Educational Semantic Web. ISSN:1365-893X

Abstract: By 2020, microprocessors will likely be as cheap and plentiful as scrap paper, scattered by the millions into the environment, allowing us to place intelligent systems everywhere. This will change everything around us, including the nature of commerce, the wealth of nations, and the way we communicate, work, play, and live. This will give us smart homes, cars, TVs, jewellery, and money. We will speak to our appliances, and they will speak back. Scientists also expect the Internet will wire up the entire planet and evolve into a membrane consisting of millions of computer networks, creating an "intelligent planet." The Internet will eventually become a "Magic Mirror" that appears in fairy tales, able to speak with the wisdom of the human race. Michio Kaku, Visions: How Science Will Revolutionize the Twenty-First Century, 1998 If the semantic web needed a symbol, a good one to use would be a Navaho dream-catcher: 
 a small web, lovingly hand-crafted, [easy] to look at, and rumored to catch dreams; but really more of a symbol than a reality. Pat Hayes, Catching the Dreams, 2002 Invited Commentary: Sims, R. (2004) Commentary on: Stutt, A. and Motta, E. (2004). Semantic Learning Webs Editors: Terry Anderson and Denise Whitelock.


Peering myopically into the future
Though it is almost impossible to envisage what the Web will be like by the end of the next decade, we can say with some certainty that it will have continued its seemingly unstoppable growth. Gi ven the investment of time and money in the Semantic Web (Berners-Lee et al., 2001), we can also be sure that some form of semanticization will have taken place. This might be superficial -accomplished simply through the addition of loose forms of meta-data mark -u p, or more principled -grounded in ontologies and formalised by means of emerging semantic we b s t a n d a rds, such as RDF (Lassila and Swick, 1999) or OWL (Mc Guinness and va n Harmelen, 2003). W h a t e ver the case, the addition of semantic mark-up will make at least part of the Web more readily accessible to humans and their software agents and will facilitate agent intero p e r a b i l i t y.
If current re s e a rch is successful there will also be a plethora of e-learning platforms making use of a varied menu of reusable educational material or learning objects. Fo r the learner, the semanticized Web will, in addition, offer rich seams of dive r s e learning re s o u rces over and above the course materials (or learning objects) specified by course designers. For instance, the annotation registries, which provide access to m a rked up re s o u rces, will enable more focussed, ontologically-guided (or semantic) s e a rch. This much is already in development. But we can go much furt h e r. Se m a n t i c technologies make it possible not only to reason about the Web as if it is one extended knowledge base but also to provide a range of additional educational semantic web services such as summarization, interpretation or sense-making, s t ru c t u re-visualization, and support for argumentation.

. 1
The nature of knowledge on the semantic web Marshall and Shipman (2003) argue that "The second perspective, the one that finds the Semantic Web in the role of a true Knowledge Na v i g a t o r, seems out of reach for both theoretic and pragmatic re a s o n s" (p. 65). By a true Knowledge Navigator they mean a "n e t w o rk of knowledge which can be used by personal agents ' (p. 59). T h e y object on the grounds that this relies on what they call decontextualized know l e d g e re p resentation. Hence, they say (p. 65) "at least in the short term -[there may be] many semantic webs rather than The Semantic We b". Mo re precisely there will be a multiplicity of community-based Semantic Learning Webs (SLWs) each with its ow n , perpetually changing ontologies, knowledge bases, repositories and ways of making sense of the world. That there will be more than one such community need not cause alarm. As writers such as Star and Griesemer (1989) have shown, our intellectual Semantic Learning Webs.

. 2 Semantic Learning Webs and Knowledge N e i g h b o u r h o o d s
Like any other semantic web re s o u rces, SLWs will depend on three things: annotated educational re s o u rces, a means of reasoning about these, and a range of associated s e rvices. Note that since one possible service is the automated identification of suitable material in un-annotated documents -for instance by using i n f o rm a t i o n e x t ra c t i o n techniques customised for the semantic web (Va r g a s -Vera et al., 2002; C i r a vegna et al., 2002), annotated materials (or learning objects) may not be strictly n e c e s s a ry. One important class of tools here are Semantic Brow s e r s, such as Ma gp i e (Dzbor et al., 2003), which use ontologies to identity important concepts in a document and provide access to re l e vant material. We use the term K n owledge Chart s for the ontologically permeated re p resentations of a community's knowledge or point of view produced and used by these tools. They are an important new re s o u rce for learning. We refer to the active processing of them as K n owledge Na v i g a t i o n.
Our notion of Knowledge Navigation is not intended as a 't ru e' Know l e d g e Navigator in Marshall and Sh i p m a n's sense. Knowledge Charts are much more context specific re p resentations and we use Knowledge Navigation only for the p rocess of traversing these (localized) re p re s e n t a t i o n s .
Briefly put, Knowledge Neighbourhoods are locations on the Web where communities collaborate to create and use re p resentations of their knowledge -K n owledge Charts. These are browsed -in a process we call Knowledge Na v i g a t i o n -using Semantic Browsers. They are constructed, communally, using Se m a n t i c C o n s t ructors. In the rest of this paper we will say more about the notions of Semantic Brow s e r s / C o n s t ructors, Knowledge Navigation and Know l e d g e Neighbourhoods. We will also examine the current state of the technologies needed to produce these, suggest ways in which they can be extended and combined and illustrate how they will act together to provide SLWs. Be f o re we do so we will say a bit about the pedagogic and practical needs of learners as they seek to interpret and navigate through the future We b. We will then say something about the pedagogic Semantic Learning Webs.  Journal of In t e r a c t i ve Media in Education, 2004 (10) Page 4 uses of argumentation and how it lends itself to support from semantic technologies. We will end with a brief discussion of how, armed with these new semantic tools, learners may be capable of becoming robustly critical thinkers, able not only to move easily through the surfeit of information sources but also to examine and critically assess the varied religious, scientific, economic, ethical and political claims and counter-claims which find fertile ground on the We b.
The aim of this paper, then, is not to predict or warn but to suggest how some c u r rent developments might be used to produce learning environments which will e xcite rather than control learners, and which will make them into responsible but critical members of civil society. T h a t's not to say that one should be unduly optimistic. The trend over the past few years has been tow a rds a view of learning as skill acquisition. This is even truer of online learning than traditional learning. Ma n y companies view e-learning as a part of their knowledge management systems. It p e rforms the role of transmitting the knowledge necessary to the company's surv i va l or success from its knowledge repositories to the members of its work force. Sa d l y, e ven governments tend to look at learning more and more in terms of information transfer and work -related skill acquisition. As Brown and Duguid (2002) say: "T h e idea of learning as the steady supply of facts or information, though parodied by Dickens 150 years ago [in Ha rd Times], still pre vails today. Each generation has its own fight against images of learners as wax to be molded, pitchers to be filled, and slates to be written on" (p. 135). Or, we might add, passive recipients of managed k n owledge. We can and should do better, as there is a great thirst for learning out t h e re. We can only hope that we don't stifle it with our learning enviro n m e n t s , semantic or otherw i s e .

Problems and opportunities for learners
Tempting as it is for technologists, let's start not from what the technology is likely to give us but from what the learner needs. We can do this at many different leve l s . Pr a g m a t i c a l l y, what learners need are things like time to study, a quiet place to study, tips on how to pass examinations, peer to peer and learner to tutor communication, someone to talk to when problems arise, and help with reading and interpre t i n g documents (taken as including written, graphical, audio and video sourc e s ) .
At a more abstract level what problems will the learner face? We note the follow i n g : information overload, and information authentication. Gi ven the amount of information available (as well as the possibilities inherent in the technology) learners could also benefit from increased customization or personalization.
Semantic Learning Webs.  In f o rmation ove rl o a d. Most of the problems which come readily to mind are to do with finding a path though the stupendous amount of information curre n t l y a vailable not only on the Web but in books, newspapers, television and films -not to mention the input from texting buddies (see Nelson, 1994, for one discussion of this phenomenon). Google currently indexes over three b i l l i o n web pages. Even 20 years ago the student learnt mainly from written material in the form of journal material or textbooks, accessed as photocopies or by visiting a library. Now, much of the journal material is available on the World Wide Web and can be accessed online ( p roviding that the student's institute has a subscription). Even books, and especially e x p e n s i ve monographs, are to be found online. Ap a rt from these digitized versions of p reviously available sources, the learner is now faced with a bewildering barrage of information intermediaries. It is too easy to overstate this howe ve r. Learners have always been faced, at least from the invention of the printing press, with a large array of material. Indeed the ability to sift and evaluate information is one of the skills which a student must acquire. Nonetheless, it is this perc e i ved mass of information which has led to many of today's most visited sites -search engines such as Go o g l e -and motivated some of the basic Semantic Web tools -semantic or ontologybased search. As an example of the latter, the On t o Broker system (Decker et al., 1999) provides an ontology-based crawling and answering serv i c e .
In f o rmation authentication. As well as more conventional indexing facilities prov i d e d by librarians and archivists, there now exist web sites where information is collated, i n d e xed and sometimes stored by a variety of actors some of whom are informal digital archivists but with similar standards of objectivity. Others, howe ve r, have an a xe to grind. Most alarmingly sites exist which present themselves as impart i a l re s e a rch conduits when in fact they are funded by commercial and other intere s t s . For example, the number of sites surrounding the issue of genetically modified organisms is immense and not a few may be indirectly or directly funded by companies with large stakes in the future of the technology.
Customization of course/information discove ry. Gi ven the amount of information a vailable, the problem of matching learner to material, which is re l e vant to his or her needs at a particular point in time, becomes more and more acute.
In f o rm a l / f o rmal support. A significant factor in any learning context is access to the s u p p o rt networks that learners need in order to be able make sense of the material which they can all too readily access. In the traditional classroom or lecture hall this is usually a human being appointed by the institute for the purpose of fielding questions. In other situations this might take the form of a tutor or mentor (who may Semantic Learning Webs.  Journal of In t e r a c t i ve Media in Education, 2004 (10) Page 6 only be slightly more experienced than the learner). This a p p re n t i c e s h i p mode of learning extends beyond trades such as plumbing and carpentry into seemingly abstract 't r a d e s' such as mathematics, as Brown, Collins and Duguid (1989) have s h own. Of greatest importance howe ver are the informal, face to face or online, g roups of learners who master a subject by asking questions howe ver trivial, by offering each other differing interpretations of more formal material, by testing these and by negotiating a consensus interpretation -in other words by c o n s t ru c t i n g a piece of knowledge. We say more about constructivism in Stutt (1997). These gro u p s , usually of a few students in face-to-face situations (including more formal seminars and tutorials), may become much larger in the online world. He re, informal gro u p s or v i rtual communities of learn i n g (see Stutt et al., 2002) may number hundre d s .

Learning and cognition
While the above re p resent the practical problems and needs of learners, it is possible to give an even more abstract account of learner needs which we can use to guide our thoughts about future learning environments. At a more cognitive level students need e n v i ronments which are congruent with what goes on in learning. From what we have said already we can distinguish between three types of learner needs: for s t ru c t u re, re l a t e d n e s s and i n t e r p re t a t i o n. These correlate more or less with the first two items in L a u r i l l a rd's (2002) characterization of learning as: a p p rehending stru c t u re, integra t i n g p a rts, acting on the world, using feedback, and reflecting on goals.

. 1 S t r u c t u r e
As Laurillard indicates, one central component of learning is coming to see stru c t u re . As the Web grows this ability will become even more important. Unless the learner can find a way to successfully navigate through and filter out irre l e va n c y, it will be m o re or less impossible to make use of the rich re s o u rces available on the we b. In our v i ew there are three main stru c t u res which can be used here and all of them can be aided by the use of ontologies: argumentation/debate, narra t i ve and a n a l o g y.
Debate includes the various scientific c o n t rove r s i e s which arise about notions such as continental drift, GM technology, global warming. These controversies are in t h e m s e l ves multi-dimensional since they often have ethical and economic/political aspects as well as scientific.
Na r r a t i ve here includes the historical narrative of how ideas change and evo l ve as we l l as the 's t o r i e s' we tell as a means of making sense of something (e.g., the story of the rise and fall of working class politics in the UK).
Semantic Learning Webs.  Journal of In t e r a c t i ve Media in Education, 2004 (10) Page 7 Analogies are an important factor in how we make sense of something. Learners can better understand something (e.g., the stru c t u re of the atom) if they can relate it to something already well understood (e.g., the planetary system). Re s e a rchers, too, often make use of analogies in making a case for something. For example, interpretations of the Anasazi culture in the American So u t h -West often rely on analogies with practices among modern Pueblo peoples.
We re c o g n i ze that in this paper we are mainly dealing with a particular kind of learning -theoretical knowledge acquisition -in which stru c t u re and interpre t a t i o n a re important. In other kinds of learning where skills and competence are more i m p o rtant (e.g., learning a language) other stru c t u res are also needed. Note, howe ve r, that at least argumentation (especially when the learner participates) has a skill-based c o m p o n e n t .
While we fore g round debate, narrative and analogy, other ontologies, for example, for the set of fundamental theories about a domain are also important -but less easy to capture. For example, cosmology appeals to a limited set of principles in it i n t e r p retation of events in the Un i verse. In other cases we may need causal models or simulations. This is true of disciplines such as ecology where it is important to see the how a model of an ecosystem changes over time.

. 2 Relatedness
Pa rt at least of the importance of stru c t u re is that it is a means of seeing something (a theory, concept, equation) as a whole. Equally important are the relations which link these to other ideas and theories. Both relate to Laurillard's recognition of the need for the integration of part s .

. Interpretation
The learner needs to be able to take a segment of learning material and situate it in a multidimensional space which includes at least: the scholarly, social, economic and political context (an obvious example here is the GM debate); its place in the metan a r r a t i ve of advancement a science tells itself (e.g., Newton supersedes Copernicus, Quantum Physics finesses Newton); and its role in an ongoing debate or conve rsation among academic stakeholders (e.g., the discove ry and articulation of plate tectonics served to refute the arguments of those who contested We g e n e r's claims about continental drift; the discove ry of Arc h a e o p t e ryx and the more recent Chinese f e a t h e red dinosaurs supports the notion that birds are dinosaur offspring).

. 4 Summary: how the semantic web can help addressing learning needs
In what follows we develop a vision of the use of the Semantic Web where : • Learning is contextualized to separable locations in the Semantic We b (i.e., it is community related rather than generic); • The stru c t u re of pieces of knowledge is given by Knowledge Charts and their Navigation tools; • The charts re p resent stru c t u res (such as narratives, arguments and analogies) using ontologies and provide access to them using graphical re p re s e n t a t i o n s ; • Relatedness is given by these objects and by the links they provide to f u rther learning re s o u rces; • In t e r p retation is facilitated by the contextual knowledge these objects p rov i d e .

The importance of argument
While it is likely that many different kinds of stru c t u re (and structuring tool) will be d e p l oyed, we will concentrate in this article on those based on argumentation and scientific controve r s y. This is for a variety of reasons.
a. Our exposition will be more focussed there by.
b. Argumentation is already the subject of much re s e a rch as the basis for learning. For instance Kirschner et al. (2002) and Andriessen et al. (2003) both contain e x t e n s i ve collections of papers on the important role of argumentation (and its visualization) in learning. Indeed an attention to critical thinking (both the analysis and construction of arguments) has long been an important component in a we l l rounded education, as books such as that by Fisher (1988) indicate.
c. Our re s e a rch in the Knowledge Media Institute has resulted in a variety of argument-based tools not least the D3E discussion tool (Sumner and Bu c k i n g h a m Shum, 1998), which provides the discussion spaces for this and other issues of JIME! Mo re import a n t l y, the ScholOnto project -see below for more details, is active l y engaged in producing the ontologies and tools needed for the construction of a semantic scholarly web (Buckingham Shum et al., 2000).
d. Fi n a l l y, if as is our view, learning should be clearly distinguished from training, then one means of doing so is by ensuring that learners are provided with the tools for accessing the claims and counter-claims of a range of voices which make up a d i s c i p l i n e's conversation with itself as well as the means for making a contribution to this ongoing conversation. Information can't be understood unless its historical context and the dialectic which produced it is understood. Thus the learning pro c e s s is as important as the product.

What exactly is the Semantic We b ?
We might as well ask what the World Wide Web is. For the explorer it is a means of communication with home-base, for the CEO a way to sell more product, for the teacher a new way to teach. Abstract definitions in terms of computer networks and h y p e rtext languages simply do not capture the range of meanings we can apply to the phrase "World Wide We b". Si m i l a r l y, we can define the Semantic Web abstractly as an extended World Wide Web where content bears it own description so that applications as well as humans can make sense and use of it. We can talk about o n t o l o g i e s as systems of concepts, pro p e rties and relations which can be used in these descriptions, about the n o t a t i o n s or l a n g u a g e s for re p resenting these, about a n n o t a t i o n as the means of adding descriptions to web pages, about p o p u l a t e d o n t o l o g i e s or k n owledge bases which capture the knowledge contained in web pages, of a g e n t s as computer programs which can reason about these descriptions in order to c a r ry out specific tasks, about s e rv i c e s which coordinate multiple agents and other s e rvices as a means of satisfying human needs. But this only gives us the technical i n f r a s t ru c t u re. As we can clearly see from Berners-Lee et al. (2001), the Se m a n t i c Web is an aspiration, a vision of what might be done, a dream even of a future we b which becomes more amenable to human needs by relinquishing much of the hard w o rk of publishing, locating and retrieving web content to automatic processes. Like any other dream, the Semantic Web is contested. T h e re are some who suggest that the c u r rent generation of web-based applications are too entrenched for such a nove l a rc h i t e c t u re to succeed (e.g., Ewalt, 2002). Yet others suggest that the re p re s e n t ational infrastru c t u re is unnecessarily complicated (Ha yes, 2002).
Note that while it often seems from accounts of the Semantic Web that the va r i o u s ontologies (debate, analogy, narrative, domain) are of the greatest importance, in fact Semantic Learning Webs. (2004) the Semantic Web will not be fully re a l i zed until a range of applications is built on top of these ontologies. Thus it is likely that semantic web services, where agents seek out suitable web services using semantic descriptions, and more part i c u l a r l y educational semantic web services will be most important in the next ten years. T h e ontologies provide the interoperable data while the services do the interesting things (like supporting argumentation).

What current research on the Semantic Web and eLearning offers
Without going into a great deal of detail we can describe current re s e a rch on learning and the (semantic) web as being centrally concerned with so-called l e a rning objectsor separable units of educational material which can be combined and reused in a variety of contexts. Central to their reusability are the descriptions which their designers provide using a variety of metadata schemes. Cu r rently there are a number of standards here but it is likely that this number will be reduced with some s t a n d a rds combining and others being discarded as commercial and other pre s s u re s come into play. Anido et al. (2002) provide a good ove rv i ew here.
Another development has been the growth in educational repositories and peer-topeer networks for sharing these. One example here is the Edutella network (Nejdl et al. 2002). At the same time as the means of sharing these objects has developed, work has also proceeded on adding detail to the metadata schemes in order that part i c u l a r learning goals, object sequences, roles and activities (in short, a pedagogy) can be specified. Wo rk on the Educational Modelling Language is key here (Ko p e r, 2000; 2 0 0 1 ) Most of this has been accomplished without the use of explicitly semantic technologies. Howe ve r, a natural development of the repositories and networks is the notion of ontology-based brokerages which match learners with learning materials (Anido et al., 2002) and course construction tools (St o j a n ovic et al., 2001 ) which attempt to automatically combine learning objects into "c o u r s e s" or sequences of objects. Fi n a l l y, more recently we have seen the development of educational semantic web services. An example here is the Sm a rt Space for Learning approach using the Elena mediation infrastru c t u re (Simon, et al., 2003.). The services here range fro m assessment, to short lectures, courses and degree programmes.
T h e re are three main points here. Fi r s t l y, none of these technologies has reached full maturity as yet or been deployed widely so it is hard to gauge their success or failure.
Semantic Learning Webs. (2004) Se c o n d l y, the Semantic Web technologies depend for their success on the viability of the strategy of depending on reusable atomic learning components, on the possibility of capturing the characteristics of these in formal descriptions using metadata schemes and the widespread acceptance of these descriptions and, finally and most i m p o rt a n t l y, on the likelihood that there will be enough incentive for learning object p roviders or others to annotate their products with the accepted metadata. T h e re h a ve been numerous criticisms of the whole project of e-learning (Dreyfus, 2001), of the suitability of Learning Objects outside an individualistic pedagogic model -one m o re over based on the needs of military trainers - (Friesen, 2003) and of the adequacy of the metadata used for descriptions (Friesen, 2002;Nilsson et al, 2002).

Stutt & Motta
T h i rd l y, while, Nejdl and his colleagues have advanced the notion of Ed u c a t i o n a l Web Se rvices, these are still based largely on Learning Objects: "Educators and (semi-) automated tutoring systems compose learning services out of learning objects and other educational re s o u rc e s" (Simon et al. 2003). To this extent, their project may fail if Learning Objects fail to delive r. Howe ve r, it seems that their arc h i t e c t u re may be generic enough to make use of any properly annotated World Wide Web content. It is also true to say that their educational services do not differ from those prov i d e d by educational institutions in face-to-face contexts. It is possible to envisage a form of educational service which can only be provided by the Semantic We b. For example it is possible to foresee a service which automatically re -c reates the chain of re a s o n i n g used in discovering, say, the cause of the SARS outbreak, using unannotated We b pages.
While much effort has been expended on Learning Objects and Learning Ob j e c t Repositories, from the Semantic Web perspective the ontological commitments of the various contending metadata schemes (LOM, SCORM, etc.) are limited. In essence the metadata is intended to describe a learning object in sufficient detail for a human or other agent to be able to select it as appropriate in some learning context. In addition metadata schemes can be used to configure or sequence a learning object as p a rt of some overall course.
What is lacking are any tags which can be used to indicate to a learner how the learning object may be contextualized. That is to say that there are no ontological relations in the learning object description which can indicate how an object should be interpreted, or how it fits into the central debates in the field. Faced with the c u r rent state of affairs a learner can successfully navigate the space of possible learning objects but cannot navigate the space composed of the far more import a n t s t ru c t u res of relations which knit topics, concepts, examples and so on into the fabric of the disciplinary field.
For example in Earth/Climate Science there is much controversy about the notion of global warming. While most scientists accept that global warming is a reality and that it is caused by increased anthropogenic CO 2 emissions (as concluded by the In t e r g overnmental Panel on Climate Change) there are others who either dispute the cause, the extent or the reality of global warming. For instance Lomborg (2001) has cast doubt on the quality of the IPCC models and suggested that the costs of limiting C O 2 emissions far outweigh the benefits. In turn others have disputed the case Lomborg makes.
Another example (Frankel, 1987;Stutt and Shennan, 1990) is provided by the various moves in the argument originating in We g e n e r's theory of continental drift. Indeed as Frankel shows, this sort of academic debate can be seen in terms of a variety of strategies. One of We g e n e r's claims was that there was once (200 million years ago) a continent, called Pangea by scientists, which broke up forming the p resent continents as the parts moved away from each other. One of the main ways of arguing against this theory was by showing anomalies in We g e n e r's accountspecifically that the Pe r m o -C a r b o n i f e rous Ice Cap couldn't have formed if the whole of the Eart h's surface land was massed around the Po l e .
These controversies are characterized by hard-fought negotiations by political, economic and business stakeholders as well as the scientists. They rarely come to a complete conclusion though there may be resolutions, closures, and abandonment ( Mc Mullin, 1987).
A semantic service could be used as a means of browsing documents and, where a p p ropriate, accessing not only services such as glossaries, but also more wideranging services along possibly many dimensions. For instance a semantic system could indicate to the climate science learner that the concept of global warming is a 'hot topic' unlike say that of the cause of anticyclonic winds. Having thus alerted the learner to the importance of the topic, the system could present material relating to the ongoing debate on the global warming issue using a sophisticated graphical i n t e rface to aid navigation through the various competing arguments.

S c e n a r i o
In the following scenario we explore the possible affordances of a Semantic Learning We b. The scenario is intended to give a more concrete form to our vision of a future learning environment based on Knowledge Charts, Knowledge Na v i g a t i o n , K n owledge Neighbourhoods and Semantic Browsing. It is important to emphasize that, while this scenario is visionary, some of the details are derived from ongoing w o rk on the modelling of argumentation and semantic browsing -see the section on Realizing the Vision below for more details of this work . We can imagine that our learner is reading a web page/document/learning object on climate change as part of some course on environmental studies. While some mention is made of alternative and competing viewpoints this is not dealt with fully in the text. As she reads, our semantic system -let's call it SWEL for Semantic We b E-Learning -automatically highlights portions of the text which it can assist with, using technologies based on KMi's ClaimSpotter (Se reno, 2003) and Magpie (Dzbor et al., 2003). In this case SWEL can offer a way into the scientific debate about global warming and/or explain the scientific concepts invo l ve d .
The learner opts to access the scientific debate. SWEL provides a graphical interf a c e to the principal components of this debate. In Semantic Learning Webs.

Stutt & Motta (2004)
Fi g u re 1: A course fragment (from the Climateprediction.net site) and its associated we b of argumentation.
Semantic Learning Webs.

Stutt & Motta (2004)
The learner clicks on the "Lomborg Sceptical En v i ro n m e n t a l i s t" node. This opens up the node to provide a more detailed version of Lomborg's argument. Ba s i c a l l y Lomborg makes two points: (a) that the models used in the IPCC's calculations about the effects of CO2 emissions are inadequate and (b) the cost of reducing the w o r l d's emissions by the negligible amount Kyoto would attain far outweighs the b e n e f i t s .

Fi g u re 2: Lomborg's arguments
Since Lomborg's argument about models is based on a view of what statistical models can do, the learner can now opt to follow a link to either a description of statistical models or a deeper view of Lomborg's argument here .

Fi g u re 3: Mo re detailed version of Lomborg's first argument
And so on. At each point in the debate model, the learner can access the original documents of which the model is a summary, using semantic search. In order for these documents to be of maximum use to the learner, in the context of this k n owledge chart, the parts in the text which are re l e vant to these argument steps have been already semantically annotated, e.g., by a SWEL crawler using ClaimSp o t t e r -Semantic Learning Webs.

Stutt & Motta (2004)
Journal of In t e r a c t i ve Media in Education, 2004 (10) Page 16 like technology or by the author himself, using scholarly semantic annotation tools, such as ScholOnto (Buckingham Shum et al., 2000).
Fi g u re 4: One web document among many where Lomborg puts forw a rd his views. In this case from the spiked science web site.
Of course, any new document or Chart could have further Knowledge Chart s associated with it which the learner can pursue in turn. For instance, Lomborg might appeal to various economic models in his reasoning. The learner could now decide to f o l l ow up links to pages or meta-models on classical and ecology-based economic m o d e l s .

. 1 Our vision part I -Navigation of Knowledge Charts using Semantic Browsers
The key role of Knowledge Charts is to provide pathways through controve r s i e s , analogies and narratives and expositions of scientific principles. Indeed, given the p re valence of documents filled with domain content, the system we envisage stands or falls on the existence of these meta-models. A whole new discipline concerned with the production of Knowledge Charts for analogical, narrative and argumentational models may spring up though it is more likely that they will be crafted by members of a particular learning community. Knowledge Charts (such as the seve r a l l e vels of argumentation and scientific controversy in Fi g u res 1-3) differ fro m s t a n d a rd learning objects in that: they are built using ontologies, they include content (summaries), annotation and associated graphical re p resentations, they have a taxo n o m y, and, they are used both for navigation (viewed hypertextually) and i n t e r p retation (viewed conceptually). Knowledge Charts will probably be prec o n s t ructed in the first instance. Howe ve r, given the number of possible metalearning objects for any course component, it is likely that we will have to find a means of automating their construction. As already mentioned, work is alre a d y p roceeding on the use of human language technologies to support the extraction of argument stru c t u res from texts (Se reno, 2003; Va r g a s -Vera and Mo reale, 2003) and it is possible that this can be extended to search the Web components of the sorts of debates or controversies common in science (see section on Realizing the Vi s i o n b e l ow for more on this).
In order for the scenario to become a reality we also need a system which can perf o r m K n owledge Navigation (i.e., a Semantic Browser). We can rely on our Se m a n t i c Browser to identify important concepts in Learning Objects using domain ontologies (perhaps in combination with information extraction techniques, see Va r g a s -Vera et al., 2002) without demanding explicit annotation. The various traversals (from text to Chart, from Chart to Chart and from Chart to new material) rely either on explicitly expressed relations among the Charts (or Chart components) or on i n f e rencing made possible by the domain and stru c t u re ontologies. For example, the system could have a set of rules which allow linkages from argument nodes where theories are used to warrant particular claims to Charts which re p resent the theory. We might also envisage a situation where agents such as SWEL could construct new K n owledge Charts. This should be possible by reasoning from the system's know l e d g e base of available re s o u rces, as well as its internal model of the sorts of components needed in a particular kind of Chart (as given by its ontology).
Note that while the illustrations given above in Fi g u res 1-3 may indicate that K n owledge Charts are fixed, this is not so. Knowledge Charts reflect the points of v i ew of an individual, a group or a community and as this knowledge may change it will be necessary for the individual or community to update their Knowledge Chart s .
In summary, SWEL needs: an ontology of types of Knowledge Chart (debate, story, a n a l o g y, causal model); ontologies for each type (claim, ground, support, refute for argument; event, actor, relationship for narrative); knowledge bases re p re s e n t i n g these; pre-designed meta-learning objects, such as an argument with all the possible links or, given the number of available re s o u rces, an automatic means of population perhaps using human language technologies; and, finally, a means of expressing the pedagogic purpose of these Charts (i.e., an extension to EML).
The real conceptual and technical difficulties will arise when we try to work out how to keep track of the complex multidimensional sets of relations for debate, community stru c t u re, community roles, practices and so on. It is likely that the populated ontologies for domain contents (i.e., the contents of learning objects) will be immense. If we combine this with the populated ontologies for all the metamodels it is possible that this will grow exponentially since for each web page or learning objects, there are many possible Knowledge Charts each of which can have in turn its own possible links to other Charts.

. 2 Graphical representation for Knowledge Charts
Taking the example of a debate or scientific controversy-based Knowledge Chart, it is important here that it can be viewed at many different levels. We need a view which encapsulates all of the moves but which also shows some of the complexities.
T h e re should also be views which select sections of the argument or, indeed, draw together themes into a single view (even if this isn't initially presented by SWEL). Fo r example, the learner could select all the arguments made by particular part i c i p a n t s . At the most detailed level, the nodes should re p resent all the significant points made in an argument. Similar points could be made about other types of Chart .
A sophisticated graphical re p resentation is thus needed as a means of navigation by the learner. The sorts of multi-dimensional stru c t u res we envisage have much in common with hypertexts. They differ in that we do not consider the path through the various types of material as a single text. The course is the central 't e x t' with the various alternative re s o u rces, debate stru c t u res, narratives and analogies mere l y d i versions along the way.
In order to provide a graphical interface to the scientific debate outlined above , SWEL must be provided with or derive at least a set of d e b a t e m oves made by debaters linked by a set of rhetorical re l a t i o n s (i.e. a debate ontology). In this case the debaters are Lomborg, the Danish Ecological Council, and contributors to Scientific American. In essence there are two main debate moves: Lomborg's attack on the notion of global warming and the subsequent rubbishing of his book by a variety of debaters. In addition there is Lomborg's detailed riposte to the Danish Ec o l o g i c a l council. The relations are all attack or refute here but we can imagine a whole set of possible relations including support, restate, and confirm. In the case of a narrative , SWEL would have a different model made up of applicable ontology instances (e.g., a narrative has events, temporality, protagonists, plots, flash-backs etc.).

Fi g u re 5: A schematic space with a course (or sequence of Learning Objects) which may be linked by a Semantic Browser to various types of Knowledge Chart s
In Fi g u re 5 we see part of a notional course which has associated with each of the learning objects sets of alternative expository material (such as web sites or individual web pages re t r i e ved by semantic search), and Knowledge Charts for narrative s , d e b a t e s / c o n t roversies and analogies. Communities are composed of a variety of members who fulfil different roles and enter into a variety of relations with each other. Members can belong to more than one community or gro u p. For example a member may perform as a l e a d e r in a p a rticular community with relations such as s e t s -t h e -a g e n d a -f o r with other members. These roles and relations may change over time. These ideas we re developed during the Alice project  where they we re applied to a customization s e rvice for ecommerce. They are even better suited to a community oriented appro a c h to eLearning. Knowledge Neighbourhoods are composed of a variety of spaces: both public (for storing important documents, for debate; for publishing; for visualizing the community) and private (personal notes, calendars).

. 3 Our vision part II -Knowledge Neighbourhoods
Communication across the boundaries of the various communities can be achieved in two main ways: (a) communities with members who also belong to other communities can rely on these members' synthesizing and translation abilities; (b) b o u n d a ry objects can be used to effect communication. Leigh Star defines these as "objects which are both plastic enough to adapt to local needs and the constraints of the several parties employing them, yet robust enough to maintain a common identity across sites. They are weakly stru c t u red in common use, and become stro n g l y s t ru c t u red in individual use. These objects may be abstract or concrete. They have d i f f e rent meanings in different social worlds but their stru c t u re is common enough to more than one world to make them recognizable, a means of translation. T h e c reation and management of boundary objects is a key process in developing and maintaining coherence across intersecting social worlds." Mo re succinctly Arias and Fischer (2000) view them as "objects that serve to communicate and coordinate the p e r s p e c t i ves of various constituencies". In our view, Knowledge Charts are boundary objects: they are both abstract re p resentations and concrete entities; they are i m p o rtant for knowledge acquisition in particular communities but will be recognizable to at least closely related domains; and, the meanings they embody ( t h rough the ontologies they employ) and give access to, will va ry across domains.
The semantic web can support these communities in a variety of ways. Fi r s t l y, by p roviding ontologies for communities, community stru c t u res, roles, relations, spaces, topics, tasks, practices and so on, they can provide an accepted lingua franca for the community as a whole. And, since these neighbourhoods are re l a t i vely circ u m s c r i b e d , t h e re will be fewer problems in formulating, negotiating and accepting these ontologies than if we attempted to provide global ontologies. Se c o n d l y, the semantic web community can provide a range of semantic web services which ensure that the community is built, maintained and flourishes. Specific services can assist with community tasks, such as intelligent search for topic-related information.
K n owledge important to the community will be annotated with ontologies re l e va n t to the community. As we said in our introduction, it is unlikely that these ontologies will be generic -that in fact there will be a single Semantic We b. It is more likely that the Semantic We b, like Ancient Athens or Me d i e val It a l y, will be composed of loosely related Knowledge Ne i g h b o u r h o o d s .

. 4 Connecting the visions
While understanding may come about through awareness of ongoing scientific debate or controversy (vision I) it often re q u i res the learner to p a rt i c i p a t e in the chains of argumentation. Since only qualified peers will be able to enter fully into the c o n t rove r s y, practical experience in arguing these points comes about by means of discussions or debates among learners.
K n owledge Neighbourhoods are an important means of contextualizing Know l e d g e C h a rts. It is unlikely that learning object providers will be able to provide the sophisticated re p resentations needed for these. It is probably unlikely that a new class of d e velopers will arise (though this is possible) dedicated to their pro d u c t i o n . K n owledge Neighbourhoods will be provided with the necessary tools for designing and storing Knowledge Charts and, at the same time, the motivation for expending the necessary energy. Building Charts may become a community enterprise. We envisage that each learning community will have its own set of specialized Chart s although it is possible that some of these may be shared (as boundary objects) with other learning communities.
Tools such as Semantic Browsers can provide the mappings from current learning object to Knowledge Chart to alternative learning materials. Other tools will be needed which deploy ontologies in constructing Knowledge Charts and in mediating community formation and intra-community discussion.

9
Realizing the visions: technologies needed for SWEL

. 1 A technology-centred analysis
One approach to e-learning, which has much in common with our notion of K n owledge Charts and their Navigation, has been developed at the Royal In s t i t u t e for Technology (KTH) in Sweden by Na e ve and colleagues. Their Ga rdens of K n owledge (Na e ve, 1997) are learning environments which can be used to explore n e t w o rks of ideas. They are also developing (Na e ve et al., 2001) the idea of the Conceptual We b -a layer above the Semantic Web intended to make it more accessible to humans using graphical context maps which include concepts and relations among concepts. Conzilla is a concept browser which allows the user to navigate through a space of context maps to access associated content.
While the idea of graphical re p resentations of domain concepts and their navigation a re similar, our approach concentrates more on the elaboration of a typology of high l e vel re p resentations (of arguments, stories and so on) which can be used for navigation and sense-making. In addition, our Knowledge Charts are embedded in the social context of Knowledge Neighbourhoods and our Knowledge Navigation is p e rformed by a tool -the Semantic Browser -which can make the necessary mappings from learning objects to learning objects as well as provide a range of o n t o l o g i c a l l y -d i rected, community-oriented services such as automated argument c o n s t ruction.
We also need another tool -a Semantic Wr i t e r / C o n s t ructor -which can constru c t or assist in the construction of Knowledge Charts. Underlying these we need va r i o u s domain and str u c t u re -related ontologies (e.g., for argumentation) and an i n f r a s t ru c t u re for creating and publishing semantic web services.
While the complete framew o rk for Knowledge Navigation does not yet exist, enough of the necessary components are available for us to be confident that it could be a vailable within the next few ye a r s .

Domain ontologies.
Both the ontologies and the means of re p resenting them are a l ready available. For instance work is ongoing to devise a detailed ontology for the climate change domain.

Se rvice In f ra s t ru c t u re .
Re s e a rch on semantic web services is concerned with d e veloping infrastru c t u re, modelling and reasoning support to allow automatic d i s c ove ry, composition and execution of web services. The IRS-II infrastru c t u re  provides service designers with the means of registering their s e rvices both in conventional and semantic registries. It also provides an enviro n m e n t for creating applications configured out of a number of these more primitive serv i c e s .
Recently we have developed a demonstration system which can schedule operations a b road for patients with arthritis and needing urgent hip replacement surgery. T h e system relies on the availability of distributed web services, such as ye l l ow pages for hospital treatments, hospital scheduling systems, ambulance providers, and curre n c y c o n ve rt e r s .

Ontologies for argumentation-based Kn owledge Chart s.
As we have indicated a b ove, there is already extensive support for scholarly argumentation in the S c h o l Onto project (Buckingham Shum et al., 2000). In order to support d i s a g reement and conflicting perspectives in academic re s e a rch fields, we need tools which support the user in making sense of the relations among documents. T h e S c h o l Onto project is developing an ontology-based digital library server to support scholarly interpretation and discourse. Re s e a rchers can articulate their view of where a document fits in the ongoing academic conversation thus creating a semantic n e t w o rk of scholarly discourse. A tool -ClaiMaker -has been developed to model the rhetorical relations (proves, refutes, is consistent with, is analogous to, and so on) among claims in re s e a rch papers, publish these on a server and make queries about the relations and the documents containing them (see Li et al., 2002). While not intended primarily as a learning tool, both the access to a web of inter-re l a t e d scholarly papers and the opportunity to add further annotations (i.e., to extend the semantic web of re s e a rch papers) have educational and well as re s e a rch applications.

Other Kn owledge Chart s.
T h e re is embryonic support for re p resenting and capturing narratives in the CIPHER project (Mulholland et al., 2002) which aims to s u p p o rt the exploration of national and regional heritage. This is accomplished by s u p p o rting online Cultural Heritage Fo rums (CHFs) where a community focussed on a specific theme or interest can browse or construct narratives relating to the theme or interest. For example, a CHF supports a community interested in communic a t i n g / re c o rding narrative accounts of re l e vant experiences at Bletchley Pa rk in Milton Keynes, UK, where the Enigma encryption machine was deciphered during the Second World Wa r. T h e re is less support currently for analogy-based and causal K n owledge Charts although there is a great deal of work in the AI literature on both of these (e.g., Ge n t n e r, 1983, on analogy; Forbus, 1990, on qualitative physics).

A means of visualizing Kn owledge Chart s.
The ScholOnto project is alre a d y exploring the use of the Compendium tool (www. C o m p e n d i u m Institute.org) for re p resenting the network of scholarly discourse.

Repositories for stru c t u re s / c h a rt s.
Again, the ScholOnto tool provides the basis for a Knowledge Chart re p o s i t o ry.

Automated Co n s t ruction of Kn owledge Chart s.
Knowledge Chart construction is the other side of the coin from Semantic Browsing, and an important activity for learning communities. Some assistance in this will be needed. While there is ongoing w o rk on the use of Human Language Technologies such as information extraction to identify concept instances in a text (Va r g a s -Vera et al. Ciravegna et al., 2002) m o re is needed for the identification of concepts from argumentation, narrative and other structural ontologies. While work here is in its early stages there are pro m i s i n g results from work on automated argumentation extraction in the Knowledge Me d i a Institute. Va r g a s -Vera and Mo reale (2003) re p o rt on experiments in extracting arguments from student essays, while there is currently ongoing work on a tool -C l a i m Spotter -for automatic discove ry of scholarly claims as part of the ScholOn t o p roject. Wo rk on story extraction from news articles is re p o rted in Va r g a s -Vera and Celjuska (2003).
A Semantic Brow s e r. This is perhaps the most important element in the idea of successful Knowledge Navigation. Our thinking here has been extensively influenced by the ongoing experiments with the Magpie semantic browser (Dzbor et al. 2003) in the Knowledge Media Institute. While Magpie can be used as a generic semantic we b b row s e r, it originated, in part, as a means of assisting in sense-making for part i cipants in the Climateprediction.net experiment mentioned above. This experiment, like the Seti@home project, makes use of the distributed computing re s o u rces of thousands of home computers, in this case, to run different versions of a climate model. Magpie provides access (via a contextual menu) to complementary sources of k n owledge, which can be used in contextualizing and interpreting the knowledge in a Web page. This is done by automatically associating a semantic layer to a Web page. This layer depends on one of a number of ontologies which the user can select. W h e n an ontology is selected, the user can also decide which classes are to be highlighted on the Web page. Clicking on a highlighted item (i.e., an instance of a class from the selected ontology) gives access to a number of semantic services. For instance the ontology could contain the class 'Pro j e c t'. Clicking on an instance of this class would p rovide access to Project details, Re s e a rch Areas, Publications, Re s u l t i n g Technologies, Members, Sh a red Re s e a rch Areas and project Web Page. In the C l i m a t e p rediction.net project access is to material which will help to make sense of statistical analyses of complex climate models as well as to the rich literature on climate modelling and climate change.
While much of the necessary infrastru c t u re is already available or could be re a d i l y adapted, there remains a great deal of work to be done on quite central components of our envisaged system.

. 2 Envisaged extensions to existing semantic t e c h n o l o g i e s
Extensions to Ma g p i e. While Magpie is already capable of semantic brow s i n g (linking from salient concepts to auxiliary material via services) it is currently unable to perform the sorts of complex linkages that would be re q u i red for Know l e d g e navigation. We need extensions to enable it to link text to Knowledge Chart s , K n owledge Charts to Knowledge Charts and Knowledge Charts to other we b re s o u rces. We also need extensions to its concept recognition abilities. Cu r rently it is able to re c o g n i ze only instances of concepts which are contained in its know l e d g e base. For example that ScholOnto is a project. T h e re are already plans to make use of Human Language Te c h n o l o g y, in part i c u l a r, information extraction tools, as a means of automatically identifying instances. This would ensure that textual items such as phrases referring to a project as well as texts in languages other than English would be identified as concept instances. Fi n a l l y, as Laurillard points out (pp. 43-4), an academic argument may extend over an entire article with its components scattere d t h roughout the text. Hence, a means of identifying and collecting the parts of complex concept instances scattered through a text is needed. This consolidation is especially important for instances of concepts such as argument, claim and p re m i s e s. The work mentioned above on ClaimSpotter could form the basis for new tools here .

Na r ra t i ve, causal and analogical ontologies and their re p re s e n t a t i o n s.
As we have a l ready said much more needs to be done on producing pedagogically re l e va n t ontologies for narrative, causation and analogy. Mo re import a n t l y, we need some means of visualizing these in a manner which is succinct enough to be used for navigation but detailed enough for the learner to make sense of the debate, narrative , causal model or analogy.

Fu rther work on repositories of Kn owledge Chart s.
Wo rk remains to be done on the exact nature of the repositories needed for the storage and re t r i e val of Know l e d g e C h a rts. For instance, deciding whether a central re g i s t ry with distributed storage nodes or a peer-to-peer network would be more appro p r i a t e .

The Semantic Wr i t e r / Co n s t ru c t o r.
This component of our system will assist in the c reation of Knowledge Charts. It will have some ability to re c o g n i ze instances of s t ructural ontologies as discussed above. Howe ve r, its main role is in providing the e n v i ronment for learners and/or their teachers (and communities more generally) to c o n s t ruct, collaborative l y, new Knowledge Charts which can re p resent the point of v i ew of the community, the group of learners, the tutor or the individual learner. As we have argued elsew h e re, the construction of complex re p resentations for know l e d g e is in itself a useful exe rcise (Stutt, 1997). As well as assisting in the construction of K n owledge Charts, the Semantic Constructor should be able to do all that is n e c e s s a ry to publish these (e.g., notify the re g i s t ry, reformat in RDF). New work just s t a rted on Magpie will result in a system which can use information extraction to c reate mark-up within documents, thus providing a basis for the Se m a n t i c C o n s t ruction of Knowledge Charts. This means that the extended Magpie will act as both Semantic Browser and Semantic Constru c t o r.
Community-based tools. Tools are needed for identifying, creating, support i n g , fostering, and tracking communities and integrating them with knowledge re p re s e ntations into Knowledge Neighbourhoods. While we have begun work on communityoriented customization in the e-shopping arena  much more w o rk needs to be done on identifying the characteristics and stru c t u res of communities as well as their dynamics.

0 Conclusion -Learning Webs and Critical Th i n k e r s
Ac c o rding to Fisher (1988, p. 1): "it is possible to rely too heavily on experts and this approach to learning and k n owledge tends to encourage passivity and re c e p t i veness rather than i n ve n t i veness and imagination… One object of this book is… to impress on the reader what a long way one can get in understanding any subject by thinking it through for oneself… We shall do this by concentrating on the arguments experts have produced for believing a wide range of things and s h owing how it re q u i res only a re l a t i vely slight knowledge of the subject to e valuate the arguments oneself." This is true whether we learn via the Web or from books. The We b, howe ve r, pre s e n t s both challenges (e.g., the danger of information overload) and opportunities (e.g., the wealth of competing viewpoints). The Semantic Web (or Webs) will provide ye t m o re opportunities for learning in the form of greater access to a multiplicity of d i verse learning objects. It can also, as we have suggested, provide the means for learners to navigate through the plethora of sources, find help in their interpre t a t i o n of material by contextualizing it to debates and narratives, and actively enter into these debates or construct these stories as members of living online communities of learners.
Instead of the oracular pronouncements of Kaku's Magic Mi r ro r, a means of t r a versing the various links possible from web document to web document by m e a n s of a meta-model or models-a Knowledge Chart -expressing the associated debate, n a r ra t i ve, analogy and so on will be most valuable to learners. By providing this in combination with a means of learner participation in these debates, using Know l e d g e Neighbourhoods, the learner becomes, not a passive recipient of knowledge, but the s o rt of critical thinker able to deal with the complexity of the material available in a k n owledge based society.