Eastgate Systems      Serious Hypertext
Hypertext With Characters

Mark Bernstein
Eastgate Systems, Inc.
134 Main Street
Watertown MA 02712 USA
email: bernstein@eastgate.com

Originally presented at the 1995 International Workshop On Hypermedia Design, Montpellier, France.


Most current hypertexts appear to the reader as a single entity, a written work that speaks with a single voice and presents a single viewpoint. Large and challenging hypertexts, whether technical, scholarly, or fictional, might benefit by introducing a dramatic multiplicity of voices and perspectives that engage the reader and each other. Characters are not merely names or pictures; to be credible and coherent, each character must be independent, persistent, and intentional. This paper proposes a framework for creating and discussing dramatic hypertexts in which separate characters participate fully and directly. The most interesting issues arise when characters are permitted to respond to other characters and to urge the reader to follow different trajectories through the hypertext.

1. Many Paths, One Voice

In most current hypertexts, the hypertext speaks with a single voice, in what Jones and Spiro called the "imagined conversation" between reader and creator [Jones 92]. In some hypertexts, the authorial voice is explicit: when reading the hypertextual version of Jay Bolter's Writing Space [Bolter 91], for example, the reader knows that the ideas on the screen are Bolter's and that the process of exploring the hypertext is, in a sense, a dialogue with the writer. In David Kolb's exploration of hypertext argumentation [Kolb 95] or George Landow's tour of his hypertext writing workshop [Landow 95], a unified authorial voice and organization mediate all the contradictory views and contrasting approaches the work presents.

In other hypertexts, like J. Yellowlees Douglas's I Have Said Nothing [Douglas 94], the dialogue occurs between the reader and a fictional character. We know, as readers, that the narrator who begins:

Remember Sherry? My brother Luke's piece: the one who got into that godawful tangle with him at Paycheck's in Hamtramck.
is not actually Professor Douglas, but while we read the hypertext, our imaginary dialogue with the imagined character is exactly as real as the dialogue with Bolter or Kolb. Even when a hypertext is the result of a collaborative scholarly project (e.g. Perseus [Crane 91]) or the collective labor of several college classes (e.g. The Dickens Web [Landow 92]), the work speaks with a single voice.

This unity of the hypertext's voice is unsurprising, since familiar print forms share it. The essay, after all, is closely tied to the lecture, just as print fiction descends from storytelling and print poetry descends from recitation and song. In drama, on the other hand, we perceive each character as a separate voice; we may know that all the characters are the creation of a single writer, yet during a performance we naturally regard each character as independent and distinct. In a scientific workshop, a television talk show, or a political debate, we enjoy and learn from the drama of distinct voices, with the attendant possibilities of education, disagreement, conflict and resolution.

How might we create hypertexts that speak with the voices of many credible and engaging characters? The multimedia answer has been straightforward: emulate the immersive qualities of performance by including video, animation, and sound. Unfortunately, hot media necessarily compete with the hypertextual experience: performance seeks to remove us from our personal experience, where hypertexts invite us to examine and to act upon our internal understandings. As an alternative to the intrinsic tension between links and pizzaz, I propose hypertext rich in character. Here we may find opportunities for dramatic engagement.

2. Hypertext Characters

Characters are independent, persistent, and intentional.

An independent character is a distinct and recognizable abstract unit, represented to the reader through a distinct and recognizable interface. Readers should be able to distinguish Hamlet from Horatio. One easy way to establish independence in a hypertext is to have each character's text (and other media) appear in its own pane.

A persistent character is perceived to possess a continuous existence, even when the character is not the center of attention. Though Horatio is not on stage, we assume that he continues to exist. Moreover, persistence implies the ability to retain state; if something happens in the hypertext that affects Horatio, the effect should persist even when Horatio leaves the stage. Tools like a Search dialog, on the other hand, may be independent but usually are not persistent.

An intentional character is perceived to possess intrinsic behaviors which seems to serve some purpose or goal. The goal may be simple or complex, and the reader need not know what it is. Indeed, the goal may not even be internally consistent; people, after all, are hard to understand. Internal rhetorical and stylistic consistency make characters seem intentional and therefore real.

We thus envision a hypertext in which different characters appear as distinct hypertextual units. Characters might be implemented as separate hypertexts that interact with each other, or as distinct types or aggregates within a single work.

Figure 1. Characters are made convincing by independence, persistence, and intentionality. Whether they are portrayed as people, organic entities, or texts is an implementation detail.

3. Conversations On Serious Matters

The utility of characters in fiction is clear, but their relevance to technical and scholarly writing may seem far-fetched. Let us consider, then, how writing with characters might strengthen conventional technical project - writing a computer language tutorial for a professional audience. A central pane in this scenario will present a concise, expository treatment of key language features, organized as a coherent and pedagogically directed hypertext. We'll call this character the Instructor. Peripheral characters, in their primary roles, comment on aspects of the central pane.
Let us describe one snapshot of this multi-character tutorial in action. The topic at hand is the C++ auto-increment operator.
The operator ++ adds one to its argument.
Advisor for C Programmers:
Avoid this operator in compound expressions. C++ does not define the sequence in which arithmetic expressions are evaluated, so x = 1; y = ++x + 2; will leave x with the value 2, but y may be either 3 or 4.
Advisor for LISP Programmers:
contemptuous silence.
Autoincrement and autodecrement operators offer useful hints to the compiler. Some processors offer special instructions for rapid autoincrement.
The autoincrement instruction was useful in the early days of C, especially since the original motivation for C was to demonstrate that a high-level language could be a practical implementation tool for compilers and operating systems.
Familiar tools can be used to represent and model characters in various ways, e.g. through typed links [Halasz 87], node classes [Nanard 91], or HDM schema [Garzotto 91].

4. Implementation: Staying In Character

As in our example, characters might be represented by placing their texts in different portions of the screen. This is a convenient spatial metaphor, establishing a connection between location and role. Because the notional voices occupy disjoint spaces, readers perceive them as distinct. Because the visible manifestation of each character - its window - remains constantly in view, it is easy to establish temporal coherence.

Dividing the screen among many characters, however, may leave each character with too little space. In this simple implementation, a character occupies the same amount of screen space whether or not it has anything to say. The placement of each character on the screen, moreover, necessarily reflects a judgement of the centrality each character, a judgement the author may be reluctant to make. Finally, the unchanging placement of characters is fundamentally undramatic; an arrangement that seems suitable for one topic may be unsuitable for another. In the worst case, large parts of the screen may be quiescent, while a spirited discussion occurs between two or three small windows scattered arbitrarily about the screen.

We can retain spatial relationships and improve the screen layout by specifying constraints rather than actual placements. When we follow a link, each character's representation might change size and position as needed to satisfy its constraints. The principle of enactment ([Laurel 91], cf. [Bernstein 93] ) suggests that the change be represented as a continuous transition rather than by redrawing the entire screen. The reader should see the movements of the characters, not merely their new positions. This concreteness of representation reinforces the independence and persistence of characters.

The logistics of specifying and maintaining these dynamic layouts may prove burdensome. In that case, we might employ other strategies for presenting character identify within a conventional screen management mechanism. We might, for example, include a portrait of the character within each of its nodes, or establish character through typographic convention, associating different characters with different textual appearances. This is, in fact, a common device in recent hypertext fiction. In Kathryn Cramer's In Small and Large Pieces, for example, one type face indicates the central narrative while a different face distinguishes a melange of nightmares and refracted images thrown off by the central story [Cramer 94].

Figure 2. Three characters discuss the usefulness of characters

5. Characters Responding to Characters

In the paradigm assumed thus far, characters respond to a central hypertext, creating an elaborate annotative halo. Of course, each character's texts will contain links, becoming interdependent hypertexts rather than mere annotation. More interesting - and more hypertextual - structures arise when characters respond not merely to the stimulus of a central hypertext, but also to each other.

Allowing characters to respond to other characters should certainly enliven the hypertext. It also adds considerably to our analytical burden in understanding the formal properties of the hypertext and predicting its behavior. A range of fascinating and pathological behaviors becomes possible. For example, consider a work in Roman history that includes (among other characters) a Marxist Historian and a Glossary. Among the words defined in the glossary is "plebiscitum," an enactment of the Roman popular assembly. The Marxist Historian responds to this definition, perhaps to illustrate a nascent sense of "class" in the ancient term. In so doing, the Marxist offers the Glossary a new opportunity to define "plebiscitum", to which the Marxist again responds. A succession of dense and vehement screens flies before our eyes.

Responsive characters can create a variety of oscillators, critical races, and deadlocks. The Trellis model [Stotts and Furuta 89] provides the analytical tools needed to model these phenomena, and could allow the system to warn the writer of potential pathological interactions.

We might simplify matters by permitting the reader to serve as a moderator. When a character has a response to the current situation, it signals to the reader that it has something to say and waits for permission to proceed. This simple solution, however, would have profound impact on the sorts of trajectories a reader would be likely take through the hypertext. A reader who thinks herself uninterested in some character might never permit that character to speak at all, an outcome that might be very contrary to the author's desires. Moreover, by giving only one character at a time a chance to change the hypertext's state, we promote depth-first exploration. Once a topic is advanced, the character to which she first responds invites the reader to pursue it, and other characters (including most especially characters who wish to suggest that this is the wrong topic to pursue) have less opportunity to get the reader's attention.

As a compromise, we might distinguish three link behaviors:

Eager Links
cause immediate transitions as soon as their preconditions are satisfied, and which should be written with care;
Plain Links
cause their character to ask for attention as soon as their preconditions are satisfied;
Timid Links
are identical to plain links unless the character is already seeking attention. In that case, plain links would change what the character wanted to say, while a timid link would merely add another option from which the reader could choose when giving the character her attention.

6. Advisors: suggesting What To Visit Next

Even where the reader plays a strong role in moderating the characters - that is, in hypertext composed predominantly of plain and timid links - characters might wish to influence the reader's choices. In our historical example, the Marxist historian might wish to express opinions not only on specific matters that arise, but also on the matters that should arise. Indeed, this sort of procedural or meta-guidance lies at the heart of pedagogy and drama. As early as 1987, Conklin noted that collaborative hypertext writers wanted explicit representational support for procedural meta-discussion [Conklin 88]. Many early efforts at introducing agents or characters into hypertexts (the guides of [Don 91], the apprentice of [Bernstein 91]) provided chiefly navigational assistance.

I suggest here a computational mechanism that will make it easy to create intentional characters that offer navigational advice. The conceptual model for the hypertext writer is very simple, but the range of behaviors that can be crafted is large and expressive.

In this approach, each character participates in a hidden game based on the reader's trajectory through the hypertext. The rules are as follows:

  1. Each character has a unique color. Characters seek to collect tokens of their own color, and ignore tokens of other colors.
  2. The author populates the hypertext with tokens of various colors and weights in selected nodes. Tokens are placed at nodes of particular interest to a the character.
  3. When the reader visits a node, all the tokens from that node are distributed to the corresponding characters.
  4. Characters offer navigational advice, seeking to maximize value of the tokens they will have collected at the conclusion of the reading. A simple min-max search to fixed depth should prove sufficient for the hypertext system to calculate each player's recommendation.
Readers, of course, don't need to know the rules or even the existence of the game; they perceive the game only through the suggestions of the characters. This simple framework appears capable of expressing a variety of behaviors. A topical guide, like [Don 91], can easily be constructed by scattering its tokens at places most relevant to its concerns. Conversely, negative token weights let us construct a character that tries to steer the conversation away from sensitive subjects, a behavior not discussed in the hypertext literature but which has clear dramatic possibilities. By placing tokens uniformly throughout the hypertext, the writer can create a character that seeks out heavily textured areas [Bernstein 92]. If we define a path through the hypertext [Zellweger 89] and place tokens along it, a character can serve as a guide to the path; if the reader departs from the path, the guide remains useful by planning and constantly updating a trajectory that will return to the planned itinerary. By placing a character's tokens at central navigational features (e.g. chapter outlines), we can design a character that steers the reader back toward a central hypertextual core and away from peripheral issues. In this way, characters can encapsulate both a rhetorical and a navigational strategy.

7. References

[Bernstein 91] Mark Bernstein, J. David Bolter, Michael Joyce, and Elli Mylonas, "Architectures for Volatile Hypertexts", Hypertext 91 Proceedings, ACM, Baltimore, 1991.

[Bernstein 92] Mark Bernstein, Michael Joyce, and David Levine, "Contours of Constructive Hypertext", ECHT 92 Proceedings, ACM, Baltimore, 1992. pp. 161-170

[Bernstein 93] Mark Bernstein, "Enactment in Information Farming", Proceedings of Hypertext 93, ACM, New York, pp.242-249

[Bolter 91] Jay David Bolter, Writing Space: The Computer, Hypertext, and the History of Writing, hypertext edition, hypertext for Macintosh computers, Lawrence Erlbaum and Associates, Hillsdale NJ. 1991

[Conklin 88] Conklin, J.; Begeman, M. L.; "gIBIS: A Hypertext Tool for Exploratory Policy Discussion", ACM Trans. Office Information Systems 6 (1988) 303-331

[Cramer 94] Kathryn Cramer, "In Small & Large Pieces", Eastgate Quarterly Review of Hypertext 1(3), hypertext for Macintosh and Windows, (1994)

[Crane 91] Gregory Crane et al., Perseus, hypertext for Macintosh computers, Yale University Press, 1991.

[Don 91] Abbe Don, Tim Oren, and Brenda Laurel, "Guide 3.0", CHI 91 Proceedings, ACM, New York, 1991. pp. 447-8

[Douglas 94] Jane Yellowlees Douglas, "I Have Said Nothing", hypertext for Macintosh and Windows, Eastgate Quarterly Review of Hypertext 1 (2) (1994)

[Furuta 89] Richard Furuta and P. David Stotts, "Programmable Browsing Semantics in Trellis", Proceedings Hypertext 89, ACM, Baltimore, 1989 pp. 27-42

[Garzotto 91] Franca Garzotto, Palo Paolini and Daniel Schwabe, "HDM- a Model for the Design of Hypertext Applications", Hypertext '89, ACM, Baltimore, 1991. pp. 313-328

[Halasz 87] Frank Halasz,Thomas P. Moran, Randall H. Trigg,; "NoteCards in a Nutshell", Proc. ACM CHI+GI 87 (Toronto, 5-9 April 1987) 45-52

[Jones 92] Robert Alun Jones and Rand Spiro, "Imagined Conversations: The Relevance of Hypertexts, Progmatism, and Cognitive Flexibility Theory to the Interpretation of Classic Texts in Intellectual History", ECHT 92: Proceedings of the Fourth ACM Conference on Hypertext, Association for Computing Machinery, New York, 1992. pp, 141-8

[Joyce 90] Michael Joyce, Afternoon, a story, hypertext document for Macintosh computers, Eastgate Systems, Watertown MA, 1990.

[Joyce 91] Michael Joyce, "Storyspace as a Hypertext System for Writers and Readers of Varying Ability". Hypertext 91 Proceedings, ACM, Baltimore, 1991. pp. 381-9.

[Kolb 95] David Kolb, Socrates in the Labyrinth: Hypertext, Philosophy, and Argumentation, hypertexts for Windows and Macintosh, Eastgate Systerms, Watertown MA, 1995.

[Landow 92] George P. Landow, ed., The Dickens Web, hypertext for Macintosh computers, Eastgate Systems, Watertown MA 1992.

[Landow 95] George P. Landow, Writing At The Edge, hypertext for Macintosh computers, Eastgate Systems, Watertown MA, 1995.

[Laurel 91] Brenda Laurel, Computers As Theatre, Addison-Wesley Publishing Company, New York.

[Moulthrop 91] Stuart Moulthrop, Victory Garden, hypertext for Macintosh and Windows, Eastgate Systems, Watertown MA, 1991.

[Nanard 91] Jocelyne Nanard and Marc Nanard, "Using Structured Types to Incorporate Knowledge in Hypertext", Hypertext 91 Proceedings, ACM, Baltimore, 1991.pp. 329-344

[Zellweger 89] Polle T. Zellweger, "Scripted Documents: A Hypermedia Path Mechanism", Hypertext 89 Proceedings (Baltimnore, ACM Press) 1989 pp. 1-14

Acknowledgements: The idea of a hypertext in which several independent and persistent characters comment upon the central action was suggested to me by Stuart Moulthrop, who was then working on a hypertext fiction of this sort. Professor Moulthrop, of course, is not responsible for the details of this proposal, and his work may bear little or no resemblance to the ideas described herein. Eric Cohen and Diane Greco made many helpful suggestions.

About the author:
Mark Bernstein

...what's new

Hypertext Tools


HypertextNow Tinderbox About Hypertext Poetry Nonfiction Fiction Storyspace Eastgate

Eastgate    Fiction    Nonfiction    Poetry    Hypertext    Storyspace    Tinderbox    HypertextNow    Order