Sam Lake on Video Game Storytelling

loladeutsch writes "What makes for a great story in a video game? Sometimes, with all the innovative development and cool graphics the actual story a game has to tell can get lost in the shuffle, or at least can seem to be an afterthought. When a game arrives on the shelves that presents one of the more engrossing stories we've seen in awhile, it's worth noting. Max Payne 2: The Fall of Max Payne has been recognized by many people with their heads screwed on straight as a benchmark in video-game storytelling. "

Comments

And the originals?

[divine_13]

How about the original games? Games like tetris don't have any story, and hell they do not need any. Chess - story? And don't say people do not play tetris anymore.

Our efforts in automating dynamic story generation

[Samir Gupta]

I'm in research and development at a well-known video game company, and among the many areas that we conduct research in to make games better, ranging from graphics to human physiology, the art of storytelling is surprisingly one of them. There have been many strides made in allowing computers to assist the game developer in storyline creation and extension, even on the fly.

Many modern implementations of computer-generated narrative -- video games being no exception -- are built from large, preset blocks of text constructed by the author, with either a set path for the plot to follow, or a significant amount of pure randomization to prompt variance in the experience. These approaches require skillful craftsmanship by the author and explicit identification of numerous story paths to an interactive and immersive experience.

Our work seeks to provide interactive narrative dynamically by using narrative theories to continuously adapt to the user's interactions while preserving dramatic content. We're investigating an architecture provides a dynamic run-time narrative, as opposed to a strict path that the user must follow to interact with the story. The main challenge of building a system like this is to preserve the story designer's dramatic vision while providing interactivity to the user. At one end of the spectrum we allow the users complete freedom to do whatever they please in their environment, without clear goals or limitations. This is much like online communities such as The Realm [Codemasters], where the users' goals consist of survival (or death if it interests them) and wandering the countryside while doing as they like to the environment and the other users. At the other end of the spectrum, we have the archetypical "good story": immersive, extremely detailed and well thought out. The reader is more or less constrained to experience the story exactly as the author intended, without deviation or variance. This is comparable to the epic film or engrossing book. We hope to straddle these two traditionally exclusive forms of narrative and provide an engaging hybrid.

In a nutshell, our system is composed of a story clip database that models the components of narratives and a set of algorithms that prescribe the combination of these components to build a story. The story clips are implemented as brief snippets of action or observation, with as little dependency on one another as possible. Each clip contains explicit pre- and postconditions that define limitations on when it can be added to the story and how the story is changed. A set of state vectors, called the story snapshot, is defined at runtime to permit fast verification of preconditions and simple accounting of postconditions. The algorithms that combine the clips form a narrative engine, which coordinates the concatenation of clips into a story that adheres to Branigan's model. In this system, we pay particular attention to the temporal relationships between story clips and define four dimensions of time that must be obeyed.

The narrative engine is the core for production of interactive stories. The run-time character, world, and narrative state vectors are compared to the precondition constraints of the clips stored in the database to extract the most appropriate story clip candidates. This comparison is accomplished by computing the vector difference between each story clip's preconditions and the current story snapshot. The difference vector is scaled by a proportional weighting vector to emphasize particular state vector components. The candidate clips specify potentially divergent plot options and some differences may rule out a clip completely, such as narrative and temporal cohesion. But variance is allowed and expected in the character states, as they describe one moment of time where each of the individuals in the story had one of many emotional combinations. The narrative engine orders the clips according to the quality of their match with the current story state.

The narrative engine uses the evaluated story cl