Focusing only on the spatial or geometric relationships among nodes and links when building a hypertext is a little like concentrating only on the spatial layout of pipes and nozzles when designing a fountain. One must also take into account water pressure, nozzle apertures and angles, and the laws of trajectory. One must be able to think in terms of process and behavior. So too with hypertext. Any good hypertext writer develops a general sense of the processes to which the work will be subjected and the behaviors with which it will respond. Yet the writer will find few aids for sharpening this sense. I would therefore like to explore some approaches for more-systematic thinking about time in hypertext.

Let's begin with the big picture. If we compare different readings of numerous hypertexts, a certain overall similarity emerges in the way they unfold over time: Most readings consist of three stages that are recognizably distinct from one another, though there is rarely a sharp boundary between one stage and the next. The way that the combinatorial textual possibilities of a hypertext solidify into a reading is similar to the way that a game of chess materializes from all the possible moves on a chess board. The three stages of hypertext reading also bear similarities to the three stages of a chess game--opening, middle game, and endgame--so it seems appropriate to adapt this chess terminology for our purposes here.

The opening lasts as long as the reader's cumulative choices result in one of several essentially predefined node sequences. A hypertext may have numerous possible openings, many of them variant versions of one another, but their number will be finite and their configurations foreseeable. The first node of a hypertext presents the reader with a limited number of options that are usually invariable from one reading to the next. (Think of chess pieces all standing in their initial positions.) Then the reader may start down one of several alternative predetermined paths or a default path. Or she may arbitrarily follow different branches along a small hierarchical structure that yields a relatively small number of combinatorial possibilities. The length of the opening depends partly upon the work's design and partly upon the reader's actions. It can last for many nodes if, for example, the reader simply follows a long default path. Or it can end within a few nodes in a very heavily linked hypertext, especially if the reader deliberately avoids any predefined paths.

The opening is the only stage of reading over which authors can easily exert a high degree of control. Authors will usually make sure that all the reader's initial options make sense as beginning points. They will also usually try to ensure that any of the early branches off of these potential beginning nodes are satisfactory, which may involve defining starting paths. A work's author is likely to have tested out every possible textual configuration that may arise from the reader's choices during the opening.

The middle reading is underway when the reader is clearly in the thick of the hypertext. The text sequence produced during this stage in one reading is very unlikely to be reproduced in any other reading. This is the stage during which the reader may frequently encounter configurations that couldn't have been foreseen by the author. Most of the theoretical commentary about a hypertext's seemingly limitless possibilities apply only to this stage.

During the final stage--the end reading--the reader is aware of having read all or most of the text and is probably trying to decide whether or not to terminate the reading. Reading strategy usually changes drastically during this stage, just as chess strategy changes when most of the players' forces have been depleted. Most of the nodes encountered have already been visited, so the reader is likely to skim large quantities of previously read material in search of something new or in the hope of getting a final structural overview. Here the reader encounters diminishing returns for further reading as a result of what could be called the Law of Diminishing Non-Returns: The further into a hypertext one gets, the more recurring material one encounters until eventually there is nothing but recurring text.

This pattern of opening, middle reading, and end reading is pretty much hardwired into most hypertexts for better or worse. An author may appreciate the opportunity provided by the opening to carefully control the reader's initial contact with the work and ensure that the reading starts off on the right foot. If an author feels that any prepared opening goes against the grain of hypertextual freedom, this stage can be radically truncated by offering the reader access to as many different textual configurations as possible at the outset. One can probably never entirely eliminate the opening--that is, drop the reader immediately into the territory of nondeterminism--since the potential initial options are always limited by the total number of nodes in the piece.

The inevitability of the end reading can be more problematic. If this stage is drawn out excessively it can prove tedious for the reader, so ideally it should be no longer than necessary. [1] There's also a certain predictability to the process of old material gradually crowding out the new during this final reading stage. When the reader encounters it in work after work, it can come to seem like a structural formula. The author who chafes against predictability may wish occasionally to replace the end reading with a true ending or one of several alternative endings that force a termination of the reading before the Law of Diminishing Non-Returns makes its effects felt. The trade-off with a forced termination is that it may prevent the reader from reading all or even most of the text during a single reading.

Considering these three stages of reading demonstrates how time alters the perceived behavior of a hypertext and correspondingly the reading strategy applied to it. It would therefore make sense for a hypertext delivery system to function differently during these three stages. Such a system should be particularly sensitive to the end reading, which it could recognize by observing the frequency of recurrences and the proportion of unvisited material remaining. A reader could also be allowed to shift the system manually into End Reading Mode. During this stage the system could, for example, increase the reader's access to diverse areas of the hypertext--and therefore new material--by activating conditional links in each node. Different link colors could indicate not just visited or unvisited nodes--following links of a particular color could take one through one or more visited nodes to reach unvisited material.

Now let's look at hypertextual time from a different perspective. Let's return to our spatial metaphor of a navigable textual topography. If hypertext is a virtual landscape, we should consider the effect that time has upon topographical features. No matter how solid they may seem, they are inexorably altered through processes of erosion and accretion. This is true whether the landscape is one of rock and soil or of words. Textual topography is formed in the reader's mind, where much of the textual content seems to erode as it fades in memory. Yet at the same time, new significance can accrete to many remembered passages as further reading sheds new light on them. The relationships that comprise topographical contours in hypertext thus change gradually during the reading in often subtle ways.

This erosion and accretion occurs during the reading of linear texts as well, of course. The dynamics of the process are complicated in hypertext, however, by the inevitable recurrence of nodes in a reading. While someone reading a printed book may on occasion flip back to reread earlier passages, recurrence is an almost unavoidable component of hypertext. In a linear text, when details near the beginning fade completely in memory during the reading they are generally gone forever. In a hypertext, any such details may have the chance of being restored in the mind by reappearing later on.

The farther back in the reading the original appearance of a node, the more significant its recurrence is likely to be to the overall perception of the text. When recurrence reverses the effect of extreme "weathering" on a node, not only are forgotten elements restored. The large quantity of text intervening between two widely separated appearances of the same node also increases the likelihood of new meaning accreting to the passage upon its second appearance, giving the second reading a different impact than the first. Certain half-forgotten passages that seemed insignificant upon the first reading may now emerge as important, thanks to new background information. The reader may then think back upon other text that surrounded the original occurrence of the node and suddenly better understand that surrounding text as well. The prominence of certain features in the topography is thus altered. On the other hand, the reappearance of a node in close proximity to its original occurrence is much less likely to have real significance in the reader's mind.

The process of textual aging is therefore a significant structural factor in literary hypertext. A link back to text that was visited just a few nodes ago is qualitatively very different from a link to text that originally appeared far back in the reading. It therefore would make sense for a hypertext system to monitor aging. Aging information could help the reader make more-intelligent link choices and it could influence when conditional links are displayed in a particular node. When a choice must be made between two previously visited nodes, the one that has aged the most could be given preference.

Let's shift perspective once again and consider one more way of thinking about the temporal in hypertext--as rhythm. In its broadest sense, "rhythm" denotes patterns of change and recurrence, which are found in most temporal phenomena. In literature, rhythm manifests itself most obviously in metrical verse, but we also talk about rhythm on a larger scale in narrative or drama, referring to patterns in plot development, scene changes, recurrence of thematic elements, and so on. These affect the "movement" or "pace" of the story as well as creating structural unity. Narrative or thematic rhythms don't define themselves precisely in real time as do musical rhythms, or even the rhythms of film montage, but they involve perceptible elements of duration, articulation, and tempo nonetheless. Like musical rhythms, they can also be complex, multilayered, and ambiguous. Writers will frequently manipulate elements of textual rhythm for specific effects, such as accelerating the tempo of plot incidents to build tension.

There are some interesting differences between the ways textual rhythm manifests itself in hypertext and printed linear writing. The node gives hypertext a distinctive rhythmic unit that is absent from print. The end of one node and beginning of another generally constitutes a structural break in the text. The continual movement from node to node therefore creates an underlying rhythm, reinforced by the physical movement of a mouse button or key and its audible click. Individual nodes may also combine to form compound rhythmic units, or periods. Any succession of nodes that cohere as a group because of their content or appearance and seem distinct from surrounding nodes is likely to be perceived as a rhythmic period. Several periods can combine to form larger ones, and periods can overlap.

The distinctness with which hypertextual rhythmic units are articulated depends primarily on the degree of disjunction that the content of a node or period has from the material preceding and following it and the degree of closure with which it ends. Navigational elements can also affect articulation, though. A sudden increase in the number of available links or in the complexity of interactive options can create a structurally significant break in the reading as the reader ponders what to do next.

The length of rhythmic units has a marked effect on hypertextual rhythm. Unusually short nodes can produce a certain emphatic or staccato quality, as in the opening "In the labyrinth/beginning" sequence of Moulthrop's Victory Garden. [2] They can also produce a sense of faster tempo than longer nodes because they result in more frequent movement from node to node. Conversely, though, the reader may perceive a decrease in node length as a slowing down. The greater frequency of breaks between nodes introduces pauses between passages that could otherwise have been presented as uninterrupted text in a single node. The sense of tempo therefore depends upon the content of the nodes as well as their length.

Nodes that don't cohere into larger periods produce a detached rhythmic effect in contrast to the more legato character of long periods. For example, the relatively self-contained nodes of Dillemuth's Omphaloskepsis [3] create a rhythm markedly distinct from those of the long multi-node narrative threads that can be found in works such as Victory Garden or Joyce's afternoon. [4] The degree of rhythmic regularity is also important to the character of a hypertext reading. Constant extreme shifts in the lengths of nodes or periods can have an unsettling affect on the reader. Regularity, on the other hand, produces a sense of stability.

Unlike the rhythms of linear text, hypertextual rhythms depend at least partly upon reader choices and often can even be deliberately varied by the reader. For example, a reader could choose to follow paths as they unfold over the course of many nodes or to jump frequently from one path to another, creating a more detached rhythm. Or she could alternate between the two approaches to vary the rhythm. Periodic backtracking or regular returns to a main map view can also have rhythmic significance. Yet despite this, certain rhythmic patterns will be more likely than others to emerge from the reading of any given piece of hypertext.

Considering these inherent rhythmic tendencies is important because rhythm can serve as a structural constant that withstands many of the permutations a hypertext might undergo. Just as the pitches of a musical passage can be altered without affecting its rhythm, a set of nodes in a hypertext can be subject to a variety of different orderings while always maintaining the same large-scale temporal pattern. For example, many hypertexts are designed in such a way that the reader will encounter in succession the fragments of different narrative threads. This may create a regular sequence of rhythmic periods articulated every few nodes by the move from one thread to another, regardless of the exact order in which fragments are encountered. This rhythm will probably cohere in the reader's mind as a dominant structural element, even if she develops no clear sense of how the threads fit into larger geometric or spatial patterns. Thus rhythm can serve as a unifying device in a hypertext that is otherwise in constant flux. Contrasting rhythms can help give distinctive characters to different sections of a work and prevent monotony.

Mark Bernstein has categorized a number of geometric patterns that can be discerned in the node/link structures of many hypertexts. [5] Distinctive inherent rhythmic tendencies can exist independently of the geometric patterns he defines. For example, whether a hypertext is structured as a contour (a set of interlocking loops), a counterpoint pattern (a systematic alternation between two types of material), or a multiple split-join pattern (the divergence and convergence of threads), the reading of it will likely be characterized by shifts in textual content that produce a rhythm. The structure may have an inherent tendency to produce regular rhythmic periods consisting of three or four nodes. Or it may naturally tend toward much longer periods punctuated irregularly by short periods of one or two nodes. Or it may tend toward another rhythmic pattern altogether. The inherent rhythmic tendencies are more likely to be determined by such factors as the length of paths and the density of links rather than whether the structure forms a contour, a counterpoint, or a split-join. Rhythm must therefore be taken into account to achieve a full understanding of differences among hypertext structures.

Another rhythmic phenomenon unique to hypertext arises from node recurrence. Both the frequency and the length of reencounters with text during a reading can establish rhythmic patterns. This is particularly noticeable as the middle reading gives way to the end reading in a hypertext. The progressively increasing recurrence creates a rallentando effect as new material is less and less frequently encountered. If the reader skims quickly over old material to reach the new, the skimming process can serve to articulate very strongly the nodes that are actually read carefully. The effect can be of a rather too ponderous rhythm.

The concepts discussed here may help writers get a better grasp of textual relationships that are always a moving target. They may also help provide a better framework for theoretical discussion of structures built upon shifting ground. Ultimately, though, writers need to be able to work with time as a structural element, just as they now work with shape or contour. There is a need for adaptive hypertext systems that take into account a reader's historical awareness of the text and let the structure itself respond appropriately to historical contexts.

[1] For a discussion of ways to achieve closure efficiently, see Robert Kendall, "But I Know What I Like," in SIGWEB Newsletter (Vol. 8, No. 2, 1999), pp. 8-13. [back]

[2] Moulthrop, Stuart. Victory Garden (Eastgate Systems, 1991). [back]

[3] Dillemuth, Jay. Omphaloskepsis: A Story (Alt-X Online Publishing Network) [back]

[4] Joyce, Michael. afternoon, a story (Eastgate Systems, 1990). [back]

[5] Bernstein, Mark. "Patterns of Hypertext" in Proceedings of Hypertext '98 (ACM, 1998). [back]