Generating Stories

by Erik Schoster
in Research

tagged research, storytelling, composition

The most difficult part of this project so far has been trying to reconcile my desire to keep the entire system fluid and surprising while maintaining a coherent and compelling narrative.

What exactly constitutes a coherent narrative though?

Dangling above the stove, an unseen spider 
sways on the path of the heated air.

Is this compelling? Is this coherent?

Just next door there was Gary. Below the ground floor unit, 
his front door was nearly buried, out of sight down a cleanly 
finished flight of concrete steps hidden behind a pair of 
overgrown bushes. The warm light from a small cluster of glass 
bricks that served as his only window would glow in the darker 
months; the exhaust from a free-standing AC choking out a 
familiar gust in the brighter months.

We all knew Gary somehow. He was gone twice a year for several 
weeks at a time -- doing research, he said -- but mostly there 
he was. Buying some lightbulbs at Grace-Ann's, or dottering 
along the street like a cat on a leash, stopping and talking 
or watching or listening. More research, probably.

Charlie once saw him carry two overfull bags of unshucked corn 
down those well-kept steps to his basement home. He wasn't 
seen for three days.

Is this compelling? Is this coherent?

The crow didn't like the way the squirrel ran and ran. The 
squirrel would run up a tree, and across a roof, and through 
the bushes and along a wire. Always running and chasing.

One day, the crow stopped the squirrel and said "you shouldn't 
run so much, you will never see where you're going, you can 
never appreciate where you are." The squirrel didn't listen, 
and ran off up a tree, along a branch and onto a high wire. 
The crow flew up above the wire and watched the squirrel run, 
until he reached an exposed patch of wire, just touching a 
nearby branch.

Later, another squirrel ran past. "You shouldn't run so much" 
said the crow.

Is this compelling? Is this coherent?

While these little stories may or may not feel coherent or be compelling, they are all fixed. None of them have any fluidity on the page -- the worlds they might create are foregone, past-tense -- they are set. Read them again and there they are.

We could point to a degree of fluidity that's always present in these fixed narratives -- especially good ones. A good story invites re-reading, and the best always seem to offer up something new despite their ultimate plasticity. Most good fixed-media pieces offer this: a musical recording, a film, a TV show, a painting. We can come back to the good stuff. Is that necessarily a quality of the work, or just as much how we change and the way reading, listening viewing are creative acts?

Generative works are another thing. The work emerges from the rules that govern it, and those rules could change over time, or modify themselves. A complex structure can emerge from incredibly simple rules. The apparent structure itself can be modulated -- rewritten -- as simple rules unfold and interact. Retelling a story is a fluid process, the common thread from telling to telling comes from the underlying structure, and the gamut of materials & material processes being worked with. The elements of the story might stay the same, but the retelling can change the story in fundamental and (satisfyingly) surprising ways.

Take this short render of one of the cues from South Seas as a simple example. There is an apparent structure that emerges from this basic rule-set: Given a sequence of events, the events are most likely short (around a second or less) but there is a small chance the event will be long (several seconds or more) -- and the next event always begins around halfway through the length of the current event. It's almost nothing, as far as algorithms go, but it creates a simple phrasing that adds a palatable structure to what would otherwise sound like a monotonous stream of events. It creates a little story, clustering the events into phrases and adding a breath which acts like the consequent of a call-and-response for each phrase. Suddenly out of almost nothing, there's a little story being told.

Here's another short render, the only change to the cue script being that the lengths of each event are uniform and much longer. All of the details are different, but the difference we really hear is in that tiny tweak to the pacing of events. This is a simple example but this sort of simple modulation of structure can be incredibly effective, and completely change the way something feels. You could say it tells a different story with the same characters, the same locations etc.

And finally just for comparison, here's another render using the same event length calculations from the original, but discarding the long events for a uniform set of short events. It has a kind of plodding, relentless feeling that the others don't have.

Coherency emerges from the structure in part as well as from the act of listening -- sound has a capacity for the superposition of coherency that I think language just can't support. Basically it's much easier to roll the dice with sound without arriving at incoherent mush. Almost any structure imposed on the sound gets promoted to narrative. This is a beautiful thing, and one of the most compelling reasons for working with generative processes in my opinion. It's much more of a straightforward process to erase yourself from the whole thing -- let the process recompose the story in surprising ways, to take on a life of its own without having to coax it here or there. The whole process feels playful.

So how do you generate a coherent story? Lets take a couple famous examples of generative story-telling. Look at madlibs -- the approach is to use the grammar of a sentence as a structure to find spots for erasure. You scramble up the particulars: names, places, descriptive modifiers within an otherwise simple & coherent framework. Even this process erodes coherency -- the simplest story can become totally absurd just by shuffling these static elements around. The process actually relies on the erosion of coherency -- that's why the result is funny. That's what makes it fun.

Another famous example that comes to mind are the "choose your own adventure" stories I grew up with. These aren't really generative stories though -- they're multiple fixed stories, a series of pre-determined alternate paths. Once you take every path, the potential for variation is completely consumed. There isn't any real affordance of agency here, just a set of fixed variations that lead to a finite number of linear stories. They're just a compressed way of writing out several variations of a story, not a real platform for fluid storytelling. Still, that's not to say they aren't fun, but they don't have the endless potential of even a madlibs approach, although they maintain coherency by lacking true freedom in variation.

Video games seem to take a sort of hybrid approach. There is the total freedom of navigating in an open world, but stories are always told in a choose your own adventure style. There are a fixed number of predetermined narrative paths that can be explored. They may be far more numerous than the traditional choose your own adventure, but there's still always the hand-off moment where the fixed story takes over -- you have the agency to choose when you experience each part of it, and which part of which story you might want to participate in, but the story itself unfolds in a goal-oriented choose your own adventure style. Open this door and you get eaten by the bear! Open the other door and you find a great treasure! That said, I have extremely limited experience with games and I suspect there are much more creative approaches to storytelling in practice out there.

My working theory at the moment is that the success of a generative story lies somewhere in its capacity for interaction -- that is, the observer-participant needs real agency for it to have freedom and fluidity without totally destroying coherency (This almost comes for free with a piece of generative music.)

I keep coming back to the dynamic I remember from my middle school days of playing Dungeons and Dragons. The dungeon-master isn't really there to tell you a story, their primary role is to facilitate the agency of the players, and to illustrate the (dice-roll-driven but also pre-composed in part) reactions of the story-world to their decisions. This feels like an appropriate role for a story-telling-system that doesn't want to dictate the story per se, but create the potential for a story -- which is ultimately constructed in its reading through the choices and observations of its readers.

This all sort of begs the question: why not just let the story become incoherent? Why not embrace the incoherency and drop the mitigating strategies -- isn't that in the end, freer? More fluid? My feeling is that allowing incoherency to bloom is a valid strategy, but just a strategy, not really a workable approach in general. If you throw away all the signposts of structure and self-logic then that lack of structure eventually becomes the foremost element, and you're left with a set of endless variations that amount to the feeling of a single state: incoherence. This is a useful place to visit -- something like Stockhausen's listening to "the beauty as it flies" -- living entirely in the moment, almost static in its constant change. A useful place to visit, and re-visit, but just one potential form in a universe of possible forms.