I’m at the Electronic Literature Organization’s annual conference in Bergen, Norway, where I hope to capture some “think aloud” readings of electronic literature (e-lit) by artists, writers, and scholars. I’ve mentioned this little project elsewhere, but it bears more explanation.
The think aloud protocol is an important pedagogical tool, famously used by Sam Wineburg to uncover the differences in interpretative strategies between novice historians and professional historians reading historical documents (see Historical Thinking and Other Unnatural Acts, Temple University Press, 2001).
The essence of a think aloud is this: the reader articulates (“thinks aloud”) every stray, tangential, and possibly central thought that goes through their head as they encounter a new text for the first time. The idea is to capture the complicated thinking that goes on when we interpret an unfamiliar cultural artifact—to make visible (or audible) the usually invisible processes of interpretation and analysis.
Once the think aloud is recorded, it can itself be analyzed, so that others can see the interpretive moves people make as they negotiate understanding (or misunderstanding). The real pedagogical treasure of the think aloud is not any individual reading of a new text, but rather the recurring meaning-making strategies that become apparent across all of the think alouds.
By capturing these think alouds at the ELO conference, I’m building a set of models for engaging in electronic literature. This will be invaluable to undergraduate students, whose first reaction to experimental literature is most frequently befuddlement.
If you are attending ELO 2015 and wish to participate, please contact me (samplereality at gmail, @samplereality on Twitter, or just grab me at the conference). We’ll duck into a quiet space, and I’ll video you reading an unfamiliar piece of e-lit, maybe from either volume one or volume two of the Electronic Literature Collection, or possibly an iPad work of e-lit. It won’t take long: 5-7 minutes tops. I’ll be around through Saturday, and I hope to capture a half dozen or so of these think alouds. The more, the better.
The Electronic Literature Organization’s annual conference was last week in Milwaukee. I hated to miss it, but I hated even more the idea of missing my kids’ last days of school here in Madrid, where we’ve been since January.
If I had been at the ELO conference, I’d have no doubt talked about bots. I thought I already said everything I had to say about these small autonomous programs that generate text and images on social media, but like a bot, I just can’t stop.
Here, then, is one more modest attempt to theorize bots—and by extension other forms of computational media. The tl;dr version is that there are two archetypes of bots: closed bots and green bots. And each of these archetypes comes with an array of associated characteristics that deepen our understanding of digital media. Continue reading →
“Non-consumptive research” is the term digital humanities scholars use to describe the large-scale analysis of a texts—say topic modeling millions of books or data-mining tens of thousands of court cases. In non-consumptive research, a text is not read by a scholar so much as it is processed by a machine. The phrase frequently appears in the context of the long-running legal debate between various book digitization efforts (e.g. Google Books and HathiTrust) and publishers and copyright holders (e.g. the Authors Guild). For example, in one of the preliminary Google Books settlements, non-consumptive research is defined as “computational analysis” of one or more books “but not research in which a researcher reads or displays substantial portions of a Book to understand the intellectual content presented within.” Non-consumptive reading is not reading in any traditional way, and it certainly isn’t close reading. Examples of non-consumptive research that appear in the legal proceedings (the implications of which are explored by John Unsworth) include image analysis, text extraction, concordance development, citation extraction, linguistic analysis, automated translation, and indexing. Continue reading →
Mark Z. Danielewski’s House of Leaves is a massive novel about, among other things, a house that is bigger on the inside than the outside. Walt Whitman’s Leaves of Grass is a collection of poems about, among other things, the expansiveness of America itself.
What happens when these two works are remixed with each other? It’s not such an odd question. Though separated by nearly a century, they share many of the same concerns. Multitudes. Contradictions. Obsession. Physical impossibilities. Even an awareness of their own lives as textual objects.
To explore these connections between House of Leaves and Leaves of Grass I have created House of Leaves of Grass, a poem (like Leaves of Grass) that is for all practical purposes boundless (like the house on Ash Tree Lane in House of Leaves). Or rather, it is bounded on an order of magnitude that makes it untraversable in its entirety. The number of stanzas (from stanza, the Italian word for “room”) approximates the number of cells in the human body, around 100 trillion. And yet the container for this text is a mere 24K. Continue reading →
I recently proposed a sequence of lightning talks for the next Modern Language Association convention in Chicago (January 2014). The participants are tackling a literary issue that is not at all theoretical: the future of electronic literature. I’ve also built in a substantial amount of time for an open discussion between the audience and my participants—who are all key figures in the world of new media studies. And I’m thrilled that two of them—Dene Grigar and Stuart Moulthrop—just received an NEH grant dedicated to a similar question, which is documenting the experience of early electronic literature.
Electronic literature can be broadly conceived as literary works created for digital media that in some way take advantage of the unique affordances of those technological forms. Hallmarks of electronic literature (e-lit) include interactivity, immersiveness, fluidly kinetic text and images, and a reliance on the procedural and algorithmic capabilities of computers. Unlike the avant garde art and experimental poetry that is its direct forebear, e-lit has been dominated for much of its existence by a single, proprietary technology: Adobe’s Flash. For fifteen years, many e-lit authors have relied on Flash—and its earlier iteration, Macromedia Shockwave—to develop their multimedia works. And for fifteen years, readers of e-lit have relied on Flash running in their web browsers to engage with these works.
Flash is dying though. Apple does not allow Flash in its wildly popular iPhones and iPads. Android no longer supports Flash on its smartphones and tablets. Even Adobe itself has stopped throwing its weight behind Flash. Flash is dying. And with it, potentially an entire generation of e-lit work that cannot be accessed without Flash. The slow death of Flash also leaves a host of authors who can no longer create in their chosen medium. It’s as if a novelist were told that she could no longer use a word processor—indeed, no longer even use words. Continue reading →
Attention artists, creators, theorists, teachers, curators, and archivists of electronic literature!
I’m putting together an e-lit roundtable for the Modern Language Association Convention in Chicago next January. The panel will be “Electronic Literature after Flash” and I’m hoping to have a wide range of voices represented. See the full CFP for more details. Abstracts due March 15, 2013.
Seeking to have a rich discussion period—which we did indeed have—we limited our talks to about 12 minutes each. My presentation was therefore more evocative than comprehensive, more open-ended than conclusive. There are primary sources I’m still searching for and technical details I’m still sorting out. I welcome feedback, criticism, and leads.
An Account of Randomness in Literary Computing
MLA 2013, Boston
There’s a very simple question I want to ask this evening:
Where does randomness come from?
Randomness has a rich history in arts and literature, which I don’t need to go into today. Suffice it to say that long before Tristan Tzara suggested writing a poem by pulling words out of a hat, artists, composers, and writers have used so-called “chance operations” to create unpredictable, provocative, and occasionally nonsensical work. John Cage famously used chance operations in his experimental compositions, relying on lists of random numbers from Bell Labs to determine elements like pitch, amplitude, and duration (Holmes 107–108). Jackson Mac Low similarly used random numbers to generate his poetry, in particular relying on a book called A Million Random Digits with 100,000 Normal Deviates to supply him with the random numbers (Zweig 85).
Published by the RAND Corporation in 1955 to supply Cold War scientists with random numbers to use in statistical modeling (Bennett 135), the book is still in print—and you should check out the parody reviews on Amazon.com. “With so many terrific random digits,” one reviewer jokes, “it’s a shame they didn’t sort them, to make it easier to find the one you’re looking for.”
This joke actually speaks to a key aspect of randomness: the need to reuse random numbers, so that, say you’re running a simulation of nuclear fission, you can repeat the simulation with the same random numbers—that is, the same probability—while testing some other variable. In fact, most of the early work on random number generation in the United States was funded by either the U.S. Atomic Commission or the U.S. Military (Montfort et al. 128). The RAND Corporation itself began as a research and development arm of the U.S. Air Force.
Now the thing with going down a list of random numbers in a book, or pulling words out of hat—a composition method, by the way, Thom Yorke used for Kid A after a frustrating bout of writer’s block—is that the process is visible. Randomness in these cases produces surprises, but the source itself of randomness is not a surprise. You can see how it’s done.
What I want to ask here today is, where does randomness come from when it’s invisible? What’s the digital equivalent of pulling words out of a hat? And what are the implications of chance operations performed by a machine?
To begin to answer these questions I am going to look at two early works of electronic literature that rely on chance operations. And when I say early works of electronic literature, I mean early, from fifty and sixty years ago. One of these works has been well studied and the other has been all but forgotten.
My first case study is the Strachey Love Letter Generator. Programmed by Christopher Strachey, a close friend of Alan Turing, the Love Letter Generator is likely—as Noah Wardrip-Fruin argues—the first work of electronic literature, which is to say a digital work that somehow makes us more aware of language and meaning-making. Strachey’s program “wrote” a series of purplish prose love letters on the Ferranti Mark I Computer—the first commercially available computer—at Manchester University in 1952 (Wardrip-Fruin “Digital Media” 302):
YOU ARE MY AVID FELLOW FEELING. MY AFFECTION CURIOUSLY CLINGS TO YOUR PASSIONATE WISH. MY LIKING YEARNS FOR YOUR HEART. YOU ARE MY WISTFUL SYMPATHY: MY TENDER LIKING.
M. U. C.
Affectionately known as M.U.C., the Manchester University Computer could produce these love letters at a pace of one per minute, for hours on end, without producing a duplicate.
The “trick,” as Strachey put it in a 1954 essay about the program (29-30), is its two template sentences (Myadjectivenounadverbverb your adjectivenoun and You are my adjectivenoun) in which the nouns, adjectives, and adverbs are randomly selected from a list of words Strachey had culled from a Roget’s thesaurus. Adverbs and adjectives randomly drop out of the sentence as well, and the computer randomly alternates the two sentences.
The Love Letter Generator has attracted—for a work of electronic literature—a great deal of scholarly attention. Using Strachey’s original notes and source code (see figure to the left), which are archived at the Bodleian Library at the University of Oxford, David Link has built an emulator that runs Strachey’s program, and Noah Wardrip-Fruin has written a masterful study of both the generator and its historical context.
As Wardrip-Fruin calculates, given that there are 31 possible adjectives after the first sentence’s opening possessive pronoun “My” and then 20 possible nouns that could that could occupy the following slot, the first three words of this sentence alone have 899 possibilities. And the entire sentence has over 424 million combinations (424,305,525 to be precise) (“Digital Media” 311).
On the whole, Strachey was publicly dismissive of his foray into the literary use of computers. In his 1954 essay, which appeared in the prestigious trans-Atlantic arts and culture journal Encounter (a journal, it would be revealed in the late 1960s, that was primarily funded by the CIA—see Berry, 1993), Strachey used the example of the love letters to illustrate his point that simple rules can generate diverse and unexpected results (Strachey 29-30). And indeed, the Love Letter Generator qualifies as an early example of what Wardrip-Fruin calls, referring to a different work entirely, the Tale-Spin effect: a surface illusion of simplicity which hides a much more complicated—and often more interesting—series of internal processes (Expressive Processing 122).
Wardrip-Fruin coined this term—the Tale-Spin effect—from Tale-Spin, an early story generation system designed by James Mehann at Yale University in 1976. Tale-Spin tended to produce flat, plodding narratives, though there was the occasional existential story:
Henry Ant was thirsty. He walked over to the river bank where his good friend Bill Bird was sitting. Henry slipped and fell in the river. He was unable to call for help. He drowned.
But even in these suggestive cases, the narratives give no sense of the process-intensive—to borrow from Chris Crawford—calculations and assumptions occurring behind the interface of Tale-Spin.
In a similar fashion, no single love letter reveals the combinatory procedures at work by the Mark I computer.
MY AFFECTION LUSTS FOR YOUR TENDERNESS. YOU ARE MY PASSIONATE DEVOTION: MY WISTFUL TENDERNESS. MY LIKING WOOS YOUR DEVOTION. MY APPETITE ARDENTLY TREASURES YOUR FERVENT HUNGER.
M. U. C.
This Tale-Spin effect—the underlying processes obscured by the seemingly simplistic, even comical surface text—are what draw Wardrip-Fruin to the work. But I want to go deeper than the algorithmic process that can produce hundreds of millions of possible love letters. I want to know, what is the source of randomness in the algorithm? We know Strachey’s program employs randomness, but where does that randomness come from? This is something the program—the source code itself—cannot tell us, because randomness operates at a different level, not at the level of code or software, but in the machine itself, at the level of hardware.
In the case of Strachey’s Love Letter Generator, we must consider the computer it was designed for, the Mark I. One of the remarkable features of this computer was that it had a hardware-based random number generator. The random number generator pulled a string of random numbers from what Turing called “resistance noise”—that is, electrical signals produced by the physical functioning of the machine itself—and put the twenty least significant digits of this number into the Mark I’s accumulator—its primary mathematical engine (Turing). Alan Turing himself specifically requested this feature, having theorized with his earlier Turing Machine that a purely logical machine could not produce randomness (Shiner). And Turing knew—like his Cold War counterparts in the United States—that random numbers were crucial for any kind of statistical modeling of nuclear fission.
I have more to say about randomness in the Strachey Love Letter Generator, but before I do, I want to move to my second case study. This is an early, largely unheralded work called SAGA. SAGA was a script-writing program on the TX-0 computer. The TX-0 was the first computer to replace vacuum tubes with transistors and also the first to use interactive graphics—it even had a light pen.
The TX-0 was built at Lincoln Laboratory in 1956—a classified MIT facility in Bedford, Massachusetts chartered with the mission of designing the nation’s first air defense detection system. After TX-0 proved that transistors could out-perform and outlast vacuum tubes, the computer was transferred to MIT’s Research Laboratory of Electronics in 1958 (McKenzie), where it became a kind of playground for the first generation of hackers (Levy 29-30).
In 1960, CBS broadcast an hour-long special about computers called “The Thinking Machine.” For the show MIT engineers Douglas Ross and Harrison Morse wrote a 13,000 line program in six weeks that generated a climactic shoot-out scene from a Western.
Several computer-generated variations of the script were performed on the CBS program. As Ross told the story years later, “The CBS director said, ‘Gee, Westerns are so cut and dried couldn’t you write a program for one?’ And I was talked into it.”
The TX-0’s large—for the time period—magnetic core memory was used “to keep track of everything down to the actors’ hands.” As Ross explained it, “The logic choreographed the movement of each object, hands, guns, glasses, doors, etc.” (“Highlights from the Computer Museum Report”).
And here, is the actual output from the TX-0, printed on the lab’s Flexowriter printer, where you can get a sense of the way SAGA generated the play:
In the CBS broadcast, Ross explained the narrative sequence as a series of forking paths.
Each “run” of SAGA was defined by sixteen initial state variables, with each state having several weighted branches (Ross 2). For example, one of the initial settings is who sees whom first. Does the sheriff see the robber first or is it the other way around? This variable will influence who shoots first as well.
There’s also a variable the programmers called the “inebriation factor,” which increases a bit with every shot of whiskey, and doubles for every swig straight from the bottle. The more the robber drinks, the less logical he will be. In short, every possibility has its own likely consequence, measured in terms of probability.
The MIT engineers had a mathematical formula for this probability (Ross 2):
But more revealing to us is the procedure itself of writing one of these Western playlets.
First, a random number was set; this number determined the probability of the various weighted branches. The programmers did this simply by typing a number following the RUN command when they launched SAGA; you can see this in the second slide above, where the random number is 51455. Next a timing number established how long the robber is alone before the sheriff arrives (the longer the robber is alone, the more likely he’ll drink). Finally each state variable is read, and the outcome—or branch—of each step is determined.
What I want to call your attention to is how the random number is not generated by the machine. It is entered in “by hand” when one “runs” the program. In fact, launching SAGA with the same random number and the same switch settings will reproduce a play exactly (Ross 2).
In a foundational work in 1996 called The Virtual Muse Charles Hartman observed that randomness “has always been the main contribution that computers have made to the writing of poetry”—and one might be tempted to add, to electronic literature in general (Hartman 30). Yet the two case studies I have presented today complicate this notion. The Strachey Love Letter Generator would appear to exemplify the use of randomness in electronic literature. But—and I didn’t say this earlier—the random numbers generated by the Mark I’s method tended not to be reliably random enough; remember, random numbers often need to be reused, so that the programs that run them can be repeated. This is called pseudo-randomness. This is why books like the RAND Corporation’s A Million Random Digits is so valuable.
But the Mark I’s random numbers were so unreliable that they made debugging programs difficult, because errors never occurred the same way twice. The random number instruction eventually fell out of use on the machine (Campbell-Kelly 136). Skip ahead 8 years to the TX-0 and we find a computer that doesn’t even have a random number generator. The random numbers must be entered manually.
The examples of the Love Letters and SAGA suggest at least two things about the source of randomness in literary computing. One, there is a social-historical source; wherever you look at randomness in early computing, the Cold War is there. The impact of the Cold War upon computing and videogames has been well-documented (see, for example Edwards, 1996 and Crogan, 2011), but few have studied how deeply embedded the Cold War is in the software algorithms and hardware processes themselves of modern computing.
Second, randomness does not have a progressive timeline. The story of randomness in computing—and especially in literary computing—is neither straightforward nor self-evident. Its history is uneven, contested, and mostly invisible. So that even when we understand the concept of randomness in electronic literature—and new media in general—we often misapprehend its source.
Bennett, Deborah. Randomness. Cambridge, MA: Harvard University Press, 1998. Print.
Turing, A.M. “Programmers’ Handbook for the Manchester Electronic Computer Mark II.” Oct. 1952. Web. 23 Dec. 2012.
Wardrip-Fruin, Noah. “Digital Media Archaeology: Interpreting Computational Processes.” Media Archaeology: Approaches, Applications, and Implications. Ed by. Erkki Huhtamo & Jussi Parikka. Berkeley, California: University of California Press, 2011. Print.
—. Expressive Processing: Digital Fictions, Computer Games, and Software Sudies. MIT Press, 2009. Print.
Zweig, Ellen. “Jackson Mac Low: The Limits of Formalism.” Poetics Today 3.3 (1982): 79–86. Web. 1 Jan. 2013.