Hyperrhiz 06: Essays

The E-ssense of Literature

Robert Kendall


Abstract

Many works of electronic literature use text generation algorithms or interactive interfaces to present the reader with a different text upon each reading. Such variable texts can be difficult to analyze and discuss because it can be prohibitively difficult to take into account all possible permutations. The standard critical methodology for approaching these texts is to discuss excerpts from different readings, perhaps comparing passages that involve alternative renderings of the same textual content. While this approach can convey a general sense of the work and its possibilities for variation, it usually doesn't allow a thorough treatment of a complex work's structural framework.

This essay presents a method for analyzing a work's source code to define the most important constant and variable properties of its constituent elements. It then applies the method to a generated electronic poem, "Snaps," by Dirk Hine. The source structure thus defined provides a springboard for critical interpretation of the work.


Introduction

When we discuss a work of electronic literature that uses text generation algorithms or nonlinear interactivity, it is often difficult to pin down the topic of discussion. Because the actual text of such a work differs upon each encounter with a reader, observations about one reading might not apply well to another. Taking all the possible permutations into account in a structural analysis can prove a daunting task. Critical discussion of nonstatic electronic writing, therefore, tends toward generalizations and brief examples rather than comprehensive close readings. The standard critical methodology is to discuss excerpts from different readings, perhaps comparing passages that involve alternative renderings of the same textual content.

This approach is certainly valuable, and there have been many effective applications of it by writers such as George Landow , Jane Yellowlees Douglas , and N. Katherine Hayles . It can convey a general sense of the work and its possibilities for variation, but it usually doesn't allow a thorough treatment of a complex work's structural framework. A more-comprehensive assessment of a variable electronic work is possible only if one systematically analyzes the underlying codebase that encompasses generative algorithms, hypertext path definitions, or any other encodings that determine the possible runtime readings. One must also cover all potential content, including generative lexicons, hypertext lexias, media libraries, and so on. This type of all-embracing structural analysis is possible only through examination of the work's source code, so we will refer to the subject of such analysis as the work's source structure.

A number of critics do provide extensive discussion of the algorithms underlying generative works. For example, Christopher Funkhouser writes in great detail about the code underlying many early works of generated poetry and Hayles discusses at length the algorithms driving John Cayley's Translation . These analyses, however, generally don't attempt to cover the full gamut of permutation that the textual content may undergo. A comprehensive and systematic analysis of source structure, however, can consider this gamut in its entirety and provide a more holistic view of a work.

Trying to catalog every possible variable detail of a large electronic work would be a daunting task likely to overwhelm the critic with indigestible minutiae. Instead, the attempt to define a work's source structure should take the opposite approach. It should focus on the elements that remain constant from reading to reading. We should start by asking, What is it that makes multiple different readings of a particular work all still recognizable as instantiations of itself rather than instantiations of a different work?

This question relates to a well-known problem in Western philosophy, so perhaps we can turn to philosophers for guidance. Plato was the first to grapple with the problem over 2000 years ago when he pondered the difference between universals or abstractions and individual instances. His response was his Theory of Forms, which postulates that one can perceive "the ever same and unchangeable" among "the changeable and manifold" . Elements such as beauty or goodness have a universal form that is unchanging. Individual instances of beauty differ from one another, but underlying them all is the unchanging idea of beauty. Similarly, Aristotle speaks of the "essence" of entities that allows us to categorize them, and this term has been adopted by later thinkers .

Debates about these ideas have occupied philosophers ever since their introduction. In particular there has been significant disagreement about whether essences actually exist or whether they are purely mental constructs developed from experience. Plato and the rationalists that followed him — among them Descartes, Spinoza, and Leibniz — argue for a world in which fundamental concepts exist a priori and can be determined by logic. More recently Noam Chomsky has put forward a rationalist case for linguistics. Empiricists — such as Locke, Berkeley, Hume, Mill, and Ayer — object that we can know nothing of the world but what reaches us through the senses. A concept such as beauty is just a collective label for all the individual beautiful things we've seen. It's unlikely we will ever resolve this issue because we will probably never have access to the universe's source code.

Defining Source Structure

The situation is a little more hopeful with electronic literature, because in this case we actually do have the source code. A variable work's program code clearly defines a priori qualities that constitute what could be considered its essence. Each reading of the work is an individual instantiation governed by this underlying essence. The standard critical approach described above, in which the critic discusses or compares runtime instantiations, could be termed an empirical approach to criticism. I would like to advocate a rationalist approach to e-literary criticism, not as a replacement to the empirical approach but as a supplement to it. The rationalist method uses the source code to delineate the work's essence, or source structure. This in turn allows the critic to forecast the full range of potential empirical experience of the work rather than just reporting on a random sampling of that experience.

A work's source structure consists of its unchanging elements. This includes obvious constants such as text that recurs verbatim in every reading, a fixed number of lines for a generated poem, or characters that appear in every version of a story. It also includes elements that remain consistent from one reading to the next in more subtle ways. For example, even though a text may change radically from reading to reading, each reading may retain certain stylistic or structural traits, such as a high degree of syntactic or narrative fragmentation or a particular type of vocabulary.

Even extensive variation can be understood as part of a work's constant framework, because the variants necessarily occur within fixed ranges determined by the source code. For example, a generated text cannot contain words or phrases not included in its encoded lexicon, and all of these text components will fall into a finite number of categories. Thus the description of a work's source structure should specify the ranges of its variable elements and categorize the possible values of those variables. This parallels a rationalist description of a universal, such as "chair," which takes into account all the variations that may occur from one instance of a chair to the next without negating its "chairness."

There might be a number of different ways to define the variables within a source structure and to categorize their values. The critic should choose the definitions and categories that have the most significance from a literary standpoint. In a large work, discussion might necessarily center on only the most important variables, but these can be explored throughout their full range to achieve a relatively holistic analysis.

In order to discuss these constant and variable factors accurately in a piece of electronic literature, we must first isolate the specific components that change or remain static from reading to reading. We must also define the hierarchies and relationships that pertain among these components. Probably the most powerful and flexible paradigm for software structure is that of object-oriented programming, so we will borrow the terminology of this paradigm. Even if a work's source code isn't explicitly object-oriented, it can be viewed conceptually in terms of objects, properties, and behaviors (or methods).

For the purposes of analysis, a literary work can be broken down into content objects (elements of text, graphics, audio, etc.), container objects (the windows, text boxes, or other container types that hold the content onscreen), and control objects (interface elements such as links, buttons, and menus). The properties of each object can then be identified and defined as either constant or variable, and the specific values of each variable can be defined. Programmatic behaviors can also be described and linked to the control objects that initiate them and the content or container objects that are effected by them.

A Practical Application

To demonstrate the approach outlined above, we will use it to analyze the source structure of "Snaps," a short generated poem by Dirk Hine . This work is written in HTML and JavaScript, so its source code is easily accessible to anyone. Every time the work is viewed it uses randomization functions to generate a 4-line poem and pair it with an image. (See Figure 1.) The program selects from 240 lines of text and 36 images, combining them in different ways.

Figure 1. An iteration of 'Snaps,' by Dirk Hine

There are several hundred thousand different possible textual combinations, and the vocabulary is fairly diverse. Comparative readings of a handful of iterations of the poem would probably do little more than convey a sense of intriguing and seemingly inexhaustible textual wanderings. We will attempt to provide a more focused and revealing view of the poem by defining its essential "Snaps-ness," as conveyed by its source structure, and subjecting that structure to literary analysis.

First we need to isolate the discrete objects that make up the poem. The highest-level object is a container object: the browser window. It remains constant in every reading, always maintaining the same visual appearance and containing the following elements in fixed positions:

  • Four lines of text, each of which constitutes a content object.
  • An image, which constitutes another content object.

There is also a single control object: A link that spans all four lines of text. This control object is constant because the link always executes the same function; it reloads the page, introducing four new lines of text and a new image.

The function attached to the poem's single control object doesn't rely entirely upon purely random selection algorithms. Although each image is chosen by means of a simple random-number generator, the lines of text are chosen by a more complex procedure. Each of the four text line objects has an array of 60 alternative text strings that can be loaded into that line of the poem. The function retrieves the system time and uses the seconds value to extract one of the 60 possible text values for Line 1 and Line 4. For example, if the clock reads "11:58:23," the poem will display the Line 1 text value with an index of "23" ("distant cities") and the Line 4 text value with an index of "23" ("mornings"). To get the text for Lines 2 and 3, the program adds the minutes value to the seconds value (58 + 23 in the example above) and uses a modulus to derive a new value between 0 and 59. This produces either identical index values or consecutive index values (such as 3 and 4) for lines 2 and 3.

The lines of the poem usually function grammatically as a pair of couplets, with the second line of each couplet modifying the first. The selection algorithm works against this pairing. The texts for Lines 1 and 4 are selected as a unit, since they use the same index values. The same is true of Lines 2 and 4, though the correspondence is often offset by one line. The correspondences between these linked lines don't have any apparent semantic significance, however; the pairings seem to have been made at random by the author.

The clock-based text-selection method achieves seemingly random results with a much lower level of text repetition than selection based on random-number generation. The rigorous association of text elements with clock times also seems to have a certain theoretical significance, explicitly subjugating the elements in the poem to the relentless and monotonous march of time.

Analyzing Graphical Content

The work's five content objects change with every reading, but they are not entirely variable. The alternative images, for example, all share certain characteristics, so they can be said to have certain properties that remain constant among all of them. We'll begin our analysis with the image object, first defining its constant properties and then moving on to its variable elements.

Table 1 outlines the most important constant and variable properties of the image object. The most obvious constant is the image size (320 by 240 pixels). This is an unusually small size for online photos and brings to mind the small, wallet-sized snapshots that are often carried around until they start to disintegrate. The look of well-worn snapshots is evoked by another notable constant: the deliberately reduced quality of the images. Each image has been processed to have the appearance of frayed, torn, smudged, or discolored edges. The images themselves always contain elements of distortion, including blurring, graininess, or pixilation. All the photos have a severely restricted color palette. Most are monochrome, and when multiple colors appear, they are highly desaturated. (See Figure 2.)

Table 1. Properties of the Image Object.
Property Type Value(s) Probability Factor
Size Constant 320 by 240
Clarity Constant Distorted
Edges Constant Damaged
Color Palette Constant Limited
Location of Scene Constant Street
Subject Matter Variable No people 15%
People not looking at camera 80%
Woman looking at camera 5%
Figure 2. An example of an image from 'Snaps'

The author explains that all the images are stills from a digital video camera. The degraded appearance of the pictures arose from his "fascination with alternative film processes such as gumoil and bromoil prints, with alternative cameras such as Lomos, toy cameras and pinhole cameras, and from a fascination with found photographs, accidental photography, and the prosaic but under-rated amateur snapshot" .

The locale of the photos is also constant, in the sense that they all depict street scenes. The author describes them as "chance captures of everyday life" .

Now we'll look at variable properties of this same image object, identifying the most significant categories of possible values for each property. Each of the properties we'll examine has a fixed number of possible values (representing value categories), and each of these values has a probability factor that determines how often that value will be used instead of another alternative value.

The subject matter of the photos can be broken down into three important categories concerning the role of people in the scenes. About 15% of the images contain no people. About 80% contain people who are not looking at the camera and appear unaware of the photographer's presence. The people in these scenes, when they are not alone, also appear to be ignoring each other's presence. Each individual seems isolated from the others. (See Figure 2.)

Two photos (about 5% of the total) stand out from the others because in them the subject looks directly at the photographer. (See Figure 3.) Both these photos appear to depict the same woman, which is the only apparent instance of a subject reappearing in more than one photo. She stands out distinctly from the anonymous people with obscured or averted faces in the other images.

Figure 3. Images of woman looking at the photographer

We can therefore say that the image object has a subject matter property with three possible values. Since about 15% of the images contain no people, and since each image has a more or less equal chance of being selected by the poem's randomization function, we can say that a 15% probability factor adheres to the "no people" value . In other words, the viewer has about a 15% chance of seeing unpeopled scenes every time a new version of the poem is loaded. The chances of encountering the other subject matter values are determined by their probability factors respectively. Isolating and analyzing the probability factors of properties in this way can reveal a great deal about the nature of the poem's content, as we shall see.

In the broader context of the poem, the loss of information in the photos through their degraded appearance seems to represent a failure of memory and communication. Tying in with this is the sense of human isolation conveyed through the subject matter of the photos. Each individual seems cut off from the others in an urban landscape drained of color and clarity. The two photos of the woman represent an infrequently occurring attempt to break the isolation by communicating with the photographer. As we will see, these themes are carried over into the text as well.

Analyzing Textual Content

Each text object has 60 alternative texts that can be loaded into it, so there are 60 different versions of each line. The different lines are combined by the quasi-randomization function described earlier.

Table 2 outlines properties that are shared by all four text lines. Just as we have a reduced color palette in the images, we have a reduced grammatical palette in the text. All versions of all four lines have the same minimal grammatical structure, consisting of sentence fragments that lack a verb. This makes it impossible for the program to create a poem containing a complete sentence. The length of lines varies between one and five words, heavily weighted toward the shortest lengths. The grammatical fragmentation and lack of verbs creates a feeling of inertia and perhaps even helplessness, echoing the sense of isolation among the people in the photos. The extreme brevity of the lines further contributes to the effect of information stripped away.

Table 2. Properties common to all four text objects.
The probability factors here vary depending on the line.
PropertyTypeValue(s)Probability Factor
Grammatical Structure Constant Sentence fragment, no verb
Length Variable 1 word 45-80%
2 words 15-40%
3 words 5-10%
4-5 words 0-5%

Now let's look more closely at individual lines. Certain properties of the four lines group them structurally into two parallel pairs. Lines 1 and 3 parallel each other syntactically, as do lines 2 and 4. The first line of each pair is always a noun phrase, the most static type of syntactic unit. (See Table 3.) The second line is usually a modifier of the first line, though occasionally it will be a noun phrase like the first line. (See Table 4.)

Table 3. Properties of Lines 1 and 3.
PropertyTypeValue(s)Probability Factor
Grammatical Structure Constant Noun phrase
Subject Matter Variable Inanimate entities

50%

"visitors" 15% (Line 1)
0% (Line 2)
Other plural categories of people 20% (Line 1)
10% (Line 3)
Nouns with possessive determiner
"our"
10% (Line 1)
35% (Line 3)
More-personal references to people 5%
Table 4. Properties of Lines 2 and 4.
PropertyTypeValue(s)Probability Factor
Grammatical Structure Variable Modifier 80% (Line 2)
95% (Line 4)
Noun phrase 20% (Line 2)
5% (Line 4)
Subject Matter Variable No reference to people 60% (Line 2)
55% (Line 4)
People or human characteristics
(impersonal)
35% (Line 2)
40% (Line 4)
People or human characteristics
(personal)
5%
Tone Variable Neutral imagery 60% (Line 2)
50% (Line 4)
Negative imagery 30% (Line 2)
35% (Line 4)
Positive imagery 10% (Line 2)
15% (Line 4)

These same line pairings hold true for subject matter as well. About half the possible values for Lines 1 and 3 refer to inanimate entities, while the remaining values make reference to people but almost always in a very generic, impersonal manner. Some of the references are to plural categories of people, such as "neighbors," "artists," or "lawyers."

About 15% of the values for Line 1 consist of the word "visitors." The instances of this repeated word appear successively in the array for Line 1, and because of the algorithm's design, "visitors" is almost guaranteed to appear in several successive iterations of the poem. This word has a lonely ring to it in the context of this poem, conjuring up thoughts of strangers far from home. There is also a touch of irony in the emphasis of this word, since visitors are often tourists who take snapshots — but the bleakly quotidian pictures displayed here are the antithesis of tourist vacation photos.

Other values for these lines consist of the possessive determiner "our" in conjunction with a noun ("our bodies" or "our pasts") to convey human possession. This use of the preposition "our" generally seems to denote belonging to all humankind rather than any specific group of individuals that includes the speaker. Only a tiny percentage of the possible values for Lines 1 and 3 depart from this impersonal representation of humanity, with phrases such as "my thoughts" or "both of us," the latter appearing once in the vocabulary arrays for both Line 1 and Line 3.

The appearance of the line "both of us" abruptly shifts the focus of the whole poem to a personal plane. After the appearance of this phrase, lines such as "our bodies" and "our pasts" no longer seem generic but seem to refer to the bodies or the pasts of two distinct people. For example, consider this sequence of two iterations of the poem:

visitors
thirsty
nights
strange

photographs
without bitterness
our eyes
speechless

Then compare it to the following version of the sequence, in which "both of us" substitutes for "visitors," and note how the second iteration takes on a more personal tone.

both of us
thirsty
nights
finally here

photographs
without bitterness
our eyes
speechless

It's an effective device for shifting gears semantically. This occasional personalizing of the vocabulary produces an effect similar to that of the photo of the woman looking at the camera.

Lines 2 and 4 echo the subject matter concerns of Lines 1 and 3. A little over half of the possible texts for these two lines make no reference to people. Nearly all the other texts refer to people or human characteristics in a generic way. Only about 5% seem to refer to a specific person and hence strike a more personal tone, with such wordings as "in my room," "in your arms," or "searching for you."

With a little imagination, the references to "you" and "I" that sometimes appear in all four lines could be construed as components of a fragmentary narrative, the characters "you" and "I" perhaps even connected by some sort of love relationship, as implied by the sexual imagery in some of the text. The photographed woman could even provide a face for one of these characters.

Most of the work's emotional content is carried by Lines 2 and 4. About 30-35% of the possible texts imply violence, sadness, or loneliness ("exploding," "dying," "weeping," "abandoned"), understandable responses to the thematic isolation. Only about 10-15% imply more-positive feelings through such phrases as "without bitterness" or "caressed."

Emergence

Discussion of any source structure should recognize it as emergent. The probability factors connected to the variable properties represent dynamic tendencies and possibilities and should not be confused with quantizations of fixed textual elements. For example, consider the subject matter properties outlined in Tables 3 and 4 above. The way these properties factor into readers' perceptions of the poem will depend partly on chance and partly on how the readers interact with the poem. The source structure has the potential to spawn a wide range of empirical reading experiences, or instantiations of the poem's essence.

Since each iteration of the poem exists as a complete and self-contained entity, a reader might view no more than one instance of the poem, presumably a perfectly valid way to experience the work. Or the reader might view multiple iterations of the poem as if they were successive stanzas of longer work. For readers who view a single instance of the poem, the elements of the subject matter we have defined as "personal" will be very unlikely to play a role. This will also be true for readers who view only a handful of different iterations of the poem.

Readers who view enough iterations of the poem to allow consumption of most or all of its textual and graphical content will have fleeting encounters with the personal elements of the poem. The effect of these encounters will vary depending on how close they occur to the beginning of the reading. Early occurrences will likely affect interpretation of the remaining text, since the subtle narrative implications of these personal elements may cue readers to parse succeeding text in a narrative context. For example, if the poem has introduced "I" and "you" into the reading, the "lovers" in the following iteration of the poem might be equated with that "I" and "you."

lovers
concealed
our meetings
gone

Similarly, if the prominently repeated word "visitors" appears in close proximity to the introduction of "I" and "you," the whole series of text and images can be construed as a record of a trip taken by those two characters together.

If the personal elements appear near the beginning of the poem, they might be perceived as representing individuals overwhelmed by the impersonal and faceless city. If they don't appear until the end, they might seem like a small victory of the individual over the anonymous. Readers who spend enough time with the poem to allow multiple encounters with the personal elements will probably come away with a sense that the personal is fleetingly but continually struggling to emerge from a sea of urban anonymity.

Future Considerations

If authors value close readings and insightful critical evaluations of their work, they can take steps to make their source structures more transparent to critics. They could avoid submerging their text and its structure within binary-format Flash files or server-side databases or scattering it throughout hundreds of HTML files. Ideally they could store their text in a single XML file tagged to indicate structural groupings and relationships, generative processes, and interactive or animated behaviors. Of course they're unlikely to do this without the aid of an XML specification for electronic literature and the software tools necessary to work with it.

The Electronic Literature Organization's fledgling X-Literature Project is an attempt to create such a transparent XML structure. The project's main goal is to provide a bulwark against software obsolescence, but it could also help to remove barriers between author and critic. Facilitating standardized, transparent approaches to structure could lead to a more standardized and expressive critical vocabulary for talking about electronic literature. Our current means of storing and preserving electronic writing are reminiscent of the fledgling state of musical notation in eighth- or ninth-century Europe. An XML file format that also served as an advanced system of electronic "notation" would let our e-works survive the ages and with luck bring to them the sort of concentrated, precise analysis that music theorists now devote to scores of Haydn or Bartók.


References and Notes

  1. Landow, George P. Hypertext 3.0: Critical Theory and New Media in an Era of Globalization.
  2. Douglas, J. Yellowlees. The End of Books — Or Books without End? Ann Arbor, MI: The University of Michigan Press, 2000.
  3. Hayles, N. Katherine. Electronic Literature: New Horizons for the Literary. Notre Dame, IN: University of Notre Dame, 2008.
  4. Funkhouser, C.T. Prehistoric Digital Poetry. Tuscaloosa, AL: University of Alabama Press, 2007.
  5. Hayles, ibid., p. 145.
  6. "The Republic" in Plato. Great Dialogues of Plato. Trans. W.H.D. Rouse. New York: New American Library, 1999. P. 281.
  7. Aristotle. The Metaphysics. Trans. Hugh Lawson-Tancred. New York: Penguin Books, 2004.
  8. Hine, Dirk. "Snaps." Cauldron & Net. Vol. 3 (2001). Accessed 30 August, 2008. «http://www.studiocleo.com/cauldron/volume3/confluence/dirk_hine/snaps/snaps.html»
  9. "Notes on Contributors." Cauldron & Net. Vol. 3 (2001). Accessed 30 August, 2008. «http://www.studiocleo.com/cauldron/volume3/contents/index.html»
  10. Ibid.
  11. All the probability factors provided in this analysis have been rounded to the nearest 5%.

DOI Permalink

https://doi.org/10.20415/hyp/006.e05