Beyond Hypertext: Adaptive Interfaces for Virtual Museums

Beyond Hypertext: Adaptive Interfaces for Virtual Museums

Douglas MacKenzie

DMC Ltd
3 La Belle Place
Glasgow G3 7LH
Scotland
Tel: 0141 333 9400
Fax: 0141 333 9490
e-mail: douglas@dmcsoft.com

INTRODUCTION

Museums are beginning to `escape the tyranny of ... truth-telling' (Sledge, 1995): curators of physical museums have steadily moved towards an understanding that visitors approach their institutions with a wide variety of expectations and questions and, indeed, motivations for asking those questions. The curatorial task in mounting an exhibition is no longer to present a God's eye-view of a subject. Of course, choices have to be made. Physical space and cost will always restrict the number of artefacts which may be placed on display and it is likely there will only be one method of connecting one object with another, that is by placing them in close physical proximity. There is an arena where these physical constraints do not apply, where technology can allow each visitor to organise elements into an individual exhibition. Sadly, though, the designers of virtual museums do not appear to have learned the lessons of their counterparts in the physical world. Most multimedia treatments of cultural and historical topics imprison the user in a structure more rigid than a Victorian museum. I believe the major reason for these failures is the role of hypertext. The ease with which it may be implemented has first seduced creators into an inappropriate design philosophy. This is bad enough. The problem is then compounded by a retrospective justification of the educational and cognitive merits of the format based on inappropriate metaphors of space and memory structure.

TAMH (Tayside: A Maritime History) is a PC-based multimedia project (MacKenzie, 1995) exploring some issues involved in allowing users, virtual visitors, to ask the questions which interest them in the way they find most appropriate. It addresses a wide audience, as a physical museum does, from primary schoolchildren to postdoctoral researchers. These different users ask different questions, ask them in different ways and require different levels and types of explanation. This has major implications for the design of the interface and the structure of the underlying data. Much of the work on the project has been to develop interfaces for different user groups and, indeed, interfaces which allow the individual user to choose what to display. It is, however, impractical to create different databases for each group so the data must be structured so that only the depth of data required for any given situation is retrieved from a common database.

HYPERSPACE: PLENTY HYPE BUT NO SPACE

We have moved from calling selectable links to different nodes from a page of information hypertext through extending the page concept to include various multimedia elements, hypermedia, to describing the entire range of elements which may be linked as hyperspace.

The number of articles devoted to the problems of `navigation in hyperspace' is legion (van Dam, 1987; Utting and Yankelovich, 1989; Edwards and Hardman, 1989; Tripp and Roby, 1990). Generally writers view hyperspace in the same way as physical space (Dillon, McKnight and Richardson, 1990) and, by analogy, offer users, or at least try to help them build, a cognitive map (e.g. de Vries and Kommers, 1993; Beasley and Waugh, 1995).

If it is difficult to navigate adequately with these maps, perhaps the maps are at fault and problems with using maps for navigation in hypermedia systems have been acknowledged for some time (Gray and Sasha, 1989; Mayes, Kibby and Anderson, 1990). The conclusions of hypertext's advocates are not encouraging. `Navigation works best for information spaces that are small enough .... and familiar enough to users to let them find their way round' (Nielsen, 1989). In other words, maps work best when no one needs to consult them. So, as one might expect, where elements can be mapped into a spatial arrangement which is readily understandable as a two-dimensional structure (exploring the on-line Warhol museum by clicking on the floorplan, for example, or in industrial training applications which mimic a dictionary (Beasley and Lister, 1992)), users have few problems with hypertext.

The reason why navigation becomes problematic when this simple spatial mapping no longer pertains is simple. Hyperspace has nothing to with physical spatial relationships. An abstract graph showing node traversal may be represented geometrically but this is just an abstraction of a a purely arithmetical construct, a Poincaré incidence matrix. Any geometric version is non-unique, certainly unnecessary, and possibly unhelpful. Only fairly recently has the notion that `navigation' is what one does when using a multimedia system been questioned (Stanton and Baber, 1994). In physical space the relationship between objects is fixed and non-subjective. In hyperspace, the relationship is quite arbitrary, determined only by the program author. The author's assumptions may not match those of the user and the more non-linear a system the greater the difficulty a user has in constructing a map. As Stanton and Baber point out, if there are multiple routes to and from a node, how can users develop a map which can show the complexity of the possible routes yet still be simple enough to allow easy recall? Indeed, their argument, that if users have an effective search strategy `why go to the additional "cognitive overhead" of developing a map?' was a central tenet in developing the user interfaces in the TAMH project, although the search tools provided are different to those envisaged by Stanton and Baber.

Instead of accepting that subjects which encompass a well-defined domain with few routes between nodes are the natural candidates for hypertext, design guides, perversely, draw the other possible conclusion: force the data to fit the model. For example, SKEIM (Kelly, 1992) demands a hierarchical knowledge base with each level defining any new terms used in its explanation until only familiar vocabulary appears in the terminal explanation. More extreme, others suggest that, for a hypertext system to be effective, the user first needs to understand the structure, content, aims and terminology of the system (Jacques, Nonnecke, Preece and McKerlie, 1993). Both sets of demands have their place. The strict hierarchy seems to work in narrow domains such as Kelly's own lecture note review project (Kelly, 1993). The willingness of users to adapt to the needs of the system is often found in training applications where users may expect improved job performance and possible reward as a result of mastering the material (Shaw, 1992).

Neither situation pertains in museum applications. We can make few assumptions about the commitment of visitors let alone their motivations. If multimedia is to be both engaging and usable in these settings it must cover a wide number of subject areas with an infinite number of routes, `enabling people to think for purposes they have defined themselves' (Sledge, 1995). A strict, visible hierarchy is the dead hand of curatorial diktat: the explicit links of hypertext answer the questions the designer wishes to ask, not necessarily the ones visitors want to.

CATAPULTS AS COGNITIVE MODEL

There are probably two reasons why hypertext has been so readily adopted in inappropriate situations. The first is that designers are seduced by the ease with which something can be created. The literature is full of examples where a development environment is chosen to reflect the developer's level of expertise rather than the needs of the users (Ess, 1991; Picciano, 1993; Skalka, 1995).

The second reason for forcing information into hypertext format is altogether more dangerous. The justification has been that hypertext resembles either the structure of the brain or human memory (Dede, 1987, Swan, 1994). A dubious claim in itself but made even more absurd by a claim that its structure necessarily leads to educational effectiveness. This view has best been characterised, and criticised, as the `homeopathic fallacy' (McKendree, Reader and Hammon., 1995) following J.S. Mill's `the most deep-rooted fallacy .... is that the conditions of a phenomenon must, or at least probably will, resemble the phenomenon itself'. John Searle (Searle, 1984) gleefully points out the tendency throughout history to choose the latest and greatest in technology as a model for the brain: a catapult to some of the Ancient Greeks, a mill for Leibniz, an electro-magnetic system, a telegraph system and a telephone switchboard as time went by.

CONSEQUENCES FOR THE TAMH PROJECT

Some might argue that the need for educational effectiveness is less important in the museum environment: many visitors are simply browsing without a specific learning motive. Fortunately this view is becoming increasingly rare and the importance of museums as educational centres recognised (Yamada, Hong and Sugita, 1995). Furthermore, exhibit designers are eschewing comprehensiveness in favour of representative items telling a story (Honan, 1990). However, such representative items tell only one story, that of the curator. A hypermedia system containing only explicit links is simply an electronic version of this. The fewer routes through the material, the easier the system is to use but the less choice visitors have to `move through information ... in ways appropriate to what they want to achieve' (Sledge, 1995).

The TAMH project sets out to let visitors choose the items which interest them rather than ones which have been chosen for them. Earlier work on the project showed that there are too many connections users wish to make between nodes for explicit hypertext-like linking to work (MacKenzie, 1995). In addition to there being more possible links than in a strictly hierarchical hypertext system, there are more objects than are typically on display in a physical museum. This is particularly so for maritime history which is about large objects: harbours, ships, whales and the like. Museums in Tayside are relatively small and what they may display severely restricted. There is a breed of curator suspicious of multimedia systems because mere representations do not give enough `respect to the artefact' (Michailidis and Loissios, 1995). Although I am not totally in sympathy with this view, it smacks of Barthes' `accumulation of goods ... so that one could at any moment do the accounts of the ineffable' (Barthes, 1957) there may be a sop to it in that the wide-ranging and ill-specified domain deliberately chosen for TAMH allows many artefacts (or representations of them) to be seen by visitors who would not ordinarily see them. Apart from the several thousand images which came from non-local sources, in the region of seventy per cent of objects photographed or scanned at local museums were not on display at the time of data capture.

Rejecting fixed, explicit links as a way of accessing items of interest is not unique to TAMH. Other attempts to make the user think first and then request information such as Microscosm (Davis, Hutchings and Hall 1993) or Perseus (Morell, Marchionini and Neuman, 1993) have already provoked howls of horror from traditionalists. `A student using Perseus is not likely to come away with a sense that someone has intended he learn something' (Eiteljorg et al., 1992). Quite so. TAMH does not wish to frogmarch its users to `learning' either. Both the teaching and learning of history are subtle things. To suggest (describing the US Civil Rights CD-ROM, Set on Freedom) that a timeline with a few selectable events leads to a `rich understanding of historical periods' (Swan, 1994) is Whig history at its worst. If we are to lead users to an understanding of history, or even better, to an understanding of what an historical explanation is, we need a richer interface than this.

Both Microscosm and Perseus have an advantage over TAMH. They are intended only for an undergraduate-level university audience. In the museum context the audience is far more diverse. One approach is to look for a typical user and design a system around that user's needs. The model of the `motivated museum visitor' as a fourteen year old challenged by a stimulating interface, uninterested in boring design, with non-specialised interests and familiar with technology is the one described and followed in the MITOS/ID project (Michailidis and Loissios, 1995). This was not the approach taken in this project. Firstly we identified too many different `motivations' in possible users: from tourists looking for a day trip to undertake by car to students undertaking specific historical research. Secondly, we wished to build on the conclusion reached in an earlier gallery project (MacKenzie, 1995) that a common database could serve the information needs of a wide audience with different perceptions and needs. To achieve this we needed both different interfaces for different user groups and, within these basic frameworks, adaptive interfaces which would alter according to the perceived needs and habits of a particular user.

INTERFACE DESIGN

Rather than attempting to design individual interfaces for, say, `a tourist', a `primary schoolchild' or a `doctoral researcher' ab initio and hope they would be right first time, the approach taken was to identify those elements which groups had in common and other elements which were specific to tasks or queries of one or more groups. This was neither as rigorous in design terms as a thorough task analysis nor, in the development sense, as an object-oriented approach though it has the flavour of both. It concentrates on building individual elements as a toolkit (MacKenzie, 1991) then, from the bottom-up, constructing a full design using these elements. There is a well-defined terminology (primitives, interface elements and forms) and methodology, the domain-specific style guide (DSSG) (Gulliksen and Sandblad, 1995) for this approach though, if truth be told, ascribing this, at least to the early days of the TAMH project, is a post hoc justification of the `suck it and see' method really used. The fact that we used this approach, in principle, without knowing that a formal method existed explains why some of the terminology here differs, or is used less rigorously, than in Gulliksen and Sandblad.

EXAMPLES OF STYLE GUIDE ELEMENTS

A general requirement for the system is the facility for a user to print out an item of information. The function is a primitive, and as a fundamental system requirement, appears as a button placed at the bottom of the screen: its appearance and function identical whether the user is using a keyboard, mouse or touchscreen. This button is shown in Figure 1 and is a sub-class of `button' a very basic instance of an interface element.

Print Page

Print Directions

Simple Interface Elements

Figure 1 - Normal Print

Figure 2 - Adapted Print
Button for `Site' screens

Some adjustment needs to be made to this interface element's properties depending on the input device. For example, the button gives the appearance of being depressed on a mouse_down/touch_screen event and popping back on the mouse_up/break_contact so, if a keyboard is the only input device the alt-P combination must trigger the same code as both the mouse_down and mouse_up. The caption for the button is another interface element (and its property may be modified to indicate the possibility of using a keyboard by underlining the first letter, P) The style guide for the `tourist' interface contains the button and the caption, without the keyboard option. Style guides for other target groups use other combinations or leave the appearance and functionality to an .INI file for which the settings are determined by a supervisor (see below).

The style guide is not just about style in the visual sense. It is also about function. So there is an extension of the Print method specific to the `tourist' style guide which allows for the printing of directions to a museum or other site of interest. If a `site' screen is currently visible, and directions to it are available, then the Print button's icon and caption change to indicate this.

On pressing the button, a local map marking the site in question is attached to an output form. If the machine has been configured to run in a particular building or town then that location is read from a configuration file and specific directions, textual or graphic, retrieved from a database and attached to the output form before printing.

This trivial example of an interface element is very close to the concept of an object in a visual programming language such as Visual C++, Delphi or Smalltalk. An interface element can be far more complex and may have been created in response to a specific user problem in testing. A major resource in the project is a database of mariners and shipping log entries from the early 16th to late 17th centuries. One way of searching this for the serious researcher uses a database query form which builds up SQL queries from the criteria entered by the user in a less rigorous format (> and < signs may be entered in any order, dates can take almost any format, `AND' and `OR' clauses can be entered directly or by using a pop-up help panel.) Most of the shipping log entries describe the cargoes and the merchants to whom the various goods were supplied. A merchant may appear in one entry as `William Anderson' in other as `Wm John Andersone'. If the query was `William' and `Anderson' the second entry here would not be retrieved, If the query looked for (`William' OR `Wm') AND `Anderson' every entry where these words occurred would be retrieved i.e. William Brown would be returned as would David Anderson. A proximity specifier was identified as a solution to this (see Figure 3). The number of words which are allowed to come between the two search terms and still allow a match is specified as part of the search criteria.

Figure 3 - Proximity Specifier

Having developed the proximity search as an interface element (combining primitives, functions, parsing and query building routines) it joins the toolkit and can be used wherever such a search constraint is needed (searching registers of vessels, passenger or crew lists, documents etc.)

ADAPTIVENESS OF INTERFACE

The TAMH interface is adaptive in two senses. Firstly there are different methods of retrieving the same data depending on user preference or administrative decision. There is a supervisor function which determines which fields are displayed in query and retrieval screens; which functions are active; which menus are displayed. This has several purposes. If the package is being used with young primary school children, it is unlikely a teacher would want them to see the full text display of shipping records: the ship, date, cargo summary, ports and master might be sufficient. Similarly that teacher might prefer children to search using visual cues rather than completing a database search mask. For an older class, the teacher might want a greater choice of search methods. In a museum setting, a curator might wish to enable only those `stories' linked to the museum's current exhibitions. A terminal on public display might have many options disabled to minimise the time spent at it to encourage visitor throughput whereas one in a private room might have every element available for personal study.

DIFFERENT METHODS OF RETRIEVAL

A researcher wishing to find out about the import of wood from Danzig to Dundee in the 16th century, and at least slightly familiar with databases, could complete the voyage part of the search screen shown in Figure 4 and view the records of voyages.

Figure 4 - Conventional Database Search

This same information, though, can be retrieved in different ways. A user viewing the map showing Tayside's main trading ports for different centuries could press the 16th century button. Selecting a good highlights the most important ports for that commodity by displaying its icon beside the port name (Figure 5). The lowest level of information, the port records retrieved in the database search, are displayed by touching the wood icon next to Danzig. In terms of the SQL search performed the methods are identical but the way search criteria are built up are quite different.

Figure 5 - Part of map search and display

Is this not just an explicit hypertext link of the type criticised above? Yes it is. So, implicit links are also available to the user. Any word or phrase in any body of text on screen can be selected and `links' searched for. Does this always result in links being found? No. Does it not sometimes return a huge number of links where it is difficult to identify what is truly useful? Yes. Does it sometimes return essentially meaningless links? Yes. Is this a design defect? No, it's a user defect. Is this not falling into the same trap identified above, that the user needs to know how the system is structured before being able to use it constructively? I think not, and I hope the last method of interrogating the database, illustrates this.

The most natural way to learn anything about history is to ask a question. It may be to ask someone's opinion, it may be a rhetorical question which starts some independent research. TAMH also allows free form questions (see Figure 6). Some of the parsing which has been done on the example question has been made visible, the `HMS' triggers a search for a vessel name to follow and `Arbroath' is recognised as a placename from an index.

Figure 6 - Asking a free-text question

Such a question is fairly easy to interpret and answer. There are many questions which are not. The list of items generated by the search may not match the questioner's expectations. Indeed, the parsing may completely misinterpret the question. There are, of course, an infinite number of improvements which can be made to the parsing though we will never adequately answer all questions. No curator nor institution will achieve that either. This is an advantage of the system rather than a limitation. It is perhaps more willing to admit to fallibility and unknowability than are many museums. It allows visitors to ask questions without fear of appearing foolish. Whether in correctly interpreting, or in misinterpreting, a question the worst that can happen is that the questioner uses the note-adding facility to tell the system designers how stupid they have been. Dialogue is rarely a bad thing. At best it can provide a pathway to discovery for the user or help formulate a new question, a new way of thinking about something. The surest way to kill someone's interest is to answer a question completely. The study of history is, as Auden says, `the study of questions: the study of answers belongs to anthropology or sociology'.

ADAPTATION BASED ON USER INPUT

The usual definition of an adaptive interface is one which changes in some way in response to its environment (Reiser, Anderson and Farrell, 1985). In TAMH, this change can be achieved both actively and passively. Actively, the user can select how something is viewed. For example, the map shown in Figure 5, can be displayed in several different ways. Political boundaries in force at various dates over the last four centuries can be overlaid. Port names can cause confusion. Most English-speaking historians use old German names for Eastern Baltic ports which may be unfamiliar to non-specialists, Reval for Tallinn for example. The user may select which names are used for ports.

When the user performs a search, by whatever method, a thesaurus automatically expands the search. For example looking for Korolevets searches for references to this first (as the choice of name implies either a particular historical interest or a reference elsewhere to a place the reader does not recognise) but will also retrieve Königsberg, Kaliningrad, Karaliaucius and Tvankste, the old Prussian form for the city. This is one way in which passive adaptation is implemented. The other way is more problematic. If the `suggestion' option is switched on in the configuration, the system offers suggestions to the user as to what else might be of interest. Similarly, when a large number of items are returned in a search, an attempt is made to list these with those thought to be of most interest to the user appearing first. The dangers of building a user model based on simple keyboard input or, indeed a few mouse clicks, (Gaines, 1987) are understood and no great claims are made for system `intelligence'. If the audit trail shows a user has seen several items on shipbuilding in Arbroath then does a search on `Stephen' and `Dundee', the first items offered will be about Alexander Stephen Shipbuilders' time in Dundee. Similarly, if a user has been reading about the sound toll at Elsinore specific links to related items in the Tayside area are offered, viz. the bowls presented to local masters calling at Elsinore for the first time and on display in Arbroath museum.

At worst with this strategy we increase the `cognitive load' on users (Yamada et al. 1995) though I am not convinced that making people think is such a bad thing. Again, at best, we may offer them a new or different insight.

CONSEQUENCES FOR DATABASE DESIGN

As might be expected from the description of searching there is a rigorous classification of items in the underlying databases. Rigorous, that is, in that the same rules of classification are imposed on similar items (all historical records of mariners and voyages have the same table structure and all short articles with accompanying multimedia elements share a structure but one different to the mariner data) but not in striving for some universal standard data model. We looked at existing data models (such as GENREG (Rold, 1995) or the CPS nomenclature (Pepermans, 1995)) which are admirable. As it happens, our database is not cataloguing objects in the traditional sense. We could shoehorn a newspaper article into classification as an artefact but it serves little purpose in our project. Also, our concept of the relationship of images to records did not, at first sight correspond to that of GENREG. We might have gone through a few hoops to achieve this. Again, though, there is no compelling reason to adopt a particular model. The only restriction placed on the database design was that it be interrogable using SQL. So, the current standalone and locally networked versions of TAMH use the Microsoft Access database and JET interface. If we produce an Internet version we will convert the database to a Unix-based INGRES variant to suit our server and search speed requirements. Previous experience of this database migration suggests a couple of days work. Given information on the structure of any other SQL-type database in the museum world we would expect to be able to interrogate it, filter the data and make use of it in our project without the user being aware that different databases were involved (except, of course, for any copyright notices which we would strictly observe !) . The same applies to others using our data. This surely is a more rational approach to database design than the museum community's endless discussion of standards which are usually technologically outdated before they are even widely accepted (e.g. Bearman, 1995).

This same mania for standards is apparent in selecting image formats (Trant, 1995). Of all the user comments on our multimedia projects, I have never heard, `I was expecting to see a TIFF but I only got a JPEG.' In the TAMH project we currently use TIFF (24-bit and grayscale), JPEG, PCX and BMP. We could incorporate images in a dozen more file formats without altering the code as the import filters automatically detect the image type and call the appropriate decoder. Some moving images are in SIF MPEG format and some in QSIF. We can make some judgments as to which format is most appropriate for any given image in the current incarnation of the project as we have fairly tight control over the delivery mechanism. Only our display routines are used. We can specify the hardware used (though monitor settings and light conditions are outwith our control). If we distribute the project on CD-ROM or give access via the Internet, that tiny bit of control disappears. People will view images on anything from 16 colour to 16 million colour displays, they will use different, totally uncalibrated monitors. They can import images into graphics packages of their own, each with different colour-space conversion and dither algorithms. To worry about graphic standards for this environment is pointless. Instead of `angels on the point of a pin' arguments about TIFF vs GIF, the sensible strategy is to keep the architecture as open as possible to allow any significant advance in image technology, wavelet compression, perhaps, to be integrated easily. With this in mind images are referred to in the main databases only by an internal `nickname' with a separate table defining the real name and location. In this way we can have the image record point to a local JPEG if the system is running on a standalone machine, to a centrally archived TIFF image if a network is in place and the highest fidelity is required or, even, to a URL somewhere on the Internet.

CONCLUSION

Time will tell if the searching and questioning techniques employed in TAMH are useful. There will certainly be better ones developed by this and by other projects but the underlying technology is not the most important aspect. TAMH will have served a useful purpose if it can move discussion away from sterile debates on which image format is `best' or which classification structure should be used and on to how museum systems can meet the needs of visitors instead of how images can be collected in a computerised form.

Achieving this will require attitudinal rather than technological change. Program developers may have to think and work a little harder to meet users' needs instead of forcing their cognitive maps on unwilling travellers, museum administrators may have to forsake standards' committees and instead investigate visitor needs and perceptions, and curators, if they are genuinely interested in public access to collections, will need to rethink their dogma of the primacy of the artefact. Making these changes is much harder than devising new interface elements. This paper has criticised the use of hypertext on the grounds that its spatial metaphor is inappropriate and that claims for it to mimic cognitive processes and, consequently, `cause' learning to take place, are absurd.

The other criticism made here, that designers have seized on hypertext systems because it is easy to create something is probably only half the story. I fear the other attraction for the traditional curatorial mind is that hypertext design reinforces dogmatic categorisation. Curators of both physical and virtual museums would do well to remember Arthur Danto's (Danto, 1992) remark that Philip IV did not have his portrait painted by Velázquez to have it hang with the Spanish School in the Musée Napoléon. `Only when the form vanished in which the King could multiply himself .... did these paintings reduce to works of art in which aesthetes could murmur about diagonals'. His observation that it is `as if the museum came into existence to celebrate the secondary value of such objects once their primary value was gone' sounds a greater warning about misplaced `respect for the artefact' than worries about whether a digital representation is appropriate or not.

REFERENCES

Barthes, R. (1957). Mythologies, Vintage edition, 1993, London.

Bearman, D. (1995). Digital Image Access Project, RLG Meeting. Conference Report. Archives and Museum Informatics: Cultural Heritage Informatics Quarterly, 9(2), 199-209

Beasley R.E. and Lister D.B. (1992). User orientation in a hypertext glossary. Journal of Computer-Based Instruction, 19, 115-118

Beasley R.E. and Waugh M.L. (1995). Cognitive mapping architectures and hypermedia disorientation: an empirical study. Journal of Educational Multimedia and Hypermedia, 4(2/3), 239-255

Danto A.C. (1992). The Museum of Museums. In Beyond the Brillo Box. Farrar, Straus, Giroux, NY, 199-214

Davis, H., Hutchings, G., and Hall, W. (1993). A framework for delivering large-scale hypermedia learning material. In Maurer, H. (ed) Educational Multimedia and Hypermedia Annual. AACE. Charlottesville, VA, 115-122

Dede, C.J., (1987). Empowering environments, hypermedia and microworlds. The Computing Teacher, 15(3), 20-24

Dillon, A., McKnight, C., and Richardson, J. (1990). Navigation in hypertext: a critical review of the concept. In Diaper, D., Gilmore., D., Cockton, G., and Shackel, B. (eds). Interact 90. North Holland. Amsterdam

de Vries, S., and Kommers P. (1993). Concept mapping as a mind tool for exploratory learning. In Maurer, H. (ed) Educational Multimedia and Hypermedia Annual. AACE. Charlottesville, VA, 139-146

Edwards, D.M., and Hardman, L. (1989). Lost in hyperspace: cognitive mapping and navigation in a hypertext environment. In McAleese, R. (ed) Hypertext: Theory into practice. Blackwell. Oxford

Eiteljorg, H., Hamilton, R., O'Donell, R., Pearcy, L.T., Wiltshire, S. (1992). Perseus Review. Bryn Mawr Classical Review, 3.5.4

Ess, C. (1991). The pedagogy of computing: hypermedia in the classroom. In Proceedings, Hypertext `91. ACM. NY, NY, 277-289.

Gaines, B. From teaching machines to knowledge-based systems: changing paradigms for CAL. In Proceedings, ICCAL 1987. University of Calgary, 17-25

Gray, S.H. and Shasha, D. (1989). To link or not to link? Empirical evidence for the design of nonlinear text systems. Behaviour Research Methods, Instruments and Computers. 21(2), 326-333.

Gulliksen, J. and Sandblad, B. (1995). Domain specific design of user interfaces. International Journal of Human-Computer Interaction, 7(2), 135-151

Honan, W. H. (1990). Say goodbye to the stuffed elephants. New York Times Magazine.

Jacques, R., Nonnecke, B., Preece, J., and McKerlie, D. (1993) Current designs in HyperCard: what can we learn? Journal of Educational Multimedia and Hypermedia, 2(3), 219-237

Kelly, A.E. (1992) SKEIM (Shell for Knowledge Engineering, Instruction and Measurement). Technical Report 09-8, Graduate School of Education, Rutgers University, New Brunswick, NJ.

Kelly, A. E. (1993) Designing instructional hypertext for use in lecture note review: knowledge engineering and preliminary testing. Journal of Educational Multimedia and Hypermedia, 2(2), 149-176

MacKenzie, D. (1991) Developing CBT Courses for End User Training. In Proceedings of the 33rd ADCIS Conference, 345-351

MacKenzie, D. (1995).Using Archives for Education, Journal of Educational Multimedia and Hypermedia, In Print

Mayes, T., Kibby, M. and Anderson, T. (1990) Signposts for conceptual orientation. In McAleese, R. (ed) Hypertext: Theory into practice. Blackwell. Oxford, 121-129

McKendree, J., Reader, W. and Hammon, N. (1995). The `Homeopathic' fallacy in learning from hypertext. Interactions, 11(3), 75-82

Michailidis, S. and Loissios, L. (1995). Hypermedia development in the Benaki museum. In Bearman, D. (ed) Hands on Hypermedia and Interactivity in Museums, AIM, Pittsburgh, PA. 107-114

Morell, K., Marchionini, G. and Neuman, D. (1993). Sailing Perseus: Instructional Strategies for Hypermedia in the Classics. Journal of Educational Multimedia and Hypermedia, 2(4), 337-353

Nielsen, J. (1989). The art of navigating through hypertext. Communications of the ACM, 33(3), 297-310

Pepermans, R. (1995). Material culture in the computer age: an assessment of the Parks Canada terminology record as a means of storing conceptual data. In Bearman, D. (ed) Multimedia Computing and Museums, AIM, Pittsburgh, PA. 19-38

Picciano, A. G. (1993) The Five Points: the design of a multimedia program on social history. Journal of Educational Multimedia and Hypermedia, 2(2), 129-147

Reiser, B.J., Anderson, J.R. and Farrell, R.G. (1985). Dynamic student modelling in an intelligent tutor for LISP programming. Proc. 9th International Joint Conference on Artificial Intelligence, 8-14

Rold, L. (1995). GENREG, a simple and flexible system for object registration at the National Museum of Denmark. In Bearman, D. (ed) Multimedia Computing and Museums, AIM, Pittsburgh, PA. 19-38

Searle, J. (1984). Minds, Brains and Science. Harvard University Press, Cambridge, MA.

Shaw, D.S. (1992). Computer-aided instruction for adult professionals: a research report. Journal of Computer-Based Instruction, 19, 54-57

Skalka, M.R. (1995). The Rembrandt Watermark Project Compact Disk. In Bearman, D. (ed) Hands on Hypermedia and Interactivity in Museums, AIM, Pittsburgh, PA. 39-52

Sledge, J. (1995). Points of View. In Bearman, D. (ed) Multimedia Computing and Museums, AIM, Pittsburgh, PA. 335-346

Stanton, N.A. and Baber, C. (1994). The myth of navigating in hypertext: how a "bandwagon" has lost its course! Journal of Educational Multimedia and Hypermedia, 2(4), 337-353

Swan, K. (1994). History, Hypermedia, and Criss-Crossed Conceptual Landscapes. Journal of Educational Multimedia and Hypermedia, 3(2), 120-139

Trant, J. (1995). Framing the picture: standards for imaging systems. In Bearman, D. (ed) Multimedia Computing and Museums, AIM, Pittsburgh, PA. 347-367

Tripp, S.D. and Roby, W. (1990). Orientation and disorientation in a hypertext lexicon. Journal of Computer-Based Instruction, 17, 120-124

Utting, K. and Yankelovich, N. (1989). Context and orientation in hypermedia networks. ACM Transactions in Information Systems, 7(1), 58-84

van Dam, A. Keynote speech. Hypertext `87

Yamada, S., Hong J-K. and Sugita, S. (1985). Development and Evaluation of Hypermedia for Museum Education: Validation of Metrics, ACM Transactions on Computer-Human Interaction. 2(4), 284-307