towle@ils.nwu.edu
wolff@cs.nwu.edu
http://www.ils.nwu.edu/~towle/
http://www.cs.nwu.edu/~wolff/
Victory Gardens [Moulthrop, 1995], a professional piece of hypertext fiction, is a good example of this style. It suffers immediately from a feeling of "lost in hyperspace". Consider this hypertext node:
...So much of it seemed contrived, spectacular. She kept telling herself it was _only_a_movie_, except it seemed to be another David Lynch film, a cinema of angst and delusion. The night was full of unfunny jokes. She was seeing a report on military intelligence delivered by a bearded tough guy named "Wolf Blitzer." What did it mean? Had there been a coup de tube by the World Wrestling Federation? Would the cameras cut now to Atlanta to show _Hulk_Hogan_ at the anchor desk?
Already we see a problem with the linking cues. If we go to "Hulk Hogan", for instance, will we get an explanation of who Hulk Hogan is? Will we hear about the narrator's opinion of Mr. Hogan? Will we hear Hulk's opinion on the Gulf War? Will we get a description of him? Will the story switch to his perspective?
Clicking on "Hulk Hogan" takes us to an imagined Hulk Hogan monologue as if he were a CNN anchor. It's interesting, but we're not sure why we're reading this at this juncture, very much because there were no expectations built up as we moved from node to node. Both pieces of text are hanging out in empty space (or rather, empty hyperspace) without anything to connect them except a shared character. While this may help to set a mood of the story, it does very little to establish a coherent narrative.
We propose a theory of linking the nodes in a hypertext fiction system that avoids this problem. Specifically, we address two issues: what should be linked, and what should the text of the link be?
We have tentatively identified three main categories of links: links on character, links on events/scenes, or links on subsidiary information. For instance, in a "Star Trek" episode, different readers might first want to hear what Captain Kirk is thinking about the current encounter with this new species of alien, find out if the crewÕs last-ditch escape attempt will work, or how the aliens are able to hold the ship motionless in space.
We propose that ASK system technology [Osgood 1994] can just as easily be applied to fiction. In a traditional ASK system, an indexer analyzes piece of text to determine what questions it raises and what questions it answers. Links are then provided from that text (or, in ASK terminology, that "story") to stories that answer the questions raised by the source. The link text, of course, is just the text of the question involved, (e.g. "What would it be like if Hulk Hogan had taken over the networks?")
In a fictional narrative, the "stories" will be scenes or scene parts, which raise questions to be answered by other scenes. We show that this approach works for fiction quite well, even though it was developed for non-fiction works.
The demonstration format seemed the best way to present these findings as fiction has a large aspect of "feel" to it---most people read fiction for pleasure, and reading a paper disembodied from the work itself would not let participants experience the impact of the medium very well, nor understand how much our approach can add to narrative coherence.
Wolff Dobson is a 3rd year graduate student at the Institute for the Learning Sciences. He is currently building software tools that help content experts write educational software.