web hosting short URLs photo sharing

1. Introduction *

1.1. Starting point: a textual migration *

1.2. An experiential approach *

1.3. Micro- vs. macrostructure *

1.4. Hypertextual vs. narrative syntax *

2. The concept of hypertext *

2.1. What is hypertext? *

2.1.1. Definition(s) *

2.1.2. History *

2.1.3. Some terminology *

2.2. Some aspects of hypertext *

2.2.1. Hypertext and its relatives *

2.2.2. A cognitive model *

2.2.3. Symbol vs. icon *

2.2.4. Informational vs. fictional hypertext *

3. Hyperfiction: formal classification *

3.1. Uni-linear structure *

3.1.1. Plain linear *

3.1.2. Annotative *

3.1.3. Tree-branching *

3.2. Multi-linear structure *

3.2.1. Plain multi-linear *

3.2.2. Braided multi-linear *

3.2.3. Nested funnel *

3.3. Idea space *

3.3.1. Rhizome *

3.3.2. Storyworld *

3.4. Non-linear *

3.4.1. Multiple User Dimension (MUD) *

3.4.2. Random elements *

4. Conclusion: toward a hyperfiction poetics *

5. Bibliography *

The aim of this dissertation is to build a realistic image of the possibilities of hyperfiction and to evaluate its potential contribution to the concept of narrativity. In the introductory remarks, we will clarify the point of departure of this work and its general approach. In the first section of the main body, we will look at some background and typical characteristics of the hypertext medium. We will try to give a definition, sketch a history of hypertext, look briefly at some terminology employed and discuss some important typicalities of the medium. The second section will be dealing with hypertext fiction more specifically. We will attempt to make a formal classification of the different types of hyperfiction, and assess the currently prevailing theory against our findings. To conclude, we will discuss the possibility of a hypertext poetics and what it could look like.

The issue of whether the book will be able to hold off electronic text or not has often been discussed, without reaching an agreement however. Many theorists believe the book’s days of solitary rule in the realm of text are numbered. Tolva claims "Despite exaggerated reports of its demise, the codex book is not dead--but, like handwriting in the age of print, it isn't likely to remain the dominant means of textual dissemination" (Tolva 1995). Others believe that reading off a screen will never become customary. Their arguments vary from the traditional "cannot read it in bathtub" to the fact that screens will never be able to offer the same resolution as the book does. Moreover, a screen is too small to give a clear and neat overview over the matter. These claims cannot be denied! However, we will now look at some arguments to the contrary effect.

First and foremost there are the economic advantages of the electronic form. Even today, when we look at the proportion of the amount of information and how many resources the medium needs to transport it, we must admit that virtual text clearly takes the cake for what concerns the price of the materials, distribution cost, and ease and cost of storage. Second, not unimportant are the ecological advantages of the electronic medium. A priori, no paper is used until the electronic text is printed which should mean less deforestation. Moreover, the transport is overall more environmentally friendly. And finally, there is not supposed to be any direct material waste due to the amount of information transmitted, which is significant when we are moving to an information society.

Third, and probably nearest to the average user, is the practical advantage of virtual text as opposed to material, especially in a network environment. We will just name a few: the ease and speed of replication of originals, the ease and speed of transmittance, the ease of manipulation and interactivity, the multimedia possibilities, the availability and up-to-dateness, the power of full-text information retrieval etc. These advantages are probably the impetus behind the exponential growth of the World Wide Web (WWW). According to Matthew Gray from the Massachusetts Institute of Technology, the WWW still doubles every three to six months. As for the Internet in general, the number of domains on Wednesday March 18, 1999 was 5,222,894.

The least we can deduce from these statistics and advantages is that they are impressive. The chance that the Internet and electronic text will play a major part in the information society of tomorrow is extremely great. Therefore, in this paper, we will avoid the hypothetical distinction between ‘if the book remains’ and ‘if the virtual text takes over’ by postulating a textual migration towards the virtual. In other words, electronic text, as opposed to printed text, will be assumed to become "the dominant means of textual dissemination" (Tolva 1995). The claim is not that we are evolving towards a paperless society. Rather, we choose for electronic texts as a working hypothesis, which--as we will see in the next section--enables us to avoid much of the hypertext 'claptrap.'

Heretofore, mainly two attitudes towards hypertext and especially towards hypertext fiction have been prevailing. On one side, we find hardcopy elegists like Sven Birkerts, who made it their life’s work to defend the book and resist to any change. Equally on this side, we find journalists like Laura Miller, who despise anything that deviates from the classical author, ''a solitary voice whispering in your ear" (Miller 1998), and who denounce what they call "the alienation of academic literary criticism from actual readers and their desires" (ibid.). Their slogan is ‘nobody reads hypertext fiction and nobody wants to!’

On the other side, we find the hypertext enthusiasts, who, especially in the early days, made ludicrous claims about hypertext going to change the whole of society with everything in it. They believed that hypertext would finally enable us to think freely, unbound from linearity. Some even saw and still see it as a counterweight to capitalist thought. Now that the early hype is over, we see that these early claims were nothing more than a cry for attention. Hypertext now has a reasonably stable basis to develop from, i.e. the WWW. The attention is there! Thus, neither as a synthesis of nor a compromise between the two views described earlier, I propose a third, which we may call an experiential approach.

As we have seen in the first section, there are reasons to believe that hypertext or some other similar device will take over or at least diminish the current dominance of paper in the domain of textuality. Therefore, as a point of departure, we have chosen to postulate a textual migration towards the virtual. This postulation has two major consequences with which we will deal now: (1) there is no longer a necessity for disproportionate claims, and (2) the theoretical angle changes from existing post-structural theory to an inductive or experiential approach.

(1) When we choose to take for granted that we will soon be working with virtual, rather than paper text, the focal point of the discussion can move from 'is electronic text desirable or not?' to 'what can we offer in electronic text, which we could not on paper?' or 'how can we carry the new hypertextual tools to their highest performance, especially for hyperfiction?'. Or in Tolva's words: "[T]he question [becomes] not if computers will transform our notion of reading and writing, but instead how?" (Tolva 1995). This allows us to avoid much of the traditional hypertext blah. We no longer have to deal with the bathtub argument. Neither do we have to prove that hypertext will change our whole way of thinking. Instead, we will try to analyse the current hypertextual devices and see how they could enhance narrative.

Moreover, we no longer have to consider hypertext as being an enemy of the book, because it is not. It is a successor or a child, of what we thus far considered the only or at least the best information carrier possible. As Bolter correctly observes: "One irony, then, is that the defenders of the printed book have not really identified their enemy. The computer as hypertext extends the possibilities of verbal rhetoric. It is computer graphics that challenge the representational basis of both printed novels and hypertexts. Virtual reality, not hypertext, is the antithesis of the printed novel" (Bolter 1996). As a result of this observation, the pressure within the kettle can eventually diminish. This should permit the discussion to become less biased and more scientific and serene. Very often the experiential approach will boil down to concretising, relativising and challenging, but not a priori discarding early claims about hypertext.

(2) Until now, a great deal of literary theory about hypertext and hyperfiction has been written starting from poststructuralist theory like that of Derrida, J. Hillis Miller, Barthes, Deleuze etc. For lack of a concrete tradition of hypertexts and hyperfictions, hypertext theorists like Landow and Moulthrop decided to put the emphasis on the similarities between hypertext and the type of text poststructuralists have been advocating. The fact that their arguments were based on their experience with paper texts did not seem to bother. And indeed, there are many similarities between the two. Many notions have been successfully transferred from one field to the other. The question is, however, if one should maintain this parallel. Is it not high time to cut the navel-string?

Again, I would like to refer back to our point of departure. When we postulate a textual migration, this implies that the aim of our theory is to be concretely applicable on the one hand, and based on experience on the other. This does not mean that all traditional hypertext theory needs debunking. What we will rather try to do is form a counterweight, a challenge to existing theory. In the first part we will therefore take a closer look at some hypertextual devices and their characteristics. In the second, these will be linked to what has up to now been established with them (on the WWW), when it comes to narrativity. This we will do by making a formal classification of the different types of hyperfiction.

Now, we will look at a comparison made by McGann in his "Rationale of Hypertext." He reminds us that the Internet is an archive of archives that "was originally designed precisely as a decentred, nonhierarchical structure." It was meant as a network that could resist partial destruction or non-availability, that could be "cut at any point, at any number of points, and still remain intact as a structured informational network (…). That kind of organisation ensures that relationships and connections can be established and developed in arbitrary and stochastic patterns" (McGann 1995). Next, he moves to the resemblance of this type of structure with that of the library as we know it. It is designed to allow indefinite expansion, it "can be accommodated to any kind of physical environment, and is neutral with respect to user demands and navigation. Moreover, the library is logically "complete" no matter how many volumes it contains."

Then McGann continues by describing even more affinities between both organisational patterns and the similar experiences of its users. In the end, however, he states somewhat hesitantly "each unit of the organization (each document and also each set of documents), like each node on the Internet, is logically defined as an independent item" (McGann 1995). Logically defined yes, but is it true that each node is independent in the same way a book in a library is? Does it relate to any other node or page of the macrostructure in the same way as it does to that of the microstructure, which it was specifically designed for? I believe that this is too optimistic.

The Internet has, indeed, a very decentred and anti-hierarchical structure, but not a non-hierarchical one. Therefore, we should distinguish between the microstructure (the book) and the macrostructure of the whole WWW (the library), as for both different rules apply. In other words, experience has proved that most Internet sites are hierarchical. It is not so that every page is equally related to every other. Rather, there are sub-webs according to the topic discussed, the country of origin, etc. In this dissertation we will be focussing on microstructure and occasionally we will refer to macrostructure.

As a final introductory remark, I would like to make a difference between two levels of abstraction, which we will need more than once throughout this paper, i.e. the level of (1) hypertextual syntax and the level of (2) narrative syntax.

(1) The level of hypertextual syntax is the one that is most difficult to define or even describe starting from other media. When we look at a paper text, we find all kinds of different tools, which divide the text into bits and pieces. There are, e.g., spaces between words, punctuation, the difference between capital and lower case letters, different fonts, paragraphs, chapters, books etc. What all these seem to do is distinguish and create a hierarchy between the different linguistic units situated on 'one long line of text'. They allow us to see not just the text but also within it, thus defining the status of each unit and giving it its place and character within the whole.

These devices are closely related to what we will describe on the hypertextual level. The only difference is that the 'line of text' is to be replaced by a large clipboard on which all the pages or nodes of the hypertext are to be attached. Then, it is up to the links to establish the connections between the different bits of text. The hypertextual syntactic level thus exceeds linearity and allows the text to become a space rather than a line. Within this space, many more different relations between units become possible. It is these relations that constitute the hypertext syntax and it is these that we will attempt to describe in our formal classification in the second part.

(2) The second level or the level of narrative syntax, on the other hand, is much less medium-specific. It primarily contains units of narrative, rather than material units. In other words, you have a kind of meta-structuring, which does not necessarily manifest itself on the first level. As an illustration, we could oppose direct speech as a second level phenomenon (a character speaks) clearly manifested on the first level (quotation marks). Indirect speech, however is the same phenomenon, but already much less articulated on the first level (word choice, sentence structure). Phenomena like pointing forward (prolepsis), however, are purely or at least for the most part ‘meta’ since they do not have a direct relation with any construction of the first level.

The transition we are experiencing now, i.e. one from paper to virtual text is primarily one to be situated on the first level. Linearity is replaced by geography. It is the medium that changes. In this way, even for hyperfiction, the shift should be seen primarily as an extension of the existing linguistic set of tools, rather than a new literary genre. It is therefore more comparable to the transition from manuscript to book culture than to that of romance to novel for example. However, this does not mean that, secondarily, the medial change will not have its consequences for the narrative level as well. In fact, many elements that were traditionally only realised on the narrative syntactic level (e.g. point of view, focalisation) can now be realised within the geographical space of hypertext (cf. infra). This may, in turn, influence what is represented on the narrative level and allow hyperfiction to evolve away from existing literary forms.

Moreover, the shift is not radical or binary, i.e., one from no to yes, but one of degree of ease. Hypertext and hyperfiction were also possible in book format, but they had nothing near the power of an electronic hypertext. Or as Landow has it: "Hypertext linking simply allows one to speed up the usual process of making connections while providing a means of graphing such transactions" (Landow 1997: 81). However, "Changing the ease with which one can orient oneself and pursue individual references within such a context radically changes both the experience of reading and ultimately the nature of what is read" (ibid.: 4). Thus, it is probable that a new genre will emerge from the new medium, but this is not the real transition at issue, it is only a concomitant phenomenon.

Probably the best known example of a hypertext in book format is the traditional codex book. Within the text you find references to other, separate pieces of text i.e. notes, which then, in turn, can refer to another text or another part of the first text or just give some additional information. These references are obviously paper counterparts of what we nowadays in electronic hypertexts call a link. Indeed, at first sight, the only significant difference between both is the fact that, when working electronically, you do not have to browse through the book manually, since the computer does it for you. Although there are some subtler differences, which will be dealt with later on, we can say that the codex is an informational hypertext avant la lettre.

However, theorists have also found fictional pre-electronic hypertexts, some that go back several centuries. We will restrict ourselves to just mentioning a few instances, since it is beyond the scope of this dissertation to deal with them more amply. Most commonly mentioned, we find "Sterne’s Tristram Shandy, Tennyson’s In Memoriam, James Joyce’s Ulysses and Finnegan’s Wake and recent French, American, and Latin American fiction, particularly that by Michel Butor, Marc Saporta, Robert Coover, and Jorge Luis Borges" (Bolter, quoted in Landow 1997: 182). Finally, Keep also makes a parallel with the book of books, i.e. the bible. Although sometimes far-fetched, he does come up with several good arguments, as there are: the different interpretations of the text which together constitute its meaning, the multivocality within the synoptic gospels and the habit of just opening the book at random and start reading. Thus, the mother of all books could also be seen as an ancestor of hypertext.

2.1.1. Definition(s)

What the term 'hypertext' exactly designates, remains a problematic issue. Theodor H. Nelson coined the word in 1965. It is a conjunction of ‘text’ and the Old Greek 'uper' (hyper) meaning 'over, above, beyond, besides'. "By ‘hypertext,’" Nelson explains, "I mean non-sequential writing--text that branches and allows choices to the reader, best read at an interactive screen." (Nelson, quoted in Landow 1997: 3). Note that Nelson meant hypertext to be read off a screen. The examples of non-electronic hypertext were attested only later, when critics started seeing the resemblance between Nelson’s ideas and similar developments in earlier experimental novel writing.

Another term invented by Nelson is ‘hypermedia’, which implies linking and navigation through material stored in many media: text, graphics, sound, music, video, etc. However, the existing hypertext systems prove that this term is generally superfluous. On the World Wide Web for example, we see that graphics and other media are handled in almost exactly the same way as text chunks. Therefore, it may be more interesting to allow ‘text,’ as in ‘hypertext,’ to contain both symbolic and iconic signs (cf. 2.2.3.) and also streaming media.

A very comprehensive and useful definition in this respect is the one we find in the Oxford English Dictionary (OED). Hypertext is "text which does not form a single sequence and which may be read in various orders; especially text and graphics (...) which are interconnected in such a way that a reader of the material (as displayed at a computer terminal, etc.) can discontinue reading one document at certain points in order to consult other related matter" (OED). We will use the OED definition as the starting point of this discussion. To be able to include non-electronic hypertexts within the scope of the term 'hypertext', we will employ 'explicit reference' as the criterion to distinguish hypertext from text with hypertextual qualities (only implicit reference).

Hypertext is a very broad phenomenon and it has been interpreted in many different ways. Within the realm of hypertext research, at least three types of definitions can be distinguished, viz. one coming from (1) literary theorists, one from (2) computer scientists and one from (3) cognitive scientists. Each type goes with its own methodology, theories and descriptions. Although there are more definitions present in the field than those we will look at, most of them can be classed under one of the three types mentioned.

(1) Nelson saw hypertext as a literary device. "Hypertext, or non-sequential writing with free user movement along links, is a simple and obvious idea. It is merely the electronification of literary connections as we already know them" (Nelson, quoted in De Bra 1998). Many literary critics agree with Nelson. They see hypertext as an opportunity to transcend the restrictions of the traditional written text, although they still see it as a derivative of the latter. "A hypertext is like a printed book that the author has himself attacked with a pair of scissors and cut into convenient verbal sizes. The difference is that the electronic hypertext does not simply dissolve into a disordered bundle of slips; the author defines its structure by establishing electronic connections among the slips" (Bolter, ibid.).

(2) Literary theorists like Landow would like to see hypertext as a convergence of contemporary critical theory and technology. Many computer scientists, however, approach hypertext as an information-delivering medium, placing the emphasis on its capacity to store and retrieve large amounts of information. "Mechanisms are being devised which allow direct machine-supported references from one textual chunk to another; new interfaces provide the user with the ability to interact directly with these chunks and to establish new relationships between them" (Conklin, ibid.). They see hypertext as a database, rather than a text. Hypertext is defined as "a database that has active cross-references and allows the reader to "jump" to other parts of the database as desired" (Schneiderman, quoted in De Bra).

The wide gap seemingly yawning between both these approaches should not be such a problem however. When talking about hypertext, computer scientists focus on the system, whereas literary theorists will refer to the composition of a hypertext when the system is already there. In other words, when considering the WWW and its HyperText Markup Language (html), computer scientists will be interested in improving the quality of the authoring tools and html itself. Of course they do care for the possibilities of creating a good hypertext, but they do this from a different perspective. For them, what is at issue is whether the connections made by the reader or author are clear and meaningful or at least that they can be so. The software is therefore generally made to produce informative rather than fictional hypertexts.

Literary theorists, on the other hand, will try to enhance the literary use of the tools offered. Thus they are not always concerned with the clarity and meaningfulness of the connection. Sometimes a good fictional hypertext deliberately employs an opaque connection to produce the right effect: to surprise the reader for example. Literary theorists are therefore not primarily concerned with hypertext software, just like they have not been with the mechanics of a typewriter. An exception on the rule has been the creation of Storyspace, a software platform specifically designed for writing hyperfiction, for which literary theorists and programmers worked together. However, lot of confusion could be avoided by using terminology more adequately and by distinguishing between hypertext systems and hypertexts themselves (cf. 2.1.3.).

(3) The third approach of hypertext research is more directly opposed to the literary viewpoint. Cognitive scientists, like computer scientists, see hypertext as a new means for structuring large databases. However, they focus on the resemblance shown by this type of structuring with the way the mind structures and accesses newly acquired knowledge. Thus, the emphasis is not so much on the storage itself, but on the associative linking between the different units. Just consider the following description: "it is appropriate to view hypertext as a method of supporting the expression of relationships among objects in a database. Hypertext should be treated as a general-purpose tool with approaches to handling nodes, links, and retrieval, that fits within the context of any application and convey common meanings to the users" (Rao, ibid.). What cognitive scientists try to do is "convey common meanings to the user." Literary theorists, on the other hand, will emphasise the aspect of interpretation. The difference between both approaches will be dealt with more extensively in 2.2.4.

Jorn Barger’s hypertext timeline (Barger 1998) goes back about 5000 years and it is divided into 11 periods, most of them situated in the last fifty years. We will use his work as a backbone for our overview of hypertext's history, and highlight the most important events. The first period starts at approximately 3000 BC with the first clay tablets and papyrus scrolls being produced. This very early beginning suggests that Barger believes that in every text there is to some degree hypertextuality, a competition between our multilinear thought and linear way of writing, together with implicit reference to other works.

Barger situates the first known hypertext in Persia in 900 BC with the 1001 Arabian Nights. Here, we should clearly distinguish between hypertextual texts or fictions and hypertexts. In the previous section we have seen that 'explicit reference' can be used as a criterion. Thus, strictly speaking, the 1001 Arabian Nights is not a hypertext. It is only in 1759 when Sterne writes Tristram Shandy, the first novel regularly containing explicit references to other pages, that we find the what is probably first fictional (non-electronic) hypertext.

In 1945 two important events occur; Eckert and Mauchly complete the first successful electronic digital computer: the ENIAC, and Vannevar Bush introduces his MEMEX (memory extender) project. In the Essay "As We May Think," Bush tries to find a solution to the growing problem of the inaccessibility of our archives.

Bush’s interest is clearly encyclopaedic; he wants to devise a new means to facilitate access to the record. Although he states "The advanced arithmetical machines of the future will be electrical in nature, and they will perform at 100 times present speeds, or more" (ibid.), he still sees his memex as a mechanical device based on microfilm and levers. "On deflecting one of these levers to the right [the user] runs through the book before him, each page in turn being projected at a speed which just allows a recognizing glance at each. If he deflects it further to the right, he steps through the book 10 pages at a time; still further at 100 pages at a time" (ibid.).

More important than the technological details, however, is the fact that Bush proposes associative indexing as a new means to retrieve information.

Thus, Bush introduces three entirely new elements directly connected with associative indexing: links, trails and sets of these trails called webs. This provides us with a new "concept of multiple textuality, since within the memex text refers to (1) individual reading units that constitute a traditional "work," (2) those entire works, (3) sets of documents created by trails, and possibly (4) those trails themselves without accompanying documents" (Landow 1997: 10). Bush should therefore be seen as the inventor of the concept of hypertext.

In 1965 Nelson wrote his first paper for the Association of Computing Machinery (ACM), it was in the same period that he coined the term hypertext. Although Nelson repeatedly refers to Bush--he even includes "As We May Think" as a chapter in his Literary Machines (1981)--they both have different conceptions of hypertext. Whereas Bush sees it as a way to manipulate and search large amounts of stored knowledge, Nelson dreams of the ultimate centralised literary archive, which he named Xanadu after the imaginary utopia in Coleridge’s "Kubla Khan."

It seems now, that his vision will remain just that, a dream of a universal library at our fingertips. Although many people argue that the Internet is attaining most of Nelson’s objectives, others zealously advocate that we still have a long way to go.

With the introduction of the ARPANET, the ancestor of the Internet in 1969, the network era commences. The first distributed hypertext appears in 1972 with ZOG at Carnegie Mellon. In 1976 Don Woods adds fantastic elements to Crowther’s cave adventure, thus creating the first interactive fiction. "Adventure in turn launched a genre. Its offspring, called "text adventures," became a mainstay of the early computer game (…), earning literary notice and praise" (Moulthrop 1995).

The micro or personal computer era starts in 1977 with the launch of the Apple personal computer. In 1978 MIT creates the first hypermedia videodisc, the Aspen Movie Map. The user can pay a virtual visit to the town of Aspen. In 1979 Truscott and Bellovin introduce Usenet news, and Bartle creates the first Multiple User Dimension (MUD). In 1980 the first text adventure, Zork, appears on the Apple computer. Infocom will produce many more after that. In 1983 the TCP/IP standard replaces NCP as the main protocol for what is to become the Internet.

In 1987 Hypercard, the first program that allowed hypertextual organisation on a simple personal computer is added to the Apple Macintosh package. In 1989 Tim Berners-Lee makes the first proposals for a World-Wide Web. In the same year, Michael Joyce writes Afternoon the first ‘serious hyperfiction,’ published by Eastgate. In 1990, the first version of 'hypertext markup language' (html) is released. In 1992 the World Wide Web is released by CERN. In only two years the WWW byte-traffic will pass that of Gopher, an older equivalent of WWW, but not hypertextual.

In 1993 the first easy-to-use web browser, NCSA Mosaic 1.0 for X Windows is released. In 1994 Clark and Andreessen form Mosaic, later that year they will release the first beta-version of Netscape. In 1995 search engines like Lycos and Yahoo are started up, Altavista lists no less than 15 million pages and Java makes applets net portable. In 1997 hypertext fiction meets with its first successes.

When we look at this history of hypertext and hyperfiction, there are two important remarks to be made.

(1) "[T]he movement from manuscript to print and then to hypertext appears one of increasing fragmentation." Whereas manuscripts were not subdivided in pages or sometimes not even in words, book culture introduced structuring (and fragmenting) devices like paragraphing, page numbering and the division of the text in chapters or even books. While going from book culture to hypertext we experience a similar phenomenon. The text is generally divided into even smaller parts and these parts achieve a certain degree of independence. This holds the danger of losing all structure. However, "the fragmentation of the hypertext document does not imply the kind of entropy that such fragmentation would have in the world of print. Capacities such as full-text searching, automatic linking, agents, and conceptual filtering potentially have the power to retain the benefits of hypertextuality while insulating the reader from the ill effects of abandoning linearity" (Landow 1997: 77).

(2) The development of electronic novels likewise seems to be the outcome of a natural process. Harris observes that "these technographic books seem the nearly inevitable extrapolation of the metafictional trajectory we can trace in the novel throughout the second half of the 20th century" (Harris 1997). He also notes that the new medium causes a new point of view on the matter of literature to come into existence. "Technographic books emerge as the history of writing itself comes to be rewritten with a new urgency, within the annals of literary theory, in the novel, and in historical accounts of print and computer textuality alike" (ibid.). Harris concludes that the novel could only follow the same course. "And so as the novel became conscious of itself as not only a medium, but as a sort of ontological play where textual worlds came into existence, it began to reflect on its own history from a viewpoint that took a retrospective turn" (ibid.).

In this section we will take a more technological viewpoint. In order to understand the implications of the change to hypertext writing better, we have to be acquainted with the different terms popping up in the discussion. The terms we will be dealing with are grouped around three main notions: (1) hypertext (system), (2) the link and (3) the network. Finally, we will briefly look at some additional notions dealing with (4) nodes and node grouping.

To begin with, let us repeat the obvious to avoid confusion later on. Following De Bra, when we speak of a hyperdocument, we will be referring to "the information content, including the information items (nodes) and the connections between them (links), regardless of the system used to read (or write) the document" (De Bra 1998). A hypertext system is "a piece of software which lets you read (and maybe also write) hyperdocuments. It does not necessarily contain a hyperdocument" (ibid.). A hypertext, finally, is "a hypertext system containing a hyperdocument" (ibid.). We will not follow Keep’s terminology of hyperbook however. The term ‘book’ refers too much to the material object to use it for hypertext. When referring to Sterne’s work e.g., we will employ the term pre- or non-electronic hypertext.

The basis of this new textuality called hypertext is of course the link. A link is a connection between two or more text units that allows the reader to traverse from one to the other. Landow sees the link as "an element that simultaneously blurs borders and bridges gaps, yet draws attention to them" (Landow 1997: 20). In order to understand the new medium better, some knowledge of this new grammatical tool is required. We will now take a look at the different kinds of links, their properties, advantages and disadvantages, which will result in a complex cross-classification.

Keep distinguishes between four link classes. (i) The first and most common one is the reference link. "When a reference link is selected, the destination node is presented on the screen, usually replacing the image of the current node" (Keep 1995). Most hypertexts on the WWW consist entirely of links of this type. Important is that "they allow the creation of non-hierarchical structures" (ibid.). When we will be looking at link types, we will see that within this link class, further subdivisions can be made.

(ii) A note link, a link class supported by software like MSWord, but not by the WWW, is the electronic equivalent of the footnote or endnote. "When a note link is selected, a small area of the screen (a "pop-up" window) displays the destination node (…) No links may be followed from the pop-up; the reader must return to the original node before navigating further" (ibid.).

(iii) A third class of links is that of the expansion link also called replacement (De Bra 1998) or stretchtext (Nelson, see Keep 1995), is not (yet) generally supported by html either. "When an expansion link is selected, the contents of the destination node are expanded in-line at the source anchor" (ibid.). In other words, when you are reading a text and you click on an expansion link, the text will break open and the node connected to the link will place itself where the word (or sign) you clicked on (anchor) used to be. This type of linking only allows a hierarchical structure since the first text inevitably assumes a superior position towards the second, the one that is inserted.

(iv) A final link class is that of the command link or action link, which is supported by the WWW. However, it has no independent status: it takes the form of a reference link. "When a command link is selected, an action is performed: the operating system is called, an external program is run, etc. In some systems, this is implemented by having the destination node contain a program script which is executed on linking" (ibid.).

Within these four link classes, several link types can be distinguished. First of all, Landow distinguishes between the unidirectional and the bidirectional link. Many hypertext systems do or did allow them both, but as we are dealing with the WWW, the distinction is meaningless. On the WWW, every link is uni- and bidirectional at the same time. Basically, the link is always unidirectional, but as most browsers have a history or a back function, these links become bidirectional. Since there are two categories of text units, to which can be linked, i.e. lexias (nodes, cf. infra) and text strings, we get four possible connections.

(i) The first is that from lexia--page if you will--to lexia. This is the most basic form of linking, and in early systems it used to be the only one. This link type is useful for continuation of a text e.g.; it is easy for the reader to orient herself.

(ii) When you want to refer more specifically to a certain passage, however, string--that is, a word or a phrase--to string linking is more appropriate. "By bringing readers to a clearly defined point in a text, one enables them to perceive immediately the reason for a link and hence to grasp the relation between two lexias or portions of them" (Landow 1997: 11). It thus allows for what Landow calls a ‘rhetoric of arrival.’ However, "The possible disadvantage of such a mode to authors--which is a major advantage from the readers point of view--is that it requires more planning or, at least more definite reasons for each link" (ibid.).

(iii) Linking a lexia to a string is not very common on the WWW although it is possible. The advantages and disadvantages are somewhat similar to those of string to string linking, except for the fact that you need a source anchor outside the text, which could be a disadvantage.

(iv) For string to lexia linking, the most common form of linking on the WWW, Landow discerns three advantages. "First, it permits simple means of orienting readers by permitting a basic rhetoric of departure" (ibid.). By rhetoric of departure, he means that the reader can be given a clear signal of where the link will lead to. By highlighting more or less text and by word choice a clear meaning can be communicated. Second, "since one can choose to leave the lexia at different points, one can comfortably read through longer texts" (ibid.). This holds, of course, for string to string linking as well.

"Third, this linking mode also encourages different kinds of annotation and linking, since the ability to attach links to different phrases, portions of images and the like allows the author to indicate different kinds of link destinations" (ibid.). This follows directly from the first remark. When employing string to lexia linking, one can e.g. link to the words 'picture of…' which is impossible when starting from a whole lexia. The greatest difficulty with string to lexia linking occurs for the reader when arriving at too large a page. Readers will often find themselves disoriented in such a position.

Another important differentiation is that of one to one linking--what we generally subsume under the term link--vs. many to one linking. "A method called many-to-one linking proves particularly handy for creating a glossary function or for creating documents that make multiple references to a single text, table, image, or other data (…) has advantages in educational or informational applications" (Landow 1997: 11). What you do is use the same information more than once so that you do not have to repeat. Very often these links are soft (or reader-created, cf. infra).

For Landow, full hypertextuality depends on the presence of the possibility of one to many linking, which lets the reader obtain more than one kind of information from one text string. Landow argues that one to many linking supports hypertextuality in many ways. "First, it encourages branching and consequently multiplies the reader’s choices" (ibid.). This is not always an advantage however; too many choices can cause cognitive overhead. "Second, attaching multiple links to a single text allows hypertext authors to create overviews and directories that serve as efficient crossroad documents, or orientation points, that help the reader navigate hyperspace. Multiple views or sets of overviews have the additional advantage of easily permitting different authors to provide multiple ways through the same information space" (ibid.).

Undoubtedly, one to many linking would be a valuable addition to current hypertext systems, but whether full hypertextuality depends on it is questionable. First of all, recent versions of html do allow some sort of one to many linking in the form of pop-up menus. Admittedly, they are not very appropriate to use in a text, they are rather meant for schemes or overviews. Moreover, if one really needed one to many linking so desperately, one could always take recourse to making an intermediary lexia. The first link from the basic anchor is a string to lexia link to another page, which then presents the reader with the required choices. Again, we must admit that this is no ideal solution, as the WWW is very often slow and therefore intermediary pages are not desirable.

One more distinction that has to be kept in mind while dealing with hypertext is that between hard and soft links. As we have already mentioned, hard links are links created by the author and that can be chosen by the reader to follow. Soft links, then, are links that are not prepared by the author, but that are evoked by the reader’s wish itself. "For example, the user can click on a word and the system can search for the occurrence of the same word or a synonymous word in other documents/nodes and traverse to that node. These can be considered non-authored link markers/links" (Balasubramanian 1995).

Concretely, when you let your browser search for a certain word or phrase on a page, you are creating a link yourself. Whether this link should be considered hypertextual is still being discussed (cf. 2.2.1.). One peculiar type of soft or (more or less) reader-created link is that generated automatically between nodes that are similar. "A test by Bernstein showed that an automated generator of hypertext links can easily achieve a 95% accuracy in generating meaningful links" (De Bra 1998). De Bra notes two difficulties with this kind of link. One is that the computer encounters problems creating meaningful anchors for the links. Another is that this procedure might lead to an undesired wealth of links, which, in turn, produces cognitive overhead.

Finally, there are two types of command links; or rather two subdivisions of one certain type of command link that need mentioning. Data-exchange links engage a certain procedure that transports information from a source to a destination device or programme. These data can then be translated into a destination file or consumed as a data stream of sound, film… Warm links are reader-activated or soft data exchange links. When you choose to download a certain file from an FTP (File Transfer Protocol) server for example, you click to start the procedure and the information is transmitted to you. In the case of "cookies" on the net, for example, it is not the user that desires to see publicity, but it is the author or provider that, so to speak, throws or pushes it onto the user’s screen. This type of data-exchange link is called a hot link.

Landow recognises at least four different meanings of the term network in current day discussions about hypertext. First, there is the possibility of transferring individual print books for example to hypertext. Thus, we get a web of nodes and links and paths, but one of the kind that we have called macrostructure in the introduction, the electronic equivalent of the real world library.

Second, there is a similar possible structure, but on a micro-plan. In other words, whilst creating a hypertext directly--i.e., not converting it from a linear text--we will get smaller lexias or nodes that will have closer relations with one another. Thus we get a new, but smaller type of network, which we will call micro-structural.

Third, there is the obvious meaning of "an electronic system involving additional computers as well as cables or wire connections that permit individual machines, workstations, and reading-and-writing-sites to share information" (Landow 1997: 42). When referring to this kind of network we will usually employ the term Internet or specify by referring to a LAN (local area network). McGann observes that hypertexts "can be distributed in self-contained forms (e.g., on CD-ROM disks) or they can be structured for transmission through the Network. In this last case, the basic hypertext structure is raised to a higher power (but not to a higher level): a networked structure (say, World-Wide Web) of local hypertexts opens out into a network of networks" (McGann 1995).

The fourth meaning, Landow recognises as being closest to matching the use in critical (literary) theory.

In other words, what Landow means by this is the covert network of interrelations that exists within the world of knowledge and literature; the relationships between our thoughts and those of others in history and, more importantly, the connections between the illustrious and their environment. Landow sees it as the task of literary theorists to discover these ties and that of hypertext to make them explicit.

Many terms have been used to designate the main textual unit or node in different hypertext systems. Keep notes a few synonyms: "frame (KMS), work space (StorySpace), card (HyperCard), and lexia (Barthes by way of Landow). On the World Wide Web, a node is simply termed a Web Page" (Keep 1995). From this group two seem to be emerging, viz. lexia and node which we will employ. When referring to the WWW, we will sometimes use the term page.

"A cluster is a subset of the complete network, a group of nodes and links which share a common concept and may be seen as an aggregate" (ibid.). Sometimes it will consist of an entire microstructure; sometimes it may be sub-micro-structural.

"A path is a sequence of nodes and links taken while navigating the network (ibid.)" In a web browser, you can retrace your path by using the history function. In some systems you can save your path and pass it on to somebody else. This is important for hypertext fictions, because then you can share your reading experience.

Finally, a filter is a device that restricts the entire network to a specified subset of nodes and links. "Filters permit the reader to limit the scope of navigation to a given area of interest. This is especially important as nodes increase in number and their contents become more heterogeneous" (ibid.).

2.2.1. Hypertext and its relatives

In this section we will try to determine some aspects of the nature of the hypertext medium by comparing it to related notions. We will successively deal with (1) hypertext as text, (2) hypertext as e-text, and (3) hypertext as a metaphor.

Many people argue that hypertext has always existed, that there is nothing new under the sun. "There never has been hypertext (Ted Nelson made it up: as Guyer has written elsewhere "It has to do with, god help us, the non-existence of abstracted dualities. By this I mean all the usual, traditional [polar] representations. The list goes on as long as consciousness itself. We make these things up!") All text is hypertext, always has been, surely is in an electronic universe" (Joyce in Feed 1995). Hypertext is considered to be the struggle of multilinear thought trying to break out of its symbolic representation. Very often, in the same respect, the implicit reference of literary works to other literary works is mentioned. Every text is, so to speak, a result of the oscillation between exterior influences and multilinear interior thought, which is then converted into linear writing. Electronic hypertext merely allows explicating the exterior influences as links, and permits to make the linear text become multilinear and therefore closer to human thought.

This idea has its merits, especially in the context of postmodern literary criticism. However, the terminological question remains whether we should call every text a hypertext. Undoubtedly, there are in every text traces of the struggle described earlier, but this is implicit. It is inherent to every text to refer to other texts by quoting or even just by using the same words, but is this hypertext? Is this not rather a quality of the concept of text that critics have overlooked for so long or, at least, not interpreted in the way they do now? I believe it is better to reserve the term hypertext for explicit reference or links as (for fiction) they start to occur in the eighteenth century with Sterne's Tristram Shandy. In this way, we can distinguish hypertext without referring to the medium it is presented by, be it paper or computer screen.

Joyce, however, seems to go a lot further than this. He claims that hypertext is somehow a kind of hyper(o)nym of text, a related yet superordinated term. He sees linearity as just one manifestation of the multilinear. "Electronic texts merely provide the occasion for the multiple consciousness which linearity has always played within and against. Graph theorists say the hierarchy is merely a special case of the network. The linear is merely a special case of the multiple" (Joyce in Feed 1995). But is this not like saying that the number 1 (one) is simply a special case of the number 2 or three? Does a network itself not consist in many linear connections? Why would we want to make the multiple into a more basic feature than the linear? There are no immediate answers to these questions, since both notions cannot be opposed. They belong to different levels of abstraction, and they are complexly intertwined; nonetheless they are both recognisable. I therefore believe that the perspective of text and thought as a struggle between the notions of linearity and multilinearity is the more fruitful one.

Multilinearity is not always recognised in text because of the linear 'outfit' of the latter. Moreover, we are so much used to book culture that we no longer recognise that it is a 'technology' like every other, and that it is no more natural than the printing press by which it was produced. Thus, the fact that many text units are in fact juxtaposed or networked to a certain degree seems sometimes an awkward idea to a naïve reader. When we take a closer look at the outline of the book, however, and especially at its evolution over the centuries, we must admit that more and more 'juxtaposing devices' were introduced. In primary school, children are taught that they should begin a new paragraph whenever a new thought occurs. In other words, paragraphing is a kind of textual tool that signals a certain parataxis or juxtaposition. The same goes for chapters and, indeed books, for when an ultimately different thought is engaged in, the author starts a new book.

Finally, as we have seen earlier in the brief discussion of Bush's "As We May Think," in a hypertextual environment, text acquires a number of additional meanings. Relations between text, be they implicit or explicit, can now be made explicit by the link. This network of intertextual relations forms itself a second text 'behind' the text as we used to know it, one that is not immediately visible to the reader; it is to be thought of as virtual. "More than any other crucial term to this discussion, text has ceased to inhabit a single world. Existing in two different worlds, it gathers contradictory meanings to itself, and one must find some way of avoiding confusion when using it." However, critics have not yet reached a verdict on this question. Landow notes "For example, in discussing that hypertext systems permit one to link a passage "in" the "text" to other passages "in" the "text" as well as to those "outside" it, one confronts precisely such anachronism" (Landow 1997: 58). For convenience' sake we will persist in employing the term 'text' in such an anachronistic fashion to refer to text as a microstructure (cf. introduction).

Landow distinguishes hypertext from "four other important kinds of electronic textuality, each of which can exist within hypertext environments but is not itself hypertextual" (Landow 1997: 309).

(i) First of all, we have graphic representations of text. They are in fact drawings--and are recognised as such by the computer--that include written symbols. The fact that the computer does not recognise the text as text has important consequences. This kind of image/text "cannot be searched, parsed, or otherwise manipulated linguistically" (ibid.). However, it can be directly altered by graphical software and manipulated as a whole by other programs who sustain it.

(ii) The second and best-known type of electronic text is simple alphanumeric digital text. "this form of electronic text, which functions linguistically, appears in electronic mail, bulletin boards, and word-processing environments" (ibid.). This text is recognised as such by the computer and allows manipulation and full-text search in many systems.

(iii) As a third type of text, Landow identifies nonlinear text which does not "enable multisequential reading (...). [V]arious forms of nonlinear textuality include (a) computer games, (b) text-based collaborative environments, such as multi-user domains (MUDs) and MUDs that employ object-oriented programming methods (MOOs), and (c) cybertext, or text generated on the fly." However, we will later see that this nonlinear textuality is dubious in many respects, especially as soliciting the status of text in the stricter sense of the word. But thus we do not want to ignore its merits, not even for hypertext.

(iv) Finally, as a fourth type of non-hypertextual electronic text, Landow extends the scope of the term text to include simulations. "Text in simulation environments can range from computationally produced alphanumeric text (and hence have much in common with the nonlinear form) to instances of fully immersive virtual (or artificial) reality." Whether this extension is desirable can be contested, but we will not pursue this any further here.

For all these different sorts of electronic text common grounds can be discovered. Murray (1997, 71), for example, identifies four main characteristics of the electronic medium: (1) procedural, it is always the result of procedures, it can never appear out of the blue, (2) participatory, it allows interaction, (3) spatial, it involves a (virtual) displacement and (4) encyclopaedic, it can contain large amounts of information. Although these fairly abstract notions do not appear absolute, they give an idea of what the characteristics of a future electronic medium could be. We will now narrow our scope and look at the relationship between electronic text (ii) and hypertext more specifically.

Jerome McGann, observes that when using hard copy texts to analyse other hard copy texts, the scale of the tools is a burden for the inquisitor. He makes the comparison with the hard sciences: "In studying the physical world, for example, it makes a great difference if the level of the analysis is experiential (direct) or mathematical (abstract). In a similar way, electronic tools in literary studies don't simply provide a new point of view on the materials, they lift one's general level of attention to a higher order" (McGann 1995). As an illustration, he compares the electronic Oxford English Dictionary to its paper counterpart.

Remarkably, McGann does not seem to make any difference between explicit hypertext and e-text. "Computerization allows us to read "hardcopy" documents in a nonreal, or as we now say a "virtual", space-time environment. This consequence follows whether the hardcopy is being marked up for electronic search and analysis, or whether it is being organized hypertextually" (ibid.). He obviously sees a strong connection between the two. Later in his essay he does appear to have a preference for marked up documents, albeit the characteristics he enumerates are inherent to the electronic medium rather than specific to hypertext.

Another important quality of this new medium resides in its ability to allow simultaneous representation. When a book is translated into electronic form, the book's (heretofore distributed) semantic and visual features can be made simultaneously present to each other. "A book thus translated need not be read within the time-and-space frames established by the material characteristics of the book. If the hardcopy to be translated comprises a large set of books and documents, the power of the translational work appears even more dramatically, since all those separate books and documents can also be made simultaneously present to each other, as well as all the parts of the documents" (ibid).

This virtual or non-real time frame poses a serious cognitive problem. McGann also recognises its dubious status. "Of course, the electronic text will be "read" in normal space- time, even by its programmers: the mind that made (or that uses) both codex and computer is "embodied." This means that, from the user's point of view, computerization organizes (as it were) sequential engagements with nonsequential forms of knowledge and experience--immediate encounters with abstract or complexly mediated forms" (ibid.). Then what is the exact advantage of this type of structuring? What do we gain by allowing text to migrate to the computer screen? "If the limits of experience remain thus untranscended through computerization's virtual enginery, however, the new tools offer a much clearer and more capacious view of one particular class or "order of things"--in this case, the order of those things we call texts, books, documents" (ibid.).

But how should we understand 'virtual enginery' of such dubious stature? What is it that makes it such a powerful tool? First of all, we should note that it is more powerful in both a quantitative and qualitative manner. Not only does it allow us to store much greater amounts of data, it also creates or allows us to create and explicate relations that would not have entered the mind of the reader if the author had not set them out. This second characteristic, the higher number of associative links, is already both a qualitative and a quantitative notion! Quantitative because there are more links than in ordinary text, qualitative because some would otherwise remain unseen. But what are its consequences for reading, concretely? On the one hand, the fact that many more connections are made, makes us read more attentively and try to conjecture where the links are leading to, even if we do not necessarily pursue them. This draws the reader's attention both to the text itself as to the ‘author's mind’, her intended message, as she is trying to understand what is meant by the connection. On the other hand, if we do pursue the link, this allows us to avoid much uninteresting material. In the following subsection (3) we will deal with some proposed metaphorical representations that are to permit us to understand better what is being developed here.

Another important argument often employed for the implementation of a new textuality is the problem we are facing managing our ever-growing knowledge stock.

This ever faster changing rate of the production of knowledge poses a reasonable threat for science. The danger exists that many important scientific articles will never reach their intended destination. Or as Bush already noted in 1945, "even the modern great library is not generally consulted; it is nibbled at by a few." Nobody knows how many world-shocking stories and theories lay buried under piles of paper, in libraries all over the world.

Since its coming into existence, many types of organisation have been proposed for the library, and just as many proved outdated after a few years or decades when the library showed considerable growth. In recent years most libraries have computerised their database. Now, the user can perform quite complex Boolean searches with key words or names or parts of titles of works. That is, when the user knows what he is exactly looking for. Many keywords, however, belong to a certain organisational system. In other words, they desire an understanding of the way books are categorised. Moreover, as Ted Nelson points out: "there is nothing wrong with categorization. It is, however, by its nature transient: category systems have a half-life, and categorizations begin to look fairly stupid after a few years" (Nelson, quoted in Landow 1997: 7).

Electronic text in general and hypertext in particular have two main advantages to these systems. First, they allow for multiple categorisation. Put differently, new categorisation systems can be devised which allow for a more organic structure. Keywords could be organised in both the classical hierarchical manner and a new kind of associative indexing at the same time, without one disturbing the other in any way. Moreover, once the traced text seems to comply with the user's wishes, it is far closer at hand for evaluation. One click of the mouse or other device and the user can see whether what she has found really corresponds to what the key words were promising. If this is not the case, a second click could inform the administrator of the hiatus or inconsistency.

The second and even more innovating change is the fact that the user can perform a full-text search. "In fact, a primitive form of hypertext appears whenever one places an electronic text on a system that has capacities for full-search retrieval or a built-in reference device, such as a dictionary or thesaurus" (Landow 1997: 312). Landow gives the example of his word-processing programme, which permits to search for a letter or text just by typing a word or string of words present in that text. The "program (...) quickly locates all occurrences of an individual word or phrase, provides a list of them, and, when requested, opens documents containing them. Although somewhat clumsier than an advanced hypertext system, this software provides the functional analogue to some aspects of hypertext" (ibid.).

Earlier, we have referred to this issue as a difference between an author and a reader determined connection, also called hard and soft link respectively. Note that one does not exclude the other. On the contrary, they need one another. On the WWW for example a user will employ a search engine to locate an interesting site dealing with a certain topic thus realising a soft or reader-determined link. Usually, this site will have a collection of links to other interesting pages. By following one of these the user pursues a hard or author-determined link. Thus the user has combined a classical search with the advantages and flexibility of associative indexing. However, the question of whether both types of linking are hypertextual is not definitively settled yet.

Recently, theorists have posed the question:

Evidently, the characteristic effects of this new medium all derive from its virtuality, the fact that computing stores information in digital codes rather than in physical marks on a physical surface. Because of the doubly layered structure of electronic text, the storage capacity of digital information (first level) can increase without affecting the second level, the size or form of the alphabetic sign. Moreover, the sequence and format in which the information is stored at the deeper level does not affect the concrete representation. This was different for microfilm for example. The fact that there was only one layer necessarily entailed that the size of the signs had to decrease when storage capacity wanted to increase and when the quantitiy of surface had to remain unchanged. Written electronic text now becomes a magic doorway looking out on immeasurable amounts of 0's and 1's, the province Gibson dubbed cyberspace. Keep describes it as:

First and foremost, a medium is something that is invented and only afterwards discovered. In other words, what we are dealing with is something we can manipulate but cannot control yet, something of which we cannot form a clear idea, simply because we have never met with the like. All we can do is try to build an image that suits our purposes. However, we should keep in mind that it is not the thing itself, not even a direct image of it. Earlier, we have attempted to apply the notions of quantity and quality to the concept of hypertextuality, thus establishing a basis for discussion. In this section we will develop this issue more profoundly by looking at the different metaphors and images we could employ to visualise how hypertext functions.

A Metaphor is probably the most powerful device to construe an image of something unknown. But what could be the metaphor to understand hypertext? How come that we are not able to construe a true image of the way it works? Is this new way of handling texts so novel that we cannot but accept it as it is? This is a question that occupies many theorists, but one that remains unanswered. All we can do is provide partial metaphors for the different aspects of hypertext. However, even together they do not fully cover the new possibilities; neither are they mutually exclusive. This is an account of a search that is still going today, and whether a final answer can be found or not remains an open question. Very likely, it will also become superfluous once we get to know the mechanism of hypertext itself better.

Hypertext could be seen as three-dimensional text, where ordinary paper text is only two-dimensional. The first dimension is purely spatial; it is a line of text, a vertical space necessary for signs to exist. The second dimension is of course also spatial. It allows for two links already, one to the preceding and one to the following word or sign. We, as readers and authors, however, experience this horizontal space primarily as temporal. One could argue that the vertical crossing of a page is also a process that takes time, but that is only so because we cross it on the horizontal second dimension. Of course, in some sign systems this is entirely different, but that is beside the point in this discussion. We are speaking metaphorically; we are trying to build a viable image of something virtual.

But what could this metaphorical third dimension be? As Keep observed in his description of cyberspace, users experience it as a place, a realm or province: somewhere you can go to that is not here. A link is often seen as a bridge crossing borders or even cliffs. Moreover, many theorists speak of the electronic frontier comparing it to where the wilderness used to start in America, where the laws of the civilised east ceased to exist and the Wild West began. But what is the value of such images, is it sheer romanticising about sitting in front of a highly sophisticated calculator? I believe there is more. The general enthusiasm among users must have a solid basis, a strong experience of something new: powerful yet unknown.

Conversely, when it comes to temporality, this is seen as a burden, rather than a blessing. Concretely, when reading--or should we say interacting or even dialoguing--the third dimension is experienced by the reader as temporal in that she has to decide whether she will engage in taking a shortcut 'into' the text or whether she simply maintains her course on the second dimension. This is negative in a certain way as it can produce cognitive overhead and slowing down of the reading process. However, in many hypertexts the possibility exists to just ignore the third dimension and only click at the end of the page, like in a book. On the other hand, links could also be seen as a positive characteristic. It draws the attention of the reader towards the text and the argument it is trying to develop. Nobody wants to miss interesting bits by overlooking a link! A second moment when this third dimension is experienced as being temporal is when the computer is preparing to present the page on the screen. Here temporality is experienced unilaterally as a burden. Of course, this second temporal moment is not interesting from a theoretical point of view as it will (hopefully) dissolve in the future when we will have more powerful hard- and software at our disposal.

If this third dimension is primarily spatial, should we consider this space as a wilderness, uncontrollable like the Wild West? Note that we are talking about hypertext here and not about copyright law for example. We are dealing with links, their sources and destinations. Proponents of hyperfiction--especially those involved with the type, which we will call rhizomes (Murray 1997)--will say yes, that we are dealing with something that transcends all structure and should remain that way. Theorists concerned with the informational capacities of hypertext structures, on the other hand, will defend a stricter structuring of where and how links should be placed and understood (cf. 2.2.4.). These two conceptions are directly opposed! In our formal classification of hyperfictions in the second part of this paper we will try to confront the two viewpoints and propose a solution.

Recently, more people have come to realise that linking should be guided and interaction controlled both by the author and the reader. "Participation in an immersive environment has to be carefully structured and constrained (Murray 1997: 106). For the author this means that she has to create only meaningful links and somehow try to guide the reader through her web. For the reader it means that there is an extra factor of the text that must be taken into account. She has to try to understand the extra dimension and interact with it, thus transgressing the pointless shoot-em-up clicking exercise some hyperfictions seem to be.

Many more metaphors have been proposed to clarify the notion of reading hypertext. The most traditional one--which many people even no longer experience as a metaphor--is the idea that we 'navigate' the internet: a direct consequence of what Murray identified as the 'spatial' quality of the electronic medium. Several subtle variations may be found like ‘wandering’ or ‘surfing’, but the underlying idea is the same. Murray advocates: "The visit metaphor is particularly appropriate for establishing a border between the virtual world and ordinary life because a visit involves explicit limits on both time and space (Murray 1997: 106). Moulthrop, however, notes

Whereas Moulthrop just takes interaction as his starting point, in a somewhat similar fashion as we have done for the distinction between warm and hot links, Balasubramanian goes even further. "[E]xisting metaphors such as electronic encyclopædia, notecards, journeys, browsing, windows, paths, guided tours, travel holidays, and survey-type maps are too restrictive and do not fully exploit the true potential of hypertext. The metaphor for hypertext should be based on "the general cognitive model of how individuals think about complex problems" (Balasubramanian 1995). As a proponent of the cognitive model, Balasubramanian seems to dream of a more direct connection between hypertext and our brain structure.

The aim of this approach is to make "hypertext systems (…) exploit the basic nature of human cognition, which is essentially organized as a semantic network of concepts linked together by associations" (Balasubramanian 1995). According to this view, a mental model should be devised "by employing self-generated metaphors in the context of a specific application while the implementation itself can be based on a general semantic model (…). Such an approach provides navigation and analysis of the underlying database independent of the specific application and the different mental models of individuals (ibid.). In the next section we will take a closer look at the reading and writing model he proposes and attempt to formulate a critique.

In the first chapter of State of the Art: Review on Hypermedia Issues and Applications (1995), Balasubramanian attempts to provide a (1) reading model and a (2) writing model for hypertext. Being largely based on early cognitive research, some parts may seem naïve and sometimes even biased. Nonetheless, it will prove a solid basis to start our discussion on the implications of hypertext on the process of reading and representing thought. As a final consideration in this respect we will briefly look at (3) the cyborg dream

Balasubramanian distinguishes four levels at which understanding takes place: lexical, syntactic, semantic and pragmatic. "At the lexical level, the user determines the definition for each word encountered. At the syntactic level, the subject, action and object of a sentence are determined. The meaning of a sentence is determined at the semantic level. The pragmatic interpretation of text depends on the integration of semantic meaning of text with the reader's knowledge of self and of the world" (Balasubramanian 1995). That a reader would determine a definition every time she reads a word is of course nonsense. Definitions do not take part in the reading process in any way. Moreover, when we determine a meaning only on the syntactic level, why would we have words then? Or how is it possible that a sentence can consist of only one word?

What Balasubramanian seems to do is transform Greimas's dialectic of knowledge acquisition into a practical framework, whereas it should be considered as a theoretical working hypothesis. Some nuances will therefore prove to be necessary. Balasubramanian apparently sensed that himself and thus adds the following utterly contradictory phrases. "While reading text, people proceed from a lexical level to the syntactic level, to the semantic and to the pragmatic levels in that order. All these levels interact continuously and they cannot be truly separated" (ibid.). Let us leave the discussion here. What should become clear from this description however, is the desire to turn hypothetical cognitive theorising into concrete practice. This is a frequently recurring phenomenon in literature about hypertext, especially in the early days.

More interesting, however, is the discussion about mental representation, which Balasubramanian sees as a form of propositions or relationships between certain entities.

Interesting is the dialogic perspective taken. Understanding and thus also acquisition of knowledge is seen as an interactive process between existing structures and new experience or theory. Expanding this interactive view to a dialogue between the text and the reader is only a small step.

Next, Balasubramanian tries to give an idea of how we should see the propositions or relationships between concepts and ideas and how it is precisely these relationships that give meaning to the concept.

Note the similarity between this representation and Derrida's notion of 'différance,' which returns in literary hypertext criticism. We should accept the dynamism and uniqueness of meaning-creation. But whereas Balasubramanian has the process of receding meaning arrest when "further activation of adjacent propositions does not change the propositions used to interpret the text," Derrida will say that the process of signification never ends.

In his writing model, Balasubramanian carries the concretising of theoretical notions even further. "Writing involves the following three phases: exploring, organizing, and encoding (...). Smith et al. call these three phases: prewriting, organizing, and writing" (ibid.). We will now briefly look at the symmetrical reading/writing model that is proposed.

The first phase in writing is pre-writing. "The writer retrieves potential content from long-term memory or external sources, considers possible relations among ideas, groups related ideas and constructs small hierarchical structures" (ibid.). Organising is the implementation of the available material into a more or less linear structure or at least one that can be represented by a linear medium. "This process involves abstract construction that involves perceiving subordinate/ superordinate relations, comparing abstractions, sequencing, proportion, and balance. Thus, the product of organization is a hierarchy of related concepts" (ibid.). Finally, encoding or writing consists in "translating the abstractions of content and the relations of a hierarchical structure into a sequence of words, sentences, paragraphs, sections, chapters, and illustrations" (ibid.).

Note that this graph is symmetrical both horizontally and vertically. On the left-hand side, we have the sequence we just described, performed by the author and on the right hand side a more or less inverted process, performed by the reader. This second process likewise consists in three stages: reading, understanding and remembering. Underneath this sequence of human actions, we follow the structure of the 'text' through the process. First, it consists of a network of ideas, then a hierarchy of concepts, then it is mediated sequentially, and finally the reader turns it into a hierarchy and a network again. Balasubramanian notes the relation between the left and the right side. "Thus, both reading and writing processes emphasize a lot on the non-linear nature of thinking, a natural process in human beings. Human cognition is essentially organized as a semantic network in which concepts are linked together by associations" (ibid.). However, after this analysis of the reading and writing process, his conclusion seems somewhat meagre. "Hypertext systems try to exploit this basic nature of cognition" (ibid.). It seems that there is more at stake than this!

How hypertext systems exploit this basic nature of cognition, Balasubramanian does not explicate, but I believe that his suggestion can hardly be misunderstood. When we look at the graph presented above and especially its symmetrical structure, and we combine this with the fact that hypertext very often assumes the form of a network, we cannot but conclude that the reason for this whole exposition is to try to find a closer connection between both poles. If ideas and concepts are generated in a network environment in our brain, then to be converted into linear writing and then back again to multilinear thought, why not attempt to find a shortcut between begin- and endpoint? If we could find a way to write as a network, this would not only make it easier for the author to produce a text, but, it seems, also for the reader to take it in. This is the idea of the cyborg or man-machine elevating human thought to unseen heights, a collaboration that would make our classical humanistic conceptions of man tremble or even collapse; a nightmare to some, a dream to others.

Traces of the cyborg dream can be found throughout the whole history of hypertext theorising. In 1945 Bush wrote "All our steps in creating or absorbing material of the record proceed through one of the senses--the tactile when we touch keys, the oral when we speak or listen, the visual when we read. Is it not possible that some day the path may be established more directly?" (Bush 1945). Bush believes that the connection between man and machine will be electrical: "In the outside world, all forms of intelligence whether of sound or sight, have been reduced to the form of varying currents in an electric circuit in order that they may be transmitted. Inside the human frame exactly the same sort of process occurs. Must we always transform to mechanical movements in order to proceed from one electrical phenomenon to another?" (ibid.). However, Bush immediately relativises by evaluating and taking a healthy distance from what he just suggested. "It is a suggestive thought, but it hardly warrants prediction without losing touch with reality and immediateness."

Moreover, in this passage Bush was not talking about his memex in which he meant to implement the first hypertext system 'avant la lettre.' The latter was more than just a dream to him. "Consider a future device for individual use, which is a sort of mechanized private file and library. It needs a name, and, to coin one at random, "memex" will do. A memex is a device in which an individual stores all his books, records, and communications, and which is mechanized so that it may be consulted with exceeding speed and flexibility. It is an enlarged intimate supplement to his memory" (ibid.). Important here is, I believe, that Bush's device was not called the bramex (brain) or mimex (mind), but the memex (memory extender). Indeed, Bush seems to cherish the cyborg dream, but he differentiates between different kinds of thought, thus making his allegations sound more realistic. "For mature thought there is no mechanical substitute. But creative thought and essentially repetitive thought are very different things. For the latter there are, and may be, powerful mechanical aids" (ibid.).

As already described in earlier sections, Bush's attempt to create a new form of classification was motivated by "our ineptitude in getting at the record (...), largely caused by the artificiality of systems of indexing" (ibid.). Diametrically opposed to traditional means of organisation, Bush places associative indexing.

Bush does not advocate associative indexing primarily to create an electrical shortcut between the two structures. He defends it because of the power he believes this sort of classification model has. The dream of the cyborg remains there on the background, but it takes a more realistic form. Why burden our memory with insignificant details when they are at our fingertips. Is it not more useful to be well able to use these fingers than to study bare facts without further theoretical importance, which is still obligatory in most curricula? Bush compares this change of perspective to a similar development that took place in the positive sciences. Note that the electronic calculator not even existed!

Of course, we should not take this comparison for granted. Memory is and always will be important. If you do not know what you are searching for, it is extremely difficult to find it, whatever system is used. Associative indexing can only provide a partial solution, since the user still needs a starting point. Moreover, are we not already cyborgs in a certain way? Plato accused writing of rendering the memory idle. Is writing not a technology we employ to help us think or at least to help us remember things? I believe an objection to this is hard to maintain. Or as Birkerts, a hypertext cynic and hardcopy elegist has it:

However, an obstacle very often overlooked in the cyborg discussion is that almost all our knowledge exists only in and through language. One of the main characteristics of human language is that it develops through time, it is therefore inherently linear. The link with the preceding and following letter, word or sentence or even text is necessary for language to have a meaning. If this is so, what could be the merit of a multilinear representation if we can only read it in a linear fashion? The change is not so much on the perceptual level, but rather on the structural. It is qualitative rather than quantitative although it bears characteristics of them both. What is about to change is not so much how we read, like many cognitivists seem to claim, but rather what we read, what the computer offers the reader according to her choices.

For thousands of years, symbolic representation has been considered superior to iconic representation, albeit not always to the same extent. In books, pictures almost always appear in the margin; they are pushed out of the main text to become a mere illustration, like a footnote. However, many believe that with hypertext a new era has commenced in which both types of communication of meaning could be given a more equal chance. Tolva compares our current day situation to that before the invention of the book. "Hypertext (...) in which visual elements (even full-motion video) are woven into the fabric of the text like a modern-day illuminated manuscript, allows visual manipulation of text blocks (called lexias) and graphical depiction of structural features. (Tolva 1995)"

The reason for this renegotiation of the power relations between the verbal and the non-verbal is the spreading computerisation of the printing process. The evolution was not started by hypertext nor even electronic text, but by modern printing techniques which made it easier and cheaper to include large pictures or drawings.

Electronic text in general and the WWW in particular are continuing this trend. Daily, new tools to furnish websites become available, and these new devices acquire more and more graphical power. The only practical objection at this moment--for the Internet i.e.--seems to be the lack of bandwidth, which makes it often cumbersome to surf to links piled with graphical features. Or as Stuebe has it: "Compounded with the slow-motion crawl of so many modems downloading so many graphics, hypertext tends to derail any coherent train of thought" (Stuebe 1996)

However, graphical elements still assume their inferior position. Nobody would say that she is reading a newspaper with quite a lot of articles or that there is an article going with the picture (one exception being photojournalism as it is practised by Life magazine since 1936). Indeed, the articles are taken for granted and the pictures are there to illustrate. But that could be about to change. When working with electronic text, the manipulation of pictures becomes as easy as the manipulation of text blocks. Moreover, as storage capacity and speed are augmenting, the 'price' to publish (in the shape of bytes and download time) is diminishing. This can easily be seen on the WWW for example, where the implementation of Java software caused a true explosion of moving images and banners. According to Zachry, "the elements of hypertext contend for the reader's attention. Hypertext is a bazaar in which multiple elements (links, control menus, animated objects, etc.) often vie with each other for attention" (Zachry 1997).

In contradistinction with Bolter, who will oppose the symbol and the icon (cf. infra), Landow approaches the opposition between symbolic and iconic signs from a different angle. He believes that writing should include both types of signs at the same time. "The expansion of writing from a system of verbal language to one that centrally involves non-verbal information--visual information in the form of symbols and representational elements as well as other forms of information, including sound-has encountered stiff resistance" (Landow 1997: 61). According to Landow, this is due to the hundreds of years of predomination of book culture. "Much of our prejudice against the inclusion of visual information in text derives from print technology. [However, l]ooking at the history of writing, one sees that it has a long connection with visual information, not least the origin of many alphabetic systems in hieroglyphics and other originally visual forms of writing" (ibid.). In handwriting the difference in expenditure and effort between writing a description and adding a quick drawing used to be a lot smaller than in printing. While printed texts became standard, this changed our conception of writing significantly.

Now, many theorists argue that we might experience a similar development with the textual migration from the book to the virtual. Critics like Landow believe that a new kind of writing including visual elements is only a logical development of our current situation. Furthermore, emphasis is placed upon the fact that hypertexts can show similarities with more iconic media like paintings as well as text. "The reader may "enter" a hypertext narrative just about anywhere, much like the viewer approaching a painting or sculpture. In such a hypertext environment, the reading process, like the gazing eye, jumps around associatively, moving not according to the work's formal structure but according to its content" (Tolva 1995). Consequently, some will see hypertext as a powerful hybrid of the symbolic and the iconic, others as a lowlife bastard child of the very same.

Hypertext writing has also been compared to this other "hybrid" form of poetry, ekphrasis. Tolva notes: "non-computerized writing, long considered a temporal art because of its paginated unidirectionality, has never been able to emulate the experience of the visual moment, though a small sub-genre of literature called ekphrasis has attempted to approximate it, archetypal examples being Homer's description of Achilles' shield, Keats' "Ode on a Grecian Urn", and Rossetti's Sonnets for Pictures" (Tolva 1995). Driven by a mimetic desire, poets wanted to create objects through words. Keats' Grecian urn e.g. has never existed outside his poem. It was an attempt to transgress the opposition between the material and the spiritual. The urn was a creation of the mind and it lived on in words, not in the physical world.

But for hypertext we have a different kind of ekphrasis. "Digital textuality effects this formal, structural, and perceptual experience of ekphrasis on a purely technical level. It is an ekphrastic medium that quite literally shapes the message [cf. hypertext syntax]. No longer a literary device or trope, ekphrasis as it applies to the computer is a practical description of the visual ways that the reader approaches the verbal text" (ibid.). This was probably the cause of the stiff resistance Landow experienced (cf. supra), a sort of 'ekphrastic fear,' "the moment of resistance or counterdesire that occurs when we sense that the difference between the verbal and visual representation might collapse, when the difference becomes a moral, aesthetic imperative rather than a natural fact that can be relied on." (ibid.).

Diametrically opposed to, but not irreconcilable with Tolva's ekphrastic fear, Bolter describes in his essay "Degrees of Freedom" a desire for the natural sign. "The power of written language to convey and convince is further undermined, as the inhabitants of cyberspace (more and more of us, though never all of us) succumb to what Murray Krieger has called the "desire for the natural sign" (Bolter, 1996). This is the desire to dispense with all arbitrary signs and to invest oneself in a second reality that approaches the first as closely as possible. "[The inhabitants of cyberspace] believe that a virtual environment rendered in immersive, three-dimensional graphics makes possible unmediated communication. Written communication, which is necessarily mediated, then becomes dispensable" (ibid.). Bolter thus opposes hypertext and text to virtual reality instead of old text (paper) to new text (electronic). Even text in the extended definition--i.e., including graphical signs like pictures and drawings--will ultimately be opposed to virtual reality. "[W]e do not see visual elements in the process of breaking out of the text, as we do in multimedia. The visual has already broken free and replaced text altogether" (ibid.).

In fact, Bolter argues, computer graphics are also symbolic in a way, but not like alphabetic signs are. The computer as alphabetic hypertext is symbolic by definition as it consists of arbitrary signs. The relation between signifier and signified can only in very few cases be accorded a degree of motivation. But even when it is motivated, it is still symbolic. Moreover, according to De Saussure, words derive their meaning from their relation to other words. Hypertext adds to these implicit links, explicit ones to other text chunks. However, these too are arbitrary acts of reference.

For Bolter, a graphic scene on a computer is viewed in a different manner, but it is symbolic. "The viewer recognizes the scene as something in the world (or in a possible world). She sees a building or a landscape or an object. And despite the fact that a computer-generated landscape has a cartoon-like geometry and brightness unlike the real thing, she can nevertheless see what the landscape is supposed to be a picture of" (ibid.). But it is still a 'picture of...;' it is not recognised as the thing itself. The desire for the natural sign is in fact a hidden desire for the actual object, for total immersion as Murray would call it.

The enthusiast of virtual representation dislikes looking at the computer screen; she prefers to look through it, to promote the illusion that the virtual world of graphics is a real world. Presently, even the most enthusiastic virtualist must at some point become aware of the screen as an artificial medium. However, he wants to limit these periods of awareness to an absolute minimum. The ultimate hope is that the computer can create a graphic environment so ‘real’ that the user can look through it without ever being reminded of the computer as the "man behind the curtain" (ibid.).

Murray believes that such a virtual experience could be valuable in a therapeutic way. "The holodeck [an imaginary virtual reality device], like any other literary experience, is potentially valuable in exactly this way. It provides a safe space in which to confront disturbing feelings we would otherwise suppress; it allows us to recognize our most threatening fantasies without becoming paralyzed by them" (Murray 1997: 25). Thus, it permits the user to hold a general rehearsal; it provides a simulation environment where one can learn how to deal with problems. Bolter argues that this is exactly what most MUD (Multiple User Dimensions) users are looking for, and that the symbolic environment they are placed in now is only due to technical restrictions. "Most MUDs do not set up an oscillation between rhetorical awareness and forgetfulness. They do not ask their users to look at the text, but only to look through it" (ibid.). As a consequence, from the moment a higher degree of immersion becomes feasible, the symbolic mode will be deserted for the iconic.

In 1992 Joyce still regarded hypertext as closely related or at least as a befriended colleague to virtual reality. When we place his description of "hypertext as virtual reality" against the background of Bolter's argument that the two are opposites, we get the impression that he as well has a desire for the natural sign.

Joyce has been too optimistic. There is indeed an emancipation of the iconic taking place, but whether this process will include virtual reality is doubtful. Thus, a new opposition--in a way again between the icon and the symbol--seems to appear. "Virtual thinking is the desire for the natural sign. Hypertextual thinking is the desire to complicate the relationship between the sign and what it stands for" (Bolter 1996). But here Bolter shows some hesitation, he realises that too extreme a separation is impossible. "When I put it this way, it sounds as if postmodern theory is hypertextual, as if postmodern writers would condemn virtual thinking as something close to false consciousness" (ibid.). However, he does not seem to find a solution.

Bolter is probably right if he says that the hypertextual mode makes demands on the reader, but not so much on the author. Whether the hypertextual mode is hard to maintain is a meaningless question when you consider our point of departure, whereby we have postulated a textual migration. Unless Bolter means that virtual reality will take over from hypertext he seems to be unnecessarily attacking the hypertext medium. But is it possible that such a shift from hypertext to virtual reality takes place? I believe not! Our whole knowledge system is based on words and texts. There is no other way to have a discussion, but in words. No philosophy can be expressed without words. We live in a discursive society and we will never be able to transgress that unless we go and live in a wordless virtual world entirely. Therefore, virtual reality should be considered as another electronic medium, one situated next to hypertext, perhaps even opposed to it, but certainly not one that could replace it.

In the early days of hypertext theory, Michael Joyce proposed the concepts of exploratory and constructive hypertext. The first type is probably the best known, since it constitutes the larger part of the WWW. It is a kind of hypertext that permits the reader to browse through fixed bodies of material without allowing her to make alterations. All she can do is search through the information available, hence the term ‘exploratory hypertext.’

Constructive hypertext, on the other hand, allows for more than one author; anyone is free to change the passage or the text she wants.

In the past, different systems have allowed different degrees of authorial control for its writers. NCSA Mosaic 1.0 for windows for example, the first really popular Internet browser, allowed the reader to alter what she wanted on someone else’s page. Because this alteration did not appear on the Internet itself but was controlled only by the working unit and its browser, the principle was later deserted when the number of pages became too big to keep a record of. It simulated a constructive environment, but failed to make it interactive.

More or less parallel with the distinction between exploratory and constructive hypertext, two different conceptions of the hypertext medium seem to have come about. In his essay "Purpose and Play in Hypertext," Mark Zachry writes: "Although it is seldom acknowledged, I believe there is a fundamental incongruity in the literature about hypertext. On one side, writers are arguing that hypertext has a strong pragmatic appeal: to facilitate the efficient creation and dissemination of complex documents and sets of documents of all kinds" (Zachry 1997). These are chiefly the theorists that follow the cognitive and computer scientific approach, rather than the literary (cf. 2.1.1.), to study hypertext. Their main aim is to make hypertext into a medium that can communicate more information and do this quicker and more easily than traditional texts. They want hypertext to be informational hypertext.

On the other side, we have the proponents of hypertext as a new literary medium. They "are claiming that writing is the creative play of signs, and the computer offers us a new field for that play" (Bolter, quoted ibid.). They want hypertext to be an idea space where the reader can wander around freely and stand still by those facets of the virtual world that interest her. For Zachry "these representative quotations betray a conflict that merits the critical attention of those who want to understand hypertext" (ibid.) Therefore, we will devote some space to this discussion to look at the claims and viewpoints of both sides and to evaluate the solution proposed by Zachry. We will first deal with (1) hypertext as a purposeful tool (cognitive approach), and then move on to (2) hypertext as a site of play (literary viewpoint).

Those theorists who approach hypertext as a purposeful tool assume that it is a new communication model that will permit the author to transmit larger amounts of knowledge with less effort. Note that the author takes initiative and that it is she that should control the communication chain, which makes this model very traditional.

For these users, the play made possible by the seducing quality of dialogic communication works counterproductive. "The dynamic functions of hypertext (e.g.: search and retrieval) are valuable, but the instability of unified meaning is a demon to be excised from the system" (ibid.). In other words, hypertext is supposed to become a highly productive polysystemic database that is consulted when a certain piece of information is needed. When this is the case, the user is supposed to be able to trace this information unit as quickly as possible and then leave the system.

In the purposeful view of hypertext, authors acquire a strictly delineated task. They have to use the available tools to design hypertexts that work in as predictable a fashion as possible. "They work with the assumption that a well-designed hypertext will produce a given set of results. When a hypertext fails to produce the results its creator envisioned, that is, when it fails to fulfil its purpose, the hypertext is a failure" (ibid.). But what should this purpose be? Is it a certain message that is to be communicated? Is it a message together with its interpretation? Or is it even useful to try to include the reader's reaction as well? This is of course too naïve! Every reader interprets the message in her own way, against her own background. Moreover, the reader's reaction can never be predicted because ‘the reader’ does not exist; there is only a reader and a model reader. What remains then is the message. Can a message be conveyed perfectly? Proponents of the purposeful view believe so or at least strive to obtain results as closely to the original objective as possible.

The objectives of an informational hypertext should be well defined. On the one hand, the reader wants to find a certain piece of information, and therefore she must be guided to it as swiftly as possible. I believe we can employ the distinction we introduced earlier between micro- and macrostructure. Here, we are clearly dealing with macrostructure. The reader must try to find his way between the different webs to the one containing what she is looking for. Once this is over, in other words, once the correct web is found, the power structure is almost always inverted. In an informational hypertext the author will try to take the reader by the hand and guide her through the text so as to make her understand it in very much the same way as the author wants to be understood. This change of direction or current in the communication is one possible criterion to distinguish informational from fictional hypertext.

The reluctance exhibited by the informational hypertext author to share power with the reader derives from the fear to be misunderstood. She is aiming at a clearly defined type of reader. "Rather than playful readers who are interested in browsing through hyperspace, ready for a serendipitous connection, a purposeful hypertext is designed to reward the sober, goal-orientated reader" (ibid.). This type of reader has a similar relation to the macrostructure of hypertext as the aforementioned author to her microstructure. "Either the reader or the author may define the objectives, but, once established, these objectives rule the communication in a well-designed hypertext. The possibility of being ‘lost in hyperspace’ is, perhaps, the most serious threat to the purposeful hypertext designer" (ibid.).

For Zachry, "the focus of these writers on ‘the needs of readers’ is inherently problematic in that it assumes that the author can predict the communicative interactions that will occur when a reader encounters a hypertext" (ibid.). Postmodern criticism, on the other hand, emphasises the dynamism of the meaning of texts as opposed to the traditional assumption that meaning is pre-established, just waiting to be discovered. Not just every text is different to every reader, but even the readings of one text by one and the same reader differ significantly according to the context. However, "Charney and many others assume that hypertext readers are unified subjects who can be neatly analyzed and identified. This analysis somehow yields the audience's ‘needs’--an equally difficult concept to identify outside a specific communicative act" (ibid.).

Finally, for the purposeful approach, hypertext as textuality is new, but at the same time considered already understood. Its proponents believe that hypertext is merely a swifter version of the conventional codex. "Links, nodes, graphics, and digitally manipulatable displays are not ignored in purposeful discussions of hypertext, but they are commonly viewed as textual enhancements that can be manipulated to serve the author's defined purpose(s). Authors are the architects of meaning who, with enough effort, can encode significance in well-constructed hypertexts" (ibid.). We will now see that literary theorists often add another dimension to this conception.

Jurgen Fauth, who believes that hypertext "has too many weaknesses and no advantages" (Fauth 1995), claims that "the co-operation (...) between the writer and the reader of a hyperfiction is not that of two collaborating artists, it is that between a game designer and a player" (ibid.). He seems to take this as an insult for the traditional reader. He sarcastically adds: "electronic literature will remain a game, just as all computer programming is a game.... In video games, the kind depicting spacecraft and deadly robots, the player competes against the programmer, who has defined goals and put obstacles in the player's path...." (ibid.).

Fauth even explains away the lack of closure from this standpoint. "No matter how competitive, the experience of reading in the electronic medium remains a game, rather than combat, in the sense that it has no finality. The reader may win one day and lose the next...." (ibid.). What is more, he believes that this playfulness will eventually effect the downfall of the new medium. "The impermanence of electronic literature cuts both ways: as there is no lasting success, there is also no failure that needs to last. By contrast, there is a solemnity at the center of printed literature--even comedy, romance, and satire--because of the immutability of the printed page" (Bolter, ibid.).

But whereas Fauth sees playfulness as an unavoidable weakness, others will regard 'play' as in 'movement or space for movement'--i.e. for the interpretation of signs-- as a positive quality. "For many of these theorists, hypertext is not so much defined by the technologies that make it possible, but by the basic nature of sign systems" (Zachry 1997). The arguments of these (mostly literary) critics are generally based on poststructuralist thought. "As Derrida argued before the advent of computerized hypertext, absolute meaning, in the twentieth century, has yielded to "the concept of play" wherein the movement of signification adds something, which results in the fact that there is always more" (ibid.). Some go even further:

Precisely due to the transition to three-dimensional hypertext many people have come to realise that the relationship between signs and meaning is not ontologically describable. It is not a given instance or unity. "Rather than being sites of stable, purposeful meaning, texts (in the broadest sense) are active--always extending into the realm of the "more." Hypertext, however, is also a dynamic, always extending realm of information. Thus, for many, the old dream seems to return. Perhaps we cannot grasp the meaning of a word or a text, but at least we can now explicitate and store store it in some way. "More so than traditional forms of publication, hypertext is a site of play in the construction of meaning" (ibid.). This highly abstract matter is of course not acknowledged any value by 'purposeful' hypertext writers. They hold on to the old assumption that meaning can be described and therefore unproblematically conveyed.

The enemies of a looser structured hypertext will claim that "play" is undesirable whenever it interferes with purpose. "[Charney] associates playful hypertext with Romanticism--an enemy of purpose that may be associated with the Enlightenment. She argues that a "Romantic view of hypertext (…) aims at enabling imaginative leaps and connections between disparate texts, facts, and images" (ibid.). Hypertexts that are not directed in a purposeful manner "put enormous technological and creative effort at the service of preserving what might be quite rare and ephemeral associations." (ibid.) For others, then, this aspect tends to be seen as favourable; they see "the decentering of meaning in all texts make communication more romantic as it leads to a more rhetorical understanding of communication than we have had in our print culture" (ibid.). We will see later that rigid hypertext structuring does not exclude decentring or even shifting of meaning, it can even contribute to it.

Another typical element of the second type of hypertext criticism is the competition between the author and reader to create meaning. Since the reader can choose what she reads and in which order, the author's message must be plain. Many traditional rhetorical devices are based on a linear format of text. They become almost dispensable in hypertext, that is to say, they can only be employed within one node. Thus, e.g., the pyramid structure of a well-written essay collapses from the moment a few links are created. The reader follows her own way through the material. The competition between reader and author can take various forms, it "may be as simple as allowing readers to add text to the original text or allowing them to filter the "texts" they receive. Conversely, it may give readers a freedom that broaches the efficacy of authorship: altering "original" texts" (ibid.).

However, there is a second, more concrete type of competition, due to what Murray calls the mosaic--this has nothing to do with the aforementioned browser--structure of modern media. For example, the fact that in newspapers more and more different articles are placed or announced on one page, very often joined together with pictures, permits the reader to commence his reading in different corners. For hypertext, this trend appears even clearer, on almost every page, links are crying out for attention. "[W]ith their many sensory appeals, potentially adventurous links, and volatile environment (...), the elements of hypertext contend for the reader's attention" (Zachry 1997). What we have here is in fact a re-emergence of rhetoric! Not the traditional verbal rhetoric, but one of colours and moving images. This, again, distorts the intended message of an informational hypertext. However, it creates a different one in a different format which is not necessarily worse. Either way, purposeful authors will condemn or at least avoid it.

Many critics of the second type also regard hypertext as a comic medium. "Its dynamic nature and mutable forms undermine the seriousness of the serious text, the revered codex. It dramatically alters the ‘rules’ of reading by encouraging the reader to participate with the text in the construction of meaning" (ibid.). This is of course a problem for purposeful authors as it diverts the reader's attention away from the intended message. "In a system where irreverence for the form of print is encouraged (if only because it is possible to reshape the written words) and where parody and satire so commonly redefine the decorum of traditional textual communication, the notion of a task-orientated, goal-directed communication is open to serious question" (ibid.).

Finally, theorists who see hypertext as a site of play often see it as a literary laboratory as well. Hypertext, with its still seemingly unlimited scope of new possibilities allows for easier experimenting than most other media. Perhaps, for paper, centuries of experimenting have exhausted the possibilities of the linear medium, making us more or less blind to the subtle differences and past achievements we encounter. Be that as it may, the shape of the message in hypertext can be altered to such an extent (by the author as well as the reader) that its actual meaning changes. "Hypertext allows readers to reshape the communicative interface (text) and to move beyond the immediate screen to new and potentially competing texts. With its links to sites unknown, hypertext provokes readers to venture into the new, to experiment with the communication process" (ibid.). Although some rules and conventions will eventually be established, the door to inventive experimenting remains open.

In this section we will attempt to classify the different sorts of hyperfiction structures formally. In the past, theorists have often suggested how a classification of the phenomenon should be regarded or, at best, how it could look like, but mostly the proposals remained vague. Landow, for example, claims in his Hypertext 2.0 that "Hypertext narrative clearly takes a wide range of forms best understood in terms of a number of axes, including those formed by degrees or ratios of (1) reader choice, intervention and empowerment; (2) inclusion of extralinguistic texts (images, motion sound); (3) complexity of network structure; and (4) degrees of multiplicity and variation in literary elements, such as plot, characterisation, setting and so forth" (Landow 1997: 180). In this way, Landow is trying to avoid 'classical' linear thinking by replacing it with a non-linear spatial representation. "I prefer to think of the organizing structures in terms of ranges, spectra, or axes along which one can array phenomena, rather than in terms of diametric oppositions, such as male-female or alphanumeric versus multimedia text" (ibid.: 180).

This type of representation of hyperfictions gives us a good idea of its complexity. In hypertext fiction many more variables are involved than in classical fiction, and they all contribute to the readability of the work. Perhaps we should view the different hypertexts by assessing them according to hypertext syntactic, narrative syntactic, and even medial qualities. For a formal classification, however, Landow's proposal is too inadequate and unscientific. He describes axes and degrees or ratios of things that can impossibly be measured or quantified objectively. How does one objectively represent or even evaluate "the degrees of multiplicity and variation in literary elements" (ibid.), for example? Is reader choice the same as intervention? And how does empowerment come into the picture? Landow's description seems to be valuable to form an idea of the complexity of the matter. For a concrete typology we will have to turn elsewhere.

Rees provides a more detailed description of what hypertext can be used for, thus attributing certain qualities and alluding to the existence of some different structures. "Hypertext may be used simply to add illustrations and footnotes (…); it may be used to present a single narrative through different viewpoints, different styles of writing, different places and different times; or it may be used to allow the reader to access a variety of material in an unstructured way and thus construct their own narrative" (Rees 1996). In this description, Rees implicitly refers to three hypertext fiction structures: annotative, multilinear and idea space (cf. infra); but it all remains vague.

Van Driel also attempts to make a classification of digital media with narrative content in general, rather than of hyperfictions in particular. For this, however, he employs such a variety of criteria (referring to comics, multiple choice, no narration), one next to and apparently of equal status as the other, that his partition is of no value to us whatsoever.

By far the best example of a classification of digital structures can be found in Katherine Phelps's web-published essay "Story Shapes for Digital Media" (1998). The purpose of her work was to create a basis in order to be able to study the mapping possibilities of standard storytelling elements in digital media. As one of her sources she quotes Shumate. "Michael Shumate in his Hyperizons site on the Internet touches on what he describes as linear, annotative and tree-branching shapes" (Phelps 1998). From these observations and from investigations of various CD-ROMS and hyperfiction sites, she has formed "a model of the possible pathing structures stories can take within digital media" (ibid.). She distinguishes between seven story shapes: Linear, Interactive, Multi-Linear, Braided Multi-Linear, Nested Funnel, Tree-Branching, and Non-Linear" (Phelps 1998).

In section 2.2.1., we have described hypertext as the explicit manifestation of the struggle between linearity and juxtaposition. In other words, from the moment that links introduce explicit reference from one text-block to another we have said we have true hypertext structuring, be it in a linear medium like the book or on a computer. From this definition we will be starting our exploration of the different types of hyperfictions. The structures we will be describing are hypertext syntactic structures; narrative syntax is only secondarily grafted on the hypertextual basis (cf. 1.4.). The order in which the classification is represented goes from more to less linear. Many of the terms and ideas are based on Phelps's story shapes. However, as she is dealing with digital media more generally, we will have to make some alterations to both the terminology and concepts in order to narrow the viewpoint to just hyperfiction. Some more distinctions will be introduced and some concepts and terms will be reinterpreted in order for them to conform to our perspective.

But before we start the actual categorisation, we have to make a few remarks. First of all, we have to refer to some terms we will be using throughout this section to differentiate between different types of structuring. These are no types of structures themselves but tools to describe the differences between them. Moreover, they do apply to hypertext syntax and not primarily to narrative syntax. Thus, when we speak of closed narrative, we are not saying that the story cannot have an open ending. Rather, we mean that it has a predetermined last page. Phelps describes a closed narrative as a structure where the storyline is set. There is only one story, of which the flow or outcome cannot be affected. This type of narrative is mostly used in hardcopy fiction. In hyperfiction, however, another dimension can be added, viz. that of open exploration. "The storyline is set, but the audience is free to move around the narrative environment in different ways" (Phelps 1996).

On the other hand, in hyperfiction we can also have an open narrative, one of which "the storyline may have diverse outcomes or entirely participant driven narrative" (ibid.). However, in a hypertext environment this is only possible "within the constraints of what is appropriate to the particular narrative environment. Creators simply cannot offer all possible choices to their audience" (ibid.). This we will refer to as closed options. Phelps adds that this problem can be overcome by developing a world whereby certain choices are more likely and more appropriate to the environment than others. But such a world is no longer strictly hypertextual. We will discuss this issue briefly in 3.1.4., when dealing with non-linear structuring and MUDs.

Moreover, we will have to distinguish between path, storyline and (narrative) universe. A path has been defined previously as "a sequence of nodes and links taken while navigating the network" (Keep 1995). It refers to the way in which certain nodes are consulted. No aspects of content are involved; a path is situated entirely on the hypertext syntactic level. The term storyline we will use in its traditional sense, referring to one line of events, mostly connected by a causal chain. In digital media, one storyline can easily be crossed in different ways and can therefore allow for more than one path. Just think of footnotes, which the reader can read or skip. They may belong to the same storyline or main thread, but the decision to read them or not entails a different reading path. A storyline is a concept situated on the narrative syntactic level, but as we will see, in hypertext it is also very often articulated on the hypertext syntactic level. A (narrative) universe, finally, consists simply in those storylines that belong to the same (narrative) world. In a hyperfiction where the user is offered choices as to how the story should continue, we get different branches that do not belong to the same universe, although they are paths and storylines of the same story.

Furthermore, to flesh out these rather theoretical considerations we will always try to graphically represent the structure referred to. These representations are by no means exact renderings of the structure of existing hyperfictions. Neither are they examples of how a hyperfiction structure can best be mapped. On the contrary, they are exemplary metaphorical depictions to help visualising what is being described. The concrete usability of the diagrams is therefore very limited due to simplification. A true mapping of hypertext structure involves plenty more parameters and more complex, preferably computer-mediated, graphs.

Finally, we also have to remark that one structuring type does not exclude the other. Different structures can be combined in one hyperfiction. How such hybrid structures should be represented or referred to has not yet been established. Perhaps a hypertextual grammar of multiple structuring may be desirable. A uniform formulary representation would greatly improve our understanding of its structuring. Equally, one structure can be interpreted in more than one way. When an annotation is as long as the main story, it is sometimes unclear which of the two is in fact the dominant frame. This is particularly problematic for those hyperfictions that offer several possible beginnings.

In this section we will look at three structures that exhibit a more or less uni-linear structure. The first type is the (1) plain linear structure, then we will move on to (2) annotative structuring or advanced footnoting and finally we will be dealing with (3) tree-branching, which is only partly linear.

In 1996, Michael Shumate wrote: "Most (…) fiction [found on the internet] is not hypertextual (some would argue that there is virtually no true hypertext fiction on the Web): it's simply print fiction on a screen with electronic page-turning devices--"click here for Chapter 3" and so on--rather than fiction intended to be read multisequentially" (Shumate 1996). Although since that time many things have changed, the situation for fiction is still very much the same. Most texts are still published on the net as if they were published on paper, i.e. in a linear fashion. We could represent this type of hypertext structure graphically by a simple line.

Since this structure is not the only existing uni-linear structure, we will refer to this type as plain linear, as in plain: not elaborate or elaborated, simple; lacking patterns.

This type of text is not very different from its hardcopy counterpart. Earlier, in the part about hypertext as a metaphor, we have referred to this kind of text as two-dimensional text. The only difference from paper text is its electronic or virtual status. Full-text search and retrieval are possible (cf. 2.2.1.). This allows the reader to create soft, reader-determined links. She can, for example, decide to read only those parts of the text that deal with the main protagonist's boyfriend. By using the 'find button' she can jump from one mentioning of the boyfriend's name to the next, skipping all the rest of the text.

This becomes more problematic when the linear text is broken up in different nodes, but not impossible. Search engines can move beyond the boundaries of nodes and look for certain strings of text.

The annotative structure is one comparable to traditional scholarly footnoting. The reader can choose to receive more information about a certain object or part of the text, but afterwards she has to return her path to the main storyline to finish the story. She can only choose to extend her path, to take a roundabout way, a detour. Annotative structuring can be represented as a line with smaller branches at the side. In this case, we have one simple side-branch, one with a fork consisting of three more annotative links and one with a web or idea space attached.

The notes outside the main line of the text do not guide the reader onto a totally other track. They merely offer the opportunity to enrich her experience of the presented material. "Any of these additional elements (…) could be automatically presented upon each new scene appearing. However, in giving that choice to the audience they have a greater sense of involvement" (Phelps 1996).

Finally, this type of structuring is sometimes also called interactive story shape (Phelps 1998) and advanced footnoting (Shumate 1996, Fauth 1995). Both of these names are not as appropriate as 'annotative' however. As we will see later, plenty of different structures are explicitly interactive. The fact that this is the most basic interactive structure does not seem very important here. For 'advanced footnoting', we could say that the term 'footnote' in itself is already an unnecessary abstraction as the notes are not at the foot of the page. Secondly, we see that not all of the annotative structures are so advanced. We may want to reserve the term 'advanced' for when the 'note' consists of a complex web and not just one node.

Large-scale exploration of the alternative possibilities of a narrative is a fairly recent phenomenon in print. It was only in 1981 that tree literature "reached the masses with the publication of the `choose your own adventure' series (beginning in 1981…) and the `fighting fantasy' series" (Rees 1996). Both series are meant for children and almost all of the books belong to the science fiction genre. Moreover, they are presented as games rather than stories, whereby the reader has to find her way to the correct outcome, most other paths leading to death or failure.

Digital tree fiction uses the same concept and techniques as these earlier examples, only on a larger scale. In this discussion we start from the presupposition that tree-branching structure (hypertext syntax) is used to allow the reader to decide what happens in the story. Consequently, every time she makes a choice, she enters a new and unique narrative universe (narrative syntax). This type of structure almost inevitably entails an open narrative--that is, one of which the outcome is not predetermined--with closed options. Theoretically, it is also possible to compose a tree fiction situated in only one narrative universe, but this would be a terrible waste of material and the reader would probably be worn out soon, since she has to return endlessly if she wants to read everything.

Tree-branching is not a truly linear structure, but neither is it a multi-linear structure or an idea space (cf. infra). The reason for this indeterminacy is that we just cannot say that it consists of only one story. In fact, with every branch that is taken, a new story is commenced. In this way, one tree fiction consists of an extremely high number of stories that are exactly similar up to a certain point. The number of different stories, in other words, depends upon the number of end nodes, which should normally coincide with the number of stories. This is not the case, however, when different stories merge. Because of the linear development of the different storylines, I have included this structure in the first group, that of linear structures.

A tree-branching structure could be graphically represented in the following way. I have added and marked one example of a merging. Note also the fact that limiting the choice to only one option results in a temporarily plain linear structure and that the number of branches added on one node can be extended almost ad infinitum.

In this simplified graph, the number of storylines differs from the number of endings due to the merging at the bottom. This merging produces that the three branches that follow it have a double and dubious status. They could either be read by someone who has followed the merging path or by someone who has taken the normal way. For both readers the three following branches can have an entirely different meaning. Thus, this tree fiction has fourteen storylines, all belonging to a different narrative universe, but only eleven endings.

Tree-branching enjoys a particular status in this classification. Not only does it hardly allow for a one-universe-story, it also inevitably has a very strong sense of playfulness to it. It is probably due to this status that tree-branching is hardly used in combination with one of the other possible structures. Another consequence is that these stories "tend to avoid having to deal with the problem of lost consequences by being structured around a fairly linear journey" (ibid.). If the character would return on her path, it is very likely that she will be confronted with the consequences of her doings or, at least, that would seem logical. When storylines have merged, however, it is impossible for the author to simulate such adaptations of context in this format.

But this is not the major problem that this type of structuring is facing. All writers choosing for this structure will come up against the exponential problem. As the number of choice-points goes up, we experience an exponential explosion of branches and endings. "With ten binary decision points, there are a thousand endings; with twenty, over a million" (ibid.). And how long is our story with ten or twenty decision points? With one page between every decision point we would still only have a short story or a very short novella. But if we want to offer the reader frequent choice, or at least the illusion of frequent choice, How can we manage it within the tree-branching framework?

Three solutions to this problem have been suggested: (a) killing off branches, (b) collaborative writing and (c) merging narratives.

(a) "Traditionally [the exponential problem] has been handled by having one 'correct' story path and all others lead to death, thus pruning their potential early into the decision making process" (Phelps 1998). This is obviously not a very satisfying solution. Moreover, it degrades the story generally to a mere gambling game. "How often in life are we offered choices of roughly equal value, it's simply a matter of committing to one and then experiencing the outcome? A basic example would be walking into an ice cream shop and choosing one flavour over another. No one right choice exists, it's a matter of preference" (ibid.). In a situation like the former, it should be possible to give the two branches an equal chance. It is most annoying that, every time the characters go eat an ice-cream, only one flavour is not poisoned, because all other branches have to be killed off as soon as possible. For a game-like fiction this solution might sometimes work, but for a true story we must turn to other solutions.

(b) A second, more attractive solution to the exponential problem is turning the tree fiction into a collaborative project.

Turning the tree fiction into a collaborative project is not a perfect solution either, however. A first problem is that even with many co-authors, writing a tree fiction is a vast undertaking. For the first Addventure 30,000 rooms had been prepared and 10,000 were filled. For the second 41119 episodes were completed out of 113534 available. And even with such high numbers of nodes, the story does not last very long. Rees has calculated: "If every English­speaking person in the world wrote a single section, together they could not complete all the branches on a tree with 28 decision points (a story in Chinese would get one decision point further)" (Rees 1996). Of course, one person can write more than one node, but nevertheless, writing a whole novel in this way hardly seems feasible.

A second problem for collaborative projects is that "collaborators will not be aware of each others' hidden assumptions and, especially in the distributed style of collaboration envisaged by some" (ibid.). Indeed, writing such hyperfiction very often seems more pleasant than reading it. It seems as if every author wants to give the story a swing of 180°. You could have, for example, a girl that meets the handsome young man of her dreams for a candlelight dinner, and just when they are starting the main dish, another author has him turn into an alien and eat her alive, killing several other people including the waiter on his way out. It should be clear that such a continuity mistake renders the story no more than a joke.

(c) Finally, a third solution to the exponential problem would be to merge different branches into one or a few threads again. But is this a good solution in a format where choice is the main principle? Is not something lost when two different threads get back together and arrive at the same destination? "One possibility," Rees notes, "is that the choice has no effect, and thus it is not really a choice" (ibid.). This is not entirely true! For the ice-cream example sketched above, a merging would not be very problematic, unless the ice-cream theme would pop back up in later nodes. But Phelps' example was only a trivial one. In tree fictions like Addventure the choices offered to the reader are a lot more substantial. "A second possibility is that the choice has a real effect all of whose consequences are then lost" (ibid.). Further, Rees observes: "Perhaps there is a third possibility, in which the continuation has two meanings, one straightforward and one ironic, depending on which path was taken. However, I find it hard to see how this could be made to work for non­trivial examples" (ibid.).

A multi-linear structure is fundamentally linear--that is, it does not allow the reader to move through the story freely, it allows only for linear reading. Nevertheless, there exists a greater freedom of exploration in this type of closed narrative than in a uni-linear structure. The reader is forced to move towards a certain endpoint, but there are different ways to get there.

In this section we will deal with two truly multi-linear structures: (1) plain multi-linear and (2) braided multi-linear. Finally, we will take a closer look at a type of structure that is only related to multi-linearity in that it simulates the former to a certain extent: i.e. (3) nested funnel.

Just like the plain uni-linear type of structuring, plain multi-linear is not restricted to digital media. In fact, this structure has often been used in hard copy fiction, but also in analogue media like film, and even music. Typically, in a plain multi-linear structure, we have several different paths (mostly storylines) which do not intersect regularly up to a certain point, where they join and lead to a common ending. On paper, this type of structuring could also be realised on the hypertext syntactic level, alongside the narrative syntactic level, by devoting one chapter to each storyline alternately for example. This traditional structure is very much comparable to that we find in plain multi-linear hyperfictions. The only difference is that on paper the different chapters are still represented in a linear fashion, one after the other, due to material restrictions. In hypertext this is not the case.

A plain multi-linear structure could be represented in the following way:

In this example we have four storylines or paths that join at a certain point and then lead to only one ending. Evidently, the multi-linear structure ends when the different threads come together. Note also that an author needs four different beginnings for this story. When the ending line would split once again, we could speak of hourglass or bottleneck structuring. When it would split and join more than once, we have a braided multi-linear structure (cf. 3.2.2.). This points to the inherent relation between the three structures.

Phelps notes that the reasons for using plain multi-linear structuring "may be that you will be travelling through the same physical spaces or the same series of events through the eyes of different characters, or the same character through different events, or perhaps even the audience is meant to compare and contrast these different lives and/or events" (Phelps 1998). Here, one question arises: why not make full use of hypertextuality and employ braided multi-linear structuring. Phelps observes in this regard: " What distinguishes digital media in the use of this form, besides just the addition of sound or moving images, is the degree to which the parallels can be made" (ibid.). This is only partly true, I believe. The reason the author chooses for plain multi-linearity in a digital medium is either that she wants the reader to finish one strand and only then pass on to the next or the fact that there are hardly parallels between the different branches. When there is no unified chronology, it is very difficult to implement braided multi-linearity. Thus, plain multi-linear structuring does not make full use of hypertext deliberately. This, however, does not mean that it has no place in hypertext narrative theory.

A braided multi-linear structure could be represented in the following way:

Phelps describes the braided multi-linear story shape as follows: "an initial situation is given which then branches out in a number of thematically related plot directions which subsequently converge upon another situation that then again offers a number of directions to choose from" (Phelps 1998). This is an adequate description of what we will call the braided multi-linear structure for hyperfiction, except for the term 'plot directions'. This term implies that we are dealing with an open narrative, where the reader can choose which way the story may go. If we allow this description, we have to include tree fictions with merging narratives in this group as well. Phelps, for example, sees 24 hours as an excellent example of braided multi-linear shape. I would prefer to leave the realm of formality here and refer to content to distinguish between the two. Those stories that confront the reader with choices are open narratives (the outcome is not fixed, the different branches belong to different narrative universes) and should therefore be seen as tree fictions with merging storylines.

A braided multi-linear structure could be described as a type of linear structure where different storylines or accounts of the same story intersect more than once so that at each intersection the reader can choose her path to the next one. This type of structure is used a lot for online hyperfiction, and very often it is mistakenly described and referred to as being a rhizome (cf. 3.1.3.). The fact that it allows for dramatic tension is probably one of its stronger points. It can "give the reader a feeling of free movement and yet even with that free movement provide a seamless story experience" (ibid.). Moreover, it makes the reader want to repeat the same story to explore the other paths. It provides her "with a framework that [she] may discover [she] enjoy[s], and then makes it possible to uniquely repeat the experience" (ibid.).

Another possible structure is what we could call an hourglass or bottleneck structure. It can be represented as follows. (Note that it can have all kinds of in- and outputs: here a web as input, and one as output.)

Hourglass is related to both the braided multi-linear structure and to nested funnels. On the one hand, when we would build a structure around one crossing in a braided multi-linear structure, we would have a simple example of an hourglass structure. The number of parallel storylines is temporarily limited to one and disperses again. On the other hand, with a nested funnel structure an hourglass has in common that it narrows down a story, and only allows one single opening to the rest of it. Hourglass, however, does not require any particular programming or complex sequential organisation of nodes. It is like a bridge between two islands of the story. There is no other way to get to the other island, but to take this bridge.

Note also that the above example, represented in the graph could be represented differently as a linear story (one path through each web is chosen to be the continuation of the line) with an annotative structure. The only annotations are then the rest of the two webs. All this exhibits that hourglass is not an independent structure type with full theoretical status. Nevertheless, it can be used very effectively.

A structure using nested funnels could also be called 'scavenger hunt' or 'multiple act shape'. It is a linear structure, but at certain points the reader has to conform to certain conditions before she can go on reading. The nested funnels themselves should be regarded as non-linear devices (cf. infra). "Within the nested funnel you must either go through all scenes, do all set activities, or at least a significant set of scenes, and/or activities within an act before you are allowed into the next act" (Phelps 1998). For hyperfiction in particular, we should add that the condition the reader has to fulfil, is not so much an activity within the story. Mostly it is simply the case that the reader is allowed to the rest of the story only when she has read a certain number of nodes from the web before the nested funnel. The Heist, for example, could easily have been structured in this way, whereby the reader is allowed to enter the Heist only when she has read about all the characters.

Because of the complexity and degree of abstraction of this type of structuring, we can only represent it metaphorically. We will be using a cone for each funnel, following Phelps (1998). Every cone represents a number of terms or conditions that need to be met with, before the reader is permitted to go on. A1, A2 and A3 are the conditions for the first cone.

In this section we eventually disband linearity for what has often been regarded as a truly hypertextual structure. Very often this structure has been named by its hyperonym hyperfiction, which is an absolute generalisation. Nowadays, it has become clear that there are many more structure types that deserve this description. We may therefore designate the truly three-dimensional web we are dealing with here by the term idea space, which has been mentioned in the literature before, but has not yet gained wide acceptance.

In his 1993 article ‘Hyperfiction: Novels for the Computer,’ Robert Coover described Moulthrop’s Victory Garden, an Eastgate production about the Gulf War, in the following way.

There are three important elements in this description. First of all, there is the connection with the "ordinary print novel," which is not denied, but conversely emphasised as being linear and chronological. We have a classical story here, although it is presented differently! Second, there is the fact that the whole narrative is behind us. This feeling of looking into the past is inherent to the genre of idea space. It diminishes the feeling of involvement for the reader, but provides something in exchange. The feeling of omniscience and empowerment permits the reader to have the impression she is digging into the ideas of the author and characters, hence the first part of the compound: idea.

Third and finally, there is what Coover calls "a kind of obscure geography" which he describes as "the reader moves about in the story as though trying to remember it." This is probably the most important innovation this genre offers. Linearity is not sacrificed, as many critics seem to think. It is replaced, extended to something that exceeds but still exhibits linearity, something that allows linear reading alongside a new type of inquisitive, dynamic, more active reading. A type of reading related to that which poststructuralist critics have been advocating zealously. The reader has finally been liberated from the author’s yoke. She no longer has to follow the path outlined before her. She can choose her own path through the (hi)story, hence the second part of the compound: space.

In the theory about this type of hyperfiction there are three important notions that appear with respect to the concrete reading process and which may seem appropriate to mention here. The first, (a) words that yield, refers to a purely text-constituting device. The second and third--which Landow addresses by the terms (b) predicative assimilation and (c) prosopopoeia--are situated on the interpretive level. This description does not aim to be exhaustive. What we will be doing is provide a brief theoretical basis, which will later serve to understand some problems and advantages when we will be dealing separately with the two types of idea space, rhizome and storyworld.

(a) In his polemic article about and against hyperfiction ‘Poles in your Face,’ Jurgen Fauth writes: "Michael Joyce's afternoon: a story (Eastgate Systems, 1990) introduced a major technical enhancement. Joyce rejected the pragmatic commands found in adventure games ("Go North"; "Take gold"; "Hit troll with axe") in favor of "words that yield:" cues to further development imbedded in the language of the story itself." The first sentence of Afternoon "I want to say I may have seen my son die this morning," for example, is tagged so that the different parts of the sentence contain links, each leading to a different part of the story. Thus, the reader may choose to click on "die," for example, or "this morning," and find her own continuation for the first sentence. Important here is, however, that both phrases--"die" and "this morning"--have different associations and almost opposite connotations. The author needs to take this into account if she wants to make linking into a meaningful grammatical and semantic device.

(b) Second, there is the interpretation of this type of text, which is at least as important as the representation. An important difference from a classical linear text is that many connections between text bits are less pronounced, less clarified by the author. Very often there exists only the link and the anchor or word to which the link is attached. When the reader decides to pursue a certain link, she is relentlessly thrown into a new text node, which she has to start interpreting from scratch, time and time again. Landow here refers to Ricoeur to explain the process: "the metaphorical imagination produces narrative by a process of what [Ricoeur] terms "predicative assimilation," which "‘grasps together’ and integrates into one whole and complete story multiple and scattered events, thereby schematizing the illegible signification attached to the narrative taken as a whole" (Landow 1997: 196).

(c) Although dealing with printed text, Landow observes that this description has affinities with the reader-author demanded by Joyce’s Afternoon and other works of hypertext fiction. "[R]eading is always a kind of writing or rewriting that is an act of prosopopoeia, like Pygmalion giving life to the statue" (ibid.: 195). Landow also compares this type of reading to what Lévi Strauss called bricolage; "every reader is eventually a bricoleur. This construction of an evanescent entity or wholeness always occurs in reading, but in reading hypertext it takes the additional form of constructing, however provisionally, one’s own text out of fragments, out of separate lexias" (ibid.: 195). This provisional construction provides a new type of wholeness, appropriate to hypertext.

Experience has shown, however, that hypertext theory can be interpreted in different ways and to different degrees. There are hypertexts that take this prosopopoeia-principle very literally and present the reader with loosely connected material. By 'predicative assimilation' the reader then has to create her own story from the bits and pieces she encounters. On the other hand, there are also hypertexts which seem to have adopted only the most interesting parts of poststructuralist theory and have added principles originating from experience and cognitive research. They allow the author to take part in the dialogue of writing. She no longer presents the reader with vague and often seemingly meaningless connections between material, but with a coherent whole. We will therefore distinguish between a non-structured idea space--which we will call (1) rhizome, and a rigidly structured idea space, which we may call (2) storyworld.

The rhizome is probably still the most prototypical of hypertext fiction structures. It is a type of fiction based on concepts derived almost directly from poststucturalist theorists like Derrida and Barthes. It consists of a finely braided web that aims to be non-hierarchical, decentred, lacking beginning and ending etc. and thus resorts voluntarily to an irregular structure.

The following figure is a highly simplified two-dimensional and to a high degree metaphorical depiction of a rhizome structure. Therefore, note that by this graph I only want to convey the fact that the structure is irregular and that it has no predetermined beginning or ending.

The term rhizome stems from poststructuralist literary theory, which, in turn, adopted the term form biology. Murray describes it as a "labyrinth derived not from Greek rationalism, but from poststructuralist theory, full of wordplay and indeterminate events, unheroic and solutionless" (Murray 1997, 132). The aesthetic vision is identified with the French philosopher Gilles Deleuze’s, who saw the rhizome as "a tuber root system in which any point may be connected to any other point" (ibid.). But whereas "Deleuze used the rhizome root system as a model of connectivity in systems of ideas; critics have applied this notion to allusive text systems that are not linear like a book but boundaryless and without closure" (ibid.).

There are, however, two important problems with which this type of text is confronted. The first is (a) the problem of confusion and the second is (b) the problem for the narrative to emerge from this type of structuring.

(a) Murray points to the former as follows: "In trying to create texts that do not 'privilege' any one order of reading or interpretive framework, the postmodernists are privileging confusion itself. The indeterminate structure of these hypertexts frustrates our desire for narrational agency, for using the act of navigation to unfold a story that flows from our meaningful choices" (ibid., 133). What the rhizome structure deliberately seems to do is refrain the reader from meaning-creation; she is constantly mislead and the choices she makes are hardly of any importance to the story, except, perhaps, for the order according to which it is told.

One important factor that may contribute to the reader’s confusion is the fact that "the semantics (…) of words that yield are unclear, and [that] the reader can never be sure if her click will influence plot, perspective, time, character, or any other element or combination of elements in the story" (Fauth 1995). This is a direct consequence of the theorem that no textual unit should have priority over another and that no hierarchy should be created. The resulting chaos of links makes it hard, and after some time also tedious, for the reader to constantly make choices that are no real choices. When the links do not have a uniformed and comprehensible structure, reading a hypertext fiction tends to become an experience of random clicking, which wears out even the most inquisitive reader after a while. Whereas Coover is "always astonished to discover how much of the reading and writing experience occurs in the interstices and trajectories between text fragments," Fauth convincingly points "to the more practical aspect of an often confused and befuddled reader who struggles to make a connection between two nodes" (ibid.).

The most common and probably the only possible solution when the author wants to hold fast to poststructuralist theory is "to revert to prose poetry or anti-fiction and the purely aesthetic. Much of the Eastgate publication list represents these approaches such as Stuart Moulthrop's Victory Garden or Michael Joyce's afternoon, a story" (Phelps 1998). Of course these works have their merits, but they also have their problems. While being forced by theory to be largely self-referential, their thematic content is limited. Moreover, most of these works exhibit the main danger for the genre, so well-described by Coover in his 1992 article "The End of Books:" as "the risk of being so distended and slackly driven as to lose its centripetal force, to give way to a kind of static low-charged lyricism--that dreamy gravityless lost-in-space feeling of the early sci-fi films" (Coover 1992). A second solution to this problem is to desert strict poststructuralist theory and allow a certain amount of structuring by the author (cf. 3.3.2.).

(b) Next to the problem of confusion, there is also the fact that from a typical rhizome structure hardly ever a complete story emerges. Landow claims that "in a hypertext environment a lack of linearity does not destroy narrative. In fact, since readers always, but particularly in this environment, fabricate their own structures, sequences, and meanings, they have surprisingly little trouble reading a story or reading for a story" (Landow 1997: 197). However, most of the literature does not agree with this statement. Even the hardiest enthusiast has to admit that concerning storytelling, a rhizome structure hardly ever satisfies the reader. For the creation of atmosphere in an immersive environment and for the rendering of poetic conjectures, the structure type is very appropriate. To develop characters and plot, however, the rhizome has to pass in most cases.

Phelps has "found it unusual to discover anything like a complete story coming out of this shape, because lack of narrative progression provides too little support for developing character and plot" (Phelps 1998). The explanation for this observation is obvious. As we have described earlier, the rhizome structure dispenses with all linearity, structure and therefore also development. The reader has become a wanderer in someone else's thoughts. She does not, however, always share the same background with the author. She cannot extract the intended story like she is supposed to. Two opposite powers are at work here. On the one hand, there is the desire to convey a story, and on the other the theoretical urge not to project any structures into the work. Thus, for a short work the audience may be willing to wade through all segments of a narrative in order to piece together a coherent story. For a longer work the tendency is for people to lose interest after only a few screens" (ibid.).

There are two important remarks to be made here. First of all, there is a solution to the problem of confusion, which at the same time evades the weakness of lack of narrative possibilities. "The most workable solution may be to create narratives for the non-linear shape much like Old Possum's Book of Practical Cats by T.S. Eliot or Our Town by Thornton Wilder which are collections of short self-contained stories held together by a common theme" (ibid.). In other words, the author tells a story held together by the main protagonist, for example, by presenting different anecdotes from her life. These anecdotes are supposed to be self-contained, that is, forming a story on their own. This is also the technique successfully employed in Adrienne Greenheart's braided multilinear hyperfiction Six Sex Scenes.

This may indeed be a good solution, but the issue at stake is whether we still have a rhizome structure then, even if the linking and the overall structure are still irregular. By bundling a number of stories into one web, the author is in fact exploring the boundaries between what we have called micro- and macrostructure. Purely formally speaking, this structure is micro-structural, just like every other rhizome structure is supposed to be. From a broader point of view, however, we see that in fact the work only makes use of the great macro-structural power of hypertext. Whereas in a linear representation the author would have had to look for a uniform principle to base the sequence of anecdotes on. Here, mere associational thought of both author and reader may create the connections. The anecdotes themselves, however, are by no means hypertextual, only the collection is. Because of the inherent sense of structuring in each node, I believe that this type of solution automatically leads us to the second group of idea spaces, i.e. storyworlds.

Note that the problems described previously have to be interpreted from the point of view of storytelling. Therefore, it does not mean that rhizomes have no chance of survival. All it means is that they are not the most appropriate medium to communicate complex factual messages such as stories. Thus, we could end by citing Murray on the subject: "the unsolvable maze does hold promise as an expressive structure. Walking through a rhizome one enacts a story of wandering, of being enticed in conflicting directions, of remaining always open to surprise, of feeling helpless to orient oneself or to find an exit, but the story is also oddly reassuring. In the rhizome, one is constantly threatened but also continuously enclosed" (Murray 1997, 133).

Perhaps we could add just one example of an experience which can impossibly be rendered more adequately by other media than it could be by a rhizome structure. What we are dealing with here is the teleological point of view. The question is: 'what could be described, communicated or depicted by the rhizome structure, that cannot be done by other, linear or analogue media?' Probably a WWW experience could be emulated almost perfectly within this mould. The reader finds herself within a cyberspace story, but one only showing virtual meta-cyberspace. She is not just the experiencer, but also the main protagonist, the focaliser and the narrator. Perhaps we could add that, here, another boundary is about to collapse. How can we distinguish a virtually virtual experience from a mere virtual one? Is there a difference between surfing through a web and surfing through an emulated web? The boundaries between fiction and non-fiction seem to be blurring…

A storyworld is a type of idea space, which developed from an entirely different angle than the rhizome. Instead of being based on poststructuralist theory, these hyperfictions have developed from Internet experience itself. The basic difference from a rhizome is that a storyworld is not denying the necessity of structuring within the three-dimensional space of the idea space structure. It is overtly hierarchic; it mostly contains a beginning (at least a starting screen) and very often an ending, and most typical problems of the rhizome are avoided by employing a strict semantics with regard to linking. Nevertheless, it also carries many of the characteristics and strong points of the idea space such as a certain degree of decentring, the possibility of collaboration, extended works with diffuse boundaries etc.

Again, if we want to present this type of structure graphically, we are confronted with the same dimensional problem as for the rhizome. Since we are not able to sketch any real structure, we will keep to the symbolical representation we have employed earlier.

The term 'storyworld' is adopted from a description by Landow. "Many hypertexts (…) exemplify what Michael Innis, head of Inscape, Inc, has termed a storyworld. Storyworlds, which contain multiple narratives, demand active readers because they only disclose their stories in response to the reader’s actions" (Landow 1997: 208). This is an excellent description of what a structured idea space is like. "Entering each environment similarly rewards the active, intrusive, curious reader. Finding ourselves projected into the cursor (or reduced to it), we probe objects until they yield stories" (ibid.). The notion of interactivity and the fact that many stories can be intermingled because of the multi-dimensionality seem to be the major contributions of this genre. "Storyworlds, in other words, take the active, intrusive critic as the paradigm of the ideal reader" (ibid.: 210).

But what could be the contribution of interactivity to the concept of a narrative text? Does it only add a playful element or does it also enhance the reader's participation in the story? Neither of both is the most significant contribution, I believe. Coover was right in saying that an idea space is automatically something of the past, an ancient ruin in which the reader is allowed to roam. This is in direct contradiction with the immersiveness of the playful element. The interaction does make the confrontation of the reader with the text more direct, but this is nonetheless not the experience of a computer game. It is rather that of exploring a bundle of love-letters found in an old cupboard in the dusty loft of your grandparents. It is not the text that differs so much; it is the environment, the movement from one place and one perspective to another. This text type would, for example, permit a very direct and more adequate description of history. A primitive instance of this is Arnold Dreyblatt's Who's Who in Central and East Europe in 1933, a typical 'who is who' reference book that represents not just the people and their descriptions, but also the relations between them.

Landow claims "in the storyworld and non-combative adventure game, reader-viewers assume the position of protagonist and their reward comes in the form of experience, not as a reward one might attain" (ibid.). Again, direct affinities with the concept of the game are denied. Playfulness is part of the experience, but it does not cover all of it. There are no high-scores, no speed contests, not even any riddles. A storyworld is still a story, a narrative, a message that is to be conveyed consisting in a number of causally related events. The search is only secondary. On the other hand, in a game we also find a story, but there interactivity is the most important element--the contest between the computer and the maker on one side and the player on the other.

How about the climax of the story? If the reader is permitted to roam around freely through the web, can there still be a climax to a certain scene or development? And if there is a climax, does the reader experience it as such? Landow claims that this form of hypertext narrative does not so much do away with climaxes as emphasize multiple ones. (ibid.: 210). By decentring the structure and the story and allowing many storylines to intermingle, several climaxes can be included in the story. Moreover, as many of these storylines intersect, some climaxes could also be witnessed and described from more than one focal point. We should add, however, that the climaxes are those of a story, a history of the past, and not those of a game or a riddle where the reader or player herself is the witness and main protagonist.

Since 253 is such an extreme example of a structured idea space, we will now look at another more moderately structured example.

The storyworld format is, in my opinion, the structure that holds most opportunities to base a true hypertext poetics on. It is the only structure permitting a particular sort of narrativity, which cannot be represented on paper with anything near the power it has in electronic hypertext. It allows combining the advantages of hypertext like decentring, collaboration, interactivity etc., with those of rigid structuring and full-text searches. Thus, it will probably benefit from developments and improvements in both fields in the near future, especially from the cognitive research that is being carried out on the subject. Finally, by means of hourglass or nested funnels between different idea spaces, a certain degree of linearity could be brought into the picture. This would allow a narrative to develop in different phases or to divide it into chronological periods thus making it into an even more powerful medium to write and read fiction.

In this section we will be dealing with what Landow has described as non-linear text. From a purely formal point of view, the difference between a hypertext and a non-linear text is the type of linking of the different text-blocks. Nelson defined hypertext as non-sequential writing, but he still saw it as writing and not so much as sophisticated programming. Hypertext is thus generally seen as fragmented text of which the different blocks are linked by connections which merely carry the reader from one node to another. Non-linear texts are structured on a different level, by more complicated devices like variables and procedures which allow the parts of the narrative to exchange knowledge and information.

But we could point to other typical qualities, which are not just situated on the hypertext syntactic level, but also influence the level of narrative syntax. Non-linear narratives allow for open narrativity, i.e. stories of which the ending is not predetermined, more than one ending is possible (just like in tree-branching hyperfiction structures). Unlike tree fictions, these non-linear texts can allow for open exploration--also of an open narrative--and even for open options. The first observation indicates that the reader is free to move through a developing narrative universe. Tree-branching always limits the choices to a certain number, and does not generally encourage retracing one's steps. Open options, on the other hand, means that--again unlike tree-fiction--non-linear structures can allow for a non-fixed ending. By interacting variables and procedures, the computer can calculate an ending according to the player/reader's behaviour.

Finally, one more distinguishing factor is the fact that for a non-linear text, it is possible that the narrative is developing real-time. In other words, the text is shaping itself while the reader is exploring the narrative universe, it is being written while it is being read and the reader herself can be one of the main characters. Unlike hyperfiction, where the story is revealed through interactivity, in non-linear text it is created by it. Moreover, the level of interactivity within a non-linear text can be of a higher plain than that of a hyperfiction. Just think of computerised characters with which the reader can attempt to communicate. And even other users can be interacted with within the story itself. For hyperfiction, interaction between readers could only take place on a meta-level.

In this section we will be dealing with two particular instances of non-linear structuring because of their relatedness to hypertext. The first structure is the (1) Multiple User Dimension (MUD), which we will not discuss in detail since such a discussion is far beyond the scope of this dissertation. We will rather give a brief description and point to its affinities with and differences from hypertext fiction. Second, we will briefly look at (2) Random elements, because of their potential usability in hyperfiction.

The acronym MUD stands for 'Multiple User Dimension,' but is also often called 'Multiple User Dungeon,' because the early MUDs were typically situated in a dungeon in a fantasy world based on that described as the dwarves' caves in Tolkien's Lord of the Rings. Nowadays, many different types of these constructs exist all over the Internet, "including variants called MOO (MUD-Object-Oriented), MUSE (Multiple-User Simulated Environment) and MUSH (where the "H" is for "Hallucination")" (Moulthrop 1995). MUD is used as the more general hyperonym, however, so we will refer to the whole group by this term.

Moulthrop describes a MUD as follows: " Roughly speaking, these creations grow out of the old Adventure game: they are virtual spaces constructed within computer memory, having the same metaphoric spatiality as hypertexts. MUD users move through the space by issuing commands. They may also manipulate objects and (most importantly) conduct transactions with other users" (ibid.). Many of these qualities have already been referred to in the discussion of non-linear structuring in general. Here, they are established concretely into a virtual space where the user is allowed to move around, interact, develop her character etc. However, whether the "metaphoric spatiality" is comparable to that of hypertexts only depends on how detailed the analysis. If you take the level of interaction into account, you cannot but admit that the metaphors of place in a MUD are much more refined than those employable in hypertext, which only allows for binary yes/no structuring.

In a MUD, a virtual world is mostly created by many writers/programmers at the same time. The world consists very often of a building or built environment with a specialised area devoted to each activity. These authors, together with other users, may then occupy the created rooms and leave behind textual traces for themselves or for others. Very often these MUDs are role-plays. Each user adopts a certain persona consisting of a name and a set of characteristics. She is then free to travel through the virtual space and interact with it. If she types a certain phrase with a certain command (e.g. quotation marks), that phrase is broadcast on the system as her utterance: X says "utterance." This utterance then becomes part of the virtual play that is being composed by all the users together. Each user becomes a character realised in the text and her environment, including the other actors, react to her actions (description based on Bolter 1996).

An interesting question in this regard is whether a MUD wizard, an author of the MUD who has greater power over the text than ordinary users, is very different from an author of an exploratory hypertext or not. Moulthrop argues that this is not the case, that both are very similar. "Both exert control over others' movements through a virtual or symbolic space. Both exploit a power gradient within the textual construct. [And b]oth represent a response to Coover's dilemma, the necessity to limit the elliptical spread of networked discourse even as one struggles against the monology of traditional writing" (Moulthrop 1995). In more sophisticated MUDs there is also the fact that the wizards choose the narrative environment and implement dangers and computer characters. Thus, they also have a great influence upon how the story will develop just like the author of a hyperfiction.

There are also differences however! First of all, there may be several wizards in a MUD, but in a distributed, constructive hypertext this is equally the case. Second, a wizard has even less control over what actually happens within the environment created by her than a hypertext author does. One important consequence of these two observations is that the multiplication of authorship may lead to conflicts between wizards. "One wizard of my acquaintance discovered that another programmer had begun to add rooms to "his" MUD, changing the nature of social interactions there. In response he created a self-replicating electronic object named kudzu, which quickly filled all the new rooms--and unfortunately the old ones as well. The MUD in question became extinct (ibid.).

"[W]hen a live-action simulation ends, the participants hold a wrap-up session in which the god of the machine—the game master or controller of the simulation--describes what has happened and solicits experiences from all the players" (Murray 1997: 180). This is remarkable in that only then the user receives an explanation of all the events occurring around her during the actual performance. Now the players "have the opportunity to see how their individual parts fit into the overall story and to understand the many processes that make up the micro-world of the simulation" (Murray 1997: 180).

In this way, improvements may be suggested and tried out during the next simulation. Since the final format for these kinds of story-environments has not yet been established, evaluations of this kind are still necessary. "Little by little we are discovering the conventions of participation that will constitute the fourth wall of this virtual theater, the expressive gestures that will deepen and preserve the enchantment of immersion" (ibid.: 125). Ironically, an account of the events occurring in a MUD environment is always recorded as linear history or as a hypertext. This may be an indication for the fact that the latter are more basic or fundamental or, at least, fundamentally different from non-linear text.

The parallels and differences between MUDs and hypertext fictions have been documented in more than one way. Alongside the formal differences, which we dealt with in the introduction to non-linear structuring, we should add Moulthrop's comparison of the interactivity that takes place in the two different environments. He believes that, in many ways, both types deliver the same kind of textual experience. "Any engagement with a MUD involves some level of interactive writing, as the user describes actions and receives passages of prose from the program in reply" (Moulthrop 1995). MUDs, however, add to this the possibility for the users "to create new spaces, objects, and even simulated persons called "NPCs" or "non-player characters" (ibid.). This creative franchise is, according to Moulthrop, a significant difference from hypertexts like Joyce's Afternoon etc.

Bolter places the opposition in another light. To him the main difference is that hyperfiction employs techniques of distancing and self-reference. "These fictions are interactive verbal texts, in which the reader follows electronic links from screen to screen and so constructs the text in the act of reading; the reader is repeatedly made aware of the artificial character of reading and writing fiction" (Bolter 1996). Although this is not always the case to such a degree as it is in the examples he quotes, Bolter has a point, when he claims that MUDs function more simply. "They embody the naive assumption of perspicacity, the idea that to read a narrative text is to look onto or enter into another world. A MUD is an example of ekphrasis; what is unusual is the collaborative character of the ekphrasis" (Bolter 1996).

Bolter bases his claim on his theory of the illusion of the natural sign (cf. also 2.2.3.). "[B]ecause MUD users accept this illusion, MUDs are stories, but no novels. Most MUDs do not set up an oscillation between rhetorical awareness and forgetfulness. They do not ask their users to look at the text, but only to look through it" (ibid.). In other words, what Bolter believes is that, in fact, the words are only in the way, that what the user really wants is to have an experience that is as close to reality as possible, but not reality itself. "So, like email and newsgroups, MUDs seem destined to become video experiences, as soon as Internet technology can support the change. MUDs will then be multi-user, networked virtual realities" (ibid.). Text, especially in the sense of descriptive text (ekphrasis), will then eventually be abandoned.

The MUD has often been named as one of the most promising and valuable frameworks to base a new, digital textuality on, together with Usenet newsgroups and Internet Relay Chat lines. Moreover, evidence has proved that also hypertext carries great potentiality, especially when we look at the growth the WWW has experienced in recent years. A very common complaint, however, is that hypertext is still too rigidly uni-directional from author to reader. Many theorists argue for a medium closer to Joyce's constructive ideal, instead of the current day exploratory texts. Moulthrop puts it in the following way: "If hypertext and other forms of electronic expression hold out any difference, it would seem to lie with constructive ventures, not such traditional offerings as electronic novels and monographs. The native country of hypertext must be a stranger place than anything we have yet imagined" (Moulthrop 1995).

By taking a closer look at the MUD, we have introduced an entirely non-linear format, outside what we regard as hypertext. However, there are also non-linear or random elements that may form a useful addition to the traditional hypertext tools. They are not hypertextual in the strictest sense, but they can be useful in that they could enhance the suspense created by the hyperfiction. "Randomness or chance can in its own right be an important added element to storytelling within several of the story shapes" (Phelps 1998). This is a field that, up to now, has hardly been explored in hypertext theory, probably because many regard it as outside the realm of simple hypertext writing and leave it to programmers. Nevertheless we will attempt to shed some light on the subject.

Phelps describes the use of random elements in other digital media as follows: "Randomness is frequently used to introduce a range of secondary characters or challenges within particular sections of a story. Which of these characters or challenges, as well as when and where they will be introduced, is left to the computer to randomly select" (ibid.). What we will attempt is to provide a brief outline of what the possibilities could be. Phelps notes the following addition potentially offered by randomness. "In stories where the audience moves around as the central character, randomness brings the pleasure of surprise and a means of developing that character" (ibid.).

She also points out that in other digital media like games or MUDs, nested funnel and tree-branching shapes are the structures in which random elements are most commonly found. "While you are choosing your course through a section of the nested funnels, occasionally the computer will select a random moment to insert one or more story events necessary to completing that section. Rarely is the randomness found in whether or not an event will take place, but rather when and where" (ibid.) Randomness should refer to the fact that the position and time of the event are random, that this is to some degree left to the computer. This does not mean that the implementation of the events should be haphazard or hit-or-miss. "The encounters themselves need an element of chance, but those chances must be influenceable with outcomes that have a real impact on the overall story in order to be satisfying aspects of the narrative (ibid.).

Now, we will briefly discuss instances of random structuring devices, which Phelps calls: (1) random cascading and (2) random shuffle.

(1) The first, random cascading, is a special type of tree-branching. "As episodes of a story proceed, at any point amongst those episodes a random event (…) can be inserted. How the audience chooses to react to that event will colour subsequent episodes" (Phelps 1998). In other words, there is one main storyline or thread, which is the trajectory primarily followed by the reader. Alongside this, however, there is a feature which allows the computer to attach a branch more or less at random to one of the nodes of the main thread, thus turning the structure into a tree-fiction.

E.g., imagine a 'Bildungsroman,' which follows the main protagonist from her birth to her death. The computer could be programmed to permit the character to win the lottery at one point in the story. This inevitably changes her whole life. In the structure, winning the lottery would entail that the reader is pulled into a branch outside the main thread. Note that this element pushes the narrative in the direction of gaming. By inserting random cascading, the user is urged to try again and again, which points to the playful element and the contest between narrative and user. However, According to Phelps "[t]his sort of shape closely mimics our experience of life. Though Grandma's death is not unexpected, it's apparently random timing can greatly effect what we do and how we feel" (ibid.).

(2) The second random element, which Phelps calls random shuffle, is a special instance of multilinear shape. "Much like the children's mechanical books, which are divided into three flippable sections whereby beginning middle and end can be randomly selected, it would be a simple matter to get a computer to perform these selections" (ibid.). The advantage of this structure is that many parts of the story can be recycled in other phases of the story. Moreover, with a limited number of nodes, the computer can generate a high number of different story sequences. From the nine nodes illustrated below, the computer can already generate no less than twenty-seven different combinations. One restriction is that for the sake of causality the narratives have to be closely parallel and that the different elements cannot go too much into detail.

One example path of this scheme is A-2-i. The result of the selection procedure is then: "It is drizzling outside. A neighbour is watching her carefully. When she reaches the restaurant, she notices that she is the first customer." Again, this type of non-linear insertion may make the hypertext much more game-like, but for recurrent trivial scenes such as the daily walk to the news-stand of one of the characters, this could be a satisfying structural solution.

Many more possible types of random elements can be found and they could be applied in many different ways. Keep (1995) for example, proposes as a possible first page for a rhizome a random choice of a node from the web, performed by the computer. In other words, since a rhizome is supposed to have neither a beginning nor an ending, a solution needs to be found as to where the reader should start. One possibility is then to let the computer cast her into the web randomly. But surely there are other possible applications for random elements like random linking etc. which are still very little documented or even experimented with. Up to now, these devices have hardly been used in hyperfiction, but in the genre of computer games they have already proved valuable. More research is therefore required in this field!

Let us recapitulate. The point of departure of this discussion on hypertext fiction was the postulation of a migration from paper to electronic text. By this we mean that it is very probable that at least a significant part, if not the major part, of textual information will become electronic. This entails a change of perspective on the matter of hypertext and hyperfiction. We should not try to develop a new medium, we should try to explore the full potential of a medium that will very likely become dominant in the future. Moreover, we have seen that every text is to a certain extent hypertextual. Hypertext already existed in print, but within the electronic medium its power increases dramatically. Every electronic text is automatically more hypertextual, since the reader can have the computer create links on demand by electronic text-search. Virtuality and hypertextuality seem to be intrinsically related. We can therefore conclude that hypertext may very well play an important role in a future electronic textuality.

Electronic textuality seems to change the prevailing conceptions of text in two important ways. First, there is the fact that a renegotiation between the symbol and the icon is taking place. Since the practical and economic advantage of symbolic text over iconic graphics is diminishing, the latter seem to be winning back a lot of terrain that was lost in print culture. The World Wide Web is becoming more and more crowded with graphical representations and moving images of all kinds. Second, the link allows textual connections to become virtual. By attaching a link to a word we get the impression that the text is becoming three-dimensional. The reader can read not only through, but also into the text. This implicates a new textual organisation, which we have situated on a new hypertext syntactic level.

The fact that hypertext is automatically networked text, and that the ideas in the human brain seem to be structured in a similar way has reawakened the cyborg dream. Could we not approximate the network in our brain and that of the text in such a way that more direct communication between both becomes possible? However, we have seen that such a claim is only partly true. Since hypertext is automatically mediated through language, be it symbolic or iconic, the linearity of the medium can hardly be surpassed on a surface level. We cannot read a network directly as such, we have to explore it sequentially. The question of how hypertext could contribute to textual performance thus shifts from 'how we read' to 'what we read.' We cannot take in the network directly; we can only take in text. What is about to change is how we select the text we desire to read. If hypertext is to make man more of a cyborg, it will happen in very much the same way as paper text has changed our idea of remembering; it will allow us to create a more powerful and accessible artificial memory board.

For hyperfiction, we could more or less draw the same conclusion. Hypertext will probably not change narrative as a metacode, nor will it affect the way in which we process it in our mind; it is the manner of exploration that can be enhanced. In the traditional hypertext literature, this issue is very often treated in an ambiguous way. On the one hand, narrative is said to be "a metacode, a human universal on the basis of which transcultural messages about the nature of a shared reality can be transmitted" (White quoted in Landow 1997: 183). On the other, it is claimed that hyperfiction will finally permit the reader to create her own narrative from the materials presented by the author. What seems to be confused here is narrative and poetry. When you present the reader with loosely connected materials and let her create her own story, this is not communication of meaning or narrative, but indulgence in a poetic dialogue.

However, apart from the way in which hypertext is used in the above-described hyperfictions--i.e. linking by association--there is a second possibility which is more appropriate to convey a narrative. On the micro-structural level, the main contribution of hypertext to textuality is its metaphorical third dimension. Whereas for the poetic reader this dimension is used for associative linking, for the reader of a narrative this third dimension has no pre-established function. Therefore, it is the author's task to make linking and the narrative space meaningful. When a reader does not know why she pursues a certain link, this link has no function and hence does not contribute in any way to the meaning of the text. When she does not have an idea of where she is going and why she is going there, there is no inherent difference from continuing to read the same node. Reading a hyperfiction becomes a mere random clicking exercise, unless the author gives the reader an idea of the 'hyperness' of the hypertext, of the structure behind the different nodes. It is precisely the enhancement of this structure that can make hyperfiction into a powerful storyteller.

In order to make the third dimension meaningful for narrative, there are two important facets of the hyperfiction that have to be taken into account. First of all, the notion of 'words that yield' has to be elaborated. Landow proposes a rhetoric or stylistics of departure and arrival. In the same way as a full stop and a capital letter signal the beginning of the next sentence, links should inform the reader of their presence and their aim, and suggest a destination. In which way anchors should contain this information has not yet been established. One possibility is to create a grammatical system consisting of different devices like colours, font, graphical elements, word-choice… indicating certain 'grammatical' relations between nodes. Perhaps, some of these devices could even be combined, thus acquiring a new meaning. Empirical research could aid in finding the best combinations of tools. Once the reader has learned this 'textual grammar', this will allow her to make meaningful choices, and facilitate reading and searching through large hypertexts.

Second, to exploit the 'hyperness' of a hyperfiction fully, a surface level rhetoric alone does not suffice. What we need is a powerful new metaphor, not to replace altogether, but to enhance the age-old 'linear' metaphor of the flow of time. Traditionally, fictions deal with a certain period of time starting with point A and ending at point B. This structure is so normal to us that we do not even have a name for it. Anomalies like pointing forward (prolepsis), and pointing backward (flashback, analepsis), on the other hand, do have a specific name. If the simple flow of time is only a metaphor due to the linearity of the medium, then what could be the predominant metaphor in a spatial medium like hypertext? I believe that hypertext can enhance traditional textuality by adding a geographical dimension to it. By using the third dimension laid bare by the electronic medium, it is possible to give the reader not just a notion of where she is in time, but also in space. By providing a clear overview and an easy-to-use navigation principle, such a complex structure becomes possible to create and control.

By making a classification of hyperfiction structures we have tried to lay a basis that could enable us to control the geographical dimension on the hypertext syntactic level. We have seen that the different structures each become suggestive of certain kinds of stories, and that the real innovation produced by hypertext is the idea space. This is not unexpectedly the format that makes use of the geographical dimension most effectively. A grammar of the hypertext syntactic level, alongside the rhetorical grammar described above, seems to be desirable. On the narrative syntactic level, different narratological devices can be attached to these basic structures. One frequently occurring example is that of point of view, but many more are possible (focalisation, narration, time-lapses…). While moving around on the hypertext syntactic level, the reader simultaneously explores the different points of view on the narrative syntactic level, without losing the temporal dimension. Thus, hypertext's major contribution to the representation of narrative seems to be the possibility to add to the traditional diachronic view of events, a synchronic dimension permitting to develop scenes more profoundly at one particular point in time.

Avalon (1998). http://www.avalon-rpg.com/.

Balasubramanian, V. (1995). State of the Art Review on Hypermedia Issues And Applications. Newark: Rutgers University: http://www.isg.sfu.ca/~duchier/misc/hypertext_review/index.html.

Barger, Jorn (1998). Timeline of Hypertext History. http://www.mcs.net/~jorn/html/net/timeline.html.

Becker, Howard S. (1996). A New Art Form: Hypertext Fiction.

http://weber.u.washington.edu/~hbecker/lisbon.html. http://www.guilford.edu/web_class_96/ppages/john/tristram/new_art.htm.

Blustein, J. (1998). alt.hypertext Frequently Asked Questions. http://www.csd.uwo.ca/~jamie/hypertext-faq.html.

Bolter, Jay David (1996). Degrees of Freedom. http://www.lcc.gatech.edu/~bolter/degrees.html.

Breitzke, Edwina (1997). Henro. http://www.glasswings.com.au/GlassWings/modern/henro/.

Burne, Philippa J. (1996). 24 hours with someone you know... Modern Adventure: http://www.glasswings.com.au/modern/24hours/.

Bush, Vannevar (1945). As We May Think. http://www.theatlantic.com/unbound/flashbks/computer/bushf.htm.

Cooper, Douglas (1994). Delirium. http://www.pathfinder.com/@@*MhFsgcAxqDcZvWs/twep/features/delirium/.

Coover, Robert (1992). The End of Books, New York Times, http://www.nytimes.com/.

Coover, Robert (1993). Hyperfiction: Novels for the Computer. New York Times: http://www.nytimes.com.

Coverley, M.D (1996). The Book of Hours: 4:00 a.m. The Lacemaker. http://gnv.fdt.net/~christys/Coverley/elys_1.html.

De Bra, Paul (1998). Hypermedia structures and systems. http://www.win.tue.nl/win/cs/is/debra/cursus/.

Dreyblatt, Arnold (1933). Who's Who in Central and East Europe in 1933. http://www.uni-lueneburg.de/memory/whoswho/Titlepage.html.

Fauth, Jurgen (1995). Poles in your Face: the Promises and Pitfalls of Hyperfiction. http://orca.st.usm.edu/mrw/06sept/06-jurge.html.

FEED (1995). Digital Thinking, Page Versus Pixel, Part One of FEED's Dialog on Electronic Text. http://www.feedmag.com/95.05dialog1.html.

Firstenberg, Allen S. (1994). Addventure. http://www.addventure.com/addventure/.

Greenheart, Adrienne (1996). Six sex scenes: a novella in hypertext by adrienne greenheart. Alt-X: http://www.altx.com/hyperx/sss/index.htm.

Grossman, Wendy (1997). Geoff Ryman's Web novel, "253," peers into the heads of a Tube train-ful of characters as they hurtle toward an uncertain fate. http://www.salonmagazine.com/march97/21st/london970320.html.

Guyer, Carolyn (1995). Written on the Web: Carolyn Guyer examines the state of hypertext fiction. http://www.feedmag.com/95.09guyer/95.09guyer.html.

Harris, Paul (1997). Hybrids. Alt-X: http://www.altx.com/ebr/EBR4/HARRIS2.HTM.

Johnson, Jeffrey & Oliva, Maurizio (1995). Internet Textuality: Toward Interactive Multilinear Narrative. http://italia.hum.utah.edu/~maurizio/pmc/.

Joyce, Michael (1992). Hypertext narrative. http://noel.pd.org/topos/perforations/perf3/hypertext_narrative.html.

Joyce, Michael (1995). Of Two Minds: Hypertext Pedagogy and Poetics; Introduction: The Comfort of Knowing We Are Not Lost. http://www.press.umich.edu/bookhome/joyce/intro.html.

Keep, Christopher & McLaughlin, Tim & robin (1995). The Electronic Labyrinth, http://web.uvic.ca/~ckeep/elab.html.

Kendall, Robert (1996). Words and Mirrors. Eastgate: http://www.eastgate.com/hypertext/kendall/Mirrors.html.

Lewis, Carroll (1997). Taking readers through the looking glass. Electronic mail&guardian: http://www.mg.co.za/mg/books/dec97/1dec-internet.html.

Mack, E. Stephen (1995). Cynthia. http://metalab.unc.edu/mal/MO/cynthia/pages/tomas.html.

McGann, Jerome (1995). The Rationale of HyperText. http://jefferson.village.virginia.edu/public/jjm2f/rationale.html.

Miller, Laura (1998). Bookend; www.claptrap.com. London Times: http://archives.nytimes.com/archives/.

Mirapaul, Matthew (1997). A Vote of Confidence for Hypertext Fiction. New York Times: http://www.nytimes.com.

Mirapaul, Matthew (1997). Hypertext Fiction on the Web: Unbound From Convention. New York Times: http://www.nytimes.com/library/cyber/mirapaul/062697mirapaul.html.

Moulthrop, Stuart (1991). You say you want a revolution?. http://xroads.virginia.edu/~DRBR/moulthro.txt.

Moulthrop, Stuart (1995). Traveling in the Breakdown Lane: A Principle of Resistance for Hypertext. http://www.ubalt.edu/www/ygcla/sam/essays/pre_breakdown.html.

Noam, Eli M. (1995). Electronics and the Dim Future of the University. http://www.columbia.edu/dlc/wp/citi/citinoam14.html.

Phelps, Katherine (1996). The Choicest Bits. http://www.cinemedia.net/GlassWings/modern/choice.htm.

Phelps, Katherine (1998). Story Shapes for Digital Media. http://www.cinemedia.net/GlassWings/modern/shapes/.

Pipsqueak Productions (1996-99). The Company Therapist: Betrayal, Office Politics, Sex...The Doctor is In. http://www.thetherapist.com/.

Rees, Gareth (1996). Tree fiction on the World Wide Web. http://www.cl.cam.ac.uk/users/gdr11/tree-fiction.html.

Rettberg, Scott (1998). The Unknown. http://www.soa.uc.edu/user/unknown/.

Ryman, Geoff (1997). 253: a novel for the Internet about London Underground in seven cars and a crash. http://www.ryman-novel.com/home.htm.

Shumate, Michael (1996). The Art World of Hypertext Fiction. http://www.duke.edu/~mshumate/hyperfic.html.

Shumate, Michael (1998). Hyperizons: Theory and Criticism of Hypertext Fiction. http://www.duke.edu/~mshumate/hyperfic.html.

Sorrells, Walter (1995). The Heist: a Hypertext Story by Walter Sorrells. http://www.mindspring.com/~walter/1.html.

Stuebe, Alison (1996). I Link, Therefore I Am (Seeking Hypertext's Meaning). New York Times: http://www.nytimes.com.

Swiss, Thomas (1998). Reviewing the Reviewers of Literary Hypertexts. http://www.altx.com/ebr/ebr8/8swiss.htm.

Tolva, John (1995). The Heresy of Hypertext: Fear and Anxiety in the Late Age of Print. http://www.mindspring.com/~jntolva/heresy.html.

Traenkner, Nick (1999). Hypertext Models. Kent State University: http://www.kentinfoworks.com/people/nick/models/.

van Driel, Hans (1998). Gedigitaliseerde Informatie over de Digitale Samenleving 1.2. Tilburg University: http://cwis.kub.nl/~fdl/general/people/drielhv/gids/98-2/index.htm.

Zachry, Mark (1997). Purpose and Play in Hypertext. http://www.public.iastate.edu/~mzachry/hypertext/.