Chris Crawford on Interactive Storytelling (66 page)

The core of the dramatic expression comes in the system for sequencing beats; their system uses something very similar to my own system for roles and inclination formulae. What I call a reaction formula, they call a precondition; what I call an inclination, they call a weight.

Mateas and Stern implement a simple drama manager in their system by adding another factor to their inclination formulae based on the dramatic tension. The storybuilder assigns to each beat an individual tension value that is applied to the selection of the succeeding beat. The basic idea is that each succeeding beat should have more tension than the previous one, until the story reaches its climax, at which point the drama manager switches to beats with diminishing tension.

The Façade technology doesn’t address issues such as gossip, secrets, lies, multiple dimensions of relationships, or variable audience requirements for executing a verb. By focusing their attention on a precisely confined dramatic context, Mateas and Stern have obviated the need for these technologies, permitting them to attack and solve all the problems of that confined context. The result is without question the best overall interactive storyworld produced as of early 2004.

Mateas and Stern also participate in an excellent blog, Grand Text Auto, at
http://grandtextauto.org
/.

One of the more charming aspects of the universe of interactive storytelling is the collection of wild and crazy ideas that have been tried. The Dr. K- Project, by Brandon Rickman, definitely falls into this category
⁵
. It was carried out as part of a master’s thesis, and so doesn’t reflect the kind of concerted effort seen in many other projects. Instead, it offers a pleasantly bizarre approach to the whole problem of interactive storytelling.

Rickman has dispensed entirely with the notion of a story as a temporal sequence of events. Instead, he takes advantage of the computer’s special capabilities to offer something that—well, it’s difficult to describe. Perhaps the sample interaction he offers best characterizes the concept:

A place. A tea chest. A body bag. A scrap of paper. An assassin. A character. Something happens.

[User selects “A character”.]

A place. A tea chest. A body bag. Some scenery. An assassin. A rascal. Something happens.

The result of the user’s input is to replace “a character” with “a rascal.” If you’re wondering where the story is, well, there isn’t any story. Rather, as best as I can tell, the software presents an unfolding vignette. The user selects phrases from the phrase list, which causes the selected phrase to expand into something more detailed. Simultaneously, ignored phrases fade into generalities.

Weird as this might seem, it demonstrates an interesting technology. Mr. Rickman has taken the concept of an object hierarchy as used in object-oriented programming and applied it to drama. He has created a hierarchy of classes based on four basic classes (props, scenes, actors, and actions) with a variety of other sub- and superclasses to build a system that can readily move in either direction in the hierarchy. The result is this “unfolding vignette.”

The DNA in cell nuclei is a biological computer. It not only stores the information that constitutes genetic heritage, but it expresses that heritage by processing the proteins that create the cells forming human bodies. Even more impressive is DNA’s ability to recombine and alter in ways that permit living creatures to adapt to changes in the environment. This biological computing system is immensely powerful; a researcher has already used DNA to solve a mathematical problem that would have taken years to solve with conventional computers. The preparation and chemical analysis of the sample took many months; the actual computation took a fraction of a second.

So it shouldn’t be surprising that computer scientists have looked to DNA for inspiration in radically new approaches to computing. A major subfield of computer science research, called
genetic algorithms
, is dedicated to this idea. Much has been learned about how to build virtual genes, how to combine them, and how to set up populations of virtual systems that learn from their environment.

It occurred to Teong Joo Ong and John Leggett of Texas A&M University that genetic algorithms might prove a useful approach to the problem of interactive storytelling, so they built a software system to do just that
⁶
. They called it HEFTI: the Hybrid Evolutionary-Fuzzy Time-based Interactive storytelling system.

At this point, I must excuse myself with the observation that the HEFTI documents are written with more than the usual academic abuse of the English language. Multiple rereadings of these documents leave many unanswered questions. My explanation of this technology will be inadequate. I must accept some responsibility, for I do not have a computer scientist’s intimate familiarity with the latest jargon, but the obtuseness of the documentation’s writing is the primary obstacle.

The HEFTI system is complex, comprising five major components. The Evolutionary Programming Subsystem (EPS) component is the most critical.

It uses genetic algorithms to assemble virtual chromosomes representing story threads. These chromosomes are compared to story threads the storybuilder provides; the best-fitting chromosome becomes the story thread used in interaction with the player.

The Fuzzy Decision Based System (FDBS) “imposes constraints, modifications, and evaluations on the solution instances with rules and variables in the knowledge base.” Translated into plain English, this means nothing more than “the FDBS crunches some numbers.” Presumably it does so using the techniques of fuzzy logic.

The Knowledge Base is a pile of storybuilder-specified data. Its fundamental component is the
story component
, the author’s term for what I call a Verb. A story component is defined as “the representation of a particular time step in the story,” but the authors don’t define “time step.” The storybuilder strings these story components together into story threads, which appear to be representative sequences of story components. The Knowledge Base also contains Contextual
Sets, used to “categorize various plot units, events, scripts, actions, and characters within the story components.” Aside from “story components,” none of these terms is explained further.

The Graphical Object Library (GOL) is a collection of graphics data and methods that provide imagery and animation for the system’s output. When the HEFTI engine has decided what it wants to do, it directs the GOL to display the appropriate imagery onscreen.

The last component of the HEFTI system is the Integrated Authoring Environment (IAE), based on XML. This development environment provided with the engine feeds data into the Knowledge Base. Here’s an example of the code used in the IAE:

This sample exemplifies the wrong approach to development systems. It’s program code for programmers, incomprehensible to the kind of creative talent needed for storytelling. So long as development environments look like this code example, only programmers will write storyworlds, and we’ll continue to get the same old clichéd stories.

The only storyworld the authors built with their technology appears to be an abbreviated version of “The Three Little Pigs.” They were able to get the wolf to attempt to eat the pigs, and they managed to get the pigs to run away. One small step for a man…

Nicholas Szilas, the developer of IDTension
⁷
, uses a model for narrative that’s based on five fundamental concepts:

Goals
: States desired by various actors in the story.

Tasks
: Procedures that must be performed to achieve goals.

Segments
: Subtasks.

Obstacles
: Factors that prevent a segment from being performed.