Storytron – Interactive Storytelling

I spent the day digging into Chris Crawford’s theoretical lectures on generative fiction, finally arriving at Storytron, the procedural narrative engine he wrote in java. I feel like I’ve stumbled on an Aladdin’s cave!

Based on his earlier software Erasmatron and a precursor to his current effort Siboot – Crawford has spent the last quarter century working on the concept, with origins going back as far as the 8-bit game Gossip he made for Atari in the 1980s. Storytron was his ambitious, quixotic stroytelling platform that promised to revolutionize interactivity to an artform, allowing authors to put the human condition at the center of the action and demanding players utilize social intelligence over puzzle-solving and shooting things.

Not only does he describe the abstract concepts of what makes a narrative game more interesting (“people, not things”), he lays out the nuts and bolts of a turn-based procedural story generator where characters act according to their own motives, reacting to crises as they arise, and influencing other characters.


Authors were to define the actions and consequences, and create characters and locations suitable to any genre; ultimately creating the rules of each story world. At the heart of the software lies an event loop that manages a single dramatic action, assigns roles to all available actors and determines each character’s reactions, and progresses time incrementally – all the while keeping track of every character interaction, every emotion, every prop and location.

Drama Loop

Each Event contains a Verb that defines a single action and assigns roles to the Actors who are present: participants and witnesses, protagonists and victims.

– Roles
–– the conditions under which an Actor may assume a Role
–– the emotional reactions of the Actor taking that Role
–– a group of Options for that Actor for reacting to the Event

– each Option specifies:
–– which WordSockets will be used construct the sentence for that Option
–– the rules for what words will be chosen to fill those WordSockets
–– the Inclination of the Actor towards executing that Option

Reaction Cycle

Then each witnessing Actor cycles through the available roles and assumes one of them if appropriate, choosing the Option highest in the list according to their Inclination (current variables). This Option is added to the Actor’s list of Plans, and will be executed in its own cycle as a later Event by the engine. The reacting Actor can also suffer Consequences: ie, be injured, die, have their reputation besmirched, etc….

Events can be Highjacked by a Role if the conditions are met by one of the Actors. Plans can be prioritized, delayed, or aborted. Characters not only keep track of their own inclinations, but also have Attributes that can be perceived by other Actors — and these perceived values are not necessarily the same as the actual values. Characters have their own opinions.

Once each Actor has accepted a role and decided on their Plan, the Event loop ends. The engine calls the next Event loop based on the urgency of pending Events, and the process starts over again.

Meanwhile the engine tracks the opinions actors have of each other, which will influence their reactions to future Events. Whether a character lies or tells the truth, another character’s perception of his honesty may be more influencial than the actual truth.

Scripts are used to break down social situations into simple mathematical formulas. Not every situation triggers a visible reaction from the Actor. Often a brief facial expression is the only clue to what they are thinking. An Actor may create a Plan that requires immediate action, or create a plan for a later time and place. If Mary sees Tom fighting with Bill, she may decide to intervene or wait and tell Joan about it later.

Some events are witnessed by everyone on the Stage, while others are exchanged “cheek by jowl” only between the Actors directly involved. Some events reflect the private mental state of an Actor, and others are completely “under the hood” making changes to the story without the Actors’ knowledge. Actors can travel to other Stages (locations) either through boredom or to accomplish a Plan, and become involved in the Events there.

Deikto: a toy language

Maybe the most ambitious aspect of Storytron is an elaborate linguistic system that pastes together complete sentences with wordsockets, Mad-Libs style. Each defined Verb action references a subject and direct object, usually an Actor and another Actor or Object assigned at runtime, plus helper words to flesh out the sentences. Remarkably the sentence strings can be chained to convey complex relationships and ideas. An Actor can report earlier events to a character who was not there, or an actor can try to influence the actions of other characters by lying, threatening, or bargaining. More on Deikto…


An omniscient, non-Actor named Fate acts as a kind of narrator and deus ex machina, opening the story with “Once upon a time”, describing the Stage and announcing the arrival and departure of characters. Fate acts as a Greek chorus to comment on the unfolding action, and ultimately deciding when the story should end once the goal of certain variables has been maintained for a set amount of time, ie: living happily ever after.

This modular approach to story creation allows any number of Actors to be replaced with players. The Options that would normally be served by the engine based on a hierarchy of roles and reactions are presented to the player as a list of choices. The engine doesn’t actually need a player and can generate an entire story procedurally.

Crawford’s Folly

Ultimately Storytron was so complex that few people could understand it, much less master the interlocking systems to create a complete story world. Crawford is amazingly candid in his online lectures and interviews. Mercurial in his references and brutally honest with his opinions, he relates the shortcomings of the game industry to grasp his larger vision, and his disappointment in games praised for their graphics over their artistic content, as well as his own inability to inspire a revolution. It’s easy to see how he became a polarizing figure, ultimately walking away from the profitable gaming industry to pursue broader truths.

Although in many of his lectures Crawford gleefully criticizes the game industry for focusing on a niche market of hardcore male gamers – the results of a narrow profit-oriented evolution and lack of sophistication – he has also worked to manifest his own ideas into a working “game”, although he is loath to call it a game. Storytron is the applied science to Crawford’s grand theories, attempting to slay his five metaphorical dragons of interactive game design:

1. facial Micro-expressions to communicate real emotions
2. Character personality models: Goodness, Honesty, and Dominance
3. Linguistics
4. Narrative engine/ AI
5. Integrated Design Environment

Crawford readily admits he hasn’t solved the issues, but only pioneered into unexplored territory. Although Storytron is now offline (in a medically induced coma, as he says on the website), he freely offers up his theories, and has scaled back his ambition to a much smaller project focused on one specific story and a simplified language of picticons used to communicate between characters. Still he leaves his many journals and diaries for the next generation, like a treasure map to undiscovered cities of gold.

Chris Crawford on Interactive Storytelling

“Interactivity is everywhere, brandished like a flag particularly where transmedia is concerned. But to what extent are the proposed narratives really interactive? How does one create a truly interactive story? What are the key elements? Pioneer video game designer Chris Crawford examines and explores interactivity, its issues, and its future.”

• INTERACTIVITY is a cyclic process in which the User and Computer alternately Listen, Think, and Speak.

First Law of Software: What does the user do?
– a Verb List of what the software does defines the software.

A more detailed version of this talk in a 2-part video:
Continue reading “Chris Crawford on Interactive Storytelling”

Believable Agents and Interactive Drama

Believable agents are autonomous agents that exhibit rich personalities. Interactive dramas take place in virtual worlds inhabited by characters (believable agents) with whom an audience interacts. In the course of this interaction, the audience experiences a story (lives a plot arc). This report presents the research philosophy behind the Oz Project, a research group at CMU that has spent the last ten years studying believable agents and interactive drama….

For believable agents, personality is king. A character may be smart or dumb, well adapted to its environment or poorly adapted. But regardless of how “smart” a character is at dealing with their environment, everything they do, they do in their own personal style. On the other hand, the focus in AI is on competence. For classical AI, this has often meant competence at complex reasoning and problem solving. For behavioral AI, this has often meant moving around in complex environments without getting stepped on, falling off a ledge, or stuck behind obstacles….

The What-If Machine – WhIM Project

WhIM is an acronym for What-If Machine, a three-year research project studying the potential of computer generated fiction. The site has a series of whitepapers that explore many aspects of creative writing, including the development of compelling character arcs, generating dynamic stories around a given topic, and even motivational slogans and poetry. The theories involve machine-learning and the analysis of vast literature libraries, to invent interesting, influential stories that resonate with three-dimensional characters struggling with internal conflicts, and  story arcs built to illustrate characters’ motivated changes.

It’s heady stuff! I’m still getting through it. The researchers generally acknowledge that self-evaluation is in its infancy, as computers have no idea whether their generated plots are at all compelling, much less coherent. A white paper on generating a single-sentence ad slogan was able to serve a tossed salad of poetic buzzwords but nothing close to grammatically-correct language – generations of machine-learning will only make this better. Another paper suggests a system whereby characters evolve, transitioning through the conflicts of seemingly contradictory traits. An arc from “good to bad” for instance, or “loser to hero”, means a character already has elements of both traits which are tested or strengthened based on story events.

Already we have games with branching narratives and non-player characters (NPC) that are influenced by, and react to, the player’s choices – even the actions of other NPC, but nothing close to a completely computer-generated story. That said, the white papers are a treasure of research ideas.
Continue reading “The What-If Machine – WhIM Project”

Designing game narrative (link)

The game designers at hitboxteam have written an essay on why all cutscenes are bad.


How do you tell a great story with a game? The answer lies not in the plot and dialogue, but in the very structure of the game design itself. In this article, we talk about why storytelling needs to revolve around the interactive nature of the medium. Come and learn how to identify great game narrative, and to understand the importance of interactive – rather than cinematic – storytelling.

[relatedpost limit=5 position=none thumbwidth=240 thumbheight=140 size=20 color=#666666 hovercolor=#FFFFFF]