Such engineering methodologies typically cover the major aspects of general ontology engineering, describing related support methods, decisions to consider, and expected results. However, these methodologies only marginally address the collaborative interaction among community members who are creating an ontology. For example, consider this two-phase methodology: first, a selected team of ontology developers creates an initial ontology; second, the development team continuously extends and refines the ontology on the basis of feedback from a panel of domain experts. 2 The interaction between the participants occurs indirectly, with the core development team acting as a mediator between the different parties suggesting changes to the ontological content. Most ontology-engineering environments also employ this indirect style. In contrast, our argumentation framework provides detailed ontology-engineering guidance explicitly supporting direct interaction between ontology engineers and domain experts. This occurs through arguments, which fosters consensus building and the creation of a truly shared ontology. The DILIGENT (distributed, loosely controlled, and evolving engineering of ontologies) argumentation framework consists of an argumentation process and a formal model, and is supported technologically by a wiki-based support tool, coefficientmakna (collaborative, ontology-engineering-efficient Makna). Here, we sketch out DILIGENT s theoretical background and an application scenario (typical to distributed ontology engineering) we envisioned when specifying the framework s functional requirements. We also describe coefficientmakna s design, architecture, and implementation. We employed several case studies to provide empirical data on the framework. We paid particular attention to lessons learned from a study in which a team of eight individuals with minimal experience in ontology engineering successfully developed a cooking ontology using coefficientmakna. Application scenario and use cases We aim to use the DILIGENT argumentation framework in collaborative ontology engineering. The tool primarily supports conceptualizing and formalizing shared ontologies, complementing native ontology editors that focus on implementation tasks. (We point readers elsewhere to learn more about the distinction between conceptualization, formalization, and implementation in ontology engineering. 1 ) In a distributed-ontology-engineering scenario, a geographically dispersed community incrementally /07/$ IEEE IEEE INTELLIGENT SYSTEMS Published by the IEEE Computer Society

2 develops an ontology or multiple versions of it reflecting the community members shared view with respect to the modeled domain (see figure 1). The participants, including domain experts and observers, don t necessarily have an ontology-engineering background. Building on changes in the target domain, evolving application requirements, and discussions on whether and how to model specific domain aspects, the engineering team continuously revises and extends the ontology and releases new versions. To effectively participate, new parties joining the community must understand the rationale behind modeling decisions and follow the ontology s release history. Ontology engineering usually starts by analyzing the domain and application requirements. The developers agree on these requirements and their importance, and propose and discuss different ways to build a model that complies with them. They must discuss whether to model specific domain information as classes, instances, attributes, or properties, to decide on an adequate granularity level for the model and on conventions for labeling and documentation. Finally, they implement the conceptualization in a formal knowledge representation language such as the Semantic Web ontology language OWL DL (Web Ontology Language Description Logic). They can revise the ontology according to user needs once new domain knowledge is available. Later, different parties might maintain and use different versions of the same ontology, while a shared version has changes integrated into it. To facilitate the shared ontology s systematic evolution and to operationalize consensus building, the developing team should discuss such issues in a controlled manner and trace the engineering process and results achieved to date. They should ensure a seamless development of the shared ontology, using instruments to resolve conflicts that arise when several parties hold irreconcilable views. We exemplify the scenario and its requirements using a discussion from a case study, in which the participants built a cooking ontology. A short passage from the study illustrates the main issues in collaborative ontology engineering: One of the participants stated the requirement that the developed ontology should include guidelines for a low-fat diet. All members agreed with this requirement. As an idea to resolve this requirement, one participant stated that this would require the inclusion of nutrition values into the ontology. At this point another participant challenged this idea and Ontology engineers #1 Board Customer s Arguments Issues Decisions stated that nutrition values were not available for the ingredients, without providing an alternative to resolve the issue. The participants informally state a common goal for the ontology and introduce ideas for its resolution on a conceptual level. While all agree on the common goal, a participant disagrees with the proposed conceptualization. We developed methods that support capturing such deliberations and help detect conflicts. Requirements From an argumentation point of view, we can categorize ontology-engineering discussions as deliberation dialogues in which the participants collaborate to decide what action or course of action should be adopted in some situation. 3 The IBIS (Issues-Based Information System) methodology provides an argumentation theory for deliberation dialogues. Developers have used IBIS in software and requirements engineering to capture design deliberations. 4,5 Researchers have developed formal models to allow for structured queries on the collected arguments. Such approaches show that formal argumentation models ease design-decision traceability, help in conflict resolution, enhance reusability, and facilitate integrating new participants. Although these are general models, they let us identify several requirements Persistent ontology versions Elaborations Ideas Positions Figure 1. A collaborative ontology-engineering scenario. Optional: acquired users from the target group Optional: ontology engineers #2 for argumentation support and formalization, which are unique to ontology engineering: You can apply general argumentation models across various disciplines, which cover many argument types. However, our studies highly recommend (and even require) limiting the set of arguments to improve such general models usability for ontology engineering. 6 You can t easily detect inconsistencies in informal argumentation models discussions because arguers don t formalize their arguments. This might be a secondary consideration for general topic discussions such as requirements analysis, but it becomes crucial for knowledge formalization tasks. Ontologies are formal models that must be semantically consistent to be of value to the applications using them. Inconsistencies in discussions on how to ontologically represent domain knowledge lead to inconsistent models of lower usability. You can build ontologies in various ways, with manual building and automatic learning as prominent examples. The latter helps produce large amounts of data but is less suited for structuring this data into manageable pieces. Argumentation models can help provide structure, offering an integrated view on manually and automatically NOVEMBER/DECEMBER

3 A r g u m e n t a t i o n T e c h n o l o g y created ontologies (or interrelated ontology fragments). We considered these requirements along with those derived from the application scenario when developing the functional requirements for the DILIGENT argumentation framework. DILIGENT s requirements correspond to its three main aspects: process, argumentation ontology, and tool support. Argumentation process The requirements for this aspect fall into three categories: Support argumentation process. The framework should support the full argumentation cycle. This includes raising issues, mediating conflicts, bargaining, clarification, and agreement. Participants also aggregate lines of reasoning to systematize their argumentation, so they must know which issues are under discussion, postponed, agreed, and discarded at any time. Support conceptualization and formalization. People might agree on the need for a certain conceptual model but not on how to implement it in a particular knowledge representation language or formalism. The framework should support argumentation for both conceptualization and formalization. Provide methodology. The framework should provide process support (for example, a process-driven methodology) for systematic ontology development. It should guide users through the necessary steps and activities, the expected intermediary results, and the optimal way to use the results to accomplish particular goals. Argumentation ontology The requirements for this aspect fall into five categories: Use established terminology. Computer science has a long history of researching argumentation and argumentation visualization. 5 The acceptance of DILIGENT s argumentation ontology and its impact beyond the boundaries of the Semantic Web community depend on the use of established terminology for this research field. Focus on relevant arguments. You should restrict the types of arguments formalized in the argumentation ontology according to empirical findings, so that the ontology-engineering discussions are focused and efficient. Focus on ontology design processes. Building on other studies results, 7 we should extend general-purpose argumentation models such as IBIS with domain-specific knowledge to apply them in specific use cases. The developed argumentation ontology should thus reflect the particularities of ontology design processes. Detecting contradictory information at the discussion level is important because such contradictions lead to ontological inconsistencies. The system should detect contradicting lines of reasoning within argumentations and discussions to speed up the engineering process. Focus ontology domain. Minimal ontological commitment ensures an ontology s extensibility for future uses. 8 An ontology should model only one domain at a time, and ontology engineers should avoid integrating many different issues in a single ontology module. So, the argumentation ontology should focus on the argumentation-related ontological entities. Avoid encoding bias. The argumentation ontology should be independent of the formalism used to model the final ontology. Each formalism allows for different sets of modeling decisions, and all can be subject to discussion. Tool support The requirements for this aspect fall into seven categories: Abstract from implementation language. Ontology-engineering environments should as far as possible abstract from a concrete implementation language such as OWL or RDF Schema and focus on modeling aspects. However, they should permit exporting the model according to different formalisms. Include inferencing. The tool should include inferencing capabilities to ensure the modeled knowledge s correctness. Support collaboration. It should also support distributed access to the ontology and discussions on issues related to ontology development. Support externalization. The tool should support externalization of the rationales underlying the modeling decisions to let new participants trace past discussions and results achieved. Provide ontology access. During the discussion, the ontology might evolve along different tracks. Although participants have agreed on some parts of the ontology, others might still be under discussion. To let participants evaluate and compare the different revision proposals, the tool should be able to visualize concurrent versions of the ontology, identify commonly agreed fragments, and export the conceptualization to a formal knowledge representation language. Detect contradictions and conflicts. The system should detect contradicting lines of reasoning within argumentations and discussions to speed up the engineering process. Later, the participants will need methods and tools to mediate conflicts and reach agreements. Provide discussion status. Requirements and modeling decisions should have a status level to help ontology developers easily assess the overall progress. The tool should group issues and ideas according to their priority and should visualize which issues are agreed on, under discussion, or postponed. DILIGENT argumentation framework The DILIGENT argumentation framework consists of two building blocks: an argumentation process description and an argumentation ontology, which are supported technologically by coefficientmakna. Argumentation process description We divide the argumentation process into five activities: 54 IEEE INTELLIGENT SYSTEMS

4 Choose moderator. The participants in an ontology-engineering discussion choose a moderator. The basic rules for moderation apply: the moderator doesn t contribute to the discussion or decision but structures the discussion and organizes the decision process. Any participant can be the moderator, and the moderator role can move from one participant to the next. All participants should accept the moderator. Choose decision procedure. The participants agree on a mechanism for reaching agreement during the discussions. They decide on a voting procedure, such as majority voting, and on the conditions triggering a new voting round such as voting within fixed time intervals or if no one has brought forward a new argument within a certain time. Specify issues. Issues correspond to the ontology s domain or application requirements. The participants must agree on a starting set of relevant issues. During the discussions, each participant can raise new issues or elaborate on existing ones. Once a discussion evolves, participants can group issues and act on them according to their priority. Provide arguments and ideas. The participants discuss issues, propose ideas, and bring forward arguments in favor of or against an issue or idea. The participants first agree on an issue s relevance and then suggest a way to formalize it. An idea describes in natural language how to formalize an issue. The participants express their agreement or disagreement with arguments and provide alternative ideas to strengthen or weaken them. The entire process s effectiveness and efficiency depends on the decisions made on the basis of the provided arguments. So, we ve analyzed this phase in detail and found that unguided discussions tend to have poor outcomes. General-purpose argumentation frameworks such as IBIS don t restrict the types of arguments that participants can provide. In unrestricted ontology-engineering discussions, participants tend to make many concessions without making much progress in the modeling. So, we analyzed the exchanged arguments with the help of Rhetorical Structure Theory. 9 (RST offers an explanation for text coherence. It assumes that each part of a coherent text has a particular function, and so, a particular argument type. Researchers have already identified and loosely defined 30 different argument types in this context.) We further evaluated the dialogues, counting the number of agreed-on issues and analyzing the discussions clarity and the participants satisfaction. Our analysis showed that restricting arguments increases all three aspects (agreement, clarity, and satisfaction). From RST terminology, we selected the arguments that proved to contribute to the discussions systematization in our experiments and that ontology developers acknowledged as valuable: Elaboration. An elaboration presents additional details about the matter of discourse. Possible elaborations include introducing members of a set, instantiations of an During the discussions, participants raise new issues or elaborate on existing ones. Once a discussion evolves, participants can group issues and act on them accordingly. abstraction, parts of a whole, stages of a process, object attributes, or specializations of a general issue. Evaluation and justification. An evaluation provides a measurable advantage for a particular matter of discourse in comparison to another. A justification gives evidence that something or someone has the authority to make a statement. Alternative. An alternative is a comparable solution for the matter of discourse. Example. An example (in RST, also called evidence) of a particular matter of discourse increases the belief in the corresponding issue or idea. CounterExample. A counterexample provides counterevidence for a particular matter of discourse and decreases belief in an issue or idea. Decide on issues and ideas. The participants agree or disagree on an issue or postpone its resolution. Agreed-on ideas become part of the shared ontology. Argumentation ontology The argumentation ontology formalizes the arguments exchanged during ontology-engineering discussions. In contrast to the work on argumentation theory in agent communication, DILIGENT s argumentation ontology doesn t aim to capture exchanged arguments in a highly axiomatized manner. It doesn t support automatic negotiation between agents, 10 but it does define the argument types that humans tend to use in ontologyengineering discussions. Our argumentation ontology s main concepts are issues, ideas, and arguments (see figure 2). The argumentation ontology extends the IBIS methodology s argumentation theory and has the semantics implied by OWL DL. The argumentation ontology adapts the original IBIS model to the ontology-engineering discussions requirements. In this context, an issue introduces a new requirement or topic in the discussion from a conceptual point of view. The issue helps explicitly separate the discussions related to a domain s conceptualization from the actual formalization or implementation. Elaborations are extensions of existing issues. Ideas respond to issues and refer to their formalization; users discuss in terms of ideas how to formally represent domain knowledge in the ontology for instance, as a class, an instance, a relation, or an axiom. Accepted ideas trigger concrete ontology change operations. To foster consensus building, participants express positions and exchange arguments on ideas, issues, and ontology entities. Justifications are arguments for (pro) an idea or issue. Challenges are arguments against (con) an idea or issue. The ontology further differentiates arguments in favor through examples and evaluations. CounterExamples and alternatives are two classes of challenges that are particularly useful in ontology-engineering discussions. Within a discussion thread, participants can state positions. They clarify their position-on an issue, idea, or argument by exclusively declaring their agreement or disagreement. Once enough arguments and positions are available, the engineering team can take decisions (associated with issues, ideas, and ontology entities). A decision s status can be underdiscussion, postponed, discarded, and agreed. A decision records the issue on which it was taken, the positions constituting the final vote on the issue (with-votes [several positions]), and the line of reasoning (a sequence of arguments) underlying the decision. A decision NOVEMBER/DECEMBER

5 A r g u m e n t a t i o n T e c h n o l o g y elaboratesonissue Decision Issue hasstatus: onissue raisedissue: givenby withvotes Idea responsestoissue formalizesidea Elaboration givenby OntologyEntity label: ontochange: Actor name: givenby summarizesargumentation argumentson 1..* Argument Argumentation hasargumentation positionon positionon 1..* Position positionon Challenge Justification Human-Argumentation Machine-Argumentation providestext: tf/idf: Disagreement Agreement Alternative CounterExample Evaluation Example Figure 2. The core of the DILIGENT argumentation ontology. can also be on-idea (the idea formalizing an issue). This allows one to focus on the relevant arguments. Arguments are given-by actors,which are either humans or machines. Different kinds of actors provide different argumentations. Humans (HumanArgumentation) tend to argue by providing strings of text stating (provides text) their reasons, and machines tend to use argumentation measures such as Frequency and TF-IDF (term frequency-inverse document frequency). For each algorithm used, developers must introduce new subclasses of argumentation to model the different kinds of measures. A reasoning mechanism can alert users if they agree and disagree on the introduction of the same ontology entity. This can happen unintentionally if concepts are interconnected through a complicated inheritance hierarchy or in the case of long-lasting discussions. Furthermore, users can introduce inconsistencies by provisioning arguments (challenge versus justification) or proposing contradicting ideas. The coefficientmakna system Wikis have received increasing attention in the Semantic Web community. 11 This popularity is probably due to the core technology s ease of use a feature not necessarily characteristic of semantic applications and to their focus on collaborative and community aspects. Existing semantic wikis primarily support the creation of semantic (instance) data expressed in Semantic Web languages such as RDF Schema or OWL. (For a recent overview of existing semantic wiki engines, see ontoworld.org/index.php/semantic_wiki_ State_Of_The_Art.) As with native ontology editors, these wikis address ontology development at the implementation level at best, assuming that the wiki s collaborative nature inherently eases the engineering team s consensus building. We conceived the coefficientmakna ( members.deri.at/~elenas/coefficient) system as a wiki-based tool for distributed ontology engineering. The system builds on the lightweight semantic wiki engine Makna ( Makna.ag-nbi.de) in terms of design, architecture, and usage principles. Like Makna, we implemented it in Java based on the wiki engine JSPWiki ( whose functionality we extended according to the application scenario s requirements. Ontology creation and management Users can create ontologies from scratch or import existing ontologies to the wiki. When importing an existing ontology, the system maps the ontology model to the wiki hypertext model according to a predefined schema. The system describes each ontological primitive using a dedicated wiki page. The wiki page corresponding to a class includes information about the direct subclasses and superclasses and instances, and lists the labels and associated comments documenting the class. The system describes properties similarly. Pages related to instances of an ontology additionally include a reference to the associated classes. Users can mark specific wiki pages as 56 IEEE INTELLIGENT SYSTEMS

6 ontology entities with the help of an Ajax assistant or by using the wiki syntax. They can edit ontology entities in the same way as regular wiki articles and add or delete statements about these entities using a dedicated assistant. Users can browse the ontology by content or structure, and the core wiki engine provides versioning support. Users can export new versions to OWL and RDF Schema for further external processing. External reasoning services can help detect inconsistencies at the content level. The DILIGENT argumentation ontology plays an important role in coefficientmakna s functionality. The system uses this ontology as a formal description of the argumentation items relevant to ontology-engineering discussions: issues, ideas, and various argument types. Collaboration We extended JSPWiki with discussion pages associated with wiki articles referring to ontological primitives either domainspecific concepts (that is, the collaboratively built ontology) or argumentation items (that is, instances of the DILIGENT argumentation ontology). Users can mark specific wiki pages as issues for an engineering context and create and elaborate ideas to solve specific issues. To ease the creation of argumentationspecific wiki content on the discussion pages, the user interface has customized buttons specific to the DILIGENT argumentation model functionality. Externalization and detection coefficientmakna captures the ontologyengineering discussions as instances of the argumentation ontology: concrete issues, ideas, elaborations, and pro and con arguments. Users can query this data to monitor the status of discussions, progress, and possible conflicts and to reconstruct the rationale behind certain decisions. An external reasoning service, which analyzes the formalized discussion data, helps detect contradictions and conflicts. Case studies We conducted several case studies to evaluate and refine the DILIGENT framework. We found that direct interaction (through argument exchange) is particularly important for fostering consensus while creating a truly shared ontology. 6 The case studies dealt with building shared ontologies in various domains ranging from research topics and legal affairs to tourist information and cooking recipes to provide IT systems with new levels of data and process interoperability. Some ontologies had initial versions available prior to the case study; others were built from scratch. To reach agreement on the target ontology, the participants discussed domain and application requirements, modeling issues, and domain-knowledge formalization. Each of the four studies needed additional methodological and technological support to trace the exchanged arguments and focus the discussions. The problems in a completely distributed environment such as the Semantic Web will likely be much more difficult. Our first case studies focused on identifying and selecting argument types likely to be Users can edit ontology entities in the same way as regular wiki articles and add or delete statements about these entities using a dedicated assistant. useful in ontology-engineering discussions. We asked participants to agree on an ontology to represent the research topics typically addressed in a computer science institute. In the first experiment, we provided only limited moderation. In the second, we followed the framework, albeit without tool support. The framework nevertheless let us identify the relevant arguments and define the collaborative ontology development process. In a subsequent study, we observed legal experts using our framework to build an ontology to represent professional legal knowledge. In contrast to the first experiment, the participants had no prior expertise in ontology engineering. The framework enabled legal-domain experts to formalize their domain, and they appreciated its detailed guidance even though tool support was limited at that time. We describe in more detail the results of our last case study, in which participants used the framework with the support of coefficientmakna. We performed the case study at the Free University of Berlin. Fifteen students built a community wiki to provide dessert recipe information using semantic technologies. They used the DILIGENT framework and an underlying ontology for concept-based search and for semantic annotation of recipe documents written in German. We supervised the students during the case study and later analyzed the arguments they exchanged using the system and collected their experiences with coefficientmakna. We started the case study with the following hypotheses: The students can agree on a shared ontology and find the guidance given by the framework useful for this purpose. Nonexperts don t require extensive training to use the argumentation model. The DILIGENT framework, and thus the application of its argumentation theory to ontology engineering, enhances design decision traceability. coefficientmakna adequately supports the argumentation process. At the beginning of the study, the engineering team, consisting of eight students, participated in a tutorial on ontology engineering and the DILIGENT argumentation framework. Afterward, they collaboratively built several versions of the ontology over four months. Prior to the requirements analysis, they chose an ontology-engineering expert as a moderator, who supervised the discussions and mediated major conflicts. The team decided to use majority voting as an instrument to resolve conflicts related to modeling decisions. During requirements analysis, the participants first agreed on a minimal list of competency questions the prospected ontology must answer, and so specified an initial set of issues. Discussions were carried out using the coefficientmakna system in particular, the argumentation model provides the set of possible argument types and a procedure to reach consensus on the raised issues. Among other issues, the participants addressed and agreed on the abstractions needed to reduce the model s complexity. They decided that the ontology would contain a classification of desserts, food products, and ingredients as well as information specific to recipes: preparation time, degree of difficulty, region of origin, calories per measuring unit, and required kitchen devices. Additional domain-specific information such as provenance or preparation procedure were considered secondary NOVEMBER/DECEMBER

7 A r g u m e n t a t i o n T e c h n o l o g y Figure 3. A DILIGENT issue represented as a wiki page in coefficientmakna. occurred; in particular, questions arose over the cost benefit of reusing classifications and over modeling units of measurement for ingredients. For the former, the moderator needed to intervene; in the latter, the participants reached an agreement. As a result of the case study, the students agreed on a shared ontology and used it as the system s conceptual backbone. The system is now used to collect and share recipes. In comparison to our previous case studies, the tool significantly sped up the provisioning of arguments and eased the methodology s application. Because most participants were already familiar with basic wiki principals, they required only a brief introduction to work with the system in addition to the DILIGENT specific training. Figure 4. A discussion page in coefficientmakna. for the application setting. The corresponding wiki page contains a natural language description of the current problem (see figure 3). By means of the two icons on the screen s upper right, users can switch between the wiki article and the discussion, which captures the related elaborations, arguments, positions, and decisions ordered chronologically. Proposing ideas resolves issues. In coefficientmakna, users added new ideas on the discussion page of the corresponding issue. Figure 4 shows an example discussion page in the Lekapidia study. Using the DILIGENT argumentation framework, it s possible to add arguments and positions to an idea or make a decision on it with the help of the assistants located on screen s upper right. A special page shows the argumentation process status. The system realizes this using a query on the statements added to the discussion pages. Later, participants identified potentially relevant information sources, which could be reused in the target recipe ontology: a set of 500 recipes covering the target domain and several general-purpose classifications, such as the Eurocode2 Food Coding System (www. ianunwin.demon.co.uk/eurocode/docmn), the ISO 3166 country name codes system (www. niso.org/standards/resources/3166.html), and the ontology UnitDim (providing definitions of physical units and quantities, nl/projectsite2/data/documents/ /UnitDim.owl). The team discussed whether reusing these sources would benefit the overall ontologyengineering process, given that two of them required considerable customization efforts. They agreed to not reuse these sources. Once the participants completed requirements analysis, the discussion continued on topics related to conceptualization and formalization (how to model or name certain entities, how to distinguish between classes and properties, and so on). More complex formalizations, such as those related to range or cardinality constraints, have also been carried on using the native ontology editor Protégé ( protege.stanford.edu). For more complex formalizations, coefficientmakna generated an OWL version of the ontology, which the team modified using the ontology editor, prior to reloading the ontologies to the wiki for further discussions. Several conflicting situations Our case studies show that the DILIGENT argumentation process guides ontology-engineering discussions and ensures an efficient agreement process. Nonexperts can quickly understand and follow the process. The theory of argumentation and the predefinition of argument types simplifies the structure of the discussions and facilitates the detection of conflicting arguments. At the same time, coefficientmakna significantly reduces the effort to capture the arguments in a structured way. However, our evaluation results are qualitative rather than quantitative. Future case studies will focus on achievable time savings following different engineering approaches. They will account for the engineered ontologies quality and reusability in terms of time needed by external parties to understand and integrate the constructed ontologies in different applications. Moreover, an issue that requires further consideration is the system s usability. Additional assistants and patterns for typical argumentation situations could offer further support. The current coefficientmakna is particularly suited for the developing simple inheritance hierarchies. Native ontology editors such as Protégé better support modeling axioms and other advanced ontology primitives. With respect to automating the argumentation process, we found that resolving natural language conceptualizations into ontology entities is challenging. However, we foresee that in different ontology-engineering projects, similar issues will emerge. It might be possible to detect patterns and propose the best modeling solution IEEE INTELLIGENT SYSTEMS

8 Our research shows that argumentation theory has much to offer current Semantic Web challenges. However, general argumentation frameworks emerging in adjacent disciplines require some customization to fit an ontology-engineering scenario s concrete requirements. In the future, we might be able to apply findings related, for example, to negotiation in ontology engineering to further facilitate building ontologies. T h e A u t h o r s Christoph Tempich is a consultant at Detecon International in Competence Practice Information Technology group. His research interests include business models for the Semantic Web, Semantic Web technologies and applications, ontology engineering, and knowledge management. He received his PhD in applied informatics from the University of Karlsruhe, Institute of Applied Informatics and Formal Description Methods (AIFB). He is a member of the TM Forum and the Open Group. Contact him at Detecon Int l, Oberkasseler Str. 2, D Bonn, Germany; tempich@aifb. uni-karlsruhe.de. References 1. A. Gómez-Pérez, M. Fernández-López, and O. Corcho, Ontological Engineering, Advanced Information and Knowledge Processing, Springer, C.W. Holsapple and K.D. Joshi, A Collaborative Approach to Ontology Design, Comm. ACM, vol. 45, no. 2, 2002, pp P. McBurney, D. Hitchcock, and S. Parsons, The Eightfold Way of Deliberation Dialogue, Int l J. Intelligent Systems, vol. 22, no. 1, 2007, pp W. Kunz and H.W.J. Rittel, Issues as Elements of Information Systems, working paper 131, Inst. Urban and Regional Development, Univ. of California, A. Selvin et al., Compendium: Making Meetings into Knowledge Events, Proc. Knowledge Technologies, AKT Technologies, 2001; 6. C. Tempich, Ontology Engineering and Routing in Distributed Knowledge Management Applications, doctoral dissertation, Institut für Angewandte Informatik und Formale Beschreibungsverfahren, Univ. Karlsruhe, C. Potts and G. Bruns, Recording the Reasons for Design Decisions, Proc. 10th Int l Conf. Software Eng., IEEE CS Press, 1988, pp M. Uschold and M. Grueninger, Ontologies, Principles, Methods and Applications, Knowledge Sharing and Rev., vol. 11, no. 2, 1996, pp W.C. Mann and S.A. Thompson, Rhetorical Structure Theory: A Theory of Text Organization, The Structure of Discourse, L. Polanyi, ed., Ablex Publishing, 1987, pp I. Rahwan et al., Argumentation-Based Negotiation, Knowledge Eng. Rev., vol. 18, no. 4, 2003, pp A. Souzis, Building a Semantic Wiki, IEEE Intelligent Systems, vol. 20, no. 5, 2005, pp Elena Simperl is the deputy director of the Digital Enterprise Research Institute at the University of Innbsbruck, and the education manager of the Semantic Technologies Institute International. Her research interests include semantic technologies, ontology engineering, collaborative applications, and business aspects of the Semantic Web. She received her PhD in computer science from the Free University of Berlin. Since 2006, she s organized the Semantic Web PhD Student Network for Berlin and Brandenburg and the Knowledge Web PhD Symposium at the European Semantic Web Conference. Contact her at the Digital Enterprise Research Inst., Leopold-Franzens- Univ. Innsbruck, ICT Technologiepark, Technikerstr. 21a, 6020 Innsbruck, Austria; elena.simperl@deri.at. Markus Luczak is a member of the Networked Information Systems Working Group at the Free University of Berlin. His research interests include Web collaboration, software engineering, ontology engineering, and corporate Semantic Webs. He received his diploma in computer science from the Free University of Berlin. Contact him at the Free Univ. of Berlin, Takustr. 9, Berlin, Germany; luczak@inf.fu-berlin.de. Rudi Studer is a full professor in applied informatics at the University of Karlsruhe, Institute of Applied Informatics and Formal Description Methods (AIFB). His research interests include knowledge management, Semantic Web technologies and applications, ontology management, text mining, and Semantic Web services. He s a member of the board of the Research Center for Information Technologies (FZI) at the University of Karlsruhe and a director of the Information Process Engineering research group at the FZI. He cofounded the spin-off company ontoprise GmbH, which develops semantic applications. He s an associate editor of the ACM Transactions on Internet Technology and an advisory board member of IEEE Intelligent Systems. He s technical director of the EU-funded Integrated Project NeOn (Lifecycle Support for Networked Ontologies). Contact him at the Institut für Angewandte Informatik und Formale Beschreibungsverfahren, Univ. Karlsruhe, D Karlsruhe, Germany; studer@aifb.uni-karlsruhe.de. H. Sofia Pinto is an assistant professor at the Lisbon Instituto Superior Tecnico s Department of Computer Science and Engineering. Her research interests include ontology engineering and ontology use, in general, including ontology learning, ontology evaluation, semantic annotation, ontology reuse, the Semantic Web, knowledge engineering, knowledge management, and virtual organizations. She received her PhD in computer science and AI from the IST. She coorganized the Building and Applying Ontologies for the Semantic Web workshop and has collaborated in several EU projects, such as SWAP, SEKT and NeOn. Contact her at the Inst. Superior Tecnico, Dept. de Eng. Informatica, Av. Rovisco Pais, Lisboa, Portugal; sofia@algos.inesc-id.pt. NOVEMBER/DECEMBER