I've been interested in dynamically generating storylines for thirty years. Originally, I wanted to create a text game (like Infocom's stuff) except with actual evolving stories rather than — effectively — puzzles.

Pen and Paper Games

Another thought along these lines (when I was working on my paper RPG ForeSight) was the idea that many interesting stories are framed by historical circumstances (real or invented), and it would be helpful for the GM of a role-playing game to be able to sketch out history on the fly as background for a long-running role-playing game taking into account player actions without simply being forced to make it all up. (This based on the observation that the invented histories for most role-playing and literary fantasy settings were abysmally dull and implausible.)

Indeed, during the design process of ForeSight (I wish I still had my notes!) I had, at one point, designed a "random history generator" for planets. (Some tiny fragments of it remain in the star system generator in the published version of ForeSight.) In essence the planet's history would start when it was first colonised, then the history would resolve a generation at a time, with events that would change societies, fragment power-bases, bring in new waves of colonization, and so on. It actually worked quite well when I tested it, but was mind-bogglingly tedious to use. A computer wouldn't have minded!

ForeSight nevertheless contained a very sophisticated procedural planet generator which not only created a planet's geography, it also populated it with a moderately fleshed out society. ForeScene — the original background for ForeSight also included a procedural ecosystem generator which started with a food chain or web and then iteratively evolved the creatures in the web based on assumed competition. Again, this was tedious to use by hand, but produced interesting and usable results.

Computer Games

In the last few years we have seen games like Spore take procedural content to undreamt-of heights. Upcoming games such as No Mans Sky, Elite: Dangerous, and Star Citizen all promise even greater quality and quantity. Sadly, all the fireworks are cosmetic. There's no hint of narrative complexity (or, in some cases, any narrative content at all).

Procedural content is nothing new, although when it was used in early games such as Akallabeth or Elite, it was at least as much an approach to dealing with limited system resources (e.g. memory and storage) as for leveraging creative resources (i.e. game designers). Speaking from personal experience, game designers of the time had no problem designing large, complex quests by hand (using flowcharts).

After GDC 1993, I was invited to the home of a fellow game developer (Andrew Leker) where I immediately saw his gigantic storyline flow chart (he was in the midway through developing Alien Logic), which bore a striking resemblance to the one I had drawn for the game I was about to release. He told me he'd adopted the approach after talking to the designer of Star Control 2.

Well, many years have passed, and while the surface content of games has improved by leaps and bounds beyond text on a screen, or rudimentary pixel graphics, the story content of games has effectively declined since it peaked around 1997 (Fallout) and 1998 (Baldur's Gate). Chris Crawford has argued that the rise of the cut scene and production values in general has led to the demise of narrative choice and experimentation in video games. (On the other hand, the more recent rise of mobile games has led to a resurgence of experimentation and widely different approaches to production value — this is one of the reasons why I am targeting mobile.)

Narrative Spectrum

Today, we essentially have at one extreme so-called "sandbox" games, such as Grand Theft Auto V, which allow players to free-roam a large space, occasionally performing pre-scripted missions more-or-less in fixed order, and at the other we have "interactive cinema" games such as Dishonored where the roaming is much less free and the missions are performed exactly in order.

Marie Laure Ryan describes approaches to dynamic narrative in games as being top-down (the designer imposes a narrative on the player) and bottom-up (the game mechanics afford the player a "toolkit" of narrative fragments from which a story can be assembled). I think her analysis is insightful, but that bottom-up approaches are omnipresent (a game always has mechanics that will create events that can be folded into narratives; indeed Celia Pearce describes basketball as creating typical narrative arcs by its nature) — what I plan to do is take a middle path that takes explicitly designed "top-down" fragments, and assembles them into coherent stories based on player actions, so that the overall story is not imposed on the player either, and "bottom-up" elements are (at least sometimes) explicitly generated by and folded into "top-down" elements.

Put like this, it's quite clear that most of today's games sit on one pretty narrow end of a possible spectrum compared with, say, a game like Arena (1994) and Escape Velocity (1996) in which the majority of a player's missions were dynamically generated. The latter had far simpler missions, but the player could permanently change the setting (e.g. by taking over space stations).

In my view, a good story engine would have several crucial qualities:

A dynamic story engine will be at a major disadvantage compared with statically scripted stories for the main reason that cut scenes and voice acting in particular cannot be generated on the fly. This gets back to the objections to "cut scenes" raised by Chris Crawford and others.

Production Quality of Dynamic Content

By targeting mobile, I hope to take advantage of players' lower expectations of production quality, but also to design content with maximizing the production value of dynamic content as much as possible (to the point where static and dynamic content can be virtually indistinguishable).

A simple example of this approach is the Mass Effect series which uses a lot of high production value cut scenes despite allowing the player to pick his/her character's name (as long as his/her surname is Shepherd) and sex, and extensively customize the avatar's appearance. The cut scenes are live-rendered, so the custom avatar appears in them, and dialog always uses the character's surname (which seems plausible in the setting). Meanwhile male and female voice acting is provided for the player, so that his/her character's gender matches the dialog.

At the other end of the spectrum, in Final Fantasy VII you can name each of your characters as they are introduced. Since dialog is merely text on the screen the names you pick are easily switched in. The huge pre-rendered cut scenes can include voice overs, or dialog that does not explicitly use names, but the overall effect does not seem stilted.

These two examples merely address the issue of changing a character's name (and in the case of Mass Effect sex). More complex cases have also occasionally been dealt with. NeverWinter Nights offers a the scripter the chance to orchestrate NPC gestures and script behaviors such as walking around (for in-engine quasi-cut scenes) as well as providing a useful set of generic voice acting (e.g. an NPC might say "Excuse me, I need a word" aubibly, and then give details of a mission with text).

Even so, where all but the most rigid stories have very thin storylines, the competition isn't fierce, and it's certainly worth experimenting with ways to make dynamically generated stories seem to have the qualities of pre-scripted content by deft use of varied generic content. (E.g. a briefing might start with voice acted canned boiler-plate, and then transition to dynamically generated text or graphics when getting to particulars.)

background markdown 12/16/2014 14:53:06

Powered by FolderMark