How can a computer generate, whether in-part or in-whole, a meaningful narrative?
Back in 1987, Chris Crawford coined the term "process intensity", or "the degree to which a program emphasizes processes instead of data." Greg Costikyan used this idea to analyze what he argued was the low-hanging fruit, the data-heavy applications the game industry was and still is pursuing, such as more polygons, more shaders and more uncompressed rendered cinematics, etc. He proposed Spore as a new hallmark in procedural generation... then two years later, we all actually played Spore and wanted to forget a lot of it.
I still think the idea is important though, and I want to use it as a lens to analyze approaches to procedural narrative.
Galatea is often cited as a milestone in the field, powered mostly by "low-hanging fruit" -- a data intensive approach, where Emily Short wrote thousands upon thousands of cases to detect the player's different inputs into the text parser, and wrote thousands and thousands of lines of branching dialogue to output back to the player. The effect is amazing, but totally dependent on Short's mastery of character and uncanny ability to predict players' reactions. We can't all be Emily Short.
Storytron is Chris Crawford's process-heavy approach that hasn't really hit it big. The one currently showcased, "Balance of Power: 21st Century," is like Fate of the World except text-only and much less robust in how it presents interaction and information. Crawford has to heavily expose the procedurality (in Storytron, the procedural parts are always enclosed in grey textboxes) in order for you to understand your effect on the narrative, which is great for gameplay but certainly isn't "immersive," whatever that word means.
Facade took Galatea and tried to add another NPC, procedural dramaturgy, a responsive 3D environment, and full voice over. So basically it tried to tackle, like, 10 more Holy Grails. It was "interesting" and innovative but still rather fragile and more-often broken. (Also as a sort-of academic researcher, I personally see it as a warning: that geniuses can spend 5+ years of their lives on something and still fall so short of what they sought to do.) It was both data-intensive in writing both Grace and Trip's lines and recording the voice actors, and also process-intensive in trying to assemble dramatic arcs with some sort of procedural narrative grammar.
All three "regular" approaches (data, process, both) assume a lot of stuff about procedural narrative. Few people have the resources or talent to successfully follow or emulate any of these approaches, and it's doubtful whether those even work all the time.
* * *
With that in mind, I'd like to highlight what I see as elegant but not always entirely convincing attempts to sidestep the problem. Most involve re-framing what "data" and "process" are, and I predict lateral thinking (rather than brute force approaches) is what's going to create more innovative work.
In defining these sidesteps, I'm simply trying to understand what other people do. These aren't discrete methodologies. The boundaries are fluid.
GAMEIFY narrative like Stephen Lavelle does in his excellent puzzle game Theatrics. Narrative design becomes a puzzle with an internal logic, transforming good strategy into what makes good storytelling. I use "gameify" in the best sense possible: not the addition of XP / badges, but the actual transposition of one system into another system of mechanics and dynamics. (If we can't take gamification back from the clutches of advertising, at least let's try to distinguish between "weak" gamification and "true" gamification.)
FILL A PLOT is what Emily Short does a lot, in addition to the data-heavy approach demanded by interactive fiction. The game might always begin with you entering a mansion, and always end with a showdown with the villain in the attic, but you can proceduralize everything else (when and where do you find clues?) based on a random seed (whodunit?) to fill this archetypal frame, as Emily Short mentions she did with "Mystery House Possessed." This isn't so much a side-step as a nice compromise that might work well with games based on genre fiction or folktales. However, if you're not working within an established narrative archetype, this might feel more artificial than authentic.
BE BUSHY, do what Deirdra Kiai did in "Chivalry is Not Dead" and don't let players see the lines between process and data; don't let them even see past the initial branches. Give them dozens of decisions with multiple options that may or may not lead somewhere or nowhere. Make the possibility space too large to fully map in your head. Fill up the "player RAM" and then some. While it's possible to trace your agency after everything happens, it fails one of the tests for a "meaningful choice" as outlined in The Rules of Play -- the player can't strategize nor predict what their choices will do. What is narrative without meaning and intention?
TELL A WORLD is Dr. Dan Pinchbeck's mental sidestep, in his replacement of "ludonarrative" with "ludodiegesis." Create a sort of story world, populated with "proto-narrative" affordances and let players cobble details together, as with environmental storytelling in BioShock. There, the proceduralism isn't in static plot / environment details or pre-authored character actions in audio logs, but rather in how the player consumes it. The problem with this route is that it doesn't account for reactive agents or NPCs actively changing stuff because it all already happened; you're doomed to exist in solitude, in the past. Could Dear Esther work with actual NPCs? And what is a narrative without characters? ("A ghost story," Pinchbeck might say. Okay fine, but only a ghost story.)
PASS THE BUCK entirely to humans to process. Computers aren't good at writing and interpreting narratives, so why make them? Let humans be the narrative computers here. I used to think this was an incredibly elegant sidestep, probably the best way of doing procedural narrative, but then I played a lot of prototypes and realized that most players aren't good storytellers. Specifically, I played Jason Rohrer's Sleep is Death and MIT Gambit's Improviso -- both games off-load "narrative processing" onto players -- and both are really, really hard to play well. After all, professional actors spend years studying and training how to improvise.
A relatively simple digital interface, like a video game, cannot help a bad storyteller become decent. It just exposes
Storybricks seems like a nice medium, a safe way of allowing player-generated content and scripting and relative freedom. I still think you'll only get a few select "power users" who use it to its fullest extent, people who would craft their own games given the tools. We'll see how that goes.
* * *
None of these sidesteps are perfect, but at least they've re-framed the holy grail problem in an interesting way and in many ways they've made it more manageable.
I still don't know what we're going to do for the thesis project. Maybe a combination of all these sidesteps?