Paper Overview: Notes on the Use of Plan Structures in the Creation of Interactive Plot (Young)
Hi everyone, this is Jean-Noël from Immersive Consulting. Today we are going to talk about a paper which is a reference in the interactive storytelling field. It was published in 2000 by Michael Young and is named “Notes on the Use of Plan Structures in the Creation of Interactive Plot”.
The Story Structure
Everything starts with the claim that, in almost every narrative theory, you can find the same key feature in the structure of the story: it is characterized hierarchically. At the lowest level, you have the Fabula, then the Story, and finally, at the highest level, the Text.
The Fabula is the set of agents populating the storyworlds, and actions these agents can perform. These actions are linked together by causal and temporal links.
Then there is the Story. This is the plot which will actually be told to the reader. More precisely, it’s the subset of Fabula’s events that will be conveyed to the reader in a particular order.
Finally, at the highest level is the Text. It is simply the medium through which the story is being told. For instance it can be a movie, or a book.
In this characterization of the story, we can therefore find these following key notions:
- The notion of agents
- The notion of actions these agents can perform
- The notion of events linked by causal and temporal relationships
And there is the magic: these notions are not new for Artificial Intelligence! That’s more or less what the planning field is about.
A very simple definition of Planning
Let’s define very simply what planning is.
Planning allows us to define a domain - meaning a set of rules and actions, and from this domain, we can then define problems with an initial state and a goal state. The job of the planning algorithm is to find a plan allowing to go from initial state to goal state, using only rules and actions defined in the domain.
In fact, the idea that planning can be used to represent plot is not new and was already suggested way back in the 70s. However, back in the days, the idea was limited by the earlier models of plan.
Young suggests a new way, based on a representation of the causal structure of the story which is at the same time rich and formal, meaning it can still be understood by a computer. The idea is the following: looking for every plan representing the plot that could satisfy the story and author’s goal, instead of just creating one plan. We talk about searching through the space of all possible plans. Knowing all the possible ways we can unfold the story, we’ve got a global context about the plot actually being created. Remember this, as it will be important later.
This approach allows two things:
- The sharing of the plot control between the user and the system.
- The modelization of suspense as experimented by the user.
To do so, Young says you need two things:
A declarative representation of the actions possible in the storyworld. A program able to use this representation in order to create/modify/maintain the plan corresponding to the plot.
Without going into too much details about the declarative representation of actions, just imagine that you need to say what/who is involved in the actions - its parameters, what preconditions need to be satisfied for the action to be performed, and what are the effect of the action.
Monitoring the story
Let’s talk about the software used to create, modify and maintain the plan representing the plot. This program is itself divided in two parts:
The first part generates the plan, meaning here the plot. The goal is of course to generate an interesting story, according to the author’s goal.
The second part is in charge of monitoring the user’s behavior inside the simulation, and check if it deviates from the planned plot! Indeed, what to do if the user starts an action which could break the planned story?
A simple example
Let’s take an example. Imagine that, in the story, you have to escort a character, let’s name it Bob to a safe house. The arrival of this character to the safe house is the planned end of the story. Now, imagine that in all the possible actions in the storyworld, is the action to fire a gun at somebody.
I think you know where I’m going with that.
If you fire a gun at Bob, and therefore, kill him, that’s a problem because the story won’t be able to reach its ending!
Young proposes two ways to manage that. The system can prevent the user to deviate from the story. Or the system can adjust the story, based on the new state of the storyworld. It means modifying the plan so that a coherent narrative is still possible even after what the user did.
In our example, what could that mean?
In the case where the system prevent the user to concretize its action, it does an intervention. A possible intervention is simply to pretend that the user missed his shot, or that the gun was misfiring.
In the case where the system adjusts the story, we can imagine multiple things, depending of the possible actions in the storyworld and the author’s goal. For instance, if healing is a possible action, the system could ask another character to heal Bob. Or the story goal could change, and the user’s goal would not be to escort Bob to the safe house anymore but instead to escape from the police and hide himself in a safe house without being caught.
Virtually everything is possible.
Of course, in reality, it opens many questions.
When the system should intervene? And how? How the system can detect that the user is performing an action going against the story’s goal?
Then, should the system prevent the user to perform his action? Or should the system adjust the story?
If the system prevents too often the user to perform his action, the risk is to break his illusion of control in the storyworld, and therefore to break his immersion. On the contrary, if the system always adjusts the plot, the global coherence of the story will be lost.
It echoes one of the fundamental question, or conflict, in interactive storytelling. What is the most important: author control or user agency? When author or narrative control is total, you are in a book or a movie, and it’s not interactive storytelling. But when the user has total control, there is no more narrative coherence. It’s a subtle equilibrium to find.
The Modeling of Suspense
Let’s finish with the second proposition of the article: the modeling of suspense!
In cognitive psychology, there is a theory that a reader reading fiction is acting as a problem solver. It means that the reader is looking for solutions to the heroe’s dilemma or conflicts.
For instance, in a mystery novel, the reader will try to find who did the crime. In Romeo & Juliette, the reader will try to find solutions to the impossible love of the two main protagonists.
Cognitive psychology tells us that suspense increases when the number of solutions the reader can find decreases.
It’s possible to translate this idea in the planning approach this way: if we start from the state of the world the user has in mind, how much plans, stories, can we generate? That’s why it is very important to not only generate one plan, but instead look inside the space of all possible plans.
The idea here is to use the cognitive model of the user, about his perception of the world and the knowledge he has of the story, in order to modelize the space of possible plans with the user knowledge as a starting point. The more solution plans are found, the less likely the user will experience suspens.
This opens up many possibilities! For instance, we can imagine a system which, when it generates the story, has the suspense curve the author wants as a constraint. The system will use this cognitive model to favor the plans respecting the most this curve. It also allows the system to think about the information it may expose to the user to maximize or minimize suspense when needed.
That’s it for the presentation of this paper!
As I said at the beginning of this article, it’s very regularly cited by researcher because it really put again planning at the front stage as a well suited tool for interactive storytelling.
Today, nearly 20 years after, planning still has this predominant place in interactive storytelling.
We will have the opportunity to talk more about it in others articles!
Thanks for reading this article. If you liked it, do not hesitate to share or to get in touch!