What is Dynamic Text?
Perhaps, having said something specific about the general area of electronic writing, I should say something general about the specific area of electronic writing I’ve been thinking about. Dynamic text hyperfiction is what I currently call the area of electronic writing that interests me the most (warnings: this area is not (yet?) rigorously defined, and it needs a better name than “dynamic text”). My interest comes, in part, because dynamic text works can skirt what I see as serious limitations in two now-current approaches to electronic fiction. The first of these to-my-mind limited approaches grows out of an artificial intelligence (AI) tradition. It attempts to create systems that understand stories, so that they can be generated (or, more modestly, adjusted) automatically. One might say that these systems attempt to “operationalize” storytelling. The primary problem with systems of this sort is that they tend to “black box” the actual production of the story text itself - to regard this as a natural language generation (NLG) problem to be solved by other researchers. However, this attempt at operationalization is not simply problematic (for those with an interest in literary electronic fiction) because it depends on the idea that NLG will eventually be good enough to produce writing of literary quality. It is also problematic because few who write fiction of literary interest view the writing process as beginning with a story specification that is made progressively more detailed until it can be used to generate language for the reader. It is, in fact, probably more widely believed that good fiction tends to begin with language, with specific passages of writing. And while literary fiction audiences have in general had little interest in the possibilities of story permutation (e.g., as in the Choose Your Own Adventure fictions) much greater interest has been evoked by work that involves permutation at the level of language (e.g., the cut-ups of William S. Burroughs). I call systems for story generation which assume later NLG will produce the text atextual electronic fiction systems.
Another major approach to electronic fiction grows out of a print writing tradition. In this approach writers create passages of text (sometimes combined with images or sounds - or made imagelike themselves through animation and typography) that are then accessed in different orders depending on the reader’s actions (or depending on even more aleatory factors). While this work is generally more satisfying as literature than anything yet produced by NLG, this is partially due to the fact that as text it represents no great departure from print literature - its screens of text nearly as fixed as a book’s pages. The algorithms at work don’t move much beyond what Raymond Queneau accomplished in print in the 1960s with works such as “One Hundred Thousand Million Poems” and “Yours for the Telling.” I call works with relatively unchanging, page-like chunks of text fixed text electronic fiction systems.
I believe it is possible to go further than fixed-text systems do. I believe it is possible to introduce literarily-interesting dynamism at the level of text in manners impossible for print writing. Let me return to Burroughs for a potentially-helpful analogy. While Burroughs distributed his completed cut-up writings, he also distributed instructions for cut-ups - so that the new cut-up, the cut-up of the reader’s particular time and place, could also be produced. At the simplest level, this is where a dynamic text approach differs from most electronic fiction work now coming from traditional writing communities - it aims to occupy a space between Burroughs’s cut-ups and his instructions, employing both text written in the past and algorithms executed in the present.
In summary, the dynamic text approach differs from atextual electronic fiction systems in that it begins with text produced by a human author. It differs from fixed text electronic fiction systems in that it tends to address text with finer granularity and use more complex algorithms in its manipulation. Given this, the current research area that is closest to a dynamic text approach is hypertext, which (as I understand it) focuses on the interconnection, combination, and presentation of human-authored materials. Prior work in what I call dynamic text has taken place in the hypertext community, though often on (or just across) its borders - including those borders adjoining the artificial intelligence, natural language processing (NLP), poetry, and game communities. Here are some examples of prior work that I consider dynamic text. The Terminal Time project includes a “style preserving” natural language system that customizes pre-written texts according to the preferences expressed by a live audience, creating an ideologically-loaded history of the last millennium that is different at every viewing. The Façade system presents a real-time interactive drama using both AI techniques of the sort pioneered at CMU’s Oz Project and pre-written text that is dynamically sequenced (and organized according to “dramatic beats”). The computer game Black and White is perhaps the most popular piece of electronic writing since Infocom’s heyday in the 1980s - within its multi-act story structure it contains over 60,000 lines of dialogue (written by James Leach) of which the reader experiences a subset, in fine-grained variations, and involving different characters and settings depending on the interactions up to that point. (Façade and Black and White were discussed at the SIGGRAPH 2002 panel that Andrew Stern and I organized.) The work of poet and translator John Cayley involves a dynamic text technique that he calls transliteral morphing - moving via sound-organized letter transformation chains from one pre-written text to another (as discussed a bit in my post about riverIsland). The Card Shark and Thespis systems combine hypertext concepts with emergent behavior (through the buildup and alteration of constraints) for the purposes of electronic fiction, and are appropriate for use with texts of a relatively fine granularity.
As a few might recognize, I first started writing about this in the article Brion Moss and I wrote (with sidebars by Adam Chapman) about The Impermanence Agent for Cybertext Yearbook 2001. I should really get around to posting that somewhere around here…