Towards Data-Driven Drama Management

Anders Drachen (IT University of Copenhagen) and Arnav Jhala (University of California-Santa-Cruz).

This was a bit of early work, but it is interesting seeing automatic annotations being put onto text transcripts of tabletop games. This could be interesting in the future, but they didn’t have much to say on what happened yet.

Content

Drama management software manages the user experience but still gives leeway for the player to feel agency in the game. Think like the “Game Master” of tabletop RPG’s.

Some examples; DODM (Nelson and Mateas), IDA (Magerko and Laird), Mimesis (Saretto, Riedl, and Young), I-Storytelling (Cavazza et. al).

This however looks at Game Masters specifically, using a D3M model collecting data from roleplaying game sessions and look for the block points of points hand-authored and also the action-reaction sequences between GM and players.

The paper specifically looks at how to gather and use the data. The progression of the game progresses through action-reaction cycles, between the GM and players. GM’s also have other layers of story management (higher level then just the current action-reaction bit), and the players have influence so the stories cannot be fully pre-planned.

Perceived storyline is what the players have seen, and the GM conceives what will happen in the future. There are multiple story threads in one story too (people fighting villains, one in a tavern and one set in a love story all at once), which is interesting to transcribe.

The GM’s have authorial control over the story – giving credibility to objects, behaviours and actions. They control the world. Only recently beginning to model this. To model, the experiemental procedure makes sure the game rooms match generally player conditions, game players are randomly selected, and many are chosen.

No observer, and using hidden cameras and wireless microphones setup in a lab. Groups divided up by experience (low, high, mixed), GM’s got given a game module to player. Sessions lasted 3-7 hours, some kicked out when they took too long 🙂

Got 150 hours of video data – how to deal with it? Can outsource to people who will transcribe the text for analysis. Need some kind of learning to get through the text data, first working on the amount of utterances – need to still go in and annotate it. Tedious by hand, so want to do it in a automated way. Currently just lists the action points and the details of the action. Finding scenes is easy to find from the text too. Also got parts not investigated yet – dramatic explanations and backtracking too.

It’s got great benefits – lots of information from the controlled experiments. There are some issues – such as granularity (such as discussing strategy within a scene), clarity (video data is noisy – people talk over/with each other), and some other bits.

Can find repeated cases, and build up hierarchically the cases.

Questions

How about games with multiple GM’s or no GM’s?

We cannot feasibly work with all kinds of roleplaying games, so working on one specific type (1 GM) to accomplish anything.

Where you using the same D20 rules for all groups? What module?

Yes, same rules for all 10 groups. A custom made module, tailored to the specific need for the tests – fairly linear, with a series of scenes. A lot of space within scenes. Not every group used all the scenes, or the same order, and some made up more scenes.

Do you have any plans of getting the internal specifics of the players?

Would like to, but needs to interview the players to do this. Takes a lot of time and money though to do these studies.

How well does the automatic annotation work?

It is ongoing work, initially categorising the text based on the word analysis, and defining objects and players, pretty basic for now.

Are you tracking the points where GM’s inflict authorical control?

It’s very rare, usually got a contract formed in the first hour to get around people pulling out lightsabres and other things.

Leave a Reply

Your email address will not be published. Required fields are marked *

Website and journal of Andrew Armstrong