Taking an axe to dialogue trees

Who was there: Marleigh Norton, lead interaction designer at Singapore - MIT GAMBIT Game Lab.

What they talked about: As sophisticated as the visuals, physics simulations, and interfaces for games have become, there are still elements of the medium that have all but eschewed advancement over the years. Norton pointed to interactive dialogue as an area that hasn't changed too much, with games most commonly leaning on the age-old multiple choice dialogue tree for a variety of reasons.

Norton emphasized that she doesn't want to replace dialogue trees entirely; she just wants to offer alternatives. She also stressed that she wasn't a programmer, so high-tech solutions like artificial intelligence that recognize and respond to natural language aren't what she was looking at, either. She structured her talk around four approaches that at least use a more complex form of multiple choice than traditional dialogue trees and won't freak out game programmers.

The first idea was small talk. The rule of thumb when meeting people is to stick to small talk; stay away from sex, politics, and religion with people you don't know well, as it can lead you into trouble quickly. Turning that into a game mechanic, Norton looked for a way for gamers to consider the eloquence with which they have to discuss a topic to avoid having it blow up in their faces. She pointed to PaRappa the Rapper as an interesting way to handle the issue, with a beat-matching game determining how effectively the player is communicating.

As an example, she brought up a conversation with a barista at a coffee shop who happens to be wearing a T-shirt featuring the Flying Spaghetti Monster (an atheist answer to creationism). Asking for another napkin would represent a low-risk interaction, and the beat matching would similarly be simple with just a few face buttons on a standard controller. However, a line like, "A fellow Pastafarian! All hail his noodle-y appendage!" would require a more complex series of button pushes and directions to reflect the riskiness of such an interaction with a complete stranger. The player's performance on a risky line could lead to a subsequent dialogue tree that gives players the opportunity to come out ahead or dig themselves in even deeper.

The second idea Norton discussed was based on eavesdropping. A different take on the standard notion of player-controlled dialogue, this approach doesn't involve dialogue with the player at all. All the player does is listen to conversations, and the actual interaction comes through other mechanics.

With a nod to Agatha Christie, Norton suggested a scenario with the player in the role of the butler, who just happens to be the killer as well. However, a detective is undercover as another servant, and the player can send the gumshoe around the room to eavesdrop on select conversations. Depending on what the detective overhears, he'll form his own idea of the means, motive, and opportunity for each suspect, so players must ensure that what the detective overhears doesn't exonerate possible suspects or implicate the player.

Norton's third idea was based on manipulating emotions but not dialogue. In the Gambit-MIT game Camaquen, players take the role of a spirit witnessing a debate between two characters over ownership of a sacred grove. While they can't control the other characters, they can change their emotional mindset to try to influence the outcome. Although the conversation is the same regardless of the play-through, the emotional state of the characters is conveyed through the characters' appearances, as well as their comic book-style text balloons.

The last alternative dialogue design Norton went over centered around context. The genesis for the idea came when Norton's boss suggested she design a game for the iPod that could be played when it was in the player's pocket. Without the use of the screen, the only interaction the player had was clicking a button, so the only meaningful dialogue interaction she came up with was clicking to interrupt a conversation at a certain point. The player doesn't know what they will say; just that they'll say something.

For example, Norton gave the audience the passage, "Tell me where the money is. We'll split it 50-50. You don't know how desperate I am." By clicking after the first sentence, the player could simply say where the money is. Clicking after the second sentence would start an argument about who gets what share, while clicking after the third could make the speaker become hostile.

Norton also showed off an early prototype of a game using this approach, with the player taking the role of a James Bond-style secret agent stuck in an evil mastermind's death trap. The player's goal is to keep the villain talking long enough for him to spill the beans on his plot for world domination and figure out a way out of the predicament along the way.

Takeaway: There's a wide spectrum of solutions between the standard dialogue tree and the totally AI-driven natural language conversations; developers just need to be a little creative with them.

Quote: "High-tech solutions, when they work, are often trading one set of problems for another."--Norton, on why she's exploring solutions along the entirety of that spectrum.