Mar 2005 Index
 

Home Page


Editor’s Note: Technology based learning objects continue to be a focus for innovation. This paper adds descriptors to standard Learning Objects (LOs) to facilitate higher levels of automation for e-Learning. It builds on SCORM and similar packaging systems that use Learning Object Management (LOM) to create customized curricula and implement these via a Learning Management System (LMS). This paper shows how addition of descriptors and data to Learning Object metadata can facilitate automation of Learning Object Management and delivery of Learning Objects on Learning Management Systems.
 

Normative Specifications of
Learning Objects and Learning Processes:
Towards Higher Levels of Automation in Standardized e-Learning


Salvador Sánchez-Alonso and Miguel-Angel Sicilia
 

Abstract

Learning object metadata records are nowadays mostly descriptive in the sense that they are intended to give information about the contents or the format of the learning object, but without entailing explicit run-time semantics for Learning Management Systems that use them. Nonetheless, normative metadata descriptions are also required in order to obtain systems that behave according to metadata records in a consistent and predictable way. This paper describes the rationale for normative specification techniques as a complement for existing descriptive metadata, which enables a higher degree of automation by precisely describing usage conditions and expected outcomes for learning objects and learning processes.

Keywords: Learning object, metadata, design by contract, ontology, semantic conformance profiles, semantic web, e-learning.

Introduction

Increasing interest in Web-based education has resulted in standardization efforts to foster portability and shared usage semantics of learning contents and learner information across vendors, platforms and systems (Anido et al., 2002). As a matter of fact, it is possible today to package a Web-oriented course according to standard formats (e.g. according to SCORM[1] packaging models) and then import and use that same content inside any learning management system (LMS) that is compliant with the given standard packaging rules. In addition, the scope of such standards and specifications is continuously expanding and covering new areas; for example, the SCORM “sequencing and navigation” specification addresses the standardization of complex navigation and sequencing strategies. Another interesting example is that of IMS “Learning Design”[2], which is targeted to model rich learning activities and their associated pedagogical considerations.

Nevertheless, progress in complexity and coverage of current specifications and standards contrasts with the lack of quality in the level of description of metadata records in existing learning object repositories, which are mostly fragmentary and unstructured, as reported recently in a study about the MERLOT repository (Pagés et al., 2003). Najjar’s study on the use of metadata in Ariadne (Najjar et al., 2003) also reported that most elements are either never or rarely used by learning object annotators. This study also points out that even those elements, for which values are more regularly provided, are used only in about a fifty percent of the total sum of cases evaluated, except for mandatory items.

In addition to the problem of completeness, current metadata schemas provide room for ambiguity and lack of precision. For example, the LOM standard (IEEE, 2002) – which is consensually considered to be the core of learning object descriptions – may lead to inconsistent usages, since it uses unstructured, natural language fields for many of its elements. For example, LOM category 4.6. “Other Platform Requirements” is aimed at describing information about software and hardware requirements that can not be expressed by the data element “4.4.Technical Requirement”, but it does not make available any value space or guidance about the expected values to be set. In this particular case, the learning object designer has, as the only help available, a pair of vague examples of “other platform requirements” such as “sound card” and “runtime X” (sic). This lack of a clear interpretation is in part due to inexistence of a complete set of consistent vocabularies, what makes most current metadata records unusable for the design and implementation of automated or semi-automated processes like learning object selection, composition or adaptation.

The main problem of LOM from the viewpoint of automation is that it is deliberately descriptive, rather than normative, with respect to the developer of software processes. Such a descriptive approach is useful for human communication, since human beings are able to understand and even disambiguate descriptions that could hardly be interpreted by current software systems (even though they are equipped with state of the art natural language understanding technology). But descriptive elements do not provide criteria to software systems to drive their actions. In other words, there is not a direct mapping from metadata values to LMS actions that could be used to implement standardized LMS behaviors. For example, how should the “language” metadata element be interpreted? Should it constrain LMS-initiated delivery to students that can proficiently “read” text in the specified languages? A notable exception for this kind of description approaches is the SCORM sequencing specification, which is written in a normative style, since it provides the details of LMS behavior for the user-content interaction. Nonetheless, the core of learning metadata elements is specified in a purely descriptive way.

The “descriptive orientation” cannot be considered as a defect of LOM as a standard, since it does not explicitly target consistent automated behavior as one of its objectives. But it certainly calls for supplementary techniques that fill the gap required to obtain LMSs that act consistently, not only for sequencing, but also for other kind of processes – e.g. composition – which would represent a significant step in standardization of e-learning content and systems.

This paper describes example metadata specification techniques – both for learning object and LMS process descriptions – in a normative style. Such or similar techniques should ideally be integrated with current standards to provide better support for learning management automation, and they would eventually remove the incompleteness and ambiguity of metadata records from annotation practices, by considering metadata completeness and precision as quality metrics for specific usages.

The rest of this article is structured as follows. First examine the current state of metadata standards, focusing on the role of LOM as a metadata communications system between learners and cataloguers. Then, requirements for normative metadata standards and their effects on learning objects are approached. Later, the runtime requirements for LMS processes are described. Finally, the conclusions derived from the previous sections are also provided.

LOM-Conformant Metadata as a Communications System

The e-learning community is defined in (Wason & Wiley, 2000) as a two-sided scenario where users and cataloguers communicate. While users discover learning objects, probably stored in public repositories, and make use of them in order to attain certain learning objectives, the work of cataloguers consists in tagging educational resources so that users can easily search, find and retrieve resources matching specific criteria. As their communications system is metadata, a consensus needs to have been reached on metadata terms, definitions and values before any fluent communication can start.

Nowadays, LOM has become the most significant and widely used communication system in e-learning applications dealing with Web-based educational resources in the form of reusable learning objects. As a descriptive standard, LOM enables cataloguers to provide metadata values on a number of different aspects, thus allowing users to decide whether a particular learning object is appropriate or not in order to reach a given learning outcome. LOM includes nine different categories covering all the current dimensions of learning objects, but as it is not a closed standard, it can be extended to host future dimensions, as structural, people, relational, etc (LTSC, 2004). Nowadays all the dimensions in LOM are only descriptive, in the sense that LMSs cannot unambiguously adapt or change their runtime behavior depending on the values in the metadata instances.

The lack of a strict formalization in LOM allows cataloguers to set very different values for the same dimension. This situation, and the fact that different cataloguers (or the same one at different moments in time), could provide different values for a given metadata element of a learning object, causes the user-cataloguer communication to be unclear. This is what Wason and Wiley refer to as “noise” in the communications system. Noise is a problem in analog communications that is considerably smaller in digital communications, since here only a set of discrete values can be transmitted. The noise problem in metadata records has been addressed in LOM through the provision of vocabularies that define a set of allowed values for (almost) each metadata element. Unfortunately, vocabularies are not available for all the dimensions in the LOM metadata space, they are not connected to commonsense knowledge representation, and a good number of elements can only be provided values through non-discrete descriptions in natural language.

If LMSs are to behave differently depending on the values of the elements in a metadata record, no uncertainty should ideally be allowed. The definition and use of vocabularies is a promising step towards the definition of precise metadata records, but it does not seem to be enough as to drive LMSs runtime behavior. In fact, ontologies have been recently proposed as substitutes for vocabularies providing richer context descriptions and enabling advanced behaviors – see, for example (Lytras et al., 2003). In addition, it is required that the metadata value establishes its degree of requirement (e.g. mandatory, optional, recommended and the like) and any additional information required like scores or parameters intended to be used by software to act according to them. Normative approaches to metadata and process specification are aimed at covering this latter problem, as described in the following sections.

Describing normative usage requirements and effects
for learning objects

When creating or adapting a given instructional material, learning content designers consider two essential elements as the drivers for the selection of style, interactivity and depth of the contents being developed, namely, the intended audience and the expected learning outcomes (Norman & Nicholson, 1999).

The description of the characteristics of the learner is addressed by “Educational” metadata in LOM, but it does not address fundamental data characteristics that are required if automated matching of learners to learning objects is needed. The following are some of these characteristics:

  • The intended LMS usage of some elements.

  • The degree of requirement for a description, i.e. whether it is mandatory or optional.

  • The degree of credibility for acquiring the expected knowledge or competencies after using the learning object.

  • The interpretation of some elements depending on their location in a conceptual representation.

Expected outcomes for a learning object can be of a diverse nature, depending on the effect that the object is intended to drive. Possibly the most common kind of outcome addressed by today’s learning objects is knowledge about some kind of subject, involving the development of mental structures. However, the development of abilities is often an objective by itself, and also competencies or social aptitudes (Lave & Wenger, 1991) can be the target of a given learning experience. For example, in a “role play” learning experience simulating a negotiation among different countries, e.g. The Versailles Experience described in (IMS, 2003), the negotiation process not only increases each learner’s knowledge on the objectives and aims of the rest of the participants, but also provides them with the ability of increasingly improving their negotiation skills. In this respect, even meta-cognitive goals may become the target of a learning object, as described in (Sánchez-Alonso & Sicilia, 2003a).

At the moment, LOM only covers the description of learning outcomes vaguely through elements like 1.5.Keyword, 1.6.Coverage and 9.1.Purpose. However, automation cannot be based only on the learning object expected outcomes as currently defined in LOM. For example, different learners with a different knowledge background could end up attaining different learning objectives after using the same educational resource. Let us consider a learning object on the genitive case in English including examples of advanced use and a final test. Such an object will be more easily assimilated by learners with a sound knowledge of English grammar, even though beginners can also benefit, at least in part, from its use. Current state of metadata specifications doesn’t allow learning content designers to clearly state the fact that different users will benefit differently, in terms of learning outcomes, from the use of such an object. Automatic systems or LMSs should consequently not deliver this object to different users on the premise that both will equally benefit from its use, since this is not always true. This situation introduces the need for normative elements that provide learning object designers and authors with the ability of defining standard machine-understandable learning object usage requirements and expected outcomes, which allow automatic or semi-automatic selection of the more appropriate resources depending on the learner’s background and other factors.

Normative approaches to learning object metadata should then provide a precise specification of the required behavior of a LMS with regards to each element. Learning object design by contract (Sicilia & Sánchez-Alonso, 2003; Sánchez-Alonso & Sicilia, 2003b) is a technique that approaches such normative effect from the viewpoint of contractual relationships between the learning object and the context in which a LMS uses it. This technique basically consists on stating, in the form of declarations called contracts, a collection of logical assertions on the requirements of use of a learning object and its expected learning outcomes. Using a recognizable syntax that facilitates automated processing, one or more contracts can be defined for each object, and published for the user community to know about it. Publishing more than one contract for a given object solves the problem of learners without a common knowledge background.

Learning object design by contract redefines the classic correctness formula {P} A {Q} meaning that “any execution of A, starting in a state satisfying P, will terminate in a state satisfying Q” and reformulates it as {C} RLO {O}[θ], to adapt it to the specificities of learning objects. The new meaning is that “the use of the learning object RLO in a learning context C (including a description of specific learner profile) is expected to facilitate the acquisition of the knowledge (or competence or abilities) O [to a certain degree of credibility θ]”. In short, a learning object contract looks like:

rlo <URI>

require

precondition1

precondition2

...

ensure

postcondition1

postcondition2

...

In this model, preconditions refer to learner profile prerequisites, augmented with platform and other technical and contextual requirements, learning object preconditions stating the constraints under which a learning object can be delivered and used. The syntax of preconditions in contracts, that uses the categories defined by LOM, supports placing information both on the category of the requirement: learner, context and system; and on the level of compromise of the requirement, which can take the values mandatory, recommended and optional. For example, a precondition stipulating that the system where an object will be delivered must be able to represent text written in Japanese, would be stated in its contract like this:

[mandatory] sys.language = jp

On the other hand, postconditions are expressed in a syntax that allows learning object authors to include learning outcomes corresponding to different LOM elements. Learning outcomes can be both represented as absolute knowledge attainments, like for example, in the following assertion:

lrn.knows(genitiveCaseEnglish)[80]

but also as relative to the previous state of the learner’s knowledge level, that is represented by ‘–1’. For example, in a simulation activity aimed at teaching emergency workers on how to handle radioactive waste, learner knowledge will increase every time the learner performs the activity:

lrn.knows(handleRadioactiveWaste) > lrn.knows(-1)(handleRadioactiveWaste)[90]

In the same way as information about sequencing of learning objects is not part of the metadata, yet introducing attributes that do not describe the content itself, formal information on both the requirements of use and the expected learning outcomes of a learning object could be added to the metadata records as normative attributes.

Other interesting aspect to think about is knowledge conceptualization. Any kind of normative approach to learning object description would ultimately require the presence of some kind of knowledge representations in order to enable richer behaviors than current linear lists of terms (vocabularies) as provided in LOM. Ontologies, understood as conceptualizations that provide an appropriate context for the interpretation of learning object metadata, can be used as:

  • A means for the representation of knowledge levels on the learner side.

  • A mechanism for the integration of learning object types, essential for the development of systems that are able to select and deliver learning objects. Previous work has addressed this aspect (Sicilia et al., 2004).

  • A way to provide reasoning facilities to LMSs, enabled by the underlying description logics (Baader et al., 2003).

This also provides results appropriate for representing postconditions in learning object contracts.

Summing up, a combination of normative descriptions with terminological knowledge representations can be used as the basis for extended learning object metadata specifications to enable a higher level of consistent automation.

Describing the run-time requirements for LMS processes

The learning processes described so far entail a content-learner (or learner-learner) setting, which can be considered as the “end” process of any LMS. But a high level of automation for learning systems would also expand to other areas that are not constrained to learner participation. In a broad, organizational view of a LMS, it should begin its functioning by some kind of materialization of the “learning needs” of the organization (which is often referred to as “knowledge gap”). Such needs may come from future projects or expected technological changes inside a company, or be part of a formal curriculum. These needs would trigger search processes and selection processes of learning objects. Such selection may involve external providers (ideally, automated learning object repositories) as well as other stakeholders or systems. Several levels of “intelligence” can also be defined to target learning objects and their delivery to the characteristics and time constraints of the employees.

In the broad view of e-learning described here, final delivery and sequencing of learning objects is only a part of the whole process. Standardization should expand its focus to the other “hidden” part of the value chain (Lytras et al., 2002). Much can be borrowed from current B2B specifications like OAGIS[3] or RossetaNet[4], since many learning processes can be considered as business processes.

The notion of “semantic conformance profile” (SCP), described in (Sicilia et al., 2004b), is a recent proposal for definition of learning processes in a broad sense, integrating the ideas of learning object design by contract and pointing to the use of ontological structures as an integral part of definition of processes. For example, the following table summarizes the main elements of a learning object composition profile (CMP-1).

Participants

Metadata

Run-time
Pre-requisites

Run-time Commitments

Required Elements

Idioms

The LMS

A collection of candidate learning objects {LOi}

LOM (9) Classifications

Content separation

 

a) Domain ontology connection with subsumption and part of relationships

b) Independence

a) Appearance merging.

b) Semantic coherence

c) Metadata coherence

Matching
Algorithm A.

The CMP-1 profile is intended to merge learning objects according to their classification inside taxonomic structures describing their contents. The participants are the LMS making the composition and a collection of candidate learning objects. The presence of LOM Classifications metadata is required, but in addition, such classifications should be connected to an ontological structure which at least represents subsumption (inheritance, “is-a” relationships) and “part-of” references. For example, “multi-dimensional arrays in Java” are a part of the subject of “arrays in Java”. These relationships are used in a concrete way described by Algorithm A (which is out of the scope of this paper), so that the behavior of the LMS is an explainable consequence of the annotations regarding Classifications. In addition to that, the learning objects being composed are required to have other properties to be composed together:

  • Their contents should be separated from their presentation, so that a given form of “appearance merging” can be done by the LMS. This requirement can be stated in terms of the obligation to use style-sheets.

  • The learning objects being composed should be stand-alone (independent), thus not requiring the recursive propagation of the process to other, dependant learning objects. This is done to keep the profile definition simple, and other more complex profile can be defined to define standardized recursive composition in the future.

  • Semantic and metadata coherence are required. While semantic coherence can be stated in terms of logical properties of consistence, metadata coherence is more difficult to characterize. For example, the difficulty levels or semantic densities for the learning objects being composed should be compatible (except in the case that differences are explicitly required, but this is not covered in CMP-1).

This way of describing processes internal to LMSs should also be complemented by the definition of a common pattern of messages exchanged between the participants (like they are specified for example, in OAGIS) in case that more than one system is involved, for example, in learning object search, retrieval or publishing processes. In addition, processes are “composable”, in the sense that they can be joined together to form more complex ones. For example, CMP-1 combined with targeting learning objects to specific users (U-SEL) and with a “search into learning object repositories” profile can be considered as a basic profile that fulfills a given learning need inside an organization.

Notations like that of Semantic Conformance Profiles complement normative approaches to metadata with normative description of learning processes of a diverse kind, broadening the scope of current learning technology specifications to the area of system integration.

Conclusions

Normative approaches to describing learning technology standards and specifications provide the required support to build automated or semi-automated software dealing with diverse aspects of the management of Web-based learning experiences. This is due to the fact that they are oriented to implementers of LMSs that behave in a concrete, predictable way. Since the current basic metadata schemas for learning objects are mostly descriptive, new techniques to complement them in normative styles are required. Learning object design by contract and semantic conformance profiles are two examples of normative techniques based on existing learning technology specifications. The former interprets basic metadata in terms of required conditions and expected outcomes for a learning object. The latter is concerned with a broader view of e-learning. It provides a technique to advance in the normative specification of diverse learning object management processes with a flexible way to specify different levels of complexity and “intelligence” in LMS behavior.

References

Anido, L. E., Fernández, M. J., Caeiro, M., Santos, J. M., Rodríguez, J. S., Llamas, M. (2002). Educational metadata and brokerage for learning resources. Computers & Education, 38(4), pp. 351 – 374.

Baader, F., Calvanese, D., McGuinness, D., Nardi, D., Patel-Schneider, P. (eds.) (2003). The Description Logic Handbook. Theory, Implementation and Applications. Cambridge.

Carey, T., Swallow, J., Oldfield, W. (2002). Educational Rationale Metadata for Learning Objects. Canadian Journal of Learning and Technology, 28(3) Fall / automne, 2002.

IEEE Learning Technology Standards Committee (2002). Learning Object Metadata (LOM), Final Draft Standard, IEEE 1484.12.1-2002.

IMS Global Learning Consortium, Inc. (IMS) (2003). IMS Learning design best practice and implementation guide.

Najjar, J., Ternier, S., Duval, E. (2003). The Actual Use of Metadata in ARIADNE: an Empirical Analysis. ARIADNE Conference 2003.

Lave, J., Wenger, E. (1991). Situated Learning: Legitimate Peripheral Participation. Cambridge University Press, Cambridge, UK.

LTSC. Joint session with CEN/ISSS Learning Technologies Workshop meeting proceedings, Madrid, 27 January 2004.

Lytras, M., Tsilira, A., Themistocleous, M.G. (2003). Towards the semantic e-Learning: an Ontological Oriented Discussion of the new research agenda in e-Learning. In Proceedings of the Ninth Americas Conference on Information Systems, Tampa, Florida.

Lytras, M., Pouloudi, A., & Poulymenakou, A. (2002). Dynamic e-Learning setting through advanced semantics : The value justification of a knowledge menagement oriented metadata schema. International Journal of e-Learning 1(4), pp. 49-61.

Norman, S., Nicholson, M. (1999). Designing in a Structured World: Instructional Design and Course Development within an SGML/Structured Information Environment - The Open School Experience. In Proceedings of ED-MEDIA 99, World Conference on Educational Multimedia, Hypermedia & Telecommunications, pp. 1448-1449.

Pagés, C., Sicilia, M.A., García, E., Martínez, J.J., Gutiérrez, J.M. (2003). On The Evaluation Of Completeness Of Learning Object Metadata In Open Repositories. In: Proceedings of the Second International Conference on Multimedia and Information & Communication Technologies in Education (m-ICTE 2003), pp. 1760-1764.

Sánchez-Alonso, S., Sicilia, M. A. (2003). Expressing meta-cognitive pre- and post-conditions in learning object contracts. IEEE Learning Technology Newsletter, 5(4),
pp. 7-10.

Sánchez-Alonso, S., Sicilia, M. A. (2003). Expressing preconditions in learning object contracts. In Proceedings of the Second International Conference on Multimedia and Information & Communication Technologies in Education, pp. 1656-1660.

Sicilia, M.A., Sánchez-Alonso, S. (2003). On the concept of learning object "Design by Contract". WSEAS Transactions on Systems, 2 (3), pp. 612-617.

Sicilia, M.A., García, E., Sánchez-Alonso, S., Rodríguez, E. (2004). Describing learning object types in ontological structures: towards specialized pedagogical selection. In Proceedings of ED-MEDIA 2004 - World conference on educational multimedia, hypermedia and telecommunications.

Sicilia, M.A., Pagés, C., García, E., Sánchez-Alonso, S. (2004). Specifying semantic conformance profiles in reusable learning object metadata. In Proceedings of the 5th International Conference on Information Technology Based Higher Education and Training: ITHET 2004.

Wason, T., Wiley, D. (2000). Structured Metadata Spaces. Journal of Internet Cataloging, volume 3 (2/3), pp. 263-277.


About the Authors

Salvador Sanchez-Alonso obtained a university degree in Computer Science from the Pontifical University of Salamanca (Spain) in 1997. He worked as an assistant professor at the Pontifical University of Salamanca in Madrid from 1997 to 2000 and from 2002 to 2005. He also worked as a software engineer at a software solutions company during 2000 and 2001. From 2005, he is a professor of the Computer Science Department of the University of Alcala, in Spain. Currently finishing his PhD thesis on the design and use of contract-based learning objects, his research interests include Learning objects reusability, Metadata, Object-Oriented Technologies and Web Engineering.

E-mail contact: salvador.sanchez@uah.es 
 

Miguel A. Sicilia obtained a university degree in Computer Science from the Pontifical University of Salamanca, Madrid, Spain, in 1996 and a Ph.D. degree from the Carlos III University in 1999. From 1997 to 1999 he worked as an assistant professor and later on as a part-time lecturer at the Computer Science Department of the same university. He also worked as a software architect in e-commerce consulting firms. From 2002 to 2003 he worked as a full-time lecturer at the Carlos III University, after which he joined the University of Alcalá. His research interests are primarily in the areas of adaptive hypermedia, learning technology, and human-computer interaction, with a special focus on the role of uncertainty and imprecision handling techniques in those fields.

E-mail contact: msicilia@uah.es
 

Both authors can be contacted at:
    University of Alcalá, Computer Science Department,
Ctra.
    Barcelona km 33.6 – 28871 Alcalá de Henares (Madrid), Spain

End Notes:

[1] http://www.adlnet.org

[2] http://www.imsproject.org

[3] http://www.openapplications.org/

[4] http://www.rosettanet.org/


 

go top

Mar 2005 Index

Home Page