Delightfully, the current issue of AI Magazine (Volume 30, number 3, Fall 2009) is on computational creativity. The number offers articles on the field overall; the history of workshops on the topic; computer models of creativity; and creative systems to generate music, stories and their tellings, moves of chess, and humor. The last article is computer-generated in high Hofstadter style.
Pablo Gervás’s contribution, “Computational Approaches to Storytelling and Creativity,” provides a clear introduction to the concept of creativity and the history of the term, analyzes the relevant features that storytelling systems can work upon, gives an outline of work in computational creativity so far, and continues with a capsule summary of several important storytelling systems. The last one of these is my system nn, which I renamed “Curveship” as I started focusing on a public release of the software.
In the nn system for interactive fiction (Montfort 2007) the user controls the main character of a story by introducing simple descriptions of what it should do, and the system responds with descriptions of the outcomes of the character’s actions. Within nn, the Narrator module [now called the Teller] provides storytelling functionality, so that the user can be “told” the story of the interaction so far. The Narrator module of nn addresses important issues in storytelling that had not been addressed by previous systems: order of presentation in narrative and focalization. Instead of telling events always in chronological order, the nn Narrator allows various alternative possibilities: flashbacks, flash-forwards, interleaving of events from two different time periods, telling events back to front. It also captures appropriate treatment of tense depending on the relative ordering of speech time, reference time, and event time. Focalization is handled by the use of different focalizer worlds [now called concepts] within the system. Aside from the actual world of the interactive fiction system, nn maintains additional separate worlds representing the individual perspectives and beliefs of different characters. These can be used to achieve correct treatment of focalization (telling the story from the point of view of specific characters). [pp. 57-58]
In discussing the systems, Gervás notes (and I agree) that the other systems he discussed, ranging from Klein’s Novel Writer and Meehan’s Talespin to The Virtual Storyteller and Riedl’s Fabulist, are system for inventing stories, while nn’s Narrator (Curveship’s Teller) is the only system for telling stories. He writes:
If the processes for inventing stories in the reviewed systems rate low in terms of creativity, the rating obtained by processes for telling stories is even sadder. The challenge of how to tell a story has received very little attention in general, and it is mostly tagged on as a final stage to systems that concentrate on inventing stories. The nn system is a notable exception in that it involves a significant effort to model computationally some of the basic elements contained in Genette’s work on narrative discourse (Genette 1980): relative order of presentation and focalization. However, all the systems that tell the stories they invent do in fact include default solutions to many of the technical challenges involved in telling a story. [p. 60]
Although Gervás has provided a good take on the system, I’ll just note one way in which Curveship (née nn) does a bit more than the article might suggest to reader and one way in which it does less.
Genette described five categories of narrative discourse: order, frequency, speed, mood (which includes focalization), and voice (which includes distance). Curveship can vary not only order and focalization; it also allows for significant variation in the other three categories. I hope this will be of practical interest to interactive fiction authors and to those seeking to teach narrative theory using Curveship. However, the main research advances that have been made so far are in the two areas that Gervás indicates: order and mood (specifically, focalization).
While Curveship can automatically creative narrative variation based on parameters, I have to note that I am not putting it forth as a creative system. This makes it unlike many of the programs discussed in Gervás’s article and in this issue of AI Magazine. Given a specification for telling (which is called a spin), the system can make the appropriate changes and generate suitable text. However, the system does not, by itself, determine how a story should be told. The code that individual IF authors and AI researchers write is needed to accomplish that task.
Of course, formalizing the elements of narrative variation is necessary for any principled system that is supposed to vary the telling of a story. I hope that Curveship’s Teller will be deeply relevant to work in the creative invention and telling of stories, and that it will be used not only to enable new sorts of learning systems and interactive fiction pieces but also, in modified or unmodified form, as a component of creative systems.