next »

[gimymblert]

Are you guy interest in generating "meaning" with procedural generation stuff?

Generally, the main complaint about procedural generation is how nonsensical some results are.

It's a pitfall that comes from the structural approach of generation. We correctly identify the "syntax" that allow endless permutation but they also generate the meaningless one. "Colorless green ideas sleep furiously" is syntactically correct but makes little sense.

This is solved generally by clever design that set interesting constrain as implicit meaning ("exit" should not be close to "start" because of "exploration", difficulty should increase with depth of level to build "tension"). The problem still remain as the level feel "goalless".

But how about generating things with a motivational heuristic? A higher set of goal that impose constrain on the selection of parameter in the syntax structure. Basically a filter to enhance meaning through coherence of element.

How we would do such a things? Any idea?

I'm currently working on a crude story generation and try to do just that.

First I define story as "problem solving process".

Since it's suppose to generate story for human, I used maslow ladder as a starting point: a human problem is an unresolved need, maslow provide the classification for needs.

I categorize these needs in 3 domains:

survival and security as protection
social and affection as recognition
self actualization and spirituality as realization.

In game terms each one can crudely be represented by some meter:

"Health point" for protection
"Lovemeter" or "faction point" for recognition
"Rank", "xp", "achievement" for realization.

I identify 4 space of parameter along two axis (internal/external - process/state):

external + state = Spatial configuration
external + process = activities
internal + process = psychological (behavior)
internal + state = moral (standard and preference)

They define where concrete problem and solution are (space + need).

For exemple: to solve a spatial problem (jumping a cliff) it may take a psychological solution (courage). The "human problem" is define by the need, it became "escaping a threat" with "protection", "attracting attention" with recognition or "challenging our limit" with "realization".

Problem create stakes on character along two axis (fear/desire - to obtain/lose something) Combine with space (localisation) and need (direction) it would define a goal (solution):

Avoid ("fear to obtain" stake) shame ("moral" localisation) with  nerd status ("recognition" social need,direction of the problem).

With that I encode a theme like "greed lead to destruction" into "Spending resource in affection protect resource in realization and protection". Theme is therefore the dynamics that link the syntactical element of story problem i had just define.

With that in mind, I can generate a series of event that pick the right tools and odds to express that thematic meaning. I think most plot structure can be derive from that. It's not generating meaning yet, but it's one step closer

.

What's your own take? What do you think? What are the solution? Discuss 

[tsameti]

This is a topic that I find extremely interesting but personally insurmountable

Can human emotion be abstracted into proto-thoughts? Can you teach a machine to string these ideas together until they take on new meaning?

I like the idea of using symbolic communication to represent the process to avoid all the ugly things language can do when you try to generate it randomly. So your game communicates to the player with heiroglyphs, essentially. The combination of symbols become the game's internal language, communicating these abstract ideas with the player.

Three minutes into the game your character's 'gf' walks up to you and says:

self-reference stop-sign  heart player-reference
self-reference hungry apple-juice

Devastating.

[Ashkin]

Some resources:
http://pcg.wikidot.com/pcg-algorithm:procedural-puzzles-and-plot-generation
http://dankline.wordpress.com/category/game-stories/
There's also a lot of good information over at the PCG wiki.
Good luck!

[tsameti]

Those are cool articles, however it seems from some of them that a lot of focus is going into the question of how to avoid the problem of explicitly telling stories. It's more like they trust the players to make their own from the sequence of events that go down (the ludo-narrative).

The Dankline blog seems closest focused on the question of how to procedurally generate human interactions.

[The Monster King]

i definitely believe in procedural story generation, i also believe that its really hard to make without it becoming repetitive after the first few times. I think its very possible if you keep a scope that ain't too big!

[Xecutor]

Procedural content=>replayablity.
"Meaning"=>predictability.

It's really hard even for a human to tell a story with finale that isn't easy to predict.
The same for other kinds of procedural content.
If content is generated according to (relatively) strict rules of logic and common sense,
it would be ... boring?
In hand crafted environments it is possible to make atractive design decisions.
But I serously doubt that it is possible to make procedurally...

[gimymblert]

Hello!
Thanks for picking the discussion 

Procedural content=>replayablity.
"Meaning"=>predictability.

It's really hard even for a human to tell a story with finale that isn't easy to predict.
The same for other kinds of procedural content.
If content is generated according to (relatively) strict rules of logic and common sense,
it would be ... boring?
In hand crafted environments it is possible to make atractive design decisions.
But I serously doubt that it is possible to make procedurally...

I don't agree here. Take Dexter saison 4 final. It makes sense but was not predictable, the event was predictable from the stake of the stories, but it was only in the realm of possibilities, WHEN and HOW was unpredictable, and better, while the new stake it create is clear cut, the way this stake would be solve is wildly unpredictable.

Meaning only close options that does not make sense or are weak.

But actually you raise a new problem, which is not meaning but interest. Can we produce interesting option procedurally? But what makes something interesting? It should be logic but unexpected and still make sense emotionally (connect strongly to stake). If we solved "meaning" procedurally we have 2/3 of the "interest" problem solved.

@Tsmati
But what if you make a difference between generation of the stories and presentation of the stories?
Storytelling generation would be a subsystem of generating a story. It's a different problem. I would start with story before the telling.

EDIT:
Meaning = WHY

When I look at common Technics to generate story the focus on the HOW part of stories. The procedural wiki is full of HOW generator.

For exemple: Imagine if you have a character who want to <cross a cliff>, HOW generator would make the character (according to his abilities) jump, build a bridge, etc... But we can't relate to that, it's trivial and feel random.

If the have a stake it became more interesting: WHY he want to <cross a cliff>? maybe he want to escape a threat? Now we can relate to the situation, if he build a bridge he would be able to escape. That's also easy to generate, for a given stake we can generate obstacle to the goal. The solution is irrelevant, the obstacle as no other meaning than preventing the goal and build tension, the obstacle is "random" and without meaning in itself.

But real and good stories don't work like that, they would make the solution to <cross a cliff> meaningful too. WHY he would choose to build a bridge over jumping? the stake should permeate every aspect of the stories. Each obstacle should represent the struggle of the THEME, which the stake illustrate, rather than mere filler to achieve true stories standard.

Personally i'm chasing the later.

EDIT2:

We can express the problem very simply: Generate the string of WHY before the string of WHAT (or who) and HOW (and when + where).

[Xecutor]

IMO:
There is huge difference between interactive and non-interactive environments.
In a movie director will show only what is allowed to see.
Characters in a movie will act in a way that do not break story line.

In a game, where there is a tiny bit of freedom it's close to impossible to implement this.

If your procedural story will plan some unpredictable turn of events,
and there is no logical connection to what happened before,
it will break everything. If there is a matherial connection, there will be a high chance
that curious player will discover it too early...
Artificially prohibiting him from discovery will break impression too.
I'm a lil too sleepy now, I'll continue tomorrow

[mirosurabu]

That's where drama manager comes in, unless you are not necessarily interested in interesting stories of course.

If you just want stories, you can focus on actor class only.
If you want interesting stories, you will have to design a drama manager.

[gimymblert]

@Xecutor

Stake is what control character actions, it create a goal and a pressure.
It create a space where all action make sense, and a good story provide focus and avoid the creation of diversion.

For example: If some character is chased by a killer, if the killer caught him he dies, if he reach the cliff he's saved. The closer to the cliff, the closer to success, the closer to the killer the closer to failure. From that space there is a lot of choice possible, but it became obvious that stopping and starting a chat let the killer came closer.

Because there is a push and pull the whole scene is kept dynamic, choice are still available and it's not static. You can make it interactive. In game what need to be seen vs what is not seen is the fog of war, we can play with that too, we can also define the fog of war as a hammer space where we spawn resource and tools to control the scene (just like game does). The outcome of the scene (result and process) can feed back the generator to output further plot with player generated "why" (the choice it made in the scene).

You can still maintain logic and meaning compatible with interactivity.

[Xecutor]

@GILBERT Timmy

Well. It might work in some scenarios.
But how to generalize this?

For example. I want to generate side quest for a fantasy based roguelike game.
And by 'quest' I mean 'QUEST', not 'TASK'.
Word 'quest' somehow degraded in crpg genre over time.
'Quest' in crpg 15 years ago was something completely different than 'quest' in modern games.
So... I want to generate a side quest. Little story initiated by some hints, rumors or something like this.
But how to make non-obvious when player will encounter similar initiation for the second time?
How to avoid using templates?
If templates are used, player (who is playing for Nth time) will work toward recognition
of templates used instead of solving the quest iteself.
Partially this can be solved by making templates base really big.
But this is awkward solution.

Generating the rest of the plot based on player's decisions sounds good, but how you are going
to evaluate player's decisions? When there is X-nary choice of possible answers, it is easy.
But when you have some kind of sandbox word, it's going to be the problem of it's own.

I'm very interested in procedural quests and plot. But so far I haven't found anything acceptable in this field.
I'm afraid that in order to make this, one need to create huge database of what is logical and have sense and what is not.
If you want to make street full of strangers, each busy with his own business, and only one stands out, such a database is a must.
As well as database of 'cause-effect relationship' information.

[gimymblert]

@GILBERT Timmy

Well. It might work in some scenarios.
But how to generalize this?

You have some very interesting point who made me wonder, I don't know if I can respond completely to those but they help uncovering new problem. 

For example. I want to generate side quest for a fantasy based roguelike game.
And by 'quest' I mean 'QUEST', not 'TASK'.
Word 'quest' somehow degraded in crpg genre over time.
'Quest' in crpg 15 years ago was something completely different than 'quest' in modern games.
So... I want to generate a side quest. Little story initiated by some hints, rumors or something like this.
But how to make non-obvious when player will encounter similar initiation for the second time?

1. Tangentially it made think about "stake" as involvement in a story. Where is the stake in common quest and "task" in common video games?

a. Generally the hero is just a tool, is not involve in the stake of the quest. It allow the player to stay free and he would see quests just like more source (power, money, items), which they are. The stake is shallow and devoid of pressure (power climbing) and it does not relate much to the world (aside from being able to pass more difficult obstacle easier). No amount of "forewarning" solve that (rumor, hint, etc...), the quality may increase but the function remain. When author translate this trope in proper storytelling, they use the "money broked" adventurer trope that allow any random event to happens and pad the length of the series with self contain plot.

b. Sometimes there is sometimes a personal stake to the hero that makes the resource gathering meaningful and engaging. "Happiness" adventure game use that a lot, the hero must solve quest in order to raise some gauge that would solve his own stake and escape his fate. The quest as the exact same function than in point "a". They can remain self contain and be loosely or not at all link to the hero. It's useful from a production perspective it allow for a lot of flexibility in how the story is handle, a lot of subplot and only major plot a some gauge level.

c. Another approach would be to link the individual local quest link or "generate" by a higher world stake (for example preventing the end of world stake)(or many level of local to global stake). For example, not only the stake bring rewards but tangible consequence, not in an unfocused simulationist but by balancing the overall push and pull that affect everything. The closest to that is game with moral system, but moral system usually does not change the world but the access of source (functionally the same). If the world stake close back to influence the hero's stake it became much more interesting, doing quest became part of managing the world and yourself, and it became a slow puzzle like game, "forewarning" became then much more meaningful than sugar "immersion" coating. It also makes naturally chained quests. I thinks the simpler approach to this is to build a "world state machine" that shift local state according to his global state. JRPG use this kind of relation but in a "baked" way (you do not have choice), the heroes personal stake always relate to the world stakes and "quest" (story section actually) impact both, good example: latter part of chrono trigger.

2. If by quest you mean story the player can engage in, well it depends on the design really. What's the difference between a "task" and a "quest" but the quality of the execution? A task is straight forward IMO, do that get that, a quest is more nuance (backstory, empathy, evolution). It's how much of a story it is over the function. Simple quest at least state the why to differentiate from a task. The more immersion tools you use the better it is. From that point of view the best quest should be involving quest full of empathy, twisted, immersive, meaningful and interactive, they should not be necessarily obvious (you may stumble on it without knowing it by engaging in some event), and they may be dynamic (they are influence by external or prior event). It's should be also a matter of messing with the quest giving (maybe it's through information puzzle) and the obtainment of the rewards (natural consequence of the plot evolution instead of the quest giver simply holding it back).

How to avoid using templates?

But should we avoid using template?

If templates are used, player (who is playing for Nth time) will work toward recognition
of templates used instead of solving the quest iteself.
Partially this can be solved by making templates base really big.
But this is awkward solution.

IMO template are simply shallow syntax, we just need to deepen them by layering abstraction until it turn into a language. Then we create a system that can speak that language to create situation, but that's where we need meaning ... to have something to say. Basically template are shallow syntax with already embedded meaning, we need to separate the syntax from the meaning (and there I understand what I actually mean in my OP without even realizing it  THANKS  ). TO be more precise we need to separate the meaning (the why) from the lexical element (who and what) and the structure (the syntax).

Regarding variety, we are already abusing template when it came to human creation. "Roguelike" is a template, some element are strongly specified (alea), some are weakly specified (fantasy theme). If you have a deep enough system you can change some variable and keep the experience fresh (like all rogue like are not the same, or any game in any genre). The other way is to change the context, the same event in different context can mean vastly different things, if I use the "I love you" template with a girl I just met, it would not have the same impact as if I'm dying in the hand of my long time friends who I know love me, but to who I never display any feeling.

Generating the rest of the plot based on player's decisions sounds good, but how you are going to evaluate player's decisions? When there is X-nary choice of possible answers, it is easy. But when you have some kind of sandbox word, it's going to be the problem of it's own.

It depend on how you design your quest and your game, how much focus action can be versus how loose the interaction are. I don't think you can make a tack-on plot generation over a generic world without thinking about how to implement tools in that world that would support the story function.

The quality of the identification of those function and how to pass them through the world is equally important as the quality of the generative data. For example, if you want forewarning, how do have information flow working in your world?

In general it's a good things to create the world and the story generation around resource source and sink, wether is in world resource (exemple, money, health, monster spawn, npc, place, etc...) or just story resource (informations, emotion, tone, etc).

The more channel you identify from the world, the stronger the generation data can be link back to that world. You may have also channel in the other way (from the world to the generative system), it's only design and where you define constrain that will help you here, I think.

It's about generating/electing and setting the right element for the right task. We need to think in term of functions!

I'm very interested in procedural quests and plot. But so far I haven't found anything acceptable in this field.
I'm afraid that in order to make this, one need to create huge database of what is logical and have sense and what is not.
If you want to make street full of strangers, each busy with his own business, and only one stands out, such a database is a must.
As well as database of 'cause-effect relationship' information.

But isn't game already such a database by definition? IMO The database does no need to be huge, it need to be flexible so we can manipulate element in different ways. We need to think of generation as language rather than database.

[Seth]

I actually have been thinking about this quite a lot in the past year, but I lack the programming knowledge to implement this in any way at the moment.

My idea was to basically focus on interpersonal relationships throughout a 'house,' or really a wealthy aristocratic family, starting as early as the 1600's in America.  Every person would have a unique relationship with every other person based on three factors:

Love
Lust
Trust

Each would be scored on a range of say, -5 to 5.  -5 in anything means the polar opposite of the factor (-5 Love is extreme hate).  This way, I figured, there would be a fairly wide range of relationship types.  So you could hate someone and be repulsed by someone but still trust them, or love them and not trust them at all, or love them and be repulsed by them.

This seems to me like it might be a good foundation for melodrama--I don't think it would be wise to start trying to generate more subtle plots until you have the sexing and murdering going on.

Every character would own items, and have values and goals in their life, from "find true love" to "collect stamps" or something.  As an American story, the items would be important, as it gives many of these rich bastards something to collect--I think motives are extremely important, and I remember Tarn Adams saying he was planning on putting them into Dwarf Fortress.  The items would also affect their actions--like say an angry baker would kill the man who slept with his wife by cooking the man alive in his oven 

[Xecutor]

@ Seth
IMO in order to make something like this we need to formalize:
 - mood (or mental condition?)
 - personality
 - interpersonal relationship (I'm not sure if this 3 is enough).


Than we need to associate an event for each person to object and object to object (with person as mediator) interaction.
And make table of how each event affects each personality type in each type of mood.
Probably list of this events could be a starting point for cause-effect relationship table.

And than such a system can be viable subsystem of general 'procedurally generated story/plot' engine

@GILBERT Timmy
And than we add (hidden?) motivations and (template?) goals to the system above,
and some kind of an AI director that is managing information and events flows...
Hm.

Probably it's good idea to get some kind of little closed society to start prototyping.
Argh... Anyone have spare time machine?

[s0]

So... I want to generate a side quest. Little story initiated by some hints, rumors or something like this.
But how to make non-obvious when player will encounter similar initiation for the second time?
How to avoid using templates?
If templates are used, player (who is playing for Nth time) will work toward recognition
of templates used instead of solving the quest iteself.

I think this is the main problem. No matter how well you implement it, the "story" is always just going to be a "frontend" for an algorithm. Once the player finds out how that algorithm works and how to use it to his advantage (and he will), any semblance of caring for the actual content of the story disappears.

[tsameti]

I guess I was having some trouble figuring out what you were trying to say, GT.

I mean, you clearly have an idea of how one might go about attacking the problem. Take a psychological theory, or economics theory or what have you, and take the principles to create a model for your game world.

But you know what? I don't know if you can separate the meaning (which is the core of what you're trying to get at) from the presentation. The presentation is the way your end user is able to understand the language of what you're doing. If my interface with the game is an ability to shoot a gun, and a key to page through npc dialogue, it doesn't really matter how complex character motivations are. There's a dude to shoot and a wall of text to tell you why you're shooting them. But if you give the player access to these abstract meaning fragments (symbols: whether a music cue, or a text string, or a hieroglyph or whatever), you give them an interface to direct a portion of the way the story goes.

Do you need a complicated procedural generator to get chased by a dude towards a bridge? Not really. Where it becomes a powerful tool is when that same guy is chasing you because he thinks you boned his girl, and you empower the player with an interface to address that man's motivations rather than simply an impetus to run as fast as possible.

So as you formalize the language to communicate to your player, you are simultaneously constructing a language that can direct your gameworld behind the scenes.



Let me leave you with this additional suggestion. Players sometimes develop associations and favorite characters in unpredictable ways. What if your 'meaning generator' keeps track of which characters the player spends the most time and attention to, and purposefully integrates them more and more deeply into the course of the story?

[Seth]

So... I want to generate a side quest. Little story initiated by some hints, rumors or something like this.
But how to make non-obvious when player will encounter similar initiation for the second time?
How to avoid using templates?
If templates are used, player (who is playing for Nth time) will work toward recognition
of templates used instead of solving the quest iteself.

I think this is the main problem. No matter how well you implement it, the "story" is always just going to be a "frontend" for an algorithm. Once the player finds out how that algorithm works and how to use it to his advantage (and he will), any semblance of caring for the actual content of the story disappears.

Well, just because you have a story generator doesn't mean that whatever game you design for it should be played to death.  The game shouldn't outlive the interest the generator provides, just as no game should outlive any interest any gameplay mechanics or story or whatever other aspect of the game provides.

but I think it's hard to talk about 'meaning' here.  I think what we should be talking about is not just how to generate a story but also how can we use those stories as something the player can interact with.

@Xecutor:

Yeah I agree that mood, personality, etc. would all be necessary, I didn't mean to imply that something as bare bones as what I wrote would be enough.

[LiquidAsh]

I agree that "meaning" is a troublesome term in this context.  I get excited about how procedural content can be used to tell different stories in new ways, but I think the "meaning" should come from the games' designer.  The other option is to bring it out of the player(s).  This is more the Ender's Game direction.  This will probably be the holy grail of game design for many years to come.  It's hard enough for an author to put together a rewarding story or experience, but to extract the bases for one from the user is at least an order of magnitude harder.  As many of you have hinted at, I think it's important to see procedural content generation as a design-tool rather than a designer-substitute.

So... I want to generate a side quest. Little story initiated by some hints, rumors or something like this.
But how to make non-obvious when player will encounter similar initiation for the second time?
How to avoid using templates?
If templates are used, player (who is playing for Nth time) will work toward recognition
of templates used instead of solving the quest iteself.

I think this is the main problem. No matter how well you implement it, the "story" is always just going to be a "frontend" for an algorithm. Once the player finds out how that algorithm works and how to use it to his advantage (and he will), any semblance of caring for the actual content of the story disappears.

Well, just because you have a story generator doesn't mean that whatever game you design for it should be played to death.  The game shouldn't outlive the interest the generator provides, just as no game should outlive any interest any gameplay mechanics or story or whatever other aspect of the game provides.

but I think it's hard to talk about 'meaning' here.  I think what we should be talking about is not just how to generate a story but also how can we use those stories as something the player can interact with.

[XRA]

I've thought about similar things, more with what might be considered interactive narrative, I believe that some good first steps to conquer are being able to create narrative in a game which doesn't wait for the player, and also focusing on a small span of time, placing the designer/creator's message more into the entire supporting environment and theme, rather than depending on specific characters for the plotline.. the message is more of the enveloping backdrop and the interactions have less requirements on lengthy scriptwriting and complex dialogue trees to spell things out.
 
As a span of time, using some sort of measurable goal that realistically would take maybe an hour or so to complete, say for example the goal you first assume is that you must deliver a message to someone in-person who lives in the next town over but the backdrop is a war or conflict, something big that is affecting everyone in the game in some way.  Keeping in mind that the game and story doesn't wait for the player in order to progress or change potential events.

Another thought is narrative that gets shaped by the Player's assumptions, adjusting pacing and story arcs based on what the Player is doing to complete their goals... avoiding situations with story options in games where you feel like you have to take a guess or are are consciously forced to determine something that is obviously going to branch the game due to an answer.  I mean, games do the subtle way all the time in terms of difficulty- adjusting it based on your actions and proficiency at certain things, why can't the same be going on with how the story is unfolding, sure a familiar Player who has mastered the game could game the system and control it to their favor.. but personally that is one of the qualities of video games that I like.

I think successful procedural/interactive narrative will be games that are set against immediately identifiable backdrops or synopsis which first grab the Player due to their interest in the subject matter. The game allows the player to re-live that setting as many times as they wish by playing it, not requiring themselves to experience the same linearly crafted story each time and also not requiring themselves to invest 10+ hours in branching generative dialogue trees, but simply living in the moment of the game, being challenged by the game system aspects but also experiencing and uncovering a deep amount of potential storytelling, having all of their actions forming the overall experience in interesting ways..

feel like i'm rambling 

[s0]

I think this is the main problem. No matter how well you implement it, the "story" is always just going to be a "frontend" for an algorithm. Once the player finds out how that algorithm works and how to use it to his advantage (and he will), any semblance of caring for the actual content of the story disappears.

Y'know, I've been thinking about this statement and I've come to the conclusion that I might have made it a bit hastily.

I've been playing Dwarf Fortress for a long time now, and through extensive Wiki reading, forum lurking and of course my own experience, I think I know the game pretty well. I have a good idea of what happens when I do something and what strategies to use to survive longer (including game mechanic "abuse").

However, despite the fact that a lot of the initial "illusion" has vanished by now, I still don't play the game in a completely analytical way. I still genuinely care about my dwarves, curse at them when they screw up (and Lucifer knows how often that happenns) and cheer when they perform well. Hell, I'm even a little sad when they die.

And I think the reason is that DF includes things that are outside of the player's control and mechanics that can't be abused (well not without hacking scripts anyway). You're never 100% in control of your dwarves. No matter how well you organize stuff, they'll always have a "will of their own". Also, Goblin Invasions, caravans, and of course the layout of the map itself are (almost) never fully predictable. So it's a healthy dose of randomness keeps the game from turning into stale number-crunching and also what helps create all those interesting/horrible/hilarious situations DF is so well known for.