Tabletop roleplaying games (TTRPGs) are story generators, let’s start there.
What do I mean by that? First off I’m going to assume you know a TTRPG when you see one. TTRPGs as a medium share the general elements of a group taking the roles of characters (roleplaying) while following certain rules. Dungeons and Dragons (D&D) may be the most common, but there are tons of games and systems such as Powered by the Apocalypse (PbtA) , Fate, Forged in the Dark, etc.
Now what do I mean by story generators? As you might guess these are systems that generate stories. Typically the phrase story generator means some kind of artificial intelligence story generation system, like for example the Google text generator GPT-2 which the internet has been enjoying, or Talespin if you’re old-school.
Basically a story generator looks like:
The little dice icon on the left indicates some random value or input (which could be many things including initial text input from a person), which some sort of function (either written by a person or machine learned) then uses to create a story.
Now what happens if we instead think about this pipeline in terms of TTRPGs? It might look something like:
In this case some random value (often the result of actual dice) is translated by players (the six abstract blobs) and the mechanics of the game (the gears) into story. For example, if I roll a “Hack and Slash” in Dungeon World (the basic attack) and get a 6 from the dice then it’s up to the mechanics (6 means failure mechanically in Dungeon World) and the players (the game master in particular) to turn that number into story (a big goblin hits me, perhaps).
Why does this matter? Well, given that we can treat TTRPGs like story generators that means that there is likely overlap in story generator design and TTRPG design. Specifically, I argue that this makes TTRPG design an example of Generative Design. Generative design is the process of creating generators that produce content (be that content stories, music, art, games, etc.). This area of design has processes, theories, and a vocabulary.
In this post I’m going to focus on generative design vocabulary, and the ways I see it relating to TTRPG design. I’m hoping this can be useful to folks interested in both TTRPG design and generative design broadly.
I originally intended this to be a pretty short blog post, but then I committed a mortal sin. I tweeted.
Hey generative design friends, what are your favorite words of phrases that we use to talk about generators?, and why? I’m talking possibility space, expressive range, grammars, etc.— Matthew Guzdial (@MatthewGuz) March 31, 2019
And I was absolutely overwhelmed with the response. I only have space to cover a few of the suggested terms that tweet prompted (this post is already huge), so I recommend reading through the responses if this intrigues you.
Possibility space is a way of conceptualizing all possible outputs of a given generator, treating each possible output piece of content as a point in a massive space. For any generator that outputs anything more complicated than a few numbers it’s impossible to actually visualize this space, but it can be a helpful framing device for considering a generator. Where is this space more sparse (less common outputs), what are the edges of this space like, and does this output space match your intentions as a designer?
In tabletop games we can think of the possibility space as representing all possible play experiences, with all possible sets of players. As a designer you can use the notion of possibility space in many ways. Maybe to talk about how “general” a particular TTRPG is, what’s the breadth of stories it can tell? How expansive or focused is this space compared to other games? Or alternatively, you can use this notion to think about the edges of the space. Maybe there’s a story you didn’t intend a particular TTRPG to be able to tell, but that’s technically possible output. Or let’s say you hack a TTRPG. How can you shift these edges to include a new kind of story?
As a quick example lets talk about the game Dungeon World (A sort of Dungeons and Dragons equivalent in the PbtA system). When a character’s hitpoints/health hits 0, they make a move called “Last Breath”. A player rolls two six-sided die (2d6) and on a 10+ they’re alive but in bad shape, on a 7-9 Death appears and makes a deal, and on a 6- they’re marked as Death’s own (they don’t die now, but soon).
It’s no fun to have a player character suddenly die, especially if it’s unexpected, dropping all the storylines and non-player characters related to them. The 6- “marked ” result attempts to get around this by leaving it up to the game master (GM) to determine a narratively-relevant death “soon”. But sometimes that’s not possible and it can interrupt the flow of a campaign.
What if we added another rule that on a 6- a GM could offer (with player permission) that an NPC close to the player character sacrificed themself in the player character’s place? This would expand the possibility space of the game (as adding choices often does) as it would lead to new possible stories/play experiences that weren’t possible before.
You could then ask: in what ways would this impact the play of a typical session? In other words, how frequently would this new mechanic appear throughout the possibility space? Is this new space a better reflection of the kind of heroic adventure story Dungeon World wants to tell? These are the kinds of questions possibility space helps us ask.
Noise is a term for some kind of random signal. This corresponds with the dice icon in the pipelines I included up above. In generative design different sources of noise are drawn on to generate different things. For example, Perlin noise, which due to its appearance lends itself to terrain generation.
A concept in generative design is choosing a source of noise that will allow you to generate the kinds of thing you want to generate more easily (like Perlin noise for terrain). Because of the particular qualities of this random noise it is better at representing different kind of things (meaning the designer has to do less work to make the output look right).
The same notion comes up in TTRPG design, generally focused around the probability distribution of different dice mechanics (shout out to Mark Nelson for prompting me to include discussion of probability distributions). Let’s take a look at a quick example.
Above I’ve included a graph comparing the probability distributions of the basic dice rolls for two games: Dungeon World (DW) and Dungeons and Dragons (D&D). DW (as previously mentioned) employs the roll of a 2d6 for most basic rolls, while D&D employs the roll of a single 20-sided die (1d20). Thus for example, it’s impossible for someone rolling a flat 2d6 to get a 1, but its most likely they’ll end up with a 7 (the minimum value required to succeed in PbtA games). But because there’s only one dice rolled in D&D, there’s an equal likelihood of any of its 20 values (thus the flat red line).
In both games players have modifiers that are added or subtracted to the final dice values, but this only shifts these probability distributions left or right, it doesn’t change their shape or size.
What do these two different types of noise do for these different games? Well, in DW (and all PbtA games), there’s a higher likelihood that the player just barely succeeds or fails, and less likely that they get the lowest or highest values. This comes from the original Powered by the Apocalypse game (Apocalypse World), which is meant to tell stories about people barely holding on after the apocalypse. As you play, you get better at surviving, so the probability distribution shifts to the right, but failure is always possible.
Comparatively, D&D is more unbiased (despite how it feels sometimes), meaning the same hero who just rolled the famous “natural 20” has an equal chance of next rolling a “critical fail”. This leads to the somewhat slapstick humor you can find in D&D.
Colors of Noise
This notion that different types of noise are better for different tasks is sometimes called colors of noise, adapted from signal processing (shout out to Isaac Karth). The idea here is that features or attributes of different sources of noise lend themself to generate specific things.
This notion comes up in TTRPGs as well, where there exist many sources of noise besides dice. For example, things like drawing from a deck of cards, like in The Quiet Year. This gives a uniform distribution for the very first draw, but with each draw the probability distribution changes. This gives an implicit shape to potential output narratives.
There are even more sources of random noise in TTRPGs outside of cards and dice.
Ten Candles is a horror game that employs the melting of actual physical candles. As an alternative Epidiah Ravachol’s Dread is a horror game that uses a literal Jenga tower (and its eventual collapse) as a primary mechanic. Bully Pulpit Games’ Star Crossed uses a Jenga tower as well, but in the context of forbidden romance. In this way the “noise” of the Jenga tower can be interpreted into two different (though related) types of stories.
There’s even a line of games that transform real life events into story creation prompts like Takuma Okada’s Conches and Cameras, PH Lee’s Alone on Silver Wings, and many of the submitted games from the Short Rest game jam.
There are no good or bad sources of noise, but there may be sources of noise that are more or less appropriate to the kind of possibility space that you may want for a game.
Instead of random noise as the basis of a generative system a generative designer can create a grammar. Like language grammars you may be more familiar with, grammars are composed of rules for how things are allowed to connect. For example, you might author chunks of a video game level and have grammar rules determine where to place each chunk. This is what happens in the game Spelunky. Check out this great video from Mark Brown on how that works.
As an alternative, in a story generators you might have rules for how different events can be strung together. For example, a character must be dead before they can be resurrected.
There are TTRPGs that use grammars instead of dice to create stories (though they don’t use this language). Employing grammars instead of dice can allow you to more closely structure potential play experiences, cutting out some or all random chance.
Grammars as Moves
Perhaps the most clear example of grammars in TTRPGs is the Belonging Outside Belonging System by Avery Adler and Benjamin Rosenbaum. In these games there’s no dice and no game master. Instead players have characters who can take vulnerable, regular, or strong moves that push the story along. However, strong moves require a token to use and vulnerable moves generate these tokens.
As an example, let’s look at Exodus, an excellent Belonging Outside Belonging game by Erika Shepherd. In this game The Guardian character has the vulnerable move “Escalate a Situation”. If they use that, they can gain a token, which will allow them to use a strong move like “Defuse a Tense Situation”. It should be fairly clear how you might string these and other moves together to create narratives, but these moves also implicitly shape the kinds of narratives a game can tell. For Exodus, that’s the story of a transangelic road trip.
Other games outside of the B.O.B. system give players moves that work like grammar rules to piece the story together. For example, John Tynes’ Puppetland gives each class (type of puppet) a specific set of moves it can do. Like the shadow puppet can “dodge things thrown at them by turning sideways” and “move quickly”. While this doesn’t quite shape the narrative in the same way (the game has a game master for that), it does clearly characterize each character and clearly specifies their narrative impact.
Alternatively, Ben Lehman’s Polaris structures play with a graphical grammar you can see here. This gives play the feeling of a ritual, which fits the tone of the kinds of stories the game is meant to tell.
Object and Action Grammars
Another way of structuring a TTRPG around a grammar is to use interacting with an everyday object that already follows particular rules. For example, Takuma Okada’s Chess: Two Kingdoms uses a literal game of chess. This naturally structures the game to tell the story of two warring kingdoms. Alternatively games like Natalie’s Commune and Mathemagicians employ the construction of a jigsaw puzzle and math puzzles (respectively) in much the same way.
Grammars for Future TTRPGs
By understanding these games as grammars there is a massive body of work and theory you can apply to them. As just a quick example, the notion of a terminal (ending a sentence) vs. non-terminal rule could be understood as referencing moves that end scenes vs. rules that keep scenes going in a grammar-based TTRPG.
There are also many types of grammars I haven’t seen in any TTRPG, but do show up in generative design. For example an L-system (often used to generate plants) or recursive grammars generally (where rules can call themselves).
As discussed above the possibility space of most generators cannot be visualized due to its size. However Gillian Smith’s Expressive Range can be used to visualize the possibility space of any generator. Essentially expressive range looks at the possibility space in terms of very particular metrics or measurements.
For example, for output of a platformer video game level generator you might care about the output levels leniency (a bit like challenge) and linearity (how much it moves along the y-axis). These two measures then give you two numbers, which you can use to plot on a standard axis (as you can see in the figure above). Every point of the figure is a number of games that have that particular leniency and linearity. So we can get a sense of the possibility space, the possible output levels of this generator (e.g. the levels from this generator tend to have a middling level of both leniency and linearity).
This is much harder in TTRPG design, since you’d have to somehow measure every possible output story. However, in the episode Tools to Make Emotions of the podcast Stop, Hack, and Roll, James Malloy uses a very similar approach to consider and visualize the emotional space covered by his in-development game Space Between. In the episode he describes tagging the cards of his game with the particular emotions they are meant to evoke, which allows him to better grasp the possible emotional space of his game and adjust things to better match his design vision.
Proposed by Graeme Richie and popularized by Simon Colton, the notion of the inspiring set is that it’s a collection of existing artifacts (story, music, etc.) that inspire you to make your generator. The idea being that you design a generator such that its possibility space contains artifacts like those from the inspiring set (or even that the generator is able to generate output exactly like these artifacts).
This idea comes up in TTRPGs, though generally as a more loosely defined notion of “inspirations” for a particular game. The important distinction here is that the inspiring set for a given TTRPG would be the artifacts from its inspirations whose stories are similar to the kinds of stories that TTRPG can be used to tell.
Perhaps the clearest example of this would be Riley Hopkins’ Interstitial: Our Hearts Intertwined. This game is directly inspired by the Kingdom Hearts series and is meant to be used to tell stories like those from that series. For example, the stories from the actual play podcast of the same name, in which Riley and their friends play the game.
The concept of inspiring set gives you a very clear way of determining when you’ve succeeded at designing your generator, when its output matches its inspiring set.
Shout out to Mike Cook for prompting me to include this!
Isaac Karth suggested what he calls landmarks to indicate particular noteworthy output in a possibility space. Perhaps you’d want to identify this output because it tells you something interesting about the generator. For example, that a TTRPG meant for one purpose can be used for an entirely different one (like a game about a medieval Kingdom being used to tell the story of a space convoy).
Notably if your output possibility space has no landmarks or too few you might have what Kate Compton calls the 10,000 bowls of oatmeal problem. Sure your generator can make a lot of things, but are any of them interesting?
Mike Treanor reminded me of the notion of reifying theories or that “a generator is a concrete implementation of a theory of the thing that it generates”. This is related to Mike and his collabor’s notion of proceduralist readings, that the mechanics of a game are themselves an argument. For example, in a game about modeling societies like Civilization or SimCity, the mechanics of the game include arguments about how these things work (or how the developers think they should work).
The same can be said about TTRPGs. For example, how a TTRPG simulates combat tells you a lot about the values of its designers. Take Dungeon World again. In Dungeon World there is no ability to block or defend. In fact, being passive in any way allows the Game Master to make a move against you. You could read this (and this is true for many PbtA games) as being an argument that “action is better than inaction”.
This does not make this or any other argument expressed by any simulation right or true, or even what the designers intended. Instead it can be a way for a designer to think about the possible readings of a particular game and how changing the game might impact these readings.
I wrote this post to survey generative design vocabulary as it applies to TTRPG design. I think this lens can be helpful to both TTRPG and generative designers broadly. While I’ve barely scratched the surface of the parallels between these areas of design I’ve hopefully at least demonstrated the ways in which this lens can be helpful in terms of communicating concepts and conceptualizing design.
Huge thanks to everyone who replied to my original tweet and to the users of the Fans at the Table, Interstitial: Our Hearts Intertwined, and Stop, Hack, & Roll Discords. They were invaluable in helping me identify particular TTRPG examples. So sorry I wasn’t able to use them all in this post, but I’ll definitely write more on this topic.
If you’re a TTRPG-focused person interested in learning more, I highly suggest Emily Short’s blog and in particular this video of her talk for PROCJAM 2016 (an annual generative design jam run by Mike Cook). Chris Martens has an excellent blog post where they include slides on a survey of story generation techniques. Further you can’t go wrong checking out the work and blogs of Kate Compton and Darius Kazemi.
If you’re a generative design or creative AI person interested in TTRPGs, I’d recommend the academic work in Analog Game Studies. However, a lot of this design community’s knowledge isn’t documented or at least not in academic venues. However, I’ve found some of it in podcasts, such as Stop, Hack, & Roll, Riverhouse Games’ Game Closet, and the RPG Design Friends podcast (on hiatus as of this writing).
If you’d like to comment, critique, question, or complain please feel free to reach me on my twitter.
Thanks for reading!