Computer Co-Authors for Fiction Nick Montfort Presentation at Computers & Writing 2000 27 May 2000, Fort Worth, TX Hi, I'm Nick Montfort. I'm a writer, and I live in Manhattan. Let me note that the full abstract of my presentation, liked to the two works I discuss, is available at my Web site, . [The actual URL: -NM] In this presentation, I will: o first, explain exactly what I mean by "Computer Co-Authors for Fiction," then o briefly discuss EddieEdit, a project of mine from the MIT Media Lab, and then o discuss Winchester's Nightmare in detail. I'm only going to speak for about ten minutes. Instead of my showing you what Winchester's Nightmare is on the screen, I'm going to use the remaining time to allow each of you here to have eight minutes or so in which you can read Winchester's Nightmare and ask me questions about it individually. I'll take further questions during the question and answer period at the end of the session, of course. To begin: The creators of electronic literature have devised fun and interesting ways for computers and people to write interactively, together. Before literary hypertext came to the computer, there was nevertheless electronic literature -- works that called on the reader to type a response or a command to the main character, continuing the text. These include conversational characters, such as the computer psychiatrist Eliza by Joseph Weizenbaum. They also include what Infocom called "interactive fiction" back in the 1980s, a category exemplified by the text adventure game. The reader was called upon to literally do writing in these works, taking on the role of an "interactor." This category is rich in literary potential, although it has been noticed by only a few critics (such as Espen Aarseth) and alluded to in more mainstream hypertext fiction only rarely (as in the works of Stuart Moulthrop.) This group of interactive fiction as more narrowly defined includes Adventure by Woods and Crowther, which Aarseth describes as the ur-text of the form, as well as the mainframe and later PC version of Zork by Blank and Lebling. Less well-known, but of interest to those who enjoy literary fiction, are two text adventures released in the mid-1980s, one by Thomas Disch, called Amnesia, and the other, a work called Mindwheel written by the U.S. Poet Laureate Robert Pinsky. Although enjoyable, occasionally artful, and even useful for educational purposes (such as language learning), the possible interactions with conversational characters and text adventures, even those of literary merit, were still quite limited. The early works in these categories allowed participation, but did not live up to the ideal of human/computer collaboration, or co-authorship. For the most part, they forced the human participant into a particular role -- a patient in therapy or a treasure-collecting adventurer, for instance. This made the interactor more of a "player" -- either in the gaming sense as one who makes decisions and controls a game token according to rules, or in the theatrical sense of an actor who is given a particular part to play. This role may be enjoyable, but it is not the same role as "co-author." It is important to note that when I say the text adventure interactor or player is doing writing, I don't mean that in the Barthian sense of the writerly reader. The text adventure interactor is literally typing words. These aren't marginal notes -- they actually become an integral part of the text. As I've previously stated, however, being a co-writer does not necessarily make one a co-author. Substantial control over story elements, or over the way in which the underlying events are narrated, would do this. This explains what I mean by "co-authorship" fairly well, I hope, but I should also make it clear that I am not talking about co-authoring a story that is for publication. I intend to provide a co-authorial experience that is enjoyable in and of itself. There has been work done on using the computer as either an autonomous composer of stories (James Meehan) or a collaborator in the compositional process (Charles Hartman). I am interested in providing a literary experience which itself is the final product, allowing the interactor to encounter an engaging and powerful story in which the interactor participates, profoundly. I believe that, starting with a form such as the text adventure that provides a genuine co-writer role to the interactor, improvements in interface can bring us closer to the goal of co-authorship. In fact, I believe Winchester's Nightmare has already helped to move us a few steps closer to the co-authorial grail: the creation of a work which would allow broad creative freedom for the human interactor, who would write in concert with a provocative computer co-author. Such works would have profound literary possibilities. First, however, I'll briefly describe a project I did in 1998, in the course of getting my Master's degree at the MIT Media Lab. In this project, EddieEdit, second and third graders conversed with a virtual kid editor by typing. The software came in two flavors: a Web-based version that I wrote in Perl and a Macintosh version, with integrated text editor, that I programmed in C. EddieEdit was supervised by Justine Cassell. The kid editor, Eddie the Editor, asked story planning questions of the children before they started the main writing process, then discussed revision with them after they wrote a story unaided. Eddie was a descendant of Eliza, in part, but also of Hugh Burns's TOPOI. Eddie was explicitly *not* a co-author. The initial conversation about story planning, and the final conversation about revision, were constructed together by the student and the computer. But Eddie got out of the way while the student actually wrote the story. From the standpoint of computer co-authorship of fiction, EddieEdit demonstrates that a "human-heavy" interaction, in which the computer was limited to a prompting and supporting role, can work well. More recently I have completed a work of interactive fiction in the literary fiction genre -- Winchester's Nightmare: A Novel Machine, which premiered at Digital Arts and Culture 99, last October. In its "hardback" form, it is a novel-length interactive fiction which includes a computer running software: a novel machine. I produced an edition of ten hardbacks, which sell for $250 each. There are still some for sale, including the ones in this room. The softback, available free, contains the entire text of the hardback edition. I want to express my thanks to John Slatin, who advised my undergraduate thesis on interactive fiction back in 1995 at UT-Austin and was the first university faculty member I encountered who would entertain the idea that text adventures could be literary. The main character of Winchester's Nightmare is the historical figure Sarah Winchester, nee Pardee, 1837-1922. Sarah is remembered for building onto her San Jose house constantly, actually 24 hours a day, for more than thirty years. The official, and rather simplistic, explanation for this eccentric enterprise is that she was following the instructions of a spiritualist, seeking redemption for the many killings effected by the Winchester rifle, made by her husband's company. Sarah was made rich by the mass production of weapons, gave her name to the Winchester hard drive (indirectly, via the street that bears her name in San Jose), and built an ever-sprawling house that serves as a metaphorical target for today's American city. In Winchester's Nightmare, which treats themes of technology and American urban life, the interactor acts and explores through her. Winchester's Nightmare is about Sarah's psyche, and does not portray her house directly. While the Winchester Mansion seems rich in narrative possibilities, Winchester's Nightmare takes place instead in the composite metropolis of Sarah's dream, United City. This city is peopled with other characters, and a plot (driven by Sarah's search for redemption) organizes the narrative. The setting, however, is the dominant element. United City is like Rockvil in Steven Meretzky's A Mind Forever Voyaging. It is an American city, one which the main character sees as home, and it is transformed through time. It is also like the landscape of Robert Pinsky's Mindwheel, in that it is a "mental map" of a character's psyche. Exploration of the world reveals aspects of the protagonist and her particular obsessions. The interaction with and completion of the text is motivated by series of challenges, as in text adventures. The puzzles presented are constructed for thematic appropriateness, and present to motivate exploration and reflection. I hope the interactor will be able to engage with the work as literature, rather staying in a jigsaw-puzzle mode of thinking during all of the interaction. Winchester's Nightmare was written in Inform, a system by Graham Nelson, and with the proper interpreter, thanks to the Z-machine by Mark Blank and Joel Berez, it runs on just about any computer: Windows, Mac, BeOS, Linux, Atari ST, etc. In Winchester's Nightmare the text is continuous, instead of being punctuated by command prompts that differentiate the interactor's input from the computer's contribution. Only English input is recognized - the abbreviated commands such as "E" (for "go east") or "L" (for "look around") are not available. This attempt to make the work more literary, and to have the human/computer collaboration produce a more continuously woven text, actually alienated the community of text adventure gamers, who found my changes crippling. But this redesign of the interface, informed by improvisational theater and conversational structure, did result in a more symmetrical interaction, with the human contribution becoming more like that of the computer narrator. In contrast to EddieEdit, the Winchester's Nightmare experience is "computer-heavy" -- the interactor is limited to completing a sentence every few lines. The interactor controls Sarah Winchester by completing her thoughts, but is not cast directly into the role of that character by being addressed in the second person, as is usually the case in text adventure games. Winchester's Nightmare is neither an exploratory hypertext nor a constructive hypertext -- since it isn't a hypertext -- but exploration and a piecing together of meaning is involved. Visiting most parts of the rich setting is optional, so different interactors will have different experiences. There is also much meaning to construct during the course of overcoming challenges. I think the ending of Winchester's Nightmare is also a first, and strikes a strong blow for the power of the interactor. By making the relationship between interactor and computer a bit more equal in these several ways, I have moved the nature of the interaction a few steps closer to co-authorship. I hope the work will also prove enjoyable and provocative. It's possible that Winchester's Nightmare has managed to fall right between the electronic literary community and the text adventure community, but some have enjoyed it, and now, you can see for yourself if you like it. I'll give you some time to read it while I am available to answer individual questions and also offer 'hints' to help you find out what types of interaction help you to progress. I want to add just one more thing, in relation to my comment about control over story elements: Winchester's Nightmare has a plot that is fixed, but the episodes that lead up to overcoming challenges, and the wandering around in the city, can vary from interactor to interactor. This is a meaningful form of control, as Greek drama demonstrates. The plot that playwrights dealt with was usually fixed, and based on well-known myths. But how they presented the episodes was under their control. ... In closing, please do take the floppy disks out of the computers you're using and take them with you, if you, or someone you know, may be interested in interacting with Winchester's Nightmare further. I'd also like to mention an even more recent work of mine, which just came out on Wednesday. This is The Ed Report, a serial hypertext novel -- one segment comes out every business day for more than a month -- and it's an example of genuine human/human co-authorship. William Gillespie of The Unknown wrote The Ed Report with me, and Dylan Meissner designed it. It's written in the form of a government report, it's a real hoot, it's what everybody's going to be talking about at ACM Hypertext this year, and it's at Thank you. ###