Stories as Systems

How emerging narrative tools are shaping the future of storytelling

Image of slide number 2

We are drawn to stories. There’s something essentially human about it.

With every new medium, every new technology, one of the first things that we see is storytelling, sometimes before the people developing the technology even really understand what it’s turning into.

Image of slide number 3

What I’m talking about today are things that already exist from an open source slant and things that could exist.

We have all of this emerging technology, we have new tools, we have new kinds of experiences through different forms of media, so there are a lot of new ways of telling stories.

At the same time, because storytelling is such a constant expectation in human experience, there’s a body of knowledge and history that we constantly refer back to.

Image of slide number 4

I think about this in terms of a paradox. Storytelling is both a systematic and highly structured activity but it also resists being put into a system.

It’s really hard to define. There’s no complete categorization, even though many many people have attempted to do this.

Although storytelling across cultures has reused common structural elements for thousands of years, this extraordinary consistency has defied explanation. There is no unified theory of narrative that can fully explain the rich complexity and affect of the stories we love.

Image of slide number 5

Often the foundation is thought about as the monomyth which originates in the kind of comparative anthropology pioneered by Joseph Campbell in the mid twentieth century.

It’s been thoroughly repudiated and rejected at an academic level but interestingly enough, it’s informed all of Hollywood and still has a huge influence.

A brilliant idea, but brilliantly flawed—that all stories can be framed as as a hero’s journey and that there are parts to that journey which are the same across all cultures and mythological traditions: the departure, the ordeal and return.

When you take this as a template and apply it to culture, you do see these patterns start to form, but when you put it under the microscope things start to disintegrate.

Image of slide number 6

It privileges a male perspective, it venerates the individual it really puts the focus on a certain kind of cultural tradition—often a patriarchal one—that that needs to be rejected a lot of the time in new forms of storytelling where people really are trying to branch out and do something different.

I think there are also some real problems with the foundation of this idea. When you start to break down the components of the hero’s journey you find that nobody really sees it in the same way.

Campbell’s original work describes eight stages to the hero’s journey. Or maybe we have eight stages? Or twelve stages?

Image of slide number 7

Then there are books that say seven basic plots. twenty master plots. One hundred and ninety five essential plots. Then two hero’s journeys.

So there’s really not a coherent consensus around this as a structure and that can be kind of hard to work with. When you look for something exact, something that you can rely on and systematize, it starts to evaporate.

Image of slide number 8

Another way of looking at storytelling is the really well-known idea of the three act dramatic structure.

This comes from Aristotle originally and again, while it’s a powerful guide to stories, it’s also really not exact. This has passed through so many different pairs of hands, and there have been so many ideas about it over time that it ends up saying everything and nothing at all.

Image of slide number 9

And yet another famous way of looking at story structure is Freytag’s pyramid which is similar to the three-act structure but it looks at rising and falling action.

It frames all stories as a kind of arc, so rather than a journey, it’s an even trajectory.

Image of slide number 10

There are thousands and thousands of books and how-to guides exploring all this. The whole field is really difficult to make sense of and everyone who tries to come up with a theory seems to see it in a slightly different way.

So storytelling resists systematization, despite being based on structure. We have this interesting paradox at its core.

Image of slide number 11

I’m looking at computational narratives in the context of games, interactive fiction and experimental art projects. These these are the aspects of media that I’m interested in understanding, and the traditions of storytelling can help guide us through these these new experiences.

Image of slide number 12

Whether our goal is to create experimental dramatic fiction or to develop game systems that dynamically respond and adapt to player input and simulation events, somehow we need to encode knowledge of narrative structures in a form that can be manipulated by our tools and algorithms.

That’s why there’s a need for stories to be systematic, because these kinds of experiences simply won’t work if we have no knowledge of storytelling. We’d end up with these flat open worlds where you can do anything and go anywhere but nothing really feels remarkable or distinct or dramatic.

Image of slide number 13

One of the big problems we’re trying to solve here is about presenting coherent and unified story structures inside dynamic or generative environments, so giving people who are entering these environments a way of actually responding and being present with the story.

People need to be able to grasp the story, understand it and follow it, and maybe even make changes to it that the author didn’t anticipate.

Image of slide number 14

We also need to have characters that act with genuine motivation, and have a personality that actually makes sense to us, both dramatically but also just in terms of reasons for their behaviour.

This is something we intuitively understand really really well. We’re wired to understand social situations and we we get frustrated very quickly when media doesn’t provide us with the sort of believability of character we find compelling.

Image of slide number 15

Then of course, we have to provide a compelling or entertaining or thought-provoking experience.

We need to to have a creative goal and we need to ensure that there’s a real purpose behind behind the media to make something that people will actually want to engage with.

Image of slide number 16

It’s really interesting when you start to look at this area of of narrative and storytelling, a lot of threads and tangents come back to this book which is a wonderful artifact of the 1990s and was really the first body of work that outlined this amazing vision that you could have these narrative worlds that people could go into and actually have agency and be able to to control and adapt and turn the stories towards their own interests and ideas, moving away from the idea that the the author or the great monomyth tradition defines what the story is and the audience are passively receiving that story.

Image of slide number 17

Let’s look at an extension of computational narrative, which is the idea of generating narratives rather than authoring them. So, using algorithmic techniques and other methods to generative narrative structures from data and patterns, rather than authoring them directly.

Image of slide number 18

What leads us here is the huge problem with the expansive story worlds and big ideas we’re talking about—that we end up with colossal state spaces, modelling environments with characters and rich detail to the point that we simply cannot afford to write enough content that would cover all the possibilities.

There’s also a tension between authoring the content and having authorial control over the story to create and shape a very particular experience, compared to an open-ended simulated world where people are going to do things that the author doesn’t expect.

Image of slide number 19

It’s not just about the quantity of what what gets generated it’s also about the variation. We could have a story world presenting unimaginably vast possibilities but if the people experiencing that story don’t notice any of the small differences between outcomes, it doesn’t really matter.

Rather than getting caught up in the state space and the range of possibilities, it’s important to think about what is meaningful and useful. What can someone get out of the story? What are the distinct things that that really make a difference?

Image of slide number 20

There’s a handful of things you could use as guidelines for the measuring expressive range of generated narratives.

When you have multiple endings, how do you reach those endings? What are the different pathways and decisions that allow you to see that variation?

You can look at the way that story beats and dramatic events are distributed. Do some stories have a lot more drama than others or is there a kind of even path where things are always happening? Are there huge differences in tragic outcomes versus comedic outcomes?

Then there’s the way that we can interact with the characters and the world and also the intensity and the difficulty in how people actually navigate through the generated spaces or story elements.

Image of slide number 21

I want to start diving deeper into the technological side of things by looking at plots and drama as a foundation for narrative.

This is, I think, a very common way that that writers and people working in linear media approach things. As we’ll see a little later on, it’s not the only way and there’s actually some surprises and some hidden stuff that hasn’t been taken into account. But I think plots are really well understood, and it’s a good way to orient ourselves in this space.

Image of slide number 22

You might be familiar with this, or vaguely remember having seen it before. It’s the infamous meme of J.K. Rowling’s original plot diagram of the Harry Potter series.

It maps out the plot as this giant grid of different events. There’s a timeline. It’s obviously not a spreadsheet, it’s on a piece of paper, and it shows a large table of interacting plot elements in this linear narrative. This is how she thought it through, this is how she made made the logic of the world that she was building work, so it’s quite often cited as a view into into how authors write these big complex stories.

Image of slide number 23

One of the first things that you notice about plots when you start to look at it systematically is that they are actually logical structures.

In order for a plot to make sense, there needs to be cause and effect. Things need to happen for a reason, and that reason needs to be driven in a temporal way (unless of course, your story is about time travel, in which case it gets really interesting and difficult, and this is also why time travel stories are so hard to get right).

We have an intuitive understanding of cause and effect and we bring that understanding to storytelling. So we need to know that an action in the story is meaningful, in terms of how it changes the future of the story. Then we feel like we can follow it as a thread.

Image of slide number 24

Interestingly enough, there’s a fairly well understood paradigm for working with these logical structures, and that’s the domain of planning algorithms, which goes back to the 1970s.

The idea of a planner is you start with a description of the initial state and the desired end state, and try to link them up. So you set up a world made from facts or propositions, you set up a bunch of logical information about how how the world is organised, and then you have an index of possible actions that you can take which can change that world state. Running it involves trying out actions in order to create an unbroken causal chain between the start and end states which can form a valid plan.

These ideas originated in the field of robotics and AI in the 1970s, but you can see that there’s a lot of application to plots and storytelling. The planner will give you a sequence of actions that lead you all the way from the start to the end goal, so to map this into a storytelling context, the goal becomes the conclusion or the the end point of the plot.

Image of slide number 25

Here’s an example of some symbolic AI planning that I’ve been working on. The ‘hello world’ in this domain is the problem of putting on your shoes. To get your shoes on, you need to put your socks on first.

Image of slide number 26

So with the propositions and actions set up, if run this, we can get one of a bunch of different plans, because all of these different orderings of the events and actions lead to the same desired outcome.

Putting on left sock, putting on right sock, putting on right shoe, putting on left shoe. That’s one way of getting to our goal. There are six ways in total, each involving a different order of actions.

Image of slide number 27

This is a visualization of the shoes and socks problem. You can see that the the partial order planning algorithm I was working with doesn’t actually define how those actions happen. It simply says there’s a causal link between these things, and here’s how you get to the goal.

But when you actually run the planner—and this is really important for generating stories— you actually do end up with a total order, so the story eventually does have to end up as a linear list, where one step happens after another.

Image of slide number 28

I didn’t spend all this time working on these planning algorithms just to solve the problem of putting on shoes and socks. I actually wanted to use it for something specific.

This is a snapshot of the current project that I’m working on, a generated novel called ‘Dreams of Golden Weather’. It’s about a secret computing project in 1970s in New Zealand with this concrete brutalist underground computer that takes up multiple floors of this building. And it’s about a technician who has to fix this computer which is malfunctioning.

The planning algorithm provides a whole bunch of possibilities for how that technician interacts with the computer, the story of the different events, how it breaks down. Then I have generators behind that that actually create the text.

I’m hoping to turn this into into a print book as well as a PDF shared online. There are many, many possible variations of this, so it’s kind of interesting to figure out how you actually distribute a book like this, when there isn’t really one definitive version of the work.

Image of slide number 29

What I have started to learn from doing this exploration—and this is very specific to this particular area of planning algorithms—is that a lot of what‘s already out there widely understood in this domain doesn’t come from art or creed or a creative practice, it comes from a computer science focus.

So there’s a lot of emphasis on optimization—finding the most efficient path to the goal, or ways of handling the combinatorial explosion of planning states in the middle of running.

Image of slide number 30

You can imagine in the 1970s, these algorithms were understood on paper but it was very hard to actually run them. Today we have a lot more memory, we have a lot more processing power—that’s a complete understatement—so it’s actually quite practical to use this kind of stuff now.

There are often weird issues when you add more and more complexity—even on modern computers, it starts to explode and slow things down, and there is huge amounts of research and understanding of that, but it’s not really focused on the other kinds of problems that you run into creatively.

Image of slide number 31

The planning algorithms aren’t set up by default to do what you need in storytelling. It works in a very baseline way, but once you start to get creative with it you run into these other issues.

With storytelling you actually need other affordances than optimization. You need to think about conflict or spectacle or dramatic arcs. Often, you don’t want the shortest path to the goal, you want a path that meanders around some interesting events. You want to have a story that that is meaningful which you don’t get when you just zoom through the state space and find a solution—those stories tend to be very boring.

Image of slide number 32

There’s a really interesting tangent of research looking into how can we extend these planning algorithms to incorporate these narrative ideas. Here’s a couple of of the fairly well-known ones—at least well-known in this incredibly niche community that I’m working in.

The idea behind the first one is to extend partial order planning to incorporate intentions, so that you can you can add logical information about the goals or desires of different characters and story elements. Often in these systems, you see you see ‘fate’ used as a kind of abstract character itself. That’s quite a heavy-handed but useful way to to drive the story—push it in the directions you want it to go in.

Then the next one is the idea of introducing conflict. That’s saying, what if we can navigate the playing space using conflict between characters or conflict in situations, so that we’re not just trying to find a path to a single goal through the story, we’re actually trying to balance the competition between the different actors.

Image of slide number 33

Here’s an example of this algorithm put to use. I haven’t got this far in my own personal work— this one is from the academic paper and it’s probably quite hard to make sense of, but we can tell the story.

It’s in a western setting. There’s a character called Timmy who gets bitten by a snake. Being bitten by the snake prompts his father Hank to travel to a general store and steal medicine from Carl. Hank wants to return to the ranch and heal Timmy, but this plan gets thwarted because the sheriff comes along and starts a shootout that kills Hank. Finally, Kyle ends up with the medicine and Timmy is dying.

There’s a tragic aspect to this ending, but there’s a lot more possibilities in the story space. You can run it multiple times and see these conflicts between characters be worked out in different ways.

Image of slide number 34

There is another way of looking at this area of of plot generation which is using a very niche form of logic programming called linear logic. I’m not going to go into too much detail on how this works as it’s well beyond the scope of this talk. But a really straightforward high-level way of understanding it is that instead of thinking about AI and planning, we can think about stories as logical or mathematical proofs.

Image of slide number 35

There’s almost like a dual—or a mapping—between between a narrative structure and a proof structure. We can potentially use logic programming with theorem-proving style logic to trace causal paths through different resources that can be built up, different actions that can be taken in the story world, and reach a goal state based on this proof search.

Image of slide number 36

So here’s an example of that. It’s rewriting ‘Madame Bovary’ which is a famous French novel by Flaubert. It’s a fairly well understood and well studied piece of literature. What the authors of this paper have done is model every major plot event in the novel in a linear logic form, then they’ve run proof searches on it to see how many different stories they can extract out of it.

It turns out they were able to generate hundreds and hundreds of different stories based on the core elements of the novel. You can see there’s a varying level of complexity and events. There’s a big set piece scene where Emma attends a ball, and in the first example she doesn’t attend the ball like in the original story, but she still has the same motivation to escape her stale, boring life and she ends up defenestrating. In the larger variation shown here, again there’s a lot of events that parallel the story but a few differences. Emma becomes sick and starts another love affair, and then she gets ruined—which is a parallel to the original story—then ingests arsenic.

So you have these story structures that resonate with the original work, but they end up in incredibly different plots and narratives.

Image of slide number 37

I want to dig a little deeper and go beyond the basic ideas of linear plots and talk about the actual atomic components of stories as they are told, through scenes and passages.

Image of slide number 38

We we talked earlier about monomyth and other high level ways of modeling narratives. They’re actually not as popular as you might think, given all of the books about stories and screenwriting and how to write a novel.

A much more practical and more recent and modern way of thinking about stories is this idea of scenes and sequels. Another thing you hear a lot is the idea of story beats—these small dramatic moments that drive the action forward. Thinking about the story in terms of what actually happens in it, rather than trying to apply this really high-level global logic.

So scene and sequel is really all about tension and release. There’s obviously parallels to music in this, and also parallels to software ideas like request and response, call and return. This is a really fruitful way of looking at storytelling because it is incredibly amenable to computational methods.

Image of slide number 39

Most people are familiar with branching narratives. This is often the first thing you think about when you hear the ideas of computational narrative or whatever thrown around.

The most important thing to know about branching narratives is that this is just one model of computational storytelling—it’s not the be-all and end-all. Often people make the mistake of thinking that anything that’s a nonlinear story, anything driven by computational media, is one of these branching narratives.

It’s not really like that. This is actually a specific form of storytelling with its own history of style, tropes and structural conventions that involve choices and alternate endings. It’s linear so it does follow that traditional idea of plots and sequences.

Image of slide number 40

One of the most popular tools for this is a wonderful open source editor called Twine which gives you a visual representation of your story and allows you to link multiple passages together. It follows the idea of the web, so it’s hypertext based, but it gives you this control over how you link the different components together and you can build enormous branches or very small looping story structures.

You have a lot of freedom with this but it’s also limited in the sense there the story is fixed. You put these nodes in, you link them together, you write the story in place, and that’s really all you do. So thinking about that in comparison to the dynamic plot generation, this is more of an authored model. It requires a lot of thinking about what you want the story to be as a fixed structure, and it doesn’t solve the problem of the combinatorial explosion because you have to end up making all of those pathways through the story work yourself, so you’ve got to write them at some point.

There probably are ways of extending it to do generative stuff but I think it would be quite complicated as it’s not really what the tool is designed for.

Image of slide number 41

The interesting thing about the model that Twine uses is that it treats the structure of a story as made of up a single common unit of scenes or passages. A lot of people like this, because when you put all these scenes or passages together as fragments of text, it’s just scripting.

This is really well understood by most people that know how to use programming languages, which is maybe why it’s so popular and widespread. Markdown and text markup languages like that are also another big influence on these methods, and there are various open source tools that are available for this kind of writing interleaved with imperative flow control.

Image of slide number 42

Ink is used for creating narrative games. It’s a text syntax for generating choices and nesting different fragments of writing that you can navigate through in a game context.

Its architecture is split into an editor and a middleware component that gets embedded into another game engine. The editor is used to write and generate scripts. The logic is fairly straightforward, based on nesting and hierarchy. So it doesn’t make any decisions about how you actually tell the story. You still need to put UI and presentation around it. But it’s a really good way of defining the choices and scenes that give you an executable map of the story.

Image of slide number 43

A similar—maybe more traditional— approach which is ChoiceScript. This is a little bit more imperative, you can see it’s got the if/else statements which are inherently part of the language and it’s influenced by Markdown with the bullet lists.

Image of slide number 44

A lot of game studios working with this technology have rolled their own. This is a big thing to do if you’re interested in visual novels or interactive comics.

There’s another open source scripting language for these types of projects called Ren’Py which is embedded in Python. Instead of using if/else branches, there’s goto statements. We can jump around from scene to scene. So if anyone who’s as old as me remembers playing around with BASIC in the 80s and writing little choose-your-own-adventure stories using goto, this is a very sophisticated and cool way of extending that approach.

Image of slide number 45

There are a lot of limitations to this the style of scripting scenes and passages, and it goes back to the combinatorial explosions. It’s a lot of hard work to manually write all the branches consistently and keep contradictions and plot holes under control. An interesting alternative for more interactive and open-eneded stories is the idea of storylets.

Instead of designing linear narrative structures, storylets are a pool of stand-alone story fragments that can be unlocked in any order. They can be used to gate gate access to certain parts of the story through scripting and decisions that are separate from a linear structure.

Image of slide number 46

Flowing from this is the idea of quality based narrative, where you have your entire story world organized into little pools of possible sub-stories that you can interact with. You can navigate through it in any order you want, using qualities, which are stats or properties that attach to you as a reader or player, as you achieve results or different things happen to you. Then those properties can be be used to unlock different elements of the story. Even though the story is fully authored, you have complete control over how you actually build that story as a reader or player, you can choose to explore whichever possibilities you want and improvise your own narrative and add your own layers of meaning to it.

A couple of examples here, one is from a game which is quite well known—it’s actually where the term quality-based narrative comes from—which is Sunless Sea by Failbetter, which is based on a storylet system called StoryNexus created for an earlier game called Fallen London. And there’s another example from a more experimental interactive fiction game Alcyone: The Last City, which extends quality-based narrative in new character-driven directions, for instance, applying it to gender, where gender actually becomes a mechanic in the game that unlocks different story possibilities.

Image of slide number 47

Let’s look at another way of framing narratives and understanding the computational media aspects of storytelling, through characters and conversation.

Image of slide number 48

There’s a traditional approach to writing parser-driven games, like the interactive fiction that you might remember from the 80s— Zork, Hitchhiker’s Guide to the Galaxy, even the early Sierra adventure games incorporated similar elements. The state of the art for this is a tool called Inform which is not open source, but you can use it for free. It’s an incredibly innovative system that uses English natural language and fact-based modelling rather than scripting.

The whole idea is that you interact with the world through a command line and operate on the world using verbs. So you type a verb in and describe the object you want the verb to act on. This is really good for moving through rooms, going north, south, east, west or opening doors, searching through containers, but it’s not really the basis of of storytelling.

As this quote by Emily Short expresses, it’s very hard to to make a compelling story out of opening boxes and going through doors. People have done it—some of the best parser games are extraordinarily emotional and affecting—but generally, there are other things that we want from stories, and a huge part of that is conversation and interacting with characters.

Image of slide number 49

The standard way this is done is through dialogue trees, which is where you just model the entire conversation in the same way as the branching narratives that we looked at earlier, but at the level of a conversation, on a statement by statement basis.

So how do you interact with with a character? Well, here’s all the possible things you can say, and then here’s all their possible reactions. Again, you can see how this this is really straightforward to understand and work with, but it doesn’t solve that that combinatorial explosion problem. Without generative methods, it involves a lot of hand authoring and the complexity can be difficult to manage.

Image of slide number 50

There are some interesting open-source tools that are coming out that are making these methods more accessible, but also addressing some of the structural problems.

Yarn and Yarn Spinner are inspired by and evolved from Twine. With this, you’re no longer modelling passages or story fragments, you’re now modelling conversations—a line of dialogue at each node, and the branching possibilities of that dialogue. This is a starting point for making really interesting narrative games. It uses that editor/middleware architecture, where the editing tool generates a script and then the middleware inside whatever game engine, whatever UI platform, basically reads from that script and represents the navigation and displays conversations to the player within the unique design vocabulary of the game.

Image of slide number 51

What I find interesting about this is what you can do with it once you have your branching dialogue modeled as a graph structure. We actually do start to see some some answers to these combinatorial explosion problems. Jon who’s been working on on the Yarn Spinner tool showed this amazing new tool that he’s been working on to take these dialogue systems and basically run checks on them. So you can jump to any state in the in the narrative, and then move on from there. It saves a lot of time. You can also statically analyze your dialogue, and automatically find unreachable branches, find bugs, find flaws in the conversation flow.

This is starting to demonstrate the potential of helping authors to collaborate with the computer and taking our possibilities further so we’re not just having to manage all these states in our heads. We can actually use the the capabilities of the data structure we put our text into to give us new kinds of superpowers. If anyone’s ever had to debug a complex dialogue tree, you’ll really appreciate what the benefits here are.

Image of slide number 52

So conversations are one of the major aspects of storytelling that bring narratives together with models of world state. There’s different ways of thinking about how we can work with these story worlds.

There’s the idea that we we can have agency that influences the story and changes the way we can select and follow the narrative, versus affect, which is really about how we influence the world state— how we handle inventory, containers and object relationships.

An another way to think about this is the interface between the world and the plot.

Image of slide number 53

In a story environment, we have actions that affect the world’s state. Decisions about which way to move, different verbs for searching, unlocking, opening, actions that modify things in the world.

Image of slide number 54

Whereas with agency, we’re talking about a single action—like asking a question or making a statement—with different ways of making it happen. We might say something, but how do we say it and how does that affect the the world in terms of influencing its social state? This is less about tweaking a world model, and more about driving a narrative.

Image of slide number 55

We can think about conversations as forms of narrative in their own right. We have scene and sequen, we have conflict and resolution, we have changing social relationships, goals, persuasion, knowledge that goes in and out depending on what is said and what is left unsaid.


There’s extraordinary possibilities to turn these kinds of things directly into story and game mechanics but it’s very hard, and I’m super interested in seeing where this can go, especially taking machine learning and natural language processing into account as well. The tool that’s potentially furtherest along here is a product called Spirit AI which is inspired by a lot of these ideas, but there’s not really anything open-source out there yet. So it presents a big opportunity and possibilities for people interested in exploring this further.

Image of slide number 56

Pride and Prejudice is a great example of the impact of character relationships and conversation. You can understand it as a plot, you can understand it as a series of events, but it makes so much more sense to look at it a social construct, a network of relationships. This social network is really the foundation of this story, the plot kind of grows out from this as the starting point.

The idea of modelling narratives in a plot-driven planning way has its limitations when it comes to stories like Pride and Prejudice. Instead, we can work through from the opposite direction, starting with the social structure and the relationships and conversation models, then realizing a plot an extension of those interactions, transforming the interactions of embedded characters into a compelling story.

Image of slide number 57

Finally, I want to talk about another area of narrative structure which is far less talked about and poorly understood, but it’s potentially the most interesting from a creative perspective.

Image of slide number 58

Rather than looking at narratives as structures of story events, we can zoom out beyond associating a perspective with a plot, and look at the totality of experience that someone has when they are engaged with a story as a subjective journey.

Image of slide number 59

But we can also flip that around and say it’s not just about the reader, but also the associations and juxtapositons suggested by the creator. Obviously we do write for readers or design for players, but what about the authoring experience? What about the ideas behind the story and what it reflects aesthetically?

How should we think about the fuzzy boundaries of creativity that goes into this work?

Image of slide number 60

Here’s a cover of Gravity’s Rainbow, which is a super complex and weird mid-20th century American Novel, some of you may be familiar with it. The structure of the story is about World War Two, the emerging military industrial complex and V2 rockets, and this cover actually reinforces this visually. The plot is famous for having this allusion to a parabolic arc, this trajectory of a rocket going up and down.

How do we get ideas like this into computational media?

Image of slide number 61

It turns out that novels are constructed in ways that are really different to what we assume, especially from what we’ve just looked at—all these different ways of systematizing stories and constructing narratives.

Here’s another example. This is The Luminaries by New Zealand author Eleanor Catton. The structure of this novel is really intricate. Each chapter and each sequence is mapped from horoscopes and the movement of the planets, and this corresponds to the characters and how they are situated in the story. It evolves through this sequence of conjunctions between these planets. As things move through the houses of the astrological star charts, we see these different different characters come in and come out and the narrative unfolds in a clockwork resonance with this this model that’s sitting behind the story

These are really difficult, weird and innovative ideas which work because they are executed to such a phenomenal level of detail. The visualization here actually shows how the astrological structure relates to the text itself, so you see these twelve parts slowly collapse down, with each section getting smaller and smaller until we reach the very end where it’s just a single passage.

It’s the height of creativity to combine this intricate and crafted structure with a dramatically interesting flow of plot-driven character interactions moving through it. It seems like this is just not really thought about by people outside of writing novels.

Image of slide number 62

One of the most interesting reference points that I found is this Australian book that came out a few years ago. It’s a series of interviews discussing how a bunch of modern Australian novels were written. It turns out the process is really fraught, really mixed up.

I guess it’s affirming in some ways, because the level of doubt and frustration these incredibly talented writers have is kind of comforting when when you’re feeling doubt and frustration about what you’re doing. To see the results they achived, you don’t necessarily think of the messiness and chaos of how they got there. Writers have a lot of really weird and interesting intuitive approaches which are worth paying attention to.

Image of slide number 63

A concept that ties all this together is what I call structural creativity. It’s developed in most part by open-ended pen and paper sketching, free association, juxtaposition of abstract shapes with narrative structures that are kind of aesthetic and vague, rather than direct and systematic. But this is a huge part of writing novels it turns out, and none of our narrative tools, none of our computational methods, really incorporate these approaches.

I think there’s a huge opportunity to incorporate some of this thinking. It’s really almost a different form of cognition, a different way of understanding storytelling—very much from the perspective of the creator and I’m interested in how make tools that tap into this and present options for writers rather than getting in their way.

Image of slide number 64

I was looking for ideas and experiements that kind of do tap into this. One of the the more interesting ones that I know of is exul mater by Jasmine Otto, framed as a recombinantorial fiction where there are these interesting UI elements, different properties and forms that you can select. By rearranging these things—both visually and in terms of text selection—you can you can explore this narrative. It’s free association, it’s wild creativity, it’s not driven by any of the narrative structures we just looked at, but it leads to some really interesting outcomes in terms of juxtapositions and flow.

Another example—more from a reading perspective than an authoring one—is The Ice-Bound Concordance by Aaron Reed and Jacob Garbe, which presents a shifting, sliding, transforming story experience that moves between print and screen.

Taking combinations of elements and recombining, rearranging, shuffling them like cards, is potentially one way that we can tap into structural creativity. Moving away from scripts and text editors, and exploring visual tools like dragging and dropping little dots into sockets and being able to rearrange words, letters and sentences as raw material—this is almost like reading tea leaves or scattering yarrow stalks or rolling dice. Not just focusing on the numbers that we get from that, but also the patterns and shapes that they form.

Image of slide number 65

Here’s another example of how stories are written. This is New Zealand novelist Pip Adam who revealing her working process, showing how she is visualizing her early work in terms of creating a novel and it’s totally distinct to her it doesn’t really mean anything to anyone else, but it shows how she’s not really thinking in terms of that JK Rowling plot structure—she’s thinking in terms of shapes and feelings and a journey that’s much more abstract and continuous rather than discrete.

Image of slide number 66

I’m interested in how can we incorporate this kind of thinking into our tools. In crossing the gap between our sketches and associative shapes and metaphors and the computational representation of stories. Obviously there are huge limitations in trying to build tools from a foundation of aesthetics. We use words like vibe and flavor to describe these vague and hard to think about things. To integrate them into computational media, we need systems, we need data, we need story shapes to be amenable to some form of logic and discrete structure, and this is very difficult.

Image of slide number 67

I think it’s worth facing these difficulties because there are huge possibilities here as well. There are no limits to how data can be transformed and mapped. We can take almost any data from any source, and if we’re creative enough, we can we can map it or rearrange it into a form that we can read from our from our computational tools.

I don’t see the limitations as as holding this back so much, I more see it as a call for thinking about how to build writing and narrative tools in a totally different way, so that we’re not necessarily writing linear text. There are lots of different ways that we can use data that we haven’t even begun to experiment with.

Image of slide number 68

The forefront of research into computational storytelling and narrative tools is about how we start to go beyond these graph structures, beyond these logical structures, and enter the realm of creativity.

Thank you!