Ontologies for Reusing Learning Object Content

Ontologies for Reusing Learning Object Content Dragan Gašević1, Jelena Jovanović2, Vladan Devedžić2, Marko Bošković3 2

1 School of Interactive arts and Technology, Simon Fraser University Surrey, Canada FON – School of Business Administration, University of Belgrade, Serbia and Montenegro 3 TrustSoft Graduate School, University of Oldenburg, Germany [email protected], [email protected], [email protected], [email protected] http://goodoldai.org.yu

Abstract. The paper proposes a framework for building learning object (LO) content using ontologies. In the previous work on using ontologies to describe LOs, researchers employed ontologies exclusively for describing LOs’ metadata. Although such an approach is useful for searching for LOs in LO Repositories, it does not provide us with features to reuse components of LOs, nor to incorporate an explicit specification of domain semantics into LO content. We propose the use of two kinds of ontologies as a solution to this problem: content structure ontologies and domain ontologies. Further, we explore necessary tools for such an approach, like Semantic Web annotation tools and specific domain authoring tools, as well as domain XML formats and transformation techniques. Additionally, we give a conceptual overview of a course authoring tool that fully support the proposed approach.

1. Introduction The term learning object (LO) is one of the main research topics in the e-learning community in the recent years [2], and most researchers pay attention to the issue of LOs’ reusability. Several standards have been developed so far aiming to improve LOs reusability. For example, IEEE Learning Object Metadata (LOM) and Dublin Core [11] are two initiatives specifying a standardized set of metadata that facilitates retrieval of Web-based resources. For both of them, XML and RDF bindings are defined, so they can be used on the Web. Many developers have already based their LO Repositories (LORs) on top of those standards. To the best of our knowledge, up to now, researchers were mainly proposing Semantic Web technologies and ontologies in particular for improving LOs’ metadata. For example, Mohan and Brooks [13] analyze relations of LOs and the Semantic Web, especially emphasizing importance of ontologies. Accordingly, they identify several kinds of ontologies related to LOs: ontologies covering domain concepts, ontologies for e-learning, ontologies about teaching and learning strategies, and ontologies about physical structuring of learning objects. In the paper [1] the authors give an example of an ontology developed in accordance with the ACM Computer Classification System (ACM CCS). This ontology is represented in RDF, and used in the Edutella system. However, these solutions do not provide us with a

2

Dragan Gašević1, Jelena Jovanović2, Vladan Devedžić2, Marko Bošković3

possibility to reuse just certain components of a LO or to use the same LO in different ways (e.g. presentational adaptivity). In order to address this issue we advocate the idea that ontologies can be used to describe LO’s content, thus providing LOs with a new dimension of reusability – content reusability. We start from the classification of ontologies for e-learning given in [17] that makes difference between the following aspects: content – what the learning material is about, context – in which form a topic is presented, and structure – as learning material does not appear in isolation. Our focus is on both structure and domain (content) ontologies. As a structure ontology we propose an ontology based on the Abstract Learning Object Content Model (ALOCoM) [19] and IBM’s Darwin Information Type Architecture (DITA) [14]. Regarding content ontologies, we propose employment of the presently available domain ontologies. Of course, for creating such LOs we need adequate tools, so we explore necessary tools like Semantic Web annotation tools and domain authoring tools (e.g. software modeling tools), as well as domain XML formats and transformation techniques. Additionally, we give a conceptual overview of functionalities that a course authoring tool should have in order to fully support the proposed approach.

2. Ontologies for learning objects We have already mentioned the approach presented [1] where the authors extend current metadata standards, like IEEE LOM, with an RDF-based ontology. Enriching LO’s metadata with keywords taken from the ontology, they provided an infrastructure able to automatically retrieve the most appropriate LOs. However, we argue that providing ontology-based descriptions of LOs’ content would make the process even more efficient. In other words, the main peculiarity of our approach comparing to the previous ones is that we are mainly focused on the LO’s content, and not on their metadata. That means we believe that semantically organized LOs content has better potentials to be repurposed. Starting from the aforementioned classification of ontologies given in [17], we base our approach on the first two ontology types, namely content (domain) and structure ontologies. Accordingly, Figure 1 depicts ontologies that we find relevant for describing learning object content: content structure ontologies (COi) and content/domain ontologies (DOi). We use domain ontologies to semantically mark up content of a LO. A domain ontology describes content of a LO in terms of concepts of the subject domain and their relationships. Concepts from a domain ontology together with the content structure ontology might be useful for specifying not only the subject (content) of a LO but also its prerequisites, expected learning outcomes in terms of acquired domain concepts and the like. LO’s metadata are described using IEEE LOM standard. This metadata can be enriched with concepts from metadata ontologies (MOi). 2.1.

Content structure ontologies

An explicit definition of a LO structure can be useful in many cases we need to reuse specific parts of a LO, rather then the LO as a whole. In such situations, current

Ontologies for Reusing Learning Object Content

3

practice is to copy & paste, in order to reuse specifically those parts of a document (e.g. a paragraph, a sentence or an illustration) that are relevant. However, this can be rather tedious and time-consuming task. More importantly, such an approach, with the manual intervention it implies, does not scale well, as it does not allow for (semi-) automated processes to assist or take over completely. A natural solution for explicit definition of LO structure is an ontology defining different types of LOs’ components, as well as a set of tools for decomposing present LOs into ontology-defined components and vice-versa (ontology-defined LOs’ components into a new LO). CO1 DO3

DO2

Content CO1

DO1

CO1

Metadata MO2 MO1

Learning Object MO3

Fig. 1. Ontologies describing learning objects: a) Metadata ontologies (MOi), b) Domain ontologies (DOi), and c) Content structure ontologies (COi)

As a content structure ontology we propose ALOCoM ontology [20]. The ontology is an extension of the Abstract Learning Object Content Model [19] with the concepts defined in the IBM’s Darwin Information Typing Architecture (DITA) [14]. The main role of the ontology is to be an integration point for different LO content model types (SCORM, CISCO RLO/RIO, NETg, etc) as well as different tool specific LO types (e.g. slides, text documents, etc). For more details about this ontology see [20]. 2.2.

Content/domain ontologies

In order to enable effective LOs reusability we have to further enhance semantics of their content. Accordingly, we argue that learning resources should be further enhanced by providing domain ontology-based descriptions of their content, or more precisely, by adding pointers to the concepts of appropriate domain (content) ontologies. We create those annotations on top of the content structure ontology. The annotation can be remote or embedded and the XML/RDF mechanism can be used to syntactically present annotations. A LO created using this principle gets a new dimension of reusability – it can be used in different ways within the same course. For example, in computer science courses, like object oriented modeling with UML, a teacher might use an UML model in a Power Point presentation, while students should try the same model in a CASE tool (e.g. Rational Rose). Similarly, this principle can be applied in other disciplines (e.g. philosophy, history). Presently there are a plethora of specific domain ontologies already available at the WWW. Actually, we can talk of libraries of ontologies created through contributions of the Semantic Web community’s members. For example, DAML Ontology Library (http://www.daml.org/ontologies/) holds nearly 300 ontologies written in the DAML

4


ontology language. An author of a LO can search one of those libraries in order to find an ontology that best describes content of the LO (s)he is creating. Also, an author might be provided with means to create his/her own ontology during a LO construction. In the latter case, the authoring tool (s)he uses should make the process of ontology development implicit (e.g. an ontology can be created from author’s annotations).

3. Ontologies in Use Here we recommend an approach that builds upon the traditional LO creational schema but further extends it to incorporate support for semantic markup of LO content. Figure 2 depicts the proposed approach for LO enhancement. The figure illustrates benefits of the approach for creating and consuming such LOs in adaptive learning courses. Course Authoring Tool

XML

Non adapted courses

Intelligent Learning Environment

XML

Learner Uses Uses LOs from LOR / Stores descriptions of created courses

Describe

Transformations (XSLT)

Defined for

CO1

Author

DO3

DO2

Authoring and annotation tool

Stores created LOs

Contains/ References

Content

DO1

Metadata MO2 MO1

LO Repository

Learning Object MO3

Fig. 2. Extended LO creational schema – LO’s content is related with a domain ontology (DOi)

However, the inclusion of ontology grounded descriptions of LO content in its markup data should not impose additional burden of attaining the basics of ontological engineering on the part of LO authors. On the contrary, we argue that it would be better if authors were ignorant of existence of the domain ontology, since one could not expect that, for example, a teacher of social sciences can be aware of the role that ontologies play in knowledge engineering. In order to overcome this problem we recommend either usage of existing (i.e. annotation tools) or development of new tools (see the next subsection) that would have an appropriate GUI for seamless creation of annotations. Later, the teacher could use these LOs within a course-authoring tool, when building courses. Thus, the process of composing a course would be primarily based on searching for adequate LOs and ‘gluing’ the retrieved LOs together according to an instructional model. Rich semantic descriptions of LOs would make the search process both easier and more accurate. This way created educational courses are used in Intelligent Learning Environments (ILE), where they become the subject of further transformations aimed at making them compliant with specific educational needs of each learner.


3.1.

5

Authoring and Annotations Tools

One possible solution for creating LOs with rich semantic descriptions is to provide a teacher with an authoring tool that would, in the background, perform all required annotations of the LO under construction (i.e. creates instances of ontology elements) while (s)he works on its development. For example, a Web browser enhanced with the support for semantic markup of Web documents can be used to create new LOs out of the content of existing Web pages. The browser would, in the background, perform all required annotations of the currently displayed Web page while the author selects some of its parts. That way created LOs could be later used by a teacher when building a course: annotated parts of Web documents can be extracted and plugged into the course structure. In fact, this idea has analogy in marking printed book parts using scriber. While reading a printed text, a teacher, uses these marks as remainders to the parts that (s)he found interesting for her/his course. An advantage that Web resources have is that denoted (i.e. annotated) parts can be automatically extracted. A first step towards building a framework that would use domain ontologies to enable reusability of LOs’ content is to have tools that would allow LOs authors to seamlessly compose LOs with semantically marked content. Since widely accepted and well-known authoring tools (e.g. text processors, MS Power Point, HTML editors) do not provide the full support, we suggest employing either additional tools in the form of annotation tools or special domain tools that would incorporate both authoring and annotation work. Furthermore, these tools should provide support for domain specific XML-based formats (e.g. the W3C Mathematical Markup Language MathML). Annotation tools. Annotation tools are a Semantic Web effort aimed at producing semantically marked up Web resources (http://annotation.semanticweb.org), in this case LOs. The primary motive is to make the process of generating semantic descriptions less burdensome and tedious, therefore stimulating, otherwise reluctant content authors to produce annotated LOs. The initial efforts to build such tools were focused on creating specialized tools that support association of semantic markups with preexisting documents. However, this approach had not proved to be efficient and accepted in content authors’ communities since it exerted an extra effort and imposed an additional burden on content authors. As a result, currently we are facing an augmenting number of tools trying to enable authors to easily annotate their documents while creating them. While the set of currently available annotation tools supports annotation of a limited number of document formats, for instance: HTML pages, Scalable Vector Graphics (SVG) and MathML document formats as well as MS Word and MS Power Point documents, future tools should implement support for other important formats of Web resources, such as PDF, Synchronized Multimedia Integration Language (SMIL), and different multimedia formats (e.g. for representing animation, sound, etc.). In Table 1 we give a comparative overview of main characteristics of four wellknown authoring and annotation tools: Amaya (a concrete implementation of the client component of the Annotea framework [6]), Ont-O-Mat (an implementation of the CREAM framework [5]), SMORE [7] and SemanticWord [18].

6


Table 1. Summary view of main features of four widely known annotation and authoring tools Annotation tool

Ont-O-Mat

SemanticWord

SMORE

Amaya

Document Format

HTML pages

MS Word documents

HTML pages, images and e-mail messages.

HTML, SVG and MathML document formats.

Annotation Form

Document’s annotation is viewed as a set of instances of DAML+OIL classes, datatype properties (attributes) and object properties (relationships).

Two types of annotations: 1) instance references – associates a text region with an instance of a class; 2) triple bags – describes a text region with a collection of DAML+OIL subjectpredicate-object triples.

Annotations are presented as standard RDF triples with the underlying subjectpredicate-object model.

A document’s annotation is composed of RDF triples based on extensible annotation RDF Schema plus document’s description as plain text.

Domain Ontology Format

DAML+ OIL

DAML+ OIL

DAML

Does not provide support for describing content using domain ontologies.

Storage of generated annotations

Both inside the original document and in the Annotation Inference Server.

Annotations are a part of the original document

Annotations are stored in a separate file; in case of e-mail messages that file is sent with the e-mail as an attachment.

As a separate file stored either on the local machine or on a server in an RDF database.

Homepage

http://annotation. semanticweb.org/ tools/ontomat

http://mr.teknowledge.com /daml/SemanticWord/ SemanticWord.htm

http://www.mindswap.o rg/ ~aditkal/editor2.shtml

http://www.w3.org/ Amaya/

Domain-specific tools. As we already mentioned, the newest annotation tools can serve as authoring tools as well. However, their focus is restricted to general domains, like creation of semantically annotated HTML pages or Word documents. More specific domain tools (e.g. Rational Rose in software engineering, AutoCAD in civil engineering etc.) currently are not extended with support for content annotation. However, if we want to better support learning in those specific domains, we would need to augment their domain tools with the ability to annotate created LOs. For example, suppose that an expert in the field of software engineering uses a domain tool augmented with annotation capabilities to build UML models for the application (s)he is developing. The created models (can be used for teaching/learning UML modeling and thus treated as LOs) would be semantically described with concepts from the domain ontologies. Since, semantically marked-up LOs can be retrieved from the Semantic Web, a teacher of software engineering might use them to prepare a course on UML-based object oriented modeling, e.g. (s)he can incorporate those UML models in slides of the PowerPoint presentation (s)he is preparing. 3.2.

Conceptual Overview of a Course Authoring Tool

Once created, a LO with its ontology-based description of content can be included in different courses. A course authoring tool (CAT) should enable seamless composition of new courses out of existing LOs available in LORs. Furthermore, it should enable an author to compose its own course model that best suits his/her pedagogical approach. Our current view on the architecture of such a tool is depicted in Figure 3. As we stated a CAT should facilitate search and retrieval of LOs available in LORs. This search should be based on concepts from standard metadata ontologies (in Figure 3 labeled as MOi) as well as concepts from domain ontologies (DOi), since it is the straightforward way to find relevant LOs for the courses’ subject domain. The search module of a CAT should use one of already available Semantic Web querying


7

languages, such as RDF Query Language (RDQL [15]) or OWL Query Language (OWL-QL [3]). However, one should keep in mind that the majority of course authors are neither familiar with the concept of ontology nor with query languages (especially those for the Semantic Web), so we envisage a rich User Interface Module that would enable users to complete the job of course building with just point-and-click and dragand-drop actions, accompanied with minimum typing inputs. User Interface Module

Instructional Metamodel (EML based)

Author Coordinator Course model building module MO2

Search Module

Course (content) composition module

MO1

MO3

R

DO2 DO3

LO it

Course Descriptions

Course Models

XML

Non adapted courses

DO1

Fig. 3. Architecture of Course Authoring Tool

When an author, using the CAT’s search mechanism, finds a useful LO, (s)he should be able to retrieve it and incorporate it into the instructional model of a course (s)he is creating. Here we consider an instructional model as an abstraction crated in accordance with an Educational Modeling Language – EML (e.g. the EML, LMML, etc) [9] that should serve as a foundation for development of a Learning Management System (LMS, e.g. WebCT, BlackBoard, etc.). The course authoring tool should also provide an author with an option of creating a new course model employing EML as a meta-language. A special module, we named it Course Model Building Module, in the CAT architecture should support this functionality. A CAT should further enable an author to “fill” his/her course model with LOs retrieved from LORs using CAT’s search capabilities. The resulting instructional model can be serialized into the XML format of the used EML (i.e. its XML bindings), and then stored in a repository of non user-adapted courses. Besides, descriptions of those courses should be stored in LORs. Inside an Intelligent Learning Environment (ILE) XML bindings of these courses should be further transformed into learner-suitable courses, both in terms of their content and presentational form. A CAT should also facilitate semi-automatic annotation of a course under construction and generation of course’s description to be deposited in a LOR. This description can be automatically composed out of semantic descriptions of its constituent LOs as it was proposed in [8]. Thus, ontology grounded semantic descriptions of LOs and their content could play the major role in semantic markup of generated courses. Further more, we consider that a course author should use the annotation mechanism to specify the complexity level of certain units within the course. This

8


could be achieved using adequate elements of the IEEE LOM standard, such as the purpose element and its educational level sub-element. We advocate the usage of an e-learning standard schema instead of a custom made schema in order to assure that the encrypted data would be properly interpreted from a system responsible for adapting the course material to each learner’s knowledge background. Finally, in the CAT’s architecture we envisage a coordination module that should act as an intermediary in communications between other CAT’s modules.

4. Technologies supporting the proposed approach This section gives an overview of technologies needed to support the proposed approach of using domain ontologies for annotating LO content. XML-based formats. Presently, XML is a widely adopted Web data sharing standard. An increasing number of tools support XML, as well as XML-based domain-specific sharing formats. Examples of XML-based formats are: MathML, Petri Net Markup Language (PNML), XML Metadata Interchange (XMI) etc. LOs based on XML can be easily converted into different formats, e.g. HTML, SVG, but also formats of general purpose software tools like are MS Power Point, MS Word, or domain specific tools. Transformations. Transformations are an important segment for achieving LOs repurposing on the Semantic Web. The most sensible approach for generating different presentations of an XML-serialized LO is to employ the eXtensible Stylesheet Language Transformation – XSLT. One should note that the Annotea annotation tool uses this XSLT-based approach. XSLT is a light-way approach for converting XML-based content of LOs into different presentation formats (e.g. SVG, HTML, SMIL, etc.). Semantic Web Query Languages. Reusability of LOs on the Semantic Web also vastly depends on query languages that are used for searching LORs and retrieving LOs. Those query languages should be expressive enough to enable formulation of precise queries over LORs, but at the same time should not pose too much computational burden on the part of the LOs providers (i.e. servers that host LORs). Currently there are few languages of this type: RDQL (RDF Query Language, [15]), TRIPLE [16], OWL-QL (OWL Query Language) [3] just to name the most important ones. It should be noted that these languages are not meant just for queering Semantic Web, but also for performing transformations between knowledge expressed using ontological languages. Further more, they could be used to enable mappings between ontologies as well as for defining views on ontologies. Main problem with those languages is that there is not a general consent which language should be accepted as the standard. A potential candidate is the recent W3C initiative for a standard transformation and query language based on OWL – OWL-QL. Although access to LORs is facilitated by Semantic Web query languages, formulation of effective queries on LO metadata with complex organization (i.e. underlying schema) may require profound understanding of the schema that is often beyond the needs of an agent (either human or software). This can be alleviated by employing the concept of views over LO metadata that enables personalization of the


9

way those metadata are viewed by the agent. More precisely, creating a view over some data on the Semantic Web (a LO metadata in our case) essentially consists of the creation of virtual metadata schemas and descriptions consistent with the agent’s perception of those data. The concept of view enables us to personalize the process of searching LORs and make it both easier and more efficient. To the best of our knowledge, currently there are only two Semantic Web view languages, both of them build upon RQL query language and aimed for RDF(S) data models: 1) the RDF View Language – RVL [10] – a view definition language capable of specifying not only views over RDF-encrypted metadata (i.e. virtual source descriptions) but also views over metadata schemas (i.e. virtual RDFS schemas), 2) the view language proposed by Volz et al in [21]. In addition, some query languages can be used for specifying views over data on the Semantic Web, as it was proposed in [12].

5. An Application Example We developed a simple educational Web application in order to illustrate the proposed approach of using domain ontologies for annotating LO content. The purpose of this application is teaching Petri nets. In this context we consider Petri net models as LOs. The application is based on the Petri net ontology.. For developing Petri net models (i.e. LOs) we use the P3 – a Petri net tool we have developed for teaching Petri nets [4]. The P3 tool is able to generate RDF description of a Petri net as well as to produce SVG description of a Petri net model. Each Web page of the application contains a graphical presentation of adequate Petri net model (RDFannotated SVG) and provides support for simulation with that model. A user can save a Petri net he is working with in the PNML (Petri Net Markup Language) format and that Petri net can be further imported into Petri net tools (e.g. P3). The same model in SVG form can be used in other Web pages, but also can be shown in a tool such as MS Power Point.

6. Conclusions In this paper we have presented an approach based on using ontologies for annotating LO content, and thus extending LO reusability. We strongly believe that the attention should be paid on two kinds of ontologies for describing LO content: content structure ontologies, and domain ontologies. As an ontology of the former type, we propose ALOCoM ontology that covers structural aspects of different types of LOs, so that LO components can be reused as well. As ontologies of the latter type, we propose currently available domain ontologies, so that semantic annotations can be incorporated into LO content, hence improving their reusability. Our next step will be to experiment with presently available authoring and annotation tools in order to practically evaluate their usefulness for creating LOs with semantically annotated content. Further, we plan to develop an Adaptive Learning System that would exploit semantic annotations of LOs content in order to adapt that content to the specific needs of each learner.

10


7. References 1. Brase, J., Nejdl, W.: Ontologies and Metadata for eLearning, In S. Staab & R. Studer (Eds.) Handbook on Ontologies, Springer-Verlag (2004) 555-574. 2. Duval, E., Hodgins, W.: A LOM research agenda, In Proc.of the 12th Int’l WWW Conf., Budapest, Hungary (2003) 1-9. 3. R. Fikes, et al: “OWL-QL - A Language for Deductive Query Answering on the Semantic Web,” KSL Technical Report 03-14, KSL, Stanford University, Stanford, USA (2003) 4. Gašević, D., Devedžić, V.: Teaching Petri nets using P3, Educational Technology & Society, Vol. 7, No.4 (2004) 153-166. 5. Handschuh, S., Staab, S.: Authoring and Annotation of Web Pages in CREAM, In Proc. of the 11th Int’l WWW Conf., Honolulu, USA (2002) 462 - 473. 6. Kahan, J. et al: Annotea: An Open RDF Infrastructure for Shared Web Annotations, In Proc of the 10th Int’l WWW Conf., Hong Kong, May, 2001, pp. 623-632. 7. Kalyanpur, A. et al: SMORE - Semantic Markup, Ontology, and RDF Editor, Available: http://www.mindswap.org/papers/SMORE.pdf (2003) 8. Keenoy, K. et al: Personalisation Services for Self e-Learning Networks, Available: http://www.dcs.bbk.ac.uk/selene/reports/SeLeNe-Personalisation.pdf (2004) 9. Koper, R.: Educational Modeling Language: adding instructional design to existing specifications, In Proc. of Works. Standardisierung im eLearn., Frankfurt, Germany (2002). 10. Magkanaraki, A. et al: Viewing the semantic web through RVL lenses, In Proc. of the 2nd Int’l Semantic Web Conf., Sanibel Island, USA, pp. 96-112 (2003). 11. McClelland, M.: Metadata Standards for Educational Resources, IEEE Computer, Vol. 36, No. 11 (2003) 107-109. 12. Miklos, Z. et al: Querying semantic web resources using triple views. In Proc. of the 2nd Int’l Semantic Web Conf, Sanibel Island, USA, pp. 517-532, 2003. 13. Mohan, P., Brooks, C.: Learning Objects on the Semantic Web, In Proc. of the IEEE Int’l Conf. on Adv. Learning Technologies, Athens, Greece, 2003, pp. 195-199. 14. Priestley, M.: DITA XML: a reuse by reference architecture for technical documentation, In Proc. of the 19th Int’l Conf. on Computer Doc., Sante Fe, USA (2001) 152-156. 15. Seaborne, A.: RDQL - A Query Language for RDF, W3C Member Submission, Available: http://www.w3.org/Submission/RDQL/ (2004) 16. Sintek, M., Decker, S.: TRIPLE - A Query, Inference, and Transformation Language for the Semantic Web,” In Proc of the 1st Int’l Sem. Web Conf, Sardinia, Italy (2002) 364-378. 17. Stojanović, Lj. et al: eLearning in the Semantic Web, In Proc. of the WWWNet Conf., Orlando, USA (2001) 18. Tallis, M.: Semantic word processing for content authors, In Proc of the Workshop Notes of the Knowledge Markup and Semantic Annotation Workshop, Sanibel, USA (2003) 19. Verbert, K. et al: Towards a Global Component Architecture for Learning Objects: an Ontology Based Approach, In Proc. of OTM 2004 Workshop on Ontologies, Semantics and E-learning, Agia Napa, Cyprus, 2004. 20. Verbert, K. et al: Ontology-based learning content repurposing, In Proceedings of the 14th Int'l WWW Conf., Chiba, Japan (2005) 21. Volz, R. et al: Views for light-weight web ontologies, In Proc. of the ACM Symposium on Applied Computing, NY, USA (2003) 1168-1173