This research is conducted within two institutions, the game research group led by prof. Craig A. Lindley at the Department of Engineering Art and New Media at Gotland University in Sweden, where I have a doctoral studentship, and within the School of Computing and Mathematics at the University of Teesside in England, which is the examining institution for my PHD work and where I have my main supervisor Clive P Fenncott. The work preceding this research has been conducted in the Interactive Institutes Zero-Game studio. The time scope for this research project is 5 years on part time, starting 2004.
Title of the investigation
Player character mind models for deep characterization and emergent story construction in massively multiplayer game worlds
Aim of the investigation
Games set in persistent virtual worlds having massive numbers of players need methods of characterization for player characters that differ from the methods used in traditional narrative media. In particular, deep characterization of the player character is a key element for creating systems for emergent narratives in massively multiplayer gameworlds.
In order to achieve deep characterisation, novel player character techniques, based in particular upon new player character mental models, will be developed and validated.
The primary thesis will be explored by iterative experimental implementation and evaluation.
Plan of work
The main focus of this work is on two closely related topics concerning persistent virtual game worlds having massive numbers of players, i.e. massively multiplayer online games (MMOGs); those topics are:
- Characterization and development of identity.
- Story construction
The central hypothesis of this project is that deep characterization of the player character is a key for creating a system for emergent narratives in MMOGs. The expression “deep” character is equivalent to Foresters (1927) term “round” character, and concerns the player character as a combination of a person playing a game and a fictive person who’s identity is continously developed. Bartle (2003) suggests that the player and the players character become one at the final level of immersion: “One individual, one persona: identity.”
The intention of this work is to design and implement a mind module (MM) drawing upon knowledge from social sciences, drama theory, AI and believable agents in order to create a system for player characters (PCs), that can result in PCs that are less generic and deeper than those typical of the current state of the art. The PC should serve as a vessel that can provide the player with means for developing a complex and deep character, or second self.
In addition to this, a framework for story construction will be designed that allows game designers and players to create raw material that results in emergent narratives when instantiated, and provides a means for players to create and experience individual discourses while playing. It is hoped that the system for creating emergent narrative will be based upon the mind module.
The role of the MM is to provide the system with emotional output from the individual player character. The MM performs computational operations on the input values, which come from virtual sensors defined at various levels of abstraction, and outputs in the form of emotional reactions and/or potential emotional reactions that in turn become inputs to the sensors of the MMs of surrounding entities, or entities that in some other way are receptive to the specific player character. (Eladhari and Lindley 2003).
Relevant previous work in the area of believable agents and AI includes recent publications from Egges et al (2003), Mateas on expressive AI (2003), Trappl and Petta on synthetic actors (1997) Bellmans thinking around the self in relation to the virtual worlds (2002) and Moffats model for personality parameters (1997).
In the area of emergent narrative, Mateas and Sterns' Fascade project (2002), Klastrups recently published thesis (2003), Bringsjord and Ferrucis BRUTUS (2000) and Szilas IDTension (2003) are useful. Some commercial games that have made significant innovations in these areas include Star Wars Galaxies, which is interesting from the perspectives of both characterization and emergent narratives. For a broader perspective on virtual gameworlds and players, Bartles 20 year long work on MUDs and MMOGs is useful (1996, 2003). Additional related work is documented in Eladhari (2002, 2003) and in Lindley and Eladhari (2002).
Research process and proposed timeline
- Research Demo, Game scenario development
In order to obtain results these modules will be devised in a game setting derived from a related EC research project, IperG. As a backup plan, the game setting will be designed to be suitable for alternative implementation using Shockwave 3D and the Macromedia Director mulituser server. The mind module and the story construction system will be built as separate components with interfaces to game clients and the multiuser server. This will make the software encapsulating research components as independent as possible from the general game environments in order to ensure demonstrable results.
- Iterative design, implementation, test an analysis phases
Due to the subjectivity of the intended game experience, measurements of results will be done both by quantitative and qualitative methods: measurments of playtime and use of functionality, survey forms, in-depth interviews, retrospective verbal reports, adapted presence questionnaires, and possible (eg. ‘think aloud’ techniques.
Criteria for the choice of test subjects will be developed. An important criterion is the players intentions and motives regarding the player character, ie. if she regards herself as being the game character in the game, rather than playing a certain fictional role. Player classification schemes like that of Bartle (1996) will be explored and potentially extended.
The method of finding test subjects will be by interviews within existing MMOGs. The ethical issue of subjects’ integrity and anonymity needs to be considered for each publication.
Bartle, Richard “Hearts, Clubs, Diamonds, Spades: Players Who Suit MUDs” in Journal of MUD Research 1996.
Bartle, Richard Designing Virtual Worlds (New Riders, 2003)
Bellman, K.L, “Emotions: Meaninful Mappings Between the Individual and Its World” in in Emotions in Humans and Artefacts, ed. R. Trappl, P. Petta, and S. Payr (Massachusetts Institute of Technology, The MIT Press, 2002) 149 – 188
Bringsjord, Selmer, Ferruci, David A. Artificial Intelligence and Literary Creativity (Lawrence Erlbaum Associates, 2000)
Creating Personalities for Synthetic Actors, Towards Autonomous Personality Agents, ed. Trappl, Robert, Petta, Paolo. (Springer 1997)
Eladhari, Mirjam Object Oriented Story Construction in Storydriven Computer Games Stockholm University, Stockholm, Sweden. June 2002.
Eladhari, M. and Lindley, C. “Player Character Design Facilitating Emotional Depth in MMORPGs” Digital Games Research Conference 2003, 4-6 November 2003 University of Utrecht, The Netherlands
Egges, A, Kshirsagar, S., Thalmann N. M., "A Model for Personality and Emotion Simulation", Knowledge-Based Intelligent Information & Engineering Systems (KES2003), 2003
Forster, Edward Morgan, Aspects of the Novel (London : Arnold, 1927)
Klastrup, Lisbeth., Towards a Poetics of Virtual Worlds - Multi-User Textuality and the Emergence of Story, IT University of Copenhagen, PHD Thesis, 2003
Lindley C and Eladhari M. 2002 "Causal Normalisaton: A Methodology for Coherent Story Logic Design in Computer Role-Playing Games". Computers and Games Conference, Alberta, Canada. 25th - 27th July 2002.
Mateas, Michael / Stern, Andrew, “Architecture, Authorial Idioms and Early Observations of the Interactive Drama Façade”, (Carneige Mellon University, 2002)
Moffat B. “Personality Parameters and Programs”, in Creating Personalities for Synthetic Actors, Towards Autonomous Personality Agents Eds Trappl R. and Petta P. 1997
Szilas, Nicolas, “IDtension: The Simulation Of Narrative”, COSIGN 2003, 10th - 12th September 2003, University of Teesside, Middlesbrough (UK)
First supervisor: Clive P Fencott, senior lecturer at School of Computing and Mathematics at the University of Teesside in England
Second supervisor: Craig A. Lindley, professor at the Department of Engineering Art and New Media at Gotland University in Sweden
Third supervisor: Paul Van Schaik, reader at School of Social Sciences at
University of Teesside, Sweden
Advisor: Michael Mateas, assistant professor at Georgia Institute of Technology, Atlanta, USA
Semi-autonomy and level of player control
The amount of control that the player has needs to be explored in order to achieve a balance resulting in good gameplay. In general we see a continuum between a rather empty “shell” avatar at one extreme, providing minimal autonomy beyond a simple mapping from player control decisions to animation and expression functions, to a highly autonomous non-player character (NPC) at the other extreme, providing minimal influences from player actions to the NPCs behaviour.
Previous research in which I was involved at the Zero-Game Studio of the Interactive Institute explored this issue for a period of two years in two projects. In the first, the avatar approach was explored, providing a player character for improvising multi-player dramas. This was a virtual on-line version of the classical Commedia del Arte,using 2D character sprites. The conclusions draw from this were that while such a minimal approach was effective for players who are skilled performers, who could effectively improvise interesting performances for hours at a time, it was not effective for unskilled performers, who would become bored within minutes. The second project involved development of a multiplayer 3D virtual drama system. This system provided the same minimal avatars as a foundation. The proposed PhD work intends to use this as a starting point to further develop concepts from this project for enhancing the simple player-driven avatar with context-sensitive expression possibilities and varying degrees of autonomous reaction to in-game situations.
The approach is to explore semi-autonomy of the player character while the player is playing. That is, the project is not concerned with autonomous character functions in a persistent world while the player is not on line. The aim is to experiment with different degrees of autonomy to determine the levels and forms of emotional engagement experienced by the player in different in-game situations. This will be done by implementing and testing a series of research demonstrators.
AI techniques for agent behavior
The state of the art in agent behaviour is to use higher level, deliberative, planning-based approaches to implement higher level cognitive functions on top of lower level, behavioural or reactive systems. This is the case when behaviour generation functions are encapsulated within an agent model. An alternative approach is to use a more global narrative manager, drama manager, or autonomous game master to orchestrate the behaviours of many agents to create dramatic and narrative action. Architectures are possible in which these strategies are mixed.
Most of these systems have been developed for autonomous characters. The problem addressed by this research, providing a player character with functionality to assist an unskilled player in achieving a dramatically effective in-game performance, has been poorly addressed to date, and raises broad questions of applicability that may change the value of the techniques adopted.
With regard to specific techniques, the first iteration of the project will use a semantic network mapping contextual information (from perception, the actions of other players, and a global world state description) onto character expression options for the player. This approach will minimize processing time (compared with a deliberative implementation) while providing explicit models for various mental phenomena, include both healthy psychic structures and unhealthy ones (eg. neuroses and phobias). Planning techniques will not be used since the player will provide those higher level cognitive functions. The hypothesis is that the lower level approach will provide a better experience of characterization for the player. If this proves in practice to be of limited effectiveness, the incorporation of higher level functions will be explored in further iterations of the project.
The “Mind” as a semantic network
The player character architecture includes an input set, a control set, and an expression set. The input set includes local perception data gathered by collision detection of game objects with character perceptual zone volumes, and control actions performed by the player. Perceived objects will transmit state data to the player character controller that includes internal information about the emotional and mental states of the characters played by other players in the game world. The expression set includes gestures, motions, speech acts, noises and visual signs; these will either be manifested directly (ie. autonomously) within the game world or presented to the player as performance options to be chosen to provide control inputs within a later control cycle. The control set includes the semantic network (including internal state machines) that will map between the input set and the expression set. The semantic network constitutes the autonomous part of the “mind” of the player character. Note however that the character in the game is the product of autonomous functions together with performance decisions made by the player. Hence the total “mind” is realized as a synthesis between the autonomous system and the player’s immersion in the role.
This is the initial approach to be taken by the research. Since this proposal is being written during the first year of a multi-year project, the approach is expected to evolve considerably in the light of issues arising during implementation and testing. Regarding testing, the effectiveness of such a system is currently envisaged in broad terms that can be measured in terms, firstly of broad playability (is this playable at all?), and then in terms of depth of engagement reflected in sustained play time in relation to the total combinatorial space of play potential within the system. A more detailed framework for evaluating these factors will be developed as the project progresses.
Story Construction and Emergent Narrative
We take emergence to be: the emergence of a higher level structure from the interaction of many simpler, lower level primitives. In this case emergent narrative must be understood as a system in which lower level elements interact to result in the emergence of a pattern of events conforming to a specific higher level pattern of narrative structure.
The initial approach is to use the AI representation and process models of semantic networks to implement an object-oriented story construction system. This will be done as an emergent narrative system, using the mind models (controllers) of the player characters, low level objects implemented as semantic networks, as simpler elements from the interaction of which (between multiple characters) it is hoped that high level narrative patterns will emerge.
This work would be an extension of the investigation conducted within my Masters thesis (Eladhari M. Objektorienterat berättande i berättelsedrivna datorspel (“Object Oriented Story Construction in Story Driven Computer Games”), Masters Thesis, University of Stockholm, http://zerogame.interactiveinstitute.se/papers.htm. 2002.).
It is anticipated that some degree of extra-character narrative control will ultimately be incorporated within the PhD research, but the starting point will be concerned with pushing this object-oriented (in fact agent-oriented) emergent storytelling approach as far as it can go. The limitations of this emergent approach will then be used to motivate the incorporation of narrative generation principles at a world level (but from the perspective of maximum variability in the generation of interesting narrative structures).