Sunday, June 7, 2009

Fallout and The Procedural Skald


To many, the appeal of games is being able to meaningfully interact with a world, making decisions with lasting consequences. Unfortunately, providing this is a very difficult challenge for developers. The combinatorial explosion that even a small number of decisions creates requires providing outcomes for all possible combinations of decisions. Just creating that content is a tremendous amount of work; creating it in such a way that all outcomes connect harmoniously and do not feel unnatural is realistically impossible.

Commonly, developers address this issue either by providing small, isolated pockets of agency that do not interact or by providing choices that are largely illusory, with the core story hitting on the same beats regardless of the player's specific decisions.

Interestingly, Fallout 1 & 2 addressed this issue cleverly, but Fallout 3 abandoned that device for a weaker solution. If Fallout 3 had augmented this instead of abandoning it, I think it could have represented a more significant step toward minimizing the complications of these issues.

Most areas in Fallout 1 & 2 provided a number of different outcomes to the player's decisions, but these outcomes were almost always isolated from each other area. Beyond adjusting the player's abstract reputation (karma), events in Shady Sands don't change what happens in The Hub. Unlike other games that adopt that model, however, Fallout remembers the outcomes of various decisions the player makes. At the end of the game, the player sees a substantial epilogue that describes the impact their actions had on the people they met and the places they visited.

Fallout 3 provided a minimal version of this, describing in binary terms how the player acted at maybe a half-dozen locations. To be honest, this disappointed me far more than the content of the ending itself. Instead of an epilogue, the bulk of response to player actions comes from the game's procedural skald, Three Dog.

(As a note, this construct came from some comments made by Marc LeBlanc at his Game Design Workshop at GDC this year. He used "procedural bard," but I'm going to use the term "skald," as being a tabletop RPG veteran, "bard" has some degree of baggage.)

Three Dog serves as a dynamic skald, providing disc jockey poetics of the player's decisions. This gives the player's actions context, demonstrating clearly that their decisions are not made in a vacuum. Karma serves a related purpose, but responding to the player's actions with a real, human voice has the potential to be more poignant. The nursery scene at the beginning of the third act of Bioshock does the same.

Similarly, this allows for the world to reaction to generally immoral acts without just making all NPCs hostile. Given the reward structures of most games (rewards being experience points, money and/or equipment), making some number of previously friendly NPCs hostile really just gives the player relatively guilt-free access to resources. This only serves to further incentivize immoral actions, rather than causing the player to reflect.

Finally, Three Dog provides live feedback, instead of only at the end of the game, as the epilogue in Fallout 1 & 2 did. I understand that some didn't connect with Three Dog, but hopefully the value of a procedural skald construct is clear. The skald doesn't necessarily need to be a specific person or even a person at all either.

As I mentioned above, the lack of a substantial epilogue in Fallout 3 left the ending feeling quite unsatisfying, and many others felt the same way. Fallout 1 has a very strict time constraint, and Fallout 2 only loosened this somewhat. Without the strict time constraints and very pronounced goals of Fallout 1 & 2, exploration is at the heart of Fallout 3. As the end of Fallout 3 terminates any further exploration (until the Broken Steel DLC, at least), without any kind of epilogue that gives the player's actions greater context and meaning, I can understand why many found the end so unsatisfying.


Players seek the ability to affect change in the worlds they inhabit. Providing meaningful outcomes with cascading consequences explodes into an uncontainable number of stories. Even the most dynamic stories still must collapse to a few certain story points. But by using both epilogue summary of decisions and active in-game feedback, we can make the player's decisions feel more substantive and influential. As our procedural skald tells epic poems of the player's actions, they're providing what many players seek. They are saying something to the world and it will speak back.

Labels: , , ,

6 Comments:

Blogger Alan Jack said...

Would you say there was a barrier other than asset generation that prevents the implementation of deeper narrative interaction in games?

As crazy as it might sound, I see the trend towards procedural content generation as a potential floodgate-opening for interactivity. Once we can generate emotion in animation, once we can perfect text-to-speech generation, and once we can put those things together, we'll have an emergent environment that will allow for any depth of interaction.

June 8, 2009 at 12:57 PM  
Blogger Nels Anderson said...

@Alan By content generation, I meant story-related content. Mapping out a different response to every possible combination of player actions, and do so in the way that is still harmonious with the story the game is trying to tell, is the insurmountable task.

With just ten binary decisions, there are 1024 possible combinations of outcomes. That's what I meant by "combinatorial explosion." I imagine Fallout 3 features a great deal more than that and most of them feature more than two possibilities. Granted, not every single decision will effect every other, but a great deal could, especially in a game where the fiction doesn't support location and people in the world being relatively isolated.

The actual creation of in-game assets (animation, VO, etc.) is a secondary, although still significant, concern. Even if the procedural techniques you mention achieve workable fidelity (and that's a pretty big "if," IMHO), that still doesn't really address the first concern.

Obviously, this only really applies to story-centric games. Games that are primarily systems build to respond to player decisions handle this (e.g. SimCity or Civ), but manage by being far more abstract.

June 8, 2009 at 2:42 PM  
Blogger Graham said...

Honestly, I think the biggest issue here is tools; consider what 3D games looked like before there were quality/accessible 3D modeling tools around.

Also, it's not purely exponential. As you allude to in the OP, most story events are actually geographically constrained, and many possible player choices (in story, or Sim City, or whatever) are shades of the same activity, rather than unique activities.

I saw the tool that Bioware used for creating the dialogue in Mass Effect; it's still effectively the same editor as in the NWN Aurora tools. It's a good tool for making branching dialogue. But it's no good for making procedural dialogue. And for that matter, branching dialogue is no good for making procedural dialogue. In that case, you do run into the exponentiation problem.

So my contention is that once we have tools that allow us to create and view procedural conversation with the kind of flexibility and acuity as our 3D modeling tools, we will be empowered to create incredibly rich procedural narratives, and the tools will both help us cover as many unique scenarios as possible, while also reigning in the combinatorial explosion.

Grand Text Auto has lots of interesting stuff on procedural narrative (though I'm having trouble finding the specific relevant article I'm looking for...), but a recurring theme there is the clunky, text-based tools they use to engineer their narratives. I analogize it to using a scripting language for 3D modeling. It's simply ineffective.

June 18, 2009 at 1:27 PM  
Blogger Nels Anderson said...

@Graham Man, I'm really unsure about that. Maybe it's just the skepticism computer scientists naturally evolve, but I would be absolutely shocked if we will see procedural dialog that will even a passable substitute for human-generated speech in our lifetimes. There's *so* much work that needs to be done in natural language processing alone to make this a reality, I just don't see it happen any time soon at all.

Now that being said, I agree that better tools and methodologies would certainly help. Pat Redding alluded to this a bit here. More focus on this area would definitely be a big step in the right direction, IMO.

June 18, 2009 at 8:44 PM  
Blogger Graham said...

@Nels Again, I would have to draw the analogy to 3D rendering; back when everything was ray tracing, if you said that we needed to make perfectly realistic humans at 60fps, I'm sure the [computer scientists] would start going of about the complexity of skin and lighting and so forth. Hooray for the polygon and the shader!

Likewise, I agree; it's far out to imagine that we will have completely procedurally assembled conversations in games in the near future.

When I say procedural, I'm thinking in more of a general case; opposed to branching dialogue. For example, verb/noun substitution, or choosing hand-crafted responses based on a truth table. These things are already possible, and even used to some extent. The problem is, as I see it, is we have no convenient ways of viewing 'the whole picture' (e.g. to discover areas of the story with content gaps), and to quickly analyze how the generated dialogue would respond to different permutations of stats and choices.

I think the example that I would have to point to for 'what I mean' is Facade. It's all crafted content (and even voice recorded!) but rather than being assembled via branching trees, and placed on a narrative line, instead it pulls responses from its database based on player actions and current character statistics. But I've seen the scripting language that they made it in, and frankly it looks like 3D modeling in notepad to me.

June 19, 2009 at 7:08 PM  
Blogger Nels Anderson said...

@Graham Ohhhh, I get what you're saying. Sorry, I was thinking fully procedural dialog based entirely on inference. What you're saying is similar to what Pat mention and is definitely a direction that bears much investigation.

Branching dialog trees need to get gone. Dialog systems a la Mass Effect and (seemingly) Alpha Protocol are steps in the right direction, but small ones. I'd definitely like to see something even further afield than that seriously attempted.

June 19, 2009 at 7:41 PM  

Post a Comment

Subscribe to Post Comments [Atom]

<< Home