Nicolas Szilas
38, rue du Père Corentin
75014 Paris
France
nicolas (DOT) szilas (AT) laposte (DOT) fr

Abstract Summary: An idea for a new kind of interactive drama is proposed which manages to maintain the dramatic intensity of the narrative.

Definitions:

Story: succession of actions that happen in the world represented by the narrative. Story lives by itself and includes actions which are not explicitly told. Narrative is only what is told.

Drama: A special kind of narrative where the actions are represented directly. Examples: movie, play, radio play, video games. Novel isn’t a drama.

Three forms of interactive drama are identified:

1. branching: author designs all branches the story can take and the user chooses. Problem: huge amount of work by author needed to create complex interactivity.
2. superposed interactivity: the course of story is not really changed, the type of interactivity is localized, where the user solves a puzzle, fights a game, etc.
3. simulations: video games involving environmental simulation, like simcity etc, where users choices have real consequences. However, these can’t be classified as stories or dramas.

Method: Principles:

1. characters as narrative functions: Character’s actions are motivated by narrative constraints rather than emotional, psychological, or social reasoning. Essentially, the right place for AI is not in character, but in narrator. (Propp, 1928) (Blumberg 1997)
2. a story can be broken down into a succession of generic processes (Bremond, 1974)
3. Conflict is at the core of the narrative. (Jenn 1991, Lavandier 1997, Parent-Altier, 1997, and scriptwriting textbooks – McKee, Campbell, Field).
4. Any narrative assumes an intention of the author towards the user; this intention is supposed to lie in the conflict itself and how it is solved.
5. For any narrative, the author is implicitly using a user model to manage the user reaction to its narrative.

Conflict: Occurs when a possible action is not compatible with the values of a character.

Architecture:

1. World of the story: current state of the world. The temporal evolution of this world is the story
2. The narrative logic: rules on how the world of story can evolve – the possible set of actions
3. The narrator: decides which actions should be proposed to the user and/or executed. It also modifies the world of the story according to the user’s responses and decides to execute an action.
4. The user model: belief on user, which could be adapted according to the user type. It has static and dynamic variables (which are adapted after each action)
5. The Theatre: interaction and display – 3D world.

Personal Criticisms: Underlying theory strong for archplots. Fails for multiplots (they don’t follow three act structure) or anti-plots. Too focused for video game environment. Overall, a very interesting paper.

Paul Bailey

Artificial Intelligence

Division of Informatics

The University of Edinburgh

80 South Bridge

Edinburgh EH1 1HN

United Kingdom

fpaulba (AT) dai (DOT) ed (DOT) ac (DOT) ukg

Abstract Summary: An approach to automatic story-generation which uses intuitive model of cognitive states and processes within a reader’s mind to heuristically search for story elements which create preferred abstract ‘storiness’ effect for the reader. It is sensitive to individual readers, but also captures the general properties of stories. Previous work has been divided into –

1. author model: model processes undergone by human author during creation of story – Lebowitz1985; Dehn 1989; Turner 1994; P´erez y P´erez & Sharples 1999).

2. story model: abstract representation of story as a structural artifact. Essentially, generation of story grammar – (Colby 1973;Rumelhart 1975; Pemberton 1989; Lee 1994, Wilensky 1983)

3. world model: construct a world and put characters in it and imbue sufficient complexity in their interactions – (Meehan 1976; Okada & Endo 1992).

He has proposed

Reader Model: Heuristic search through the space of possible stories with selection depending upon the preferences for appropriate story effects within the cognitive processing of the generated story upon the mind of the reader.

Assumption: Ability to gain access to the cognitive states and processes in the reader’s mind – essentially, ability to study the effects of the story upon the reader.

Model Description: Reader-response classified into following two categories and each component of reader’s KB is assigned a strength (which leads to assigning strength values to the following two).

1. Expectations – logical inferences from story so far

2. Questions – arising out of the story so far – What? Why? When?

A best first search which uses the numbers and strengths of expectations and questions controls the story generation.

Algorithm: The generation of the story follows a cyclical scheme. The cycle has four steps:

1. search space of possible next segments created from reader’s knowledge base

2. the effect of these segments using expectations and questions is analyzed

3. the segment which best fits the patters preferred is chosen

4. the reader’s expectations and questions are updated to take account of the new segment.

Knowledge Base: Uses first order predicate logic

Generation of Search Space: Search-space is intended to represent everything the reader is capable of conceiving, and is generated by manipulation of KB using four operators: generalize, specialize, detach, join (why use these operators?)

Personal Criticisms: The model doesn’t take into account the following:

1. story ideas which shatter expectation sometimes form the best stories – since this uses KB to select the next segment, it can’t produce unexpected storylines

2. operators for creating search-space doesn’t make too much sense.

Idea: The model should be used in conjunction with the world model and/or the story model.

People, the first AlgoMantra video has been released. DOWNLOAD MPEG [16MB]

In the video, we played the scenario of a hive-mind police control room where someone is trying to send the message that 500 grams of cocain is being sent in a blue truck at a particular time from Powai. The game had people walking around like automatons on a tile-gird, so they shared packets of encrypted information and figured out the intended message due to crystallisation by hive activity.