HYPERTEXT FEATURES 21
Selection 22
Nodes and Links 22
Nodes 22
Text 25
Simulations 25
Symbols 25
Icons 28
Anchors 29
Network 29
Node Types 31
Link Types 32
Link types and Instruction 33
Browsing 34
Hypergraphs 35
Control and Navigation 35
SemNet 37
Cyberspace 37
Cognitive Overload 38
Hypermedia 39
Storage 39
Index 40
Embedded Menus 40
Hot-Keys versus Point-and-Click 41
Computing on HyperText Structures 41
StrathTutor 41
Typographical Cuing 42
Unresolved issues in HYPERTEXT research. 42
Classifications of HYPERTEXT SYSTEMS:43
The Function of Hypertext Systems 44
Programming and Design Environments 45
Information Retrieval 45
Interface Systems 46
Knowledge Representation 47
An incredible new terminology has grown up around this new technology, and much of the terminology is still unstable and settling into more precise definition, so do not be surprised to see others using terms in a slightly different way than they are described here. The terminology is undergoing fairly rapid evolution as our understanding of hypertext continues to improve.
HYPERTEXT FEATURES
Hypertext is based on dynamic interaction betrween users and text and graphics. Hypertext tries to fulfill the basic wish that many of us have of seeing something that we don't understand, and just touching it to get a fuller picture of what is going on. Hypertext lets us do that. It is based on a fundamental datatype in the computer, usually called a selection.
Selection
A selection marks out a piece of the computer screen through some sort of direct manipulation: by pointing or buttoning, or mousing at an area of the screen. It is extraordinarily useful and central to hypertext. The area of the screen defined by the selection can be a piece of text or part of a bitmap, or any other data structure the computer knows how to deal with. Most modern computer systems let you select a character, sets of characters, words, phrases, lines, or rectangular areas of the screen. Irregular areas can be selected too, but at the expense of much greater computation or memory. Objects in draw-based graphics, or windows containing spreadsheets, video, sound, or any other display entity, can all be selected.
Usually a selected item is a temporary or evanescent thing. It can be manipulated in some convenient way, and then it is forgotten, dropped as a selection. In hypertext, however, the selected item can be made semi permanently selected, so that whenever it is pointed at in the future, something happens. So instead of selecting something to copy or delete it, an item can be selected in order to make it permanently interactive. Usually, when this is done, the item is altered in some way so that the change is also permanently visible, by highlighting, quoting, placing an asterisk near it, or leaving some general icon or mark. The selected content then becomes an anchor or an active hotspot, hotword, or hotbutton.
Nodes and Links
One convenient way of talking about the content of hypertext systems, in a way that makes the content explicitly hypertext, is to move away from simply text or graphics, or spreadsheets, or even designs, and speak in the abstract terms of nodes and links.
Nodes
Nodes are sources or destinations of links. They are the content of hypertext. As a source, they are the active hotspots, hotwords, or hotbuttons of a hypertext; and they are also the destinations of the links from a hotspot. It is curious that something so basic as the destination of a link in hypertext does not have its own term, except the general term of a node. A node is said to be anchored, if it has a hotspot button that leads to it. It is also said to be anchored to the destination of a link, if that link leads to marked text or icons in the destination. My preference is to use the word anchor only for that special case where where a particular word or phrase in a text, or a particular spot in a display, is the marked destination of a link.
Hotspots, hotwords, or hotbuttons:
By selecting (buttoning or mousing or hotkeying) a hotspot node, you immediately provoke the computer into displaying the destination node or anchor. This simple and convenient method for moving through text and graphics is the essence of hypertext. Underlying this user navigation function can be a myriad of different computer implementations for how to store and retrieve the text and graphics; but these very important, sytem-level techniques will remain outside the scope of this book. For the interested reader, the NIST conferences on hypertext (1989, 1990) will provide an excellent starting source.
Nodes are the content parts of the hypertext, the things that people want to link together. You can think of them as the nouns and verbs of hypertexts. Sometimes, they are in fact just single words, and obviously they can be any word in a hypertext, or any picture, or part of a picture, or any other object, video, or simulation. But in terms of their function think of them as the nouns or verbs, even when they are an adjective. For instance, in the example from Jay Bolter's Storyspace hypertext system for writing, three windows are shown. The top window is a note, that says that , "It is curious that something so basic as the destination of a link in hypertext does not have its own term, except the general term of a node. " This note is anchored in two ways. In the Chart window, below it is anchored as an icon, a small box (™) that is split (indicating a window and title bar). Buttoning this icon opens the note. It is also anchored in the text of the window as a wire-frame around the word "Nodes" in the window titled "Nodes". Again, mousing this word lets you read the note and return to the text afterwards.
Nodes within a hypertext can also be classified by the kind of source or destination they are. As either source or destination, the selected node can be coarsely defined as a large file or document, as a smaller card or paragraph; or even finer, as a word, phrase, or sentence.
As a source, the node is usually marked in some carefully defined way to attract the user's attention, and call out, "Hey, I'm here! Ready to help you find out more!" Some hypertext systems attract attention by using special icons or marks for the node, or beside the node. Asterisks and boxes are frequently used. The same things happen in some systems (sych as Guide) immediately and automatically whenever the spot is overridden by the standard cursor, only the cursor changes its shape. As a target or destination, they are defined more dynamically, by selection after the link has navigated to the proper place.
Nodes can also be thought of in terms of the information they represent. Three common typ[es of information that can be represented are text, simulations, or symbols. Each of these representations can be complexly subtyped (analyzed into many different but related kinds of stuff):
Text: written, spoken, discourse, fiction, drama, plays, mysteries, comedies, outlines, dictionaries, thesauruses, hierarchical, multimedia dictionaries, encyclopedias, almanacs, etc.
Simulations: movies, bitmaps, animations, cartoons, cad/cam, models, paintings, drawings, 3-D graphics, sounds, music, operas, cyberspaces, etc.
Symbols: mathematical equations, spreadsheets, calendars, statistics, demographics; maps, charts, figures and diagrams, and tables; timelines, tours, hypergaphs of nodes and structures, arguments, semantics, grammars, relational databases, frames, hierarchies, etc.
It seems especially noteworthy that the last category of symbolic structures seem to be diversifying and proliferating particularly quickly because of hypertext systems. That is to say, hypertext as a technology seems to be generating and promoting the development of new symbolic systems, whereas text and graphics within hypertext sytstems are being recombined in very similar and traditional formats. ( Although, for a real exception, see the information on George Miller's WordNet, below). In part, this is because hypertext is a symbolic system for communication in its own right, and it carries out this communication by guiding the reader through special navigational structures that link information in the many different nodes that are available. So timelines offer access and guidance to browse through the hypertext database, and so do tours and hypergraphs of nodes and structures. New access methods probably are waiting for novel symbolic systems of expression, perhaps from the rich domain of graphic and visual pattern recognition and visual abstraction. In fact, how to create iconic forms from realistic imagery may be one of the exciting challenges that takes us from purely relational expression to a new conception of visual semantics. However, this is an aside that will be explored more carefully in a further section on visual semantics.
Fig. 2.3 A Timeline of Geological History: Buttoning any label takes
you to a desription of that era.
Links
The things which connect together the nodes are called links, and when the whole content is richly interlinked it will form a network of nodes interconnected with links. It is hard to talk about links independently of nodes, since they carry information and knowledge symbiotically, together. Links may be typed (be restricted to a particular kind of node, such as graphic or numeric) or have attributes (such as a superordinate link or a synonym link), and they may be uni- or bi-directional. The user accesses the information in the nodes by navigating the links.
In constructing or traversing a rich network of linked nodes, (that is to say, for both the designer and the end user of a hypertext) it is very important to have bidirectional links. That way, if one end is changed or removed, it is possible to follow the link to the other end to examine and decide what changes need to be made. Otherwise, updating and making sure that a hypertext is coherent and consistent, without a lot of loose fragments hanging all around becomes an incredibly difficult chore. Also, it makes it much easier for the user to retrieve the context for a particular sequence of ideas she may be following. Being able to backtrack right from the node without having to move to a separate history list relieves much of the burden from short term memory constraints. Very few of the popular hypertexts have this backtracking capability directly. Usually, you have to do a special search of the hypertext database to find the source.
The MultiMedia editor being developed by Elliot Soloway and his group at the University of Michigan simply uses annotations. Because they are dealing more closely with hypermedia, or even multimedia, than hypertext, they can resort to a very special solution. Rather than creating hots spots in the text, on the words of the text themselves, they have a special scrolling field attached to the text, where icons for multimedia hotspots can be placed.
Figure 2.2 The Soloway Productions MultiMedia Editor (Ver. .023)
Icons
Graphic representations of visual reality, in a miniature form; icons have spawned a whole industry. They are the mainstay of the graphical computer interfaces created by Xerox PARC and commercialized by the Apple Macintosh and Microsoft Windows. As the examples above show, icons are heavily favored in most hypertext systems. As multimedia systems become more powerful, the icons begin to have more dynamic and dramatic features. Some show animations, or cycling video clips. These can be called videocons. Others produce sounds and noises when disturbed: audiocons. To distinguish them from the earlier static class of icons, these new things oculd be called dynacons.
Anchors
As a destination, the node is generally called an anchor. Again, the anchor can be a document, card, or some smaller component. When a link is traversed to an anchor, it has to show itself in the same ways that the source node did; by highlighting, boxing, or changing the cursor when it is moused. Anchors have been particularly championed by Norman Meyrowitz and his colleagues on Intermedia, but their notion of an anchor is much broader than just as a constant, fine-grained destination for a link. In fact they speak of anchors as any " persistent or sticky selection" in a text. Implicit in this notion is an idea that anchors are some small part of the text, such as a word or phrase. However, it is much more common for hypertext systems to have small source nodes than it is for them to have small destination nodes, so I will use anchor only to refer to small (word or phrase) destination nodes.
Some systems only have fine-grained sources, and coarse-grained anchors. Others, the other way around. In fact, general purpose retrieval systems can be thought of as having no source nodes at all, and only anchors that are found by the search and retrieval mechanisms, or the indexes available in the database.
Network
When many nodes or cards of information are linked together in consistent ways, they form a network. A network is an interconnected and interrelated group or system of nodes and links. The content of the hypertext network is both in the nodes and in the rich interconnection of the links . The figure below shows an example of the network built in The Learning Tool, an idea processor and outliner developed by Robert B. Kozma and John Van Roekel.
One of the ongoing controversies in the hypertext field is the nature of the structure imposed on the hypertext. This network of hypertext ideas may be constructed by the designer or the user to resemble in some coherent way the way the mind works or structures knowledge. This is then a semantic network of ideas . On the other hand, one of the strengths and conveniences of hypertext is that any node can be connected to any other node, arbitrarily. Anything can be connected to anything else! When done imaginatively, this can result in creative and unique arrangements of ideas that yield new perspectives and insights. It can also lead to garbage! The strength of these arbitrary networks , in other words, probably derives from the kind of serendipity that occurs when you are looking through the stacks of a library for one book, and come across the perfect book for the topic, even though you didn't know it existed.
However, the kind of serendipity that occurs in real world searches like this, really depends often on some kind of structured knowledge. For instance, in a library , the books are not randomly arranged, but ordered in a very complex way that derives from expected usage. In hypertext systems, there are many strengths and uses that come from an ordered arrangement of links. The coherent arrangement of nodes into complex networks that somehow capture the structure of mental information or knowledge, is particularly useful for instructional design, and will be articulated in the sections on AI and Intelligent Tutoring Systems. However, with both kinds of networks (arbitrary and semantic), the contrast between the network representation of ideas to the serial presentation of ideas in traditional text or media is profound and stimulating.
Node Types
Some hypertext systems allow only a few node types ( usually graphic and text), while others leave this as wide open as possible, and encourage the creativity of the user ( e.g. Notecards). For instance Storyspace has only Text and Display nodes types. The types of nodes that are available to the hyper-author really depend on how principled she wants to be and the flexibility of the system. At the moment, there are very few studies that have even begun to examine the kinds of nodes that users create for different purposes. Perhaps the most interesting evidence comes from the kinds of nodes created in IDE by Russell, Pirolli, and Greeno (1988). Another interesting example comes from gIBIS and IBIS. IBIS is a problem-solving environment that provides for issue, position, and argument nodes. Individuals involved in problem solving can state an issue, positions can be added and anyone can state arguments for those positions. An instructional hypertext might have rationale, examples, media, prerequisites, instructor, explanation, cause, models, and test nodes (among many more that one could derive from a formal analysis of instructional design). In typing nodes, you are constraining the authors and modeling the structure of the hypertext for the users. Most systems that type nodes signal the type of node to the user through color, background, icon, or shape of the nodes on the screen.
An even richer typing of nodes can be envisioned by thinking of nodes as specific content rather than general classification. Instead of dealing with the overall content of a card or page of a document, the destination of a link can be a phrase or a word or a graphic feature by itself. Naturally the source node can also be the actual icon or anchor of the link (i.e. a word or phrase or graphic feature). So one can think of these nodes not just in terms of whether it is a graphic or a word, but what its function is overall in the hypertext. More specifically one can try to envision whether it is a verb or a noun ( an action or an object) and what level of abstraction it has. Is it very abstract, akin to the very generic word "thing" or the very generic act "is" or "do"? Or, is the node at an intermediate level of abstraction, such as "dog" or "run"? Or is it a very concrete instance of something, such as " my cat, Licorice", or "the 100-meter hurdles"?
As Nelson has asked, is it a public node, or a private one? Can everyone edit it or only the author? Is it just for reading or for writing too? Who owns it? Who should be paid copyright for it? What are its attributes? What kinds of attributes can it be described by?
As Botafogo (1990) has shown, popular nodes that are visited often are very different from rarely frequented nodes. His nodes were cards with several paragraphs of text or graphics. These coarse kinds of nodes that are linked to many other nodes are central in some important way and have some unique characteristics. They are also different depending on whether they are popular source nodes or popular destinations. Indexes are great source nodes, but not very popular as destinations (why bother going to an index in a hypertext, when you can go straight to the node of relevance?). Important themes that run throughout a topic are popular destinations, but they usually do not have specific destinations of their own. Instead, people are expected to return to the source that brought them there. Prerequisites that establish the validity of other theories and contentions, or abstractions that generalize over an entire domain become very important sources with few destinations. But actions that many people hold as valuable, and that are complex, and difficult to understand will have many destinations but be the source for very few further nodes. However, this information does not yet get direectly at the issue of importance: how does one define a node within a conceptual space so that it is navigable?
The link type itself determines what sort of nodes are popular. Prerequisite links will point to many causes or foundational knowledge; whereas causal links will point to many complex phenomena as consequent of more primitive and simpler states.
Link Types
If nodes are the content of a hypertext, then links are the relations, and not just the poor relations. Links are every bit as important as nodes. If nodes can be thought of as the nouns and verbs of hypertext, then links are the attibutes, the adjectives and adverbs. In fact, buttoning on a node generally invokes the a general navigation link and has much the same meaning as saying "go to ..." the next destination.
Landow (1988) postulates the following heuristics for guiding the construction of hypertext links, in attempting to derive a rhetoric of hypertext. Landow develops a travel metaphor for links, and asks how links can be made to provide departure and arrival information to help guide the weary traveller. From the wealth of experience using Intermedia in an English course at Brown University, Landow has concluded that hypertext links stimulate the users to:
- expect purposeful, important relationships between nodes
- think relationally
- resent unimportant or uninformative links
In discussing these expectations, Landow has a very specific kind of relationship among knowledge components in mind. He thinks of relationships in terms of a concept map, expressed graphically. In this concept map, one topic is placed centrally like the spoke of a wheel, and other information is cycled around it, with links graphically connected like the spoke of a wheel. This spoke and wheel concept map is more or less intermediate between a purely arbitrary organization, and a very strict, disciplined hierarchical organization, such as a description of the job positions in a bureaucracy, or a semantic network. In fact, without some guiding hierarchical construction, the concept map may all too easily degenerate into a purely arbitrary linkage.
Conklin (1987) describes two methods for linking two nodes in a hypertext--referential and organizational links. Referential links connect a source in the current node to a referent in the destination node. The user is typically allowed to return via the same link. Organizational links impart hierarchical information. Organizational links connect a parent node to its children in a tree fashion. Users are presented special links to parent, child, or sibling nodes. Hierarchical systems are simpler with fewer navigation problems, but they constrain the user to pre-conceptualized organization.
Link types and Instruction
The use of links is closely connected to the power of the system for creating instructional paths through the web of knowledge. Rather than pre-determining the sequence of instruction for the user, exploratory microworld systems enable the user to choose the kind of instruction they interact with. One of the earliest explorations of this power occurred in the TICCIT system, where, for instance, the computer based instructional system provided definition keys, example keys and many other options to the learner at any point in the program. Lesgold (1988) has suggested that a prerequisite link is one of the most important for laying out instructional material.
The power of instructional hypertext has been best explored in this TICCIT environment over the past decade in a large project for the Army that has created a basic skills instruction program. The system now has more than 200 hours of instruction on many fundamental topics ranging from reading and writing, to motivational skills and learning strategies. In large scale tryouts it has led to 2 standard deviation improvements in instruction over traditional paper and pencil environments.
Browsing:
Browsing has many definitions and meanings. Animals browse by eating the best tidbits. Humans browse information by walking through stacks of books or flipping through pages until something strikes their fancy. Obviously, these can at best be metaphors for browsing through digital hypertext. Trying to be more precise is at best wordy and at worst confusing. One psychological perspective is to view browsing as the intellectual process of acquiring knowledge in an individualistic way.
Within hypertext, browsing means to find information. that is individually meaningful. Marchionini and Shneiderman (1988) describe a framework for a browsing system. The user or information browser is the key to the system. But his or her ability is affected by the setting, and the most important aspect of that is the content of the task domain or the subject matter. What are its characteristics? How is it organized? Finally, the user is affected by the search system that is used (see Jones, 1989) and this interacts with the user's background, characteristics, and search skills. Browsing systems are complex because of the support systems that must be built into them to adapt to all of these important variables.
Notecards is largely based on a large graphical browser, a hypergraph or hyper map. The Notecard browsers are each a structured hypermap of the network of notecards in the system. The browser illustrates the notecards and the type of links between them. Each of the cards that are illustrated in the browser are also links. The browser may be supplied by the author or generated by the user.
Larson ( 1986) suggests four basic operations to prepare for useful user browsing. Structuring pre-processes the information into an organized form (hierarchy, map, index, alphabetic list, web, network, outline, etc.) to facilitate browsing by a user. Filtering selects out relevant information for a user so that there is less information at any particular level of the hypertext. Planning provides the navigator with road signs of what is up ahead (when following a link or entering a node). Zooming determines the detail of the information that is displayed.
Furnas (1987) has created a unique envisionment environment for navigating complex nodes, the Fisheye View. This technique has been used in the SemNet project, described below. In it, nearby or semantically related nodes are enlarged, and distant or unrelated nodes are diminished dynamically, so that while the user is browsing through the system there is a constant dynamic zooming and refocussing of the neighborhood of nodes,
Hypergraphs:
Many hypertexts provide graphical views of the structure of nodes and links contained in the hypertext. These usually take the form of concept maps and are frequently referred to as graphical browsers. Some hypertexts use hypermaps as the main access vehicle to information.
Beware! Hypermaps will not help all users. They depend upon certain spatial skills which many users do not possess. It is a common fallacy to think that people are born innately with the ability to interpret graphs. A brief history of the use of graphs and charts, show that they are a modern phenomenon, created only in the last few hundred years. Many people are hopelessly inept at reading maps and graphs. For them, hypermaps may even impede their navigation of the database. On the other hand, hypermaps and hypergraphs offer a rich area for creative research, such as that underway by Roy Pea and colleagues at the Institute for Research on Learning (see also Simon and Larkin; Kosslyn and Pinker).
Control and Navigation
The most commonly identified user problem is navigating through hypertext. Shneiderman has coined this phenomenon "Hyperchaos". It is also known was being "lost in hyperspace" (cf. Conklin, 1987). Many hypertexts consist of hundreds and even thousands of nodes with a potentially confusing array of links connecting them. It has been well documented that in such systems, users can easily become lost, not knowing where they came from, where they should go next, or even how to exit the part of the program they are in. Users are frustrated by this experience. Often, they give up without acquiring any information from the hypertext.
Many different kinds of aids have been used to overcome this problem. Hyperties uses an index and online search functions. HyperCard uses an iconic overview of the last 28 cards seen. Notecards and many lisp-based systems use overviews in hypergraphic browsers. None of these techniques have really solved the problem, which may depend on semantic navigators that understand the knowledge-based hypertext and can provide interactive guidance. Certainly this is an important area for future developments.
Hypertext enables the user to make decisions about how to sequence information. By selecting or creating the associative links, the user may browse or examine information in many ways from different perspectives. Users in many hypertext systems may also amend or add to the nodes that they access in order to make them more complete or personally meaningful. Users may make comparisons, rearrange, or illustrate the information in many hypertext systems. The type and level of control may be changed by the user so the control is dynamic. This was a fundamental notion that Nelson had about hypertext, that the user could speed up, slow down, expand or elaborate the information. At any point, the user could ask for a quiz or example, argue with the text, or move onto another topic. This type of control is still unavailable in most traditional instructional environments.
Carolyn Foss (1987), has analyzed the problem of getting lost in complex materials into two phenomena she calls 'The Embedded Digression Problem' and 'The Art Museum Problem'. In the former, the user loses track of her original goals and interests and is sidetracked by ever new information. This is a common problem even with paper materials, such as an encyclopedia with cross-referencing. The Art Museum Problem occurs where, after spending a long period browsing through many images, everything starts to look the same, and loses its distinct meanings. A similar loss of meaning occurs when the same word is repeated over and over again: it soon looks or sounds strange. Try repeating the word "strange" about fifty times. The rapid succession of meanings that occur when hypertext lets users fly through text may lead to a similar flattening or leveling of meaning, not at all what designers may have intended.
Other researchers have provided some insight into the general process of navigation that might provide analogies or metaphors for understanding hypertext navigation. Several experiments have examined navigation and cognitive mapping in subjects experiencing a new, large-scale physical environment such as a town or city. The most notable of these was by Siegel and White (1975) who proposed a developmental sequence of cognitive representation whereby an individual's knowledge of the environment alters in form through a series of four stages, with each new form being superior to the last. They proposed that subjects initially recognised landmarks, these being objects that for some reason were prominent or notable in the environment, then formed route maps consisting of routes connecting the landmarks, followed by the creation of 'minimaps', which are survey-type maps of small areas, and finally developed full survey maps of the whole area by joining these minimaps together.
Tools for constructing these maps in hypertext documents seem a high priority. Currently, it is impossible to create "landmarks" consistently in hypertext for navigation purposes. There is not even a good theory of what a landmark should look like or how it should be implemented in hypertext. A good landmark in navigating a city is a prominent building or hill in the landscape that can be seen from afar. What does that mean in a hypertext system that shows only on card or page at a time? What sort of minimap could be constructed to string landmarks together, when we don't know what landmarks should look like in hypertrext?
SemNet
Another system for 3-dimensional navigation through large data spaces was the SemNet system (a semantic network editor) from MCC. SemNet uses a more traditional output medium than a head-mounted display however, as it shows a 2-D picture of the 3-D data structure on a traditional color screen. The idea is to allow navigation through a large knowledge base by a "helicopter metaphor". As you "fly" through 3-D space the view of the data base is animated: Some nodes shrink to the background while others show up larger and larger in a closeup. You can also get the computer to move for you along one of the links between data nodes or you can ask it something that cannot be done in nature: To "reel in" all the other nodes that are linked to some selected node so that you can get an idea of the structure of the knowledge space. The other nodes zip in from all directions towards the selected node (and they zip back to their proper 3-D positions when you are done). This user interface is certainly neat, but whether it will really help users grasp the structure of a large knowledge base is not really certain. Some experiments are now being carried out at MCC to test this.
Cyberspace:
Perhaps the most intriguing new development that can move hypertext into real hyperspace, is the creation of systems using datagloves and helmet mounted displays and goggles, for creating virtual realities. As Michael Benedikt put it during the first conference on cyberspace, "In cyberspace, information-intensive institutions and businesses take on a form, identity, and working reality --- quite literally, an architecture --- that is both counterpart and different to the form, identity, and reality they have in the everyday physical world." Through the awesome power of computers and these interfaces, geographic space can be distorted, compressed, and reshaped at will. People can literally walk and peruse artificial space as if it were real. And the fantastic can be made to be part of that reality. Touching an object can literally explode it into a new world: any object can be made the doorway of yet another reality. A physical - looking object can be turned into a hotspot that links immediately to another time and space. Only imagination can tell us where this development will lead. Already architects use it to explore building designs; businesses are exploring the techniques for teleconferencing; and others are beginning to use it for data analysis of complex relationships. As a natural expansion of the powers of hypertext, cyberspace interfaces adn technologies will let people experience the ultimate abstraction of navigating pure information, information that only exists within a computer and someone's imagination.
The most imaginative notions about cyberspace stem from the first book to hold the word, "Neuromancer", by William Gibson (Ace Books, 1984).
Cognitive Overload
Dede (1988) also cites the problem of cognitive overhead. The exponentially greater number of learning options available to learners places increased cognitive demands upon the learners that they are often unable to fulfill. Effective browsers must be able to monitor their own comprehension of the information presented in the hypertext, select appropriate strategies for rectifying any misconceptions, develop information seeking strategies that facilitate integrating information. These are known as meta-cognitive strategies, and they require additional effort on the part of the browser. We know that good learners use them and poor learners do not. "The richness of non-linear representation carries a risk of potential intellectual indigestion, loss of goal-directness, and cognitive entropy" (Dede, 1988, p.8). Browsing hypertext places significant demands upon the user.
Since hypertext is a new information source, users have not developed skills in navigating, integrating information , and so on. Most novice browsers report discomfort. Many browsers are afraid and don't want to commit the effort. This is partially because they do not have a schema for using hypertext and partly because learners are not inured to the types of information processing stimulated by hypertext. It is a novel form of study. Marchionini (1988) points out that if large amounts of our reading in the future will be by unguided and unconstrained electronic text, new strategies (a hypertext literacy) will be needed. Use with college students has shown that they soon adapt however. Sophisticated browsers develop strategies for searching, comparing information, and solving problems. Effective browsers need to develop hypertext processing strategies in order to effectively use hypertext.
Hypermedia:
Hypertext is not always comprised of text only. Hypertext is often integrated with other technologies which are capable of producing and displaying sound and speech, graphics, video, and so on. This multi-media information constitutes the nodes, so that choosing a link to another node may result in a video display or a speech. This combination is known as HYPERMEDIA.
Hypermedia systems usually interface a computer to a graphic/sound device. Popular combinations include a videodisc player, a CD-ROM player, or CD/I. Videodisc players can provide still frame and motion video sequences. CD-ROM systems provide text only. For large hypertexts, CD-ROM is used to store the hypertext because of its large memory capacity (600 megabytes). The ultimate hypermedia technology is CD/I (compact disk interactive). This emerging technology can store sound, graphics, still and motion video, and text in any combination on a 4.72 inch compact disk. Hypermedia systems often run on advanced graphics workstation (eg. SUN) which are capable of producing
Storage:
Compact disk interactive was announced in 1986 and is now being delivered by Philips and Sony. Consumer versions are expected in 1990. CDI uses the same technology as CD-ROM but permits the encoding of a variety of signals. CDI disks may contain any combination of the following:
-650 megabytes of text (approximately 150,000 pages)
-up to 7800 still video frames
-1.2 hours of high fidelity audio
-19 hours of voice quality (low fidelity) audio
-several minutes of partial-screen motion video
Consumer CDI's should provide interactive encyclopedias, games and simulations, tour guides, and many other applications. CDI, at this time, appears to be the ideal hypermedia medium. It provides mass storage of a variety of information forms.
Digital video interactive is a new technology that uses sophisticated data compression routines to provide greater capacity to CDI and CD-ROM systems. Its major addition is full screen motion video. It provides up to 72 minutes of motion video, although the image quality is somewhat poorer. DVI also requires a more powerful and capacious computer as well as additional computer hardware to decompress the image. This additional hardware investment may affect its growth in the consumer market. However, corporate, agency, and educational hypermedia systems may profit from its denser information storage capabilities. Although prototype DVI systems exist, it probably will not be generally available until 1991.
Index:
A subject index lists the topics in a document in an arbitrary sequence (ie. alphabetic ). Given the computational use of search and query systems, such a purely arbitrary index seems unconscionable. Other sources for structured indexes such as on-line thesauruses, and hierarchical dictionaries( e.g. WordNet) provide intriguing models for potentially more effective indexes.
Often, the user may want to jump to a term that is not present on the screen. An index would permit the user to search the index to identify a term, and then jump to it. The hypertext designer could constrain the possible connections through the use of a syndetic structure (cross references, SEE and SEE ALSO). This would prevent the user from making random and completely meaningless connections. The index may show all the possible connections available from a specific part of the hypertext rather than represent the structure in an alphabetic listing. Another option is to list the related index terms and/or cross references in a window at the bottom of the screen to let the user know what indexed terms are related to the current topic.
Embedded Menus
The grain size of a system seems to determine whether or not the hotspots or buttons feel like embedded menus. The card-based systems, for example, generally use a whole card as the unit for linking to, while they allow single words or smaller units to be used to link from. The scrolling systems, especially those with anchors, can use single words or icons as the destinations of links. Certainly there is something of the feel of using menus - albeit menus embedded in text - with the card-based systems, whereas text-oriented systems such as Intermedia or Notecards retain much more of the feeling of reading linear text. In fact, just as there is a continuum between hypertext and semantic networks, there is one between hypertext and menu systems. In this case, the dimensions which vary are; the size of the displayable unit (i.e. what constitutes the display of a linked-to node), the explicitness of the marking of links in the text, and the amount of embedding of the linked text in non-linked text.
Hot-Keys versus Point-and-Click
A continuing controversy in the development of user interfaces, particularly hypertext systems (which can be thought of as one consistent interface
Computing on HyperText Structures:
In order to bring the strength and dynamism of computed interactivity to hypertext, it is necessary to have the nodes and links defined in such a way that an interpreter can make decisions on the basis of their type and content. The computability over nodes and links in hypertext depends on a consistent semantic structure applied to the hypertext. Although some initial attempts to define this structure is evident in the work of Lenat on the CYC project (Lenat, 1987) and in much other research on machine learning and knowledge representation (e.g. Porter and Bareiss, Proteus Project [1988]), there is little consensus on how to bring semantic uniformity to hypertext structures. This remains one of the most important unresolved issues that would lead to the introduction of AI techniques and a marriage of AI and hypertext.
Perhaps the best example of the issues and problems involved comes from Dan Russell's IDE Interpreter (Russell, 1988). In his work, the link types for instruction were variations on a general Prerequisite link, and on a semantic Causal link, and the hierarchical hyponymic, hypernymic, meronymic, and holonymic links. With these links it became very important to be able to retrieve only those nodes connected by particular kinds of links. In effect, it became important to have a structural kind of retrieval, with certain kinds of input links (say "prerequisite" and "exploratory" ) along with certain kinds of output links (say, "test_negative") attached to the node that is searched for and retrieved.
Doing this repeatedly within the Notecards environment proved quite limiting; and so recent exploratory versions of IDE have been built on top of a commercial database system, yielding somewhat limited text processing capabilities, but vastly improved speed of access and retrieval of structured nodes, rather than simple strings.
StrathTutor
A truly generative vision of how to use computational power on links comes from the work of Kibby and Mayes at Strathclyde.
In the StrathTutor, each frame is described with a limited set of attributes. the system can then automatically compute the 'relatedness' between nodes in the hypertext, on the basis of conceptual connectivity or semantic proximity. A simple function will take the user automatically to the most appropriate node without the need for the link between the two nodes to be explicitly drawn by the designer.
This work is loosely based on the holographic and associative ideas of Richard Semon (1923). It is more directly linked to more resent research by Hintzman. His contribution has been to cast this theory into the form of a computer model, MINERVA, He uses a three-state attribute system: the attribute is present, present in a negated form, or not-present represented by the values +1, -1, or 0
Tanimoto similarity measure Application to CAL: StrathTutor
™ Authoring without programming and no coding at the level of the organization of the learning material. No links between frames explicitly represented and no sequencing necessary.
™ Knowledge represented as attributes;. Up to 60 attributes used throughout a tutorial to code each frame of text and graphics.
A frame may be deleted and the system will still operate without the need to remove links now undefined. Similarly, a frame together with its attributes may be added and the system automatically
The intelligent system they propose may well require large sets of attributes to describe extensive amounts of material. However, it is likely that attributes may be redefined in 'local' areas, or have a structure imposed on them.
Typographical Cuing
The way the nodes in the NoteCard Browser are displayed uses a bitmap icon to represent the 'node'. A bitmap can be altered to give different forms. Many different types of icons can be used in a browser to indicate different functions of HYPERTEXT elements. The Filing Box icon can be used to suggest a collection of further items. Other icons are not as descriptive, in this instance, but can be associated with different levels of detail or items containing bibliographic information, or media, or guidelines. The lines used to indicate the path from node to node are also differentiated. Lines of different strength and format can be used to indicate main paths, minor paths, etc.
Unresolved issues in HYPERTEXT research.
Halasz (1987, 1989) showed that there are enough unresolved problems to keep research going for some time yet. Halasz' seven issues were:
1) How to do search and query.
2) Composition of the basic nodes and links to higher structures.
3) Virtual structures.
4) Using computation to change HYPERTEXT networks from being passive to being active.
5) Versioning.
6) Support of collaborative work.
7) Extensibility and tailorability.
In summary, Halasz felt that items 1-3 on his list were the most important. In the long term he would assume that HYPERTEXT would become a standard inside computer systems.
Classifications of HYPERTEXT SYSTEMS:
According to Nielsen (1989), hypertext systems can be divided into on the one hand the "original" generation of
Memex [Vannevar Bush], NLS/Augment [Engelbart], Xanadu [Ted Nelson], and on the other hand the "current" generation consisting of e.g.
™ Research systems: InterMedia [Brown University], NoteCards [Xerox], Janus (Colorado), Hyperties (University of Md.)
™ PC Product systems: Guide [Owl], HyperCard [Apple], Hyperties [UMd.], LinkWay (IBM),Architext, Plus, Supercard, Black Magic
™ Workstation products: Document Examiner [Symbolics], Hyperties (Sun), Neptune [Tektronix], GIBIS (MCC).
Further dimensions for classifying hypertext systems are:
Scope of the user target:
™ Single user: Guide (IDEX), HyperCard, Hyperties, Architext, Plus, Supercard, the original NoteCards, Ntergaid's HyperWriter
™ Work group: InterMedia, IDEX, NoteCards recent version.
™ Corporate division: Augment, (ZOG, KMS [Carnegie-Mellon]).
™ Whole World: Xanadu.
Browsing vs. authoring:
™ Focus on information presentation: Document Examiner, HyperTies [Shneiderman], Architext,Guide (IDEX), KMS, Plus, Supercard.
™ Focus on knowledge representation and manipulation: Augment, Neptune [Tektronix], NoteCards.Hierarchical Databases
Hierarchical Databases
Many databases now integrate hypertext components
Expert Systems
Expert system shells for the PC world routinely integrate hypertext and multimedia management systems.
Knowledge Garden's Knowledge Pro
Nexpert
Kee
1st Class
Intelligent Developer
™ Task specific: Document Examiner[Documentation], GIBIS (Argumentation), Janus (Colorado), Thoth (argumentation)
The Function of Hypertext Systems
The various systems that have been developed can not really be classified in any formal way that is generally agreed upon. All of these systems have been developed without the help or guidance of anything like an overarching theoretical framework. By and large, we still have no good notions of how to use hypertext. The best ideas come from the historical, pre-hypertext environments of paper-based tools; and these are obviously biased and inadequate. Yet these are the sources for the design and use of current hypertext systems. One of the most obvious impediments to novel hypertext uses are the existing skills and preferences of the users. Since they are particularly expert at paper-based environments, they often want those features (Outlines, indexes, notes, headings, abstracts, sequential structure, etc.) in the new hypertext systems. Obviously this presents a basic conflict whose resolution depends not just on the technology, but on the gradual change of user's skills as well; and that may mean one or more generations of gradual transformation. Users may change much more slowly than the technology!
Nevertheless, here is a kind of typology of hypertext systems, with some of the more prominent systems arranged under their dominant characteristics.
Programming and Design Environments
It is instructive that HyperCard does not advertise itself as a hypertext system, although it does call itself a hypermedia system in several advertisements. Like Linkway, HyperPad, Supercard, Plus, and a swelling number of other systems, HyperCard is really a programming environment. Their strength comes from an ability to bring many tools into the same environment so that text editing, drawing, and programming computer languages can all be done harmoniously without tediously switching from one system to another. At the same time, they offer the ability to do programming indirectly, by creating or modifying objects directly and having the underlying code automatically updated. The ability to copy and edit ;the objects themselves rather than the code that creates them makes this programming appear effortlessly easy to acquire, and in contrast with traditional languages it is easy to acquire. Perhaps the most advanced programming by example, or programming by direct manipulation (as it might also be called), can be found in the NextStep environment on the NeXt machine. However, it lacks many of the interactive hypermedia components of the other environments and is much more purely a programming by selection environment. All of these systems have an underlying functional language with some object oriented components, that can be used by more expert and daring users. However, these underlying languages (the languages of advanced resort) are generally not very powerful. They usually do not have the features of more traditional languages, often missing compilers and complex data structures like records, arrays, frames, objects, and inheritance.
Information Retrieval
To many people, hypertext represents a database of information that facilitates searching and accessing information. In many ways it is. Its organizational structure is similar to that of a database, which organizes information in records, tables, or higher order objects such as frames or hash tables. In fact, many hypertexts store the information contained in the nodes in a database, which facilitates access to the information when the user links to it. Unlike a database, which usually provides operators to retrieve the database in record or table form, hypertext retrieves in more complex representations. Hypertext can represent information multi-dimensionally. That is, the links are not constrained by the two-dimensional data structure of a database.
Instead of the very complex query languages found in most databases, retrieval is simplified ultimately to the act of buttoning a spot or word on the screen. Instead of the complexities of database programming languages, very simple Do As I Did (DAID) direct manipulation or programming by doing concepts are used. Of course, these are also potential techniques for database systems, and as DBMS adopt these techniques hypertext and databases will become more intertwined.
Certainly the issues of text retrieval are central to many effective hypertext systems. Many hypermedia systems are belatedly adding additional retrieval capabilities that go beyond simple string searches. Thus conditional searches are being added to Hyperties; graphic searches to Hyperties and hypercard; HyperKRS adds boolean and wildcard searches to HyperCard; Notecards is adding full text search to its original capabilities of searching titles; etc.
The relationship between hypertext and standard information retrieval systems is being actively explored ( see also the retrieval effects of hypertext in Fischer, McCall, and Morch's Janus system). Handcrafting links among nodes adds both to the accuracy and specificity of the retrieval process, improving it greatly over standard techniques. For instance, in large databases the probability of finding something in the first place you look is around 20 % (Furnas, Landauer, Gomez, and Dumais, 1987). At the same time the proportion of relevant documents retrieved by the standard boolean search is usually well below 50 % (Salton and McGill, 1983). Given these abysmal figures, when even a 99% figure may be too low in very large databases, the specificity and power of pre digested hypertext may become extraordinarily important. If, as futurists are excited to tell us, the whole world's knowledge base may be at our fingertips over glass fiber lines within our lifetimes, information retrieval will be no small problem and hypertext may be an important part of the solution.
Interface Systems
Hypertext and hypermedia systems can readily be viewed as interface systems that provide a common front end to many disparate functionalities. The most prominent hypertext system that has also inspired the view of an interface system is Notecards. Notecards created a variety of card types -- text cards, graph cards, browser cards, simulation cards, animation cards, etc., etc. --- all having a common menu and visual structure. To some extent we can also see this phenomenon continuing with HyperCard. As each new developer brings out a special product for hypercard, the product automatically inherits the global properties of HyperCard, its menus, abilities to cut and paste, draw, write, etc., and the special functions such as find, home, index, and the ability to create buttons and fields. This creates a consistency among different applications that certainly makes the user's task much easier, and perhaps helps ensure that the developers actually incorporate as much functionality in their product as they reasonably can. At the least it acts a as a reminder that certain capabilities are indeed possible.
<< interface in HAWK MACH - III >>
Knowledge Representation
Some hypermedia environments are really mainly complex knowledge representation systems. Notecards falls best into this category. With its graphic browser and tools for computing on the titles of cards, it appears to create strong pressure to use titles that uniquely specify one concept. Then by ordering these concepts in the browser into hierarchical structures, Notecards provides an easy access to the generic knowledge structures. Even inheritance is possible by developing card types that suit conceptual categories. These card types then can create defaults that are conceptually specific for each new subconcept. Notecards encourages the exploration of conceptual structures and relationships. Each node can be a unique concept, and each link or link type can be an attribute of that concept.
Reading and Writing Systems
Hypertext as Multimedia