Game Design Theory - Institute for Logic, Language and Computation

Game Design Theory - Institute for Logic, Language and Computation
Engineering Emergence
Applied Theory for Game Design
Joris Dormans
Joris Dormans, 2012.
ISBN: 978-94-6190-752-3
This work is licensed under the Creative Commons Attribution-NonCommercial
3.0 Netherlands License. Visit http://creativecommons.org/licenses/by-nc/3.0/nl/
or send a letter to Creative Commons, 171 Second Street, Suite 300, San Francisco, California, 94105, USA.
Engineering Emergence
Applied Theory for Game Design
ACADEMISCH PROEFSCHRIFT
ter verkrijging van de graad van doctor
aan de Universiteit van Amsterdam
op gezag van de Rector Magnificus
prof. dr. D.C. van den Boom
ten overstaan van een door het college voor promoties ingestelde
commissie, in het openbaar te verdedigen in de Aula der Universiteit
op vrijdag 13 januari 2012, te 11.00 uur
door
Joris Dormans
geboren te Geleen
Promotiecommissie:
Promotor:
prof. dr. ir. R.J.H. Scha
Co-promotor:
dr. ir. J.J. Brunekreef
Overige leden:
dr. ir. A.R. Bidarra
prof. dr. P. van Emde Boas
prof. dr. P. Klint
prof. dr. ir. B.J.A. Kröse
dr. M.J. Mateas
prof. dr. E.S.H. Tan
Faculteit der Geesteswetenschappen
Contents
Map of the Land of Games
vi
Preface
ix
1 Introduction
1.1 Games . . . . . . . . . . . . . . . .
1.2 Mechanics . . . . . . . . . . . . . .
1.3 Game Classification . . . . . . . .
1.4 Emergence and Progression . . . .
1.5 Emergence . . . . . . . . . . . . . .
1.6 Progression . . . . . . . . . . . . .
1.7 Approach and Dissertation Outline
1.8 Terminology . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
2
6
10
13
16
19
21
22
2 Rules, Representation and Realism
2.1 The Iconic Fallacy . . . . . . . . .
2.2 Indexical Simulation . . . . . . . .
2.3 Symbolic Simulation . . . . . . . .
2.4 Less Is More . . . . . . . . . . . .
2.5 Designing Emergence . . . . . . . .
2.6 Conclusions . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
25
27
30
31
33
37
40
3 Game Design Theory
3.1 Design Documents . . . . . .
3.2 The MDA Framework . . . .
3.3 Play-Centric Design . . . . .
3.4 Game Vocabularies . . . . . .
3.5 Design Patterns . . . . . . . .
3.6 Mapping Game States . . . .
3.7 Game Diagrams . . . . . . . .
3.8 Visualizing Game Economies
3.9 Industry Skepticism . . . . .
3.10 Conclusions . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
43
44
46
48
49
51
53
57
59
61
64
4 Machinations
4.1 The Machinations Framework . . . . . . .
4.2 Flow of Resources . . . . . . . . . . . . .
4.3 Flow of Information . . . . . . . . . . . .
4.4 Controlling Resource Flow . . . . . . . . .
4.5 Four Economic Functions . . . . . . . . .
4.6 Feedback Structures in Games . . . . . .
4.7 Feedback Profiles . . . . . . . . . . . . . .
4.8 Feedback Analysis and Recurrent Patterns
4.9 Implementing Machinations Diagrams . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
67
68
71
75
78
80
83
87
90
93
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
iv
Joris Dormans | Engineering Emergence
4.10 Randomness and Nondeterministic Behavior . . . . . . . . . . . . 96
4.11 Case study: SimWar . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.12 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5 Mission/Space
5.1 Level Layouts . . . . . . . . . . . .
5.2 Tasks and Challenges . . . . . . . .
5.3 The Mission/Space Framework . .
5.4 Mission Graphs . . . . . . . . . . .
5.5 Space Graphs . . . . . . . . . . . .
5.6 Level Analysis: The Forest Temple
5.7 Mission-Space Morphology . . . . .
5.8 A Software Tool for Level Design .
5.9 Conclusions . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
109
110
113
115
117
121
125
129
130
133
6 Integrating Progression and Emergence
6.1 From Toys to Playgrounds . . . . . . . . . . . .
6.2 Progression through Structured Learning Curve
6.3 Economy Building Games . . . . . . . . . . . .
6.4 A Mismatch Between Mission and Space . . . .
6.5 Mechanics to Control Progression . . . . . . . .
6.6 Feedback Mechanisms for Locks and Keys . . .
6.7 Progress as a Resource . . . . . . . . . . . . . .
6.8 Conclusions . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
135
136
137
142
146
149
152
154
159
7 Generating Games
7.1 Game Design as Model Transformation . .
7.2 Formal Grammars . . . . . . . . . . . . .
7.3 Rewrite Systems . . . . . . . . . . . . . .
7.4 Graph Grammars . . . . . . . . . . . . . .
7.5 Shape Grammars . . . . . . . . . . . . . .
7.6 Example Transformation: Locks and Keys
7.7 Generating Space . . . . . . . . . . . . . .
7.8 Generating Mechanics . . . . . . . . . . .
7.9 Procedural Content in Games . . . . . . .
7.10 Adaptable Games . . . . . . . . . . . . . .
7.11 Automated Design Tools . . . . . . . . . .
7.12 Conclusions . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
161
162
164
166
168
170
173
174
180
185
187
191
193
8 Conclusions and Validation
8.1 Structural Qualities of Games . . .
8.2 Validation . . . . . . . . . . . . . .
8.3 Teaching Game Design . . . . . . .
8.4 Building Prototypes . . . . . . . .
8.5 Academic and Industry Reception
8.6 Omissions . . . . . . . . . . . . . .
8.7 Future Research . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
197
198
200
201
205
208
209
210
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Contents
8.8
v
Final Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
A Machinations Overview
B Machinations Design Patterns
B.1 Static Engine . . . . . . . . .
B.2 Dynamic Engine . . . . . . .
B.3 Converter Engine . . . . . . .
B.4 Engine Building . . . . . . . .
B.5 Static Friction . . . . . . . . .
B.6 Dynamic Friction . . . . . . .
B.7 Attrition . . . . . . . . . . . .
B.8 Stopping Mechanism . . . . .
B.9 Multiple Feedback . . . . . .
B.10 Trade . . . . . . . . . . . . .
B.11 Escalating Complications . .
B.12 Escalating Complexity . . . .
B.13 Playing Style Reinforcement .
215
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
217
217
220
223
227
229
231
234
238
240
242
244
246
248
C A Recipe for Zelda-esque Levels
253
Summary
259
Nederlandse Samenvatting
263
Index
267
Bibliography
273
Ludography
285
vi
vii
viii
Preface
My road to the summit of Mount PhD located in the heart of the Land of
Games has been a long and wondrous journey. During its many detours I met
many great and inspiring people. Some I met briefly, others accompanied me
for a while, none failed to brighten up my journey with valuable advice and
unwavering support. Looking back at the curvy route, I now know that all
twists and turns were necessary to bring me to the place I am today.
One might say that I am a native of the Land of Games. During my childhood
I spend a great many hours playing games of all sorts: board games, pen-andpaper roleplaying games, and video games. And we invented our own from the
very start. Countless are the sheets of paper I used to design boards for overly
ambitious games, dungeons for endless roleplaying sessions, and plans for games
that I never quite finished programming on our Commodore 64.
Growing into adulthood and becoming a student, I left the land of games on
what would prove to be the longest detour of all. Years I spent in the fine halls
of education on the far side of the Alpha Ocean. During this period I gained
knowledge and skills that would prove to be trusty tools on the journey that
finally led me here.
Returning to these shores was a joy. I had grown wiser, and my vision had
been sharpened. Some things had changed, many others had remained the same.
Immediately I knew where I wanted to go. Finding the right road did not always
prove easy, but I can now say I succeeded. I must thank many people for the
support, advice and company along this journey.
Claartje, my mother, showed me at an early age that through strength of will
and firm determination you can shape your destiny and climb any mountain you
will find on your path. Mom, no matter what happens, your inner strength will
be my guiding light forever.
Marije has been my loving companion since long before I set out on this path
and has been with me all the way. Now you stand beside me at this journey’s
conclusion as my paranimf. Without your moral support and textual advice I
would never have reached this high.
My promotor, Remko fulfilled the role of mentor with excellence. Despite
being a relative stranger to the Land of Games, you never failed to ask all the
right questions when I needed to reset my course.
My co-promotor, Jacob, provided with me with the right map to successfully
navigate the Shores of Engineering. Without that map, I would never have been
able to travel as far and wide as I did.
I also thank the other members of the committee: Ben, Ed, Michael, Paul,
Peter and Rafael. Some of you I met in distant lands, many of you I met along
the way. I am very grateful that you took the time to accompany me during the
final ascent.
I could not wish for a merrier band of travel companions than my colleagues
at the Hogeschool van Amsterdam. We shared many adventurous days on the
winding paths that led me to this mountain. Your trust in the course I set
and support in paving the roads so that others might easily follow must not go
x
Joris Dormans | Engineering Emergence
unmentioned.
My students never failed to surprise me, especially when they were replicating
parts of my journey. They always picked their own route, which for me was a
source of inspiration.
I must also thank my family and friends whose interest in my progress was
both sincere and supportive. A deep bow to all of you who where once my
opponents in so many well-played games.
From all people along my journey I will thank two more persons. Firstly,
Carla who introduced me to all the right people which finally allowed me make
landfall on these shores after being lost at sea for a while. And finally, last but
not least, Jasper my other paranimf whose friendship goes back all the way back
to our childhood. The many games we enjoyed kept the memory of these lands
alive and paved the way for my return.
There is no single sentence describing what
makes games attractive.
Jesper Juul (2003)
1
Introduction
Designing games is hard. Although games have been around for a very long
time, it was the rise of the computer game industry over the past few decades
that caused this problem to become prevalent. During its short history the
computer game industry has grown from individual developers and small teams
towards multi-million dollar projects involving hundreds of employees. In the
contemporary game industry there is little room for mistakes: the financial stakes
have grown too high. Today, more than any time before, there is a need for a
better understanding of the process of designing a successful game in order to
prevent such mistakes; there is a need for better applied theory and intellectual
tools to aid game designers in their task.
At the same time, more people are playing video games than ever before. A
wider audience means that there is an ever increasing demand for games with
quality gameplay. As game players get more experienced they grow an appetite
for ever more sophisticated games. Compared to other forms of art and media,
computer games are a fairly recent invention. There is still plenty of room for
development and innovation.
The general premise of this dissertation is that the difficulties in designing
games lie within the nature of games as complex rule based systems that exhibit
many emergent properties on the one hand, but must deliver a well-designed,
natural flowing user experience on the other. In facing these difficulties, the
game designer’s tool box is quite empty. The nature and emergent behavior of
games is poorly understood. Level design has been one way to harness a game’s
emergent behavior, by restricting the gameplay to a series of relatively simple
tasks loosely strung together by a storyline. However, high-quality content is
expensive to produce. Games with many hand-crafted levels are expensive to
produce, and fail to exploit the true expressive power of open game worlds that
emerge from rule systems.
This dissertation examines the nature of emergence in games in order to construct applied theory to deal with emergence in games head-on. This theory will
enable the designer to get more grip on the elusive process of building quality
games displaying emergent behavior. The theory developed in this dissertation
applies to game mechanics and levels. However, where many scholars and designers treat levels and mechanics as two vastly different elements of game design,
Joris Dormans | Engineering Emergence
2
this dissertation attempts to integrate the two: for both rules and levels, this
dissertation seeks to find formal, abstract representations through which the process of designing these aspects of games might be elevated and unified. Through
these representations the material that game designers work with should become
more tangible. This leads to the central research question of this dissertation:
what structural qualities of game rules and game levels can be used in the creation of applied theory and game design tools to assist the design of emergent
gameplay?
The development of software tools to assist game designers in their task is
an important aspect of this dissertation. All too often, design theory has been
created in isolation from the practice of game design. The creation of software
tools that implement the theories presented here, means that these theories
have to be very concrete and applicable. In addition, it allows the automation
of certain parts of the design process. By automating these parts, designers will
be assisted in their work and can focus on those aspects of design that require
the most of human creativity and ingenuity.
This chapter introduces the central notions of this dissertation: games, gameplay, mechanics, levels, emergence, and progression. It also outlines the general
approach and the contents of the chapters that follow.
1.1
Games
What are games? Many people play them, but only a few stop to contemplate
their nature. The study of games, especially the study of games in its current
form, is very young. It was only in 2001 that Espen Aarseth declared that
year to be “year number one” of game studies (Aarseth, 2001). That year saw
the launch of the first peer-reviewed online journal and the first international
academic conference dedicated to games. Games were studied before, but it was
not until 2001 that game studies gained enough momentum to be recognized as
a separate academic discipline.
The study of games has been a multi-disciplinary affair from the beginning,
with researchers from different fields studying games from different perspectives. The first few years of game studies were characterized by a fierce debate
between narratologists and ludologists. The former group comprised academics,
often with a background in literature, who had been studying games from that
perspective for a while. They regarded games as a new medium for storytelling
and placed games in the context of literature and media studies (Laurel, 1986;
Murray, 1997; Ryan, 2001). The ludologists opposed this position, for them
games are rule-driven play experiences first and foremost. The story and visuals
are secondary to rules which are the most critical factor for game quality. Their
argument is that good rules with visuals and story of lesser quality still make
for a good game, whereas the opposite is not true (Eskelinen, 2001; Juul, 2005).
Today both positions are considered rather extreme. It is difficult to find
somebody who would maintain that the paradigms used to study stories in literature or cinema apply directly to games. You cannot ignore rules, interactivity
Chapter 1 | Introduction
Figure 1.1: Raph Koster’s hypothetical Mass Murderers Game
based on Tetris.
3
Figure 1.2: Shooting missiles at terrorists in
the satirical September 12.
and gameplay in any study of games. On the other hand, the reskinning of
Tetris into the hypothetical “Mass Murderer Game” by Raph Koster (see figure 1.1) where the player tries to fill pits with awkwardly shaped dead bodies,
clearly illustrates that story and visuals do affect the experience of play (Koster,
2005b, 166-169). The biting irony of a game like September 12 where the
player is invited to shoot missiles at terrorists in an Arabic city and to explore
the consequences of that action is only made possible by the sharp contrast between rules that support simple, typical gamelike shooting action and the game’s
meaningful reference to a very real situation outside the game (see figure 1.2).
Games do not exist in isolation but are part of a heterogeneous media landscape
and the social structures from which stories derive their meaning. In this case
it is worth noting that September 12’s developer, Gonzalo Frasca, is also a
prominent ludologist and was in fact the first game researcher that coined the
term ludology (Frasca, 1999).
The examples of the Mass Murderer Game and September 12 also illustrate nicely that in games, what matters most is what the player does. These
actions are determined by rules on the one hand, but on the other a game’s
art and story frame these actions and give them meaning. In a game, rules
set up possible interactions, but through clever design of levels developers have
some control over the order in which players encounter game elements and the
challenges they pose. It is through levels that developers primarily control the
sequence of actions.
The general consensus in game studies is that games are rule based artifacts
designed to be experienced by one or more players, in which they strive to achieve
some sort of goal. Without rules there would be no game, but the structured
experience of the player is not unlike the structures and experience encountered
in other media.
After close examination of eight different definitions of games, from histo-
Joris Dormans | Engineering Emergence
4
rian Johan Huizinga to game designer Greg Costikyan, Katie Salen and Eric
Zimmerman define games as follows:
“A game is a system in which players engage in artificial conflict,
defined by rules, that results in a quantifiable outcome.” (2004, 80)
In their definition system, players, artificiality, conflict, rules and quantifiable
outcome are the key notions. All games are systems consisting of many parts
that form a complex whole (Salen & Zimmerman, 2004, 55). The system is
defined by rules that determine what players can and cannot do. Following
those rules, players engage in conflict against each other or against the game
system. The conflict is artificial in the sense that the game is set apart from real
life in both time and space, a space where the players submit to the rules of the
game. In this sense, games are often said to take place within a “magic circle”,
after the work of Johan Huizinga (1997). Finally, a game has a quantifiable
outcome: players can win or lose, or measure their performance with some sort
of score.
Salen and Zimmerman’s definition resembles many other definitions, even
those not investigated by themselves. Mark J. P. Wolf uses the elements conflict,
rules, player ability and valued outcome to define games (2001, 14). Alexander
Galloway states that: “a game is an activity defined by rules in which players
try to reach some sort of goal” (2006, 1). Ernest Adams and Andrew Rollings
identify rules, play, goals and pretending as key elements of games. The latter
element is linked to the magic circle and by extension to Salen and Zimmerman’s
notion of artificiality (Adams & Rollings, 2007, 5-11). For Tracy Fullerton a
game is “a closed, formal system that engages players in structured conflict and
resolves its uncertainty in an unequal outcome” (2008, 43).
Jesper Juul examines many of the same definitions of games, including the
definition of Salen and Zimmerman. He concludes that the following six features
define games (Juul, 2005, 36):
1. Games are rule based,
2. and have variable, quantifiable outcomes,
3. which are affected by the effort of the player,
4. and which are assigned different values,
5. and to which the player is emotionally attached,
6. and which consequences are negotiable.
Of these six features only the first three are properties of the game as a formal
system. The other three are either properties of the relation between the game
and the player or the relation between the game and the rest of the world (Juul,
2005, 37).
Compared to Salen and Zimmerman, Juul’s definition incorporates a few
extra elements. First, in Juul’s definition the outcomes of a game are not only
Chapter 1 | Introduction
5
quantifiable, they are also variable. Games must have different outcomes to work
as game. As soon as a game will always have the same outcome, there is little
point in playing. This can happen when two players of vastly different levels
are competing. When one of them is sure to win, when the outcome is known
before the start, the game will cease to function as a game. Second, for Juul the
player must be able to affect a game by putting in an effort. Without effort, the
player’s actions are meaningless and the player will never become emotionally
attached to the outcome of the game. For Juul this makes all games of pure
chance, where players cannot affect the outcome in any way, borderline cases
(Juul, 2005, 44). Thirdly, Juul pays more attention to the relation between the
game and the player, and to the relation between the game and the world as is
indicated by his last four points.
At the same time, Juul leaves out the artificiality of Salen and Zimmerman’s
definition. In Salen and Zimmerman’s definition artificiality plays a similar role
to Johan Huizinga’s concept of the ‘magic circle’ through which games create a
reality outside real life (Huizinga, 1997). Although the magic circle is arguably
porous (Copier, 2007) the artificial nature of games is without question. Game
rules are made by designers and upheld by players to create an experience;
players submit to these rules to experience the game. Games are also constrained
by ‘rules’ that exist prior to the game, such as the law of gravity which constrains
almost any sport (Juul, 2005, 58), but all games add rules to set up artificial
goals, conflict and challenges. The game creates a space where the game is
played, whether or not that space has clear boundaries.
For this dissertation I choose to build on Salen and Zimmerman’s definition
of games, although I do add Juul’s player effort and ability to affect the variable
outcome of the game. I choose to disregard Juul’s other additions as this dissertation focuses on games as formal systems and not on the relation between
games and players. Thus, for this dissertation, games are defined as follows:
A game is a system in which players engage in artificial conflict, defined by rules, that results in a variable, quantifiable outcome affected
by player effort and ability.
Gameplay, a key notion associated with the players actions’ and experience of
play, is more difficult to define. Gameplay somehow consists of what the player
does. At the same time, the term is used to describe a quality possessed by
games themselves. Reviewers of games often talk about gameplay in this sense.
Used in this way, “gameplay has become synonym with good gameplay” as Niels
’t Hooft once remarked.1 That is to say, whether or not a game has gameplay,
has become an assessment of its quality: good games have gameplay.
For this dissertation I will use gameplay in this sense. When designers are
working to create gameplay, they are always working to create a compelling
game experience. The game, as a product, is the prime source for this experience. What follows is that gameplay somehow emerges from the way a game is
1 Niels ’t Hooft is a freelance game journalist who works for Basher.nl and the Dutch newspaper NRC Next. He made this statement during a GameLab meeting on gameplay on February
2, 2011 in Pakhuis de Zwijger, Amsterdam.
Joris Dormans | Engineering Emergence
6
constructed. It is these structural qualities of games as rule based systems that
are the focal point of this dissertation.
1.2
Mechanics
When the game design community talks about game systems, they prefer the
term “game mechanics” over “game rules”. “Game mechanics” is often used as
a synonym for rules but the term implies more accuracy and is usually closer to
an implementation. Although implementation here is still relatively independent
from any platform or medium. Game designers Ernest Adams and Andrew
Rollings explain the difference between the two with the following example: the
rules of a game might dictate that in a game caterpillars move faster than snails,
but the mechanics make the difference explicit; the mechanics specify how fast
caterpillars move and how fast snails move. Mechanics need to be accurate
enough for game programmers to turn them into code without confusion or for
board game players to execute them without failure; mechanics specify all the
required details (Adams & Rollings, 2007, 43).
In a similar vein, Morgan McGuire and Odest Chadwicke Jenkins state that
“Mechanics are the mathematical machines that give rise to gameplay; they
create the abstract game” (2009, 19). With that they point out that mechanics
are media-independent: they are amongst those parts of games that are separable
from images and sounds and might actually be transposed from one medium to
another: a board game might be recreated as a computer game with different
art and a different theme without altering the mechanics.
Game designers are perfectly comfortable talking about a “game mechanic”
in the singular form (McGuire & Jenkins, 2009; Brathwaite, 2010). With this
they are not referring to a person who is skilled in dealing with game mechanics,
as the common use of the singular form “mechanic” would imply.2 Instead, they
are referring to a single game mechanism that governs a certain game element.
One such mechanism might include several rules. For example, the ‘mechanic’
of a moving platform in a side-scrolling platform game might include the speed
of the platform’s movement, the fact that creatures can stand on it, the fact
that when they do they are moved along with it, but also the fact that the
platform’s velocity is reversed when it bounces into other game elements, or
perhaps after it has traveled a particular distance. In this dissertation I prefer
to use “mechanism” as the singular form indicating a single set of game rules
associated with a single game element or interaction.
Some mechanics may be more central to a game than others. The term “core
mechanics” is often used to indicate those mechanics that the player interacts
with most frequently and have the biggest impact on the gameplay (Adams &
Rollings, 2007; McGuire & Jenkins, 2009). Moving and jumping, for example,
are core mechanics of most platform games. In contrast, the mechanics that
specify that players gain one extra life for every hundred stars they collect,
might or might not be considered to be the core of a game. For a game where
2 The Oxford Advanced Learner’s Dictionary lists “a worker skilled in handling or repairing
machines” as the sole meaning of the word mechanic.
Chapter 1 | Introduction
7
the extra life is just a nice bonus, it probably is not a core mechanism, but
for a game where stars are abundant and players lose lives easily it probably is.
The distinction between core mechanics and non-core mechanics is not clear-cut;
even for the same game, interpretation of what is core and what is not can vary
between designers or even between different moments within the game.
Mechanics have come to indicate many different types of rules in games.
The term sometimes denotes mechanics in the physics sense: the science of
motion and force. In games characters commonly move, jump or drive vehicles.
Knowing where a game element is, in what direction it is moving and whether
or not it is intersecting or colliding with other elements make up the bulk of all
calculations in many games. Here mechanics might be interpreted quite literally
as the implementation of the physical laws that govern motion and force within
the game. At the same time, games also include mechanics that have nothing
to do with physics: for example, mechanics that specify how many coins need
to be collected to gain an extra life. The mechanics that deal with power-ups,
collectibles and other types of game resources constitute something that might
be called an internal economy (Adams & Rollings, 2007, 331-340). The nature
of economic mechanisms and game physics is different in a number of crucial
ways. One problem of using the term mechanics for both is that it obscures
these crucial differences.
Physics in modern games tends to be simulated with accurate mechanics that
create near continuous game simulations. A game object might be positioned
half a pixel more to the left or right and this might have a huge effect on the result
of a jump. In contrast, the rules of an internal economy tend to be discrete; game
elements and actions are a finite set that do not allow any gradual transitions:
in a game you usually cannot pick up half a power-up. This continuous nature of
game physics versus the discrete nature of game economies has consequences for
the medium (in)dependence of games, the nature of the player interaction, and
even for the opportunities for design and innovation. These effects are discussed
below.
Due to its continuous nature, the implementation of physics tends to be much
more closely tied to the medium or platform than a game economy is. Economic
mechanics are indeed separable from a game’s medium, but physics not to the
same extent. For example, a game that relies heavily on physics can not be
easily mediated as a board game. Creating a board game for Super Mario
Bros. (see figure 1.3) where the gameplay originates from moving and jumping
from platform to platform is very difficult. The continuous physics of a platform
game translate poorly to the discrete nature of board games. A die only has so
many sides, and to keep the game accessible overly complex calculations are best
avoided. In platform games physical dexterity matters, just like a whole myriad
of physical skills determine whether or not somebody is good at playing reallife football; those skills would be lost in a board game. Super Mario Bros.
is probably better mediated as a physical course testing players’ real running
and jumping abilities. The point is, a rule that states you can jump twice as
high after picking up a certain item, can be easily translated between different
media, whereas rules that implement the physics of a jump cannot. The physical
8
Figure 1.3: The gameplay in Super Mario Bros. emerges in
large part from continuous physics
for running and jumping.
Joris Dormans | Engineering Emergence
Figure 1.4: The physics in Boulder Dash
is implemented through discrete system. Objects such as diamonds and boulders are always
aligned to the game’s grid of tiles.
mechanics of a game seem to be bound more closely to the medium than the
discrete rules that govern a game’s economy.
Interestingly, when we look back at the early history of platform games and
other early arcade games, physics were often handled quite differently, much
more discrete, one might say. The moves in Donkey Kong were much less
continuous than they were in Super Mario Bros.. In Boulder Dash (see
figure 1.4) gravity is simulated by moving boulders down at constant speed of
one tile every frame. It might play slowly, but it is possible to create a board
game for Boulder Dash. In those days the rules that created the game’s
mechanics (in the physical sense) were not that different from other types of
game rules. But times have changed. Today the physics in a platform game
have grown so accurate and detailed that they have become impossible, or at
least inconvenient, to represent with a board game.
With discrete rules it is possible to look ahead, to plan moves, create and
execute complex strategies. Although this does not need to be easy, it is possible
and players are encouraged to do so. Player interaction with this type of rules is
much more on a strategic level. On the other hand, once players grasp the physics
of a game (whether simulated or not), they can intuitively predict movements
and results, but with less certainty. Skill and dexterity become a more important
aspect of the interaction. This difference is crucial when you are using a game
to educate players. Angry Birds (see figure 1.5) won a serious game award for
teaching players a thing or two about physics in a fun way. While there is no
doubt that Angry Birds is fun and involves physics, I doubt that players really
learn about the application of forces, gravity or momentum in any conscience way
that is applicable to science education. Players of Angry Birds are involved
with those aspects mostly on the level of skill, rather than strategy; they might
develop an intuitive feel for the effects of forces, gravity and momentum, but that
is not quite the same thing as truly understanding them. Strategy in Angry
Birds involves those aspects of the game that are governed by discrete rules.
Chapter 1 | Introduction
Figure 1.5: In Angry Birds players
catapult birds to destroy pigs protected
by stone, wood and glass structures.
9
Figure 1.6: In World of Goo players
construct towers, bridges and other structures from a limited supply of ‘goo balls’.
Players will have to plan how to use number and types of birds available to
attack the pigs’ constructions most effectively. This requires identifying weak
spots and to formulate a plan of attack, but the execution itself is based on skill
and the effects can never be foreseen in great detail. Compare that to World
of Goo (see figure 1.6) where players need to build constructions from a limited
supply of goo balls. Physical notions such as gravity, momentum and center of
mass play an important role in the mechanics of this game. Indeed, players
might form an intuitive understanding of these notions from playing World of
Goo. But more importantly, players learn how to manage their most important
(and discrete) resource: goo balls, and use them to build effective constructions.
The difference between Angry Birds and World of Goo becomes very clear
when one considers the effects of continuous physics. Where in Angry Birds
the difference of a single pixel can translate into a critical hit or complete miss,
the effects are less felt in World of Goo. In the latter game, placement is not
pixel precise: releasing a goo ball a little more to the left or right usually does
not matter as the resulting construction is the same, and the spring forces push
the ball into the same place. The game even visualizes what connections are
going to be made before the player releases a ball (as can be seen in figure 1.6).
Without trying to argue which game is more fun, I would say that players learn
much more about construction in the World of Goo than they learn about
physics in Angry Birds.
Physics and economy in games also affect design and innovation differently.
One might say, that as games and genres evolve, the physical mechanics are all
evolving into a handful of directions that correspond closely with game genres:
most of the time there is little point in completely changing the physics of a firstperson shooter.3 In fact, as games increasingly use physics engine middleware to
handle these mechanics, there is less room to innovate in that department. On
the other hand, every game is trying to create unique content, and many first
person shooters do create an unique system of power-ups or economy of items to
collect and consume to make their gameplay different from their competitors. If
3 Although certain games, like Portal, have successfully introduced innovative physics systems to established genres.
10
Joris Dormans | Engineering Emergence
there is room for creativity and innovation it is with the mechanics that govern
these economies, and not with the physics of the game.
Still, looking back at four decades of computer game history, one must observe that physics has evolved much faster than any other type of mechanics in
games. Physics is relatively easy to evolve because we have access to Newtonian
mechanics and increasingly more computing power. The same solution does not
apply to other types of mechanics. Calling all game rules ‘mechanics’ might
distract developers from the fact that not all types of rules can be understood
in the same way. Worse, developers might falsely assume that those other types
of mechanics will turn out right, as long as we keep throwing more detailed
rules and more processing power at it. The term mechanics is an unfortunate
misnomer exactly because it might be holding back the development of proper
understanding of different types of game rules. It might cause us to turn a blind
eye to the artificial, discrete nature of those rules that are not part of the physical mechanics, but which are an equally important aspect of what makes games
truly clever and unique. Without a solid theoretical framework for non-physical,
discrete mechanics it is hard to evolve mechanics of that type beyond a certain
point.
I will still use the term mechanics throughout this dissertation, as is customary within the game industry. However, in using the term, I will refer to the
discrete mechanisms that generate a game economy more often than I will refer
to continuous physical mechanics of motion. When appropriate I will differentiate between these and other types of rules, as mechanics do not impact all types
of games equally.
1.3
Game Classification
What type of rules drives the gameplay of a particular game varies a lot
between games and genres. Some games derive their gameplay mostly from their
economy, others from physics, level progression, tactical maneuvering or social
dynamics. Categorizations of games in different genres by the game industry
and game journalists is usually based on the type of gameplay (Veugen, 2011,
42), and thus by extension on the different types of rules that feature more or
less prominently in these genres. Figure 1.7 provides an overview of a typical
game classification scheme and how these genres and their associated gameplay
relate to different types of rule systems. Note, however, that this classification is
one of many. There is a serious lack of consensus among the several classification
schemes in use. The point here is not to present a definitive genre classification.
Rather, it is to indicate how different types of rules correlate to different types
of gameplay. There are many more genres and sub-genres that can be derived
from this basic classification. For example, first-person shooters are a particular
sub-genre of action games, whereas action-adventure games are common hybrids
of the action and adventure game genres.4
In figure 1.7 I distinguish between five different types of mechanics. The
4 In fact, action-adventures are so common that they constitute a separate genre in most
other genre classifications.
Chapter 1 | Introduction
11
Figure 1.7: Games genres taken from Adams & Rollings (2007) and correlated to five
different types of game rules or structures. The thickness and darkness of the outlines
indicate relative importance of those types of rules for most games in that genre.
12
Joris Dormans | Engineering Emergence
boundaries between these types of mechanics are not very hard, and a single
game can have multiple types of mechanics. The figure indicates typical configurations of these types of mechanics as they are frequently encountered across
game genres, but it should be clear that each individual game can have its
own, unique configuration of game mechanics. The mechanics of physics and
economy were already discussed in detail in the previous section. Progression,
tactical maneuvering and social interaction are new and will be discussed below.
Progression deals with those aspects of gameplay that stems from quality
level design and mechanics that control player progress through these levels. In
these games, designers have created levels in which players need to overcome a
predefined set of challenges. Completing a particular challenge will often unlock
other challenges, and this way players progress towards a particular goal. For
most of these games, the goal is to reach a particular location (where usually
the final challenge awaits in the form of a “boss fight”). For this type of game,
careful lay-out of the levels creates a smooth experience. They tend to take
longer to complete than games that do not rely on level progression, but once
they are completed, they offer little replay value: many players play through this
type of game only once. Because the play experience and progress through a
progression-driven game can be tightly controlled by a designer, this type of game
lends itself particular well to games that also deliver stories. Typical examples
of level-driven games include action-adventure games such as The Legend of
Zelda or Assassins Creed, first-person shooter games such as Half-Life or
Halo, and role-playing games such as Baldur’s Gate or The Elder Scrolls
IV: Oblivion.
Tactical maneuvering involves those mechanics that deal with the placement
of game units on a map for offensive or defensive advantages. Tactical maneuvering is critical in most strategy games, but also features in certain role-playing
games and simulation games. The mechanics that govern tactical maneuvering
typically specify what strategic advantages units gain from being at a particular
location. These mechanics might be continuous or discrete, but discrete, tile
based mechanics still seem to be common. Tactical maneuvering is important
in many board games such as Chess and Go but also computer strategy games
such as StarCraft or Command & Conquer: Red Alert.
Much social interaction that emerges from playing a game is not captured
with mechanics. As soon as a multiplayer game allows direct, in-game interaction, social interaction outside the rules emerges. Some games include mechanics
that deal with that sort of interaction more explicitly. For example, role-playing
games might have rules that guide the play-acting of a character, and a strategy game might include rules that govern the forming and breaking of alliances
between players.
This dissertation mostly zooms in on the discrete mechanics of economy and
progression. Three reasons for this are:
1. As should become clear from figure 1.7 these types of mechanics play a role
in most game genres. They are more common than tactical maneuvering
and social interaction.
Chapter 1 | Introduction
13
2. As was discussed above, there is usually more freedom for design in those
mechanics that are, to a certain extent, discrete. Continuous physics generally aim to accurately simulate a real or imagined setting, the required
knowledge can be taken directly from real physics. For discrete mechanics and game economy, there exist far fewer off-the-shelf solutions. This
dissertation aims to contribute to the development of applied theory to
improve this.
3. In order to control the scope of this dissertation, not all types of mechanics
can be discussed in equal detail. Tactical maneuvering and social interaction in games are both very large topics that would warrant independent,
detailed study.
1.4
Emergence and Progression
Mechanics of progression correspond to what Jesper Juul calls “structures of
progression” in games which he separates from “structures of emergence” (Juul,
2002). His classification is very influential within game studies and provides a
relevant framework for the study of mechanics in this dissertation. Put simply,
emergence indicates that relatively simple rules lead to much variation, whereas
progression indicates that many predesigned challenges are ordered sequentially.
According to Juul, “emergence is the primordial game structure” (Juul, 2002,
324) that is caused by the many possible combinations of rules in board games,
card games, strategy games and most action games. Games of this type can be
in many different configurations or states: all possible arrangements of playing
pieces in a Chess constitute different game states as the displacement of a
single pawn by even one square is a critical difference. The number of possible
combinations of pieces on a Chess board is huge, yet the rules easily fit on a
single page. Something similar can be said of the placements of residential zones
in the simulation game SimCity or the placement of units in the strategy game
StarCraft.
Progression, on the other hand, relies on a tightly controlled sequence of
events. Basically, a game designer dictates what challenges a player encounters
by designing levels in such a way that the player must encounter these events
in a particular sequence. The use of computers to mediate games have made
this form possible. Progression requires that the game is published with much
content prepared in advance, for board games this is inconvenient.5 As such,
progression is the newer structure, starting with the text-adventure games from
the seventies. In its most extreme form, the player is ‘railroaded’ through a
game, going from one challenge to the next or failing in the attempt. With
progression the number of states is relatively small, and the designer has total
5 Published scenario’s for for pen-and-paper role-playing games are examples of non-digital
games of progression. They take the form of books specifying setting, characters and possible
storylines. However, they cannot be considered to be older forms of progression in games than
computer games, as pen-and-paper role-playing originate from the same period as computer
games of progression.
14
Joris Dormans | Engineering Emergence
control over what is put in the game. This makes games of progression well
suited to games that tell stories.
In the original article Jesper Juul expresses a preference for games that include emergence: “On a theoretical level, emergence is the more interesting
structure” (Juul, 2002, 328). He regards emergence as a structure that allows
designers to create games where the freedom of the player is balanced with the
control of the designer: with a game of emergence designers do not specify every
event in detail before the game is published, though the rules may make certain
events very likely. In fact, a game with an emergent structure often still follows
fairly regular patterns. Juul discusses the gun fights that almost always erupt in
a game of Counter-Strike (Juul, 2002, 327). Another example can be found
in Risk where the players’ territories are initially scattered all over the map,
but over the course of play their ownership changes and the players generally
end up controlling one or a few areas of neighboring territories. Despite these
emerging patterns Juul acknowledges that most games combine emergence and
progression. The main example in Juul’s article, EverQuest, is “a game of
emergence, with embedded progression structures” (Juul, 2002, 327).
In his book Half-Real, Juul is more nuanced in his discussion of emergence and
progression (Juul, 2005). Most modern games fall somewhere between games
of emergence and progression. Grand Theft Auto: San Andreas has a
vast open world, but also a mission structure that introduces new elements and
unlocks this world piece by piece. In the story-driven first-person shooter game
Deus Ex the storyline dictates where the player needs to go next, but players
have many different strategies and tactics available to deal with the problems
they encounter on the way. It is possible to write a ‘walkthrough’ for Deus Ex,
defining it as a game of progression according to Juul’s classification, but there
are many possible walkthroughs for Deus Ex. Just as, at least in theory, it is
possible to create a walkthrough for a particular map in SimCity, instructing
the player to build certain zones or infrastructure at a particular time in order
to build an effective city. It would be hard to follow such a walkthrough, but
creating one is possible. Pure games of emergence and pure games of progression
represent two extremes on a bi-polar scale, but most games have elements of
both. Yet at the same time, emergence and progression are presented as two
alternative modes of creating challenges in games, that might co-exists in a game,
but are hard to integrate. This dissertation questions this perspective and seeks
strategies to merge structured level design and emergent, rule-based play more
effectively.
One trajectory towards an answer is that emergent behavior thrives somewhere on the border of chaos and order (cf. Salen & Zimmerman, 2004, 155).
A true chaotic system will seem random and meaningless to most observers,
whereas in games it helps if the player can make sense of what is going on.
Where rules push games towards chaos by introducing dynamic behavior, levels
pull games back towards order by imposing structure. If games are pulled too far
back, they become games of progression where the spatial structure dominates
the rules and little dynamic play remains.
This dissertation acknowledges that most games display complex, emergent
Chapter 1 | Introduction
15
Figure 1.8: Civilization III is a good example of a game with a vast open world for
the player to explore, conquer and shape.
behavior. Many games structure this behavior through level design, but some
games can do without. Many casual games, such as Bejeweled, are purely
games of emergence. Many other casual games, such as Angry Birds, have
many pre-designed levels that confront players with new challenges, but they
are more puzzles than structured, story-like play experiences. For these games,
once the mechanics are in place, many new puzzles and levels can be generated
endlessly, not unlike the game of Tangram.
On the other hand, pure games of progression are quite rare. The most
typical examples of these games are text-adventures such as Colossal Cave
Adventure or Zork. But that game genre became almost extinct over two
decades ago. Today adventure games are almost always action-adventure games;
they almost always include some form of mechanics-driven, emergent action as
part of the gameplay. So even though games can have both emergence and
progression, it seems that modern games cannot do without the first, but can
do without the latter.
So-called ‘sandbox games’ create an open, virtual world that is not designed
to guide the player towards a particular goal. Sandbox games roughly correspond
with the management simulation genre in figure 1.7. In this type of game, players
are free to explore as they see fit, whether this is from a first person perspective
as in Grand Theft Auto III or from the god-like perspective in a game like
SimCity or Civilization (see figure 1.8). In a typical sandbox game there are
few restrictions and many optional goals for the player to pursue, some of these
goals are set by the player, not the game. Will Wright, the designer of SimCity
and many other simulation games, is quoted to have stated that his games are
more like toys as they do not dictate any goals (in Costikyan, 1994). These
games do not define a variable, quantifiable outcome. Instead, players set and
value their own goals.
As a medium for telling stories or delivering a concise play experience, vast
open worlds are not always the best option. Worlds have gotten so large that
Joris Dormans | Engineering Emergence
16
the player can easily lose track of the main storyline. Hunting down the story
and making your way through yet another dungeon can be experienced as quite
tedious. It is a flaw that large games such as The Elder Scrolls IV: Oblivion or Fallout 3 suffer from. It is also the reason why Chris Crawford does not
put too much faith in this structure for interactive storytelling (2003b, 261-262).
Yet, the artificial worlds found in games seem to grow larger and more detailed
with every new release, indicating that progression remains a relevant aspect of
game design.
1.5
Emergence
The use of the term emergence in games, which predates Juul’s categories
(for example see Smith, 2001), is often in reference to the use of the term within
the sciences of complexity. There it refers to behavior of a system that cannot be
derived (directly) from its constituent parts. At the same time Juul cautions us
not to confuse emergent behavior with games that display behavior the designer
simply did not foresee (Juul, 2002). In games, as in any complex system, the
whole is more than the sum of its parts. While the active agents or active
elements in a complex system can be quite sophisticated in themselves, they
usually can be simulated as rather simple models. Even when the study is about
the flow of pedestrians in different environments, great results have been achieved
by simulating them with only a few behavioral rules and goals (Ball, 2004, 131147). Similarly, the elements that make up games can be a lot more complex
than the elements of a typical system studied by the science of complexity, but at
least some games (such as Go and Chess) are famous for generating enormous
depth of play with relatively simple elements and rules. The active substance of
these games is not the complexity of individual parts, but the complexity that
is the result of the many interactions between the parts.
The main assumption of this dissertation is that the particular configurations
of elements into complex systems that contribute to emergence in other systems
also cause interesting gameplay. In other words: gameplay is an emergent property of a game system defined by its rules. For game designers this means that
understanding the structural characteristics of emergent systems in general, and
in their games in particular, is essential knowledge.
One of the simplest systems that show emergent behavior is a particular
class of cellular automata studied by Stephen Wolfram (2002). The cells of
cellular automata are relatively simple machines that abide only to local rules.
The algorithm that defines their behavior takes input only from its immediate
surroundings. In this particular class of cellular automata, the cells are aligned
on a line, the state or color of each cell is determined by the previous state
of that cell and its two immediate neighbors. With only two possible colors,
this creates eight possible combinations. Figure 1.9 displays one set of possible
rules (on the bottom) and the resulting, surprisingly complex pattern (on top).
This pattern is created because each iteration of the system is displayed on a
new horizontal line below the previous iteration. Wolframs’s extensive study has
revealed three critical qualities of systems that exhibit dynamic behaviors: 1)
Chapter 1 | Introduction
Figure 1.9:
Stephen Wolfram’s
http://www.stephenwolfram.com
17
’Rule
30
Automaton’,
taken
from
They must consist of simple cells whose rules are defined locally, 2) the system
must allow for long-range communication, and 3) the level of activity of the cells
is a good indicator for the complexity of the behavior of the system. These
qualities are discussed below.
The easiest way to implement a cellular automaton on a computer is to
program a simple state-machine that takes its own state and the states of its
immediate neighbors as input for the function that determines its new state after
each iteration. This communication of each cell’s state plays an important role in
the emerging behavior, without such input all cells would behave individually,
and system-wide behavior would not be possible at all. In order to get more
dynamic behavior communication between neighboring cells must lead to longrange communication in the system. This type of long-range communication is
indirect and takes time to spread through the system. Systems that show pockets
of communication with little or no communication between the pockets will show
less complex behavior than systems in which such pockets do not occur or are less
frequent (Wolfram, 2002, 252). Connectivity is a good indicator of long-range
communication in the system. A special case of long-range communication is
feedback: a cell or group of cells produce signals that ultimately feed back into
its own state somewhere in the future. Long range communications travel over
long distances through the system or, alternatively, through time and produce
delayed effects. As we shall see throughout this dissertation, this sort of feedback
is very important for games.
The number of cells that are active (cells that change their state) is important
18
Joris Dormans | Engineering Emergence
for the behavior of the system as a whole. Complex behavior, that is behavior
that is hard to predict but still seems to follow some sort of logic or hidden
pattern, occurs mostly in systems with many active cells (Wolfram, 2002, 76).
Cellular automata show us that the threshold for complexity is surprisingly
low. Relatively simple rules can give rise to complex behavior. Once this threshold is passed introducing extra rules does not affect the complexity of the behavior as much (Wolfram, 2002, 106).
In another study of emergence, Jochen Fromm builds a taxonomy of emergence that consists of four types of emergence (types I, II, III and IV). These
types can be distinguished by the nature of communication, or feedback, within
the system (Fromm, 2005). Feedback is created when a closed circuit of communication exists within a system; in effect, when a state change of a particular
element directly or indirectly affects the state of the same element later on.
Feedback is called positive when these effects strengthen themselves, as is the
case with guitar feedback where strings are vibrated to produce sound, and amplification of the sound causes the strings to vibrate in turn. Feedback is called
negative when the effect dampens itself. A thermostat is a typical example, a
thermometer detects the temperature of the air, when it becomes too low it will
activate a heater, the heater will then cause the temperature to rise which in
turn will cause the thermostat to turn off the heater again. Negative feedback
is often used in this way to maintain balance in a system.
In the simplest form of emergence, nominal or intentional emergence (type
I), there is either no feedback or only feedback between agents on the same level
of organization. Examples of such systems include most man-made machinery
where the function of the machine is an intentional (and designed) emergent
property of its components. The behavior of machines that exhibit intentional
emergence is deterministic and predictable, but lacks flexibility or adaptability. Both the guitar feedback and the thermostat are examples of this type of
predictable feedback.
Fromm’s second type of emergence, weak emergence (type II), introduces
top-down feedback between different levels within the system. Flocking is an
example he uses to illustrate this type of behavior. A flock-member reacts to
the vicinity of other flock-members (agent-to-agent feedback) and at the same
time perceives the flock as a group (group-to-agent feedback). The entire flock
constitutes a different scale than the individual flock-members. A flock-member
perceives and reacts to both.
One step up the complexity ladder from weakly emergent systems we find
systems that exhibit multiple emergence (type III). In these systems multiple
feedback traverses the different levels of organization. Fromm illustrates this
category by explaining how interesting emergence can be found in systems that
have short-range positive feedback and long-range negative feedback. It propels
the appearance of stripes and spots in the coat of animals and the fluctuation
of the stock-market. John Conway’s Game of Life is also an example of this
type of emergence (Gardner, 1970).6 The Game of Life can easily be shown
6 Although called the Game of Life, Conway’s cellular automaton does not fall in the
category of games as defined earlier in this chapter. It does not have any quantifiable goal and
Chapter 1 | Introduction
19
to include both positive feedback (the rule that governs the birth of cells) and
negative feedback (the rules that govern the death of cells). The Game of Life
also shows different scales of organization: at the lowest end there is the scale of
the individual cells, on a higher level of organization we can recognize persistent
patterns and behaviors such as gliders and glider-guns.
Fromm’s last category is strong emergence (type IV). His two main examples
are life as an emergent property of the genetic system and culture as the emergent property of language and writing.7 Strong emergence is attributed to the
large difference between the scales on which the emergence operates and the existence of intermediate scales within the system. Strong emergence is multi-level
emergence in which the outcome of the emergent behavior on the highest level
can be separated from the agents on the lowest level in the system. For example,
it is possible to set up a grid of the cellular automata used for the Game of
Life in such way that on a higher level it acts as a Turing Machine which in
itself also displays emergent behavior. In this case causal dependency between
the behavior displayed by the Turing Machine and the Game of Life itself is
minimal.8
From this brief discussion a number of important observations on the nature
of emergence comes forward. Within this dissertation emergent behavior is attributed to feedback loops in the system, and preferably multiple feedback loops.
Only one would only lead to nominal (type I) feedback. Therefore, emergent systems must consist of multiple elements that act more or less independently. A
sufficient level of activity is required; a system with only a few active elements
tends to be too stable and predictable to make for interesting games. Communication (or interaction) must exist between these elements at a local scale
and this local communication must indirectly enable long range communication.
Feedback, a form of communication where information and actions are fed back
to the source, often causes emergent behavior, especially when more than one
feedback loop affects the system. Finally, emergent systems often show different
scales of organization, with communication and feedback traversing these scales.
1.6
Progression
Despite the importance of emergence in games, no professional game designer
can turn a blind eye towards level design and mechanics of progression. To
subject yourself to game rules is to cross the boundary into the magic circle
and to immerse yourself in the game’s fictional space. Within that space the
does not require any effort by the player. However, as we have seen with toys, players of the
Game of Life can set goals themselves, such as finding configurations that will live for a very
long time, or grow into stable systems. These goals are quantifiable and do require effort to
reach.
7 One could question whether in both cases one follows from the other, or whether they
have evolved in unison. Emergence might not be the best way to describe their mysteries. In
fact, some researchers express serious doubts whether or not strong emergence can exist at all
(Chalmers, 2006).
8 Although this is not the same as claiming that a Turing Machine could emerge from a
particular, seemingly random, starting condition for the Game of Life.
20
Joris Dormans | Engineering Emergence
Figure 1.10: In Half-Life 2 the player arrives in the game by train, but never leaves
the rails.
player starts to explore the game and its possible states. The number of rules,
interface element and gameplay options of a modern retail video game is usually
larger than most players can grasp at once. Even smaller games found on the
Internet frequently require the player to learn a multitude of rules, to recognize
many different objects and to try out different strategies. Exposing a player
to all these at the same time can result in an overwhelming experience, and
players will quickly leave the game in favor of others. The best way to deal
with these problems is to structure the game experience with clever level design
that teaches the player the rules in easy-to-handle chunks. In many cases games
include special tutorial levels to introduce a player to the core concepts, and
even then they will introduce new concepts with extreme care.
The use of tutorials and level design to train the player is an illustration
of one of the strengths of the medium of games: the use of game space to
structure player experience. Unlike literature or cinema, which are well suited
to depict events in time (histories), games are well suited to depict space. Henry
Jenkins places games in the tradition of spatial stories, an honor they share with
traditional myths and hero’s quests as well as modern works by J.R.R. Tolkien
(Jenkins, 2004). Simply by traveling through the game space, a story is told.
A similar sentiment is found in Ted Friedman’s essay on Civilization (1999)
where the drama of that game directly stems from the player’s journey through
and conquest of a virtual world.
Many games have utilized this capacity to great effect. The Half-Life series
stands out as a particular good example. The games from this series are firstperson shooter action games in which the player traverses a virtual world that
seems to be vast but which in reality is confined to a quite narrow path. The
Chapter 1 | Introduction
21
whole story of Half-Life is told within the game, there are no cut-scenes that
take the player out of the game, all dialog is performed by characters inside the
game, and the player can choose to listen or ignore them altogether. HalfLife has perfected the art of guiding the player through the game, creating a
well-structured experience for the player. The practice is often referred to as
‘railroading’; in this light it is probably no coincidence that in Half-Life and
Half-Life 2 the player arrives in a train (see figure 1.10).
1.7
Approach and Dissertation Outline
Designing games for emergent gameplay and coherent progression using discrete mechanics presents designers with many problems. Perhaps the biggest
handicap for the game industry is the lack of formal, theoretical tools to deal
with the complexity of emergent gameplay and to assist the design of games.
Several prominent designers and academics have answered Doug Church’s call
to develop “formal abstract design tools” (Church, 1999). This dissertation also
answers that call. By developing applicable design theory for discrete game mechanics and level progression I hope to help designers understand the complex
nature of game systems and to get a better grip on the elusive notion of gameplay and to create progression efficiently. In this respect this dissertation does
not discriminate between board games and computer games, both are essentially
rule-based artifacts and that can have emergent gameplay. The games discussed
in this dissertation come from both categories in more or less equal measure.
The development of prototype software tools to assist or automate the design process is an important aspect of this dissertation. In many ways, the
development of these prototypes was an important step in validation of the theory’s applicability. At the same time, implementation invariably led to further
improvements of the theory in what turned out to be a highly iterative process.
In the next chapter I will explore games as rule based systems in more depth.
Games share some relations with simulations, which are also rule based systems.
However, where simulations aim to model a source system accurately, games
have a different goal: they aim to create an interesting experience. This means
that games can deal with rules differently.
Chapter 3 discusses game design theories from the game industry and from
academia. Although all these theories have their own merits, no theory has
emerged as an industry or academic standard. In fact, some people doubt that
any theoretical or methodological approach to the design of games can work, as
none of these can do justice to the creativity involved in designing games. Chapter 3 will also address the arguments put forward by the people who subscribe
to this opinion.
In Chapter 4 I will present my Machinations framework as an a alternative
theory of game design focusing on internal economy and emergent gameplay.
The design of this framework takes into account the concerns that have been
discussed in Chapter 3. It utilizes an abstract, visual notation to represent
discrete game mechanics. The digital version of these Machinations diagrams
can be run in order to simulate a game. The Machinations framework aims
Joris Dormans | Engineering Emergence
22
to foreground those structural qualities of game mechanics that contribute to
(good) gameplay.
The Machinations framework focuses on game economy and neglects level
design. In Chapter 5, game levels and mechanics of progression take center
stage. In this chapter I will develop the Mission/Space framework, which offers
a structural perspective on these elements of game design. As an illustration of
how the Machinations framework and the Mission/Space framework can be used
to inform game design on a theoretical level, I will leverage both frameworks in
Chapter 6 to explore how emergence and progression might be, and occasionally
have been, integrated in games.
In Chapter 7 I will unite the formal perspectives on games presented in Chapters 4 and 5 under the notion of game design as a series of model transformations.
Model transformation, a concept taken from software engineering, describes how
models, each representing a different perspective on the subject matter, can be
transformed into other models through the use of formal grammars and rewrite
systems. Model transformations for game design provide theoretical leverage to
automate certain aspects of the design process, as will be illustrated with the
discussion of some experimental software prototypes that generate game content.
In Chapter 8 I will evaluate the applied theory and the tools developed during
this research and answer the main research question: what structural qualities
of game rules and game levels can be used in the creation of applied theory
and game design tools to assist the design of emergent gameplay? I will discuss
the result of integrating them into the game development courses taught at the
Hogeschool van Amsterdam (Amsterdam University of Applied Sciences) and
the workshops I have hosted at several industrial and academic conferences.
The reception of the tools and theory presented in this dissertation by students,
industry veterans, and academic peers is an important aspect of its validation.
1.8
Terminology
The following terminology is used throughout this dissertation:
• A game is a system in which players engage in artificial conflict, defined
by rules, that results in a variable, quantifiable outcome affected by player
effort and ability.
• Gameplay is an emergent property of the game as defined by its rules.
Gameplay is a qualitative measure of player actions and experience; good
games are said to have gameplay.
• The mechanics of a game is a set of rules governing the behavior of a
single game element. These rules are specific: for example, a mechanism
specifies exactly how fast a character moves, how high a character jumps
and how much energy this costs. In this dissertation I will prefer using
“game mechanism” over the slightly awkward, but commonly used, singular form of “game mechanic”. This dissertation focuses on discrete game
mechanics, not on continuous physics, as these mechanics generally offer
Chapter 1 | Introduction
23
more design-freedom and are less well understood because the reference
systems available to model discrete mechanics on are less established.
• The core mechanics of a game are those mechanics that players interact
with most frequently and which affect the gameplay the strongest. The
boundaries of this set are fuzzy.
• The internal economy of a game is constituted by the production, flow
and consumption of game resources. These resources include, but are not
restricted to, power-ups, collectibles, points, items, ammunition, currency,
health and player lives. These resources can be tangible or abstract. The
structure of the mechanics that determine the production, flow, and consumption play an important role in the emergent gameplay of the game.
• A level is a particular spatial and/or logistical structure in a game that
dictates what challenges players encounter. Typically, a level contains a
set of positioned game elements and/or scripts to control special events
and players’ progress through the game.
• Emergence in games refers to the fact that the behavior of certain games is
the result of a complex and dynamic system of rules. This means that for
these games the number of possible states is huge: relatively few, and often
discrete mechanics can create a large number, sometimes even infinite, of
possible states. Emergence is an important source of gameplay and replay
value, but it is also very hard to predict, design and control.
• Progression in games refers to the structures in games where a designer
outlined the possible game states beforehand, usually through level design
or through some form of game scripting. Progression offers much control
of the play experience but has the disadvantage that it generates relatively
low replay value.
• Feedback occurs when a change in the current state of a particular element
in a system affects the state of the same element at a later time. Feedback
requires that a closed circuit of causality causes the effects of the state
change to ‘feed back’ into the new state. Feedback plays an important
role in those structural qualities of game mechanics that contribute to
emergence and emergent gameplay.
24
Joris Dormans | Engineering Emergence
It seems that perfection is reached not when
there is nothing left to add, but when there is
nothing left to take away.
Antoine de Saint-Exupéry (1939)
2
Rules, Representation and
Realism1
Within the entertainment game industry, much effort is spent on making games
more realistic. Game productions get bigger every year as graphics, physics
modeling and artificial intelligence take huge strides to ever more realistic simulations. Although these developments advance our understanding of the medium
of games, few developers show an active interest in alternative approaches to
games. Certain type of games, such as serious games, suffer from this trend.
Compared to other educational media, serious games are already quite expensive to produce. If the players and producers of these games expect a level of
realistic sophistication that approaches the level found in triple-A titles these
games will fail to keep up.
This chapter presents an alternative perspective on games that breaks away
from realism and investigates games as a form of abstract, non-realistic, and rulebased representation. From this perspective, the strength of games does not lie
in the accurate modeling of fantasy worlds but in capturing complex systems
with relative simple rules, while still retaining the overall dynamic behavior of
the original system a game is trying to model. In this chapter I argue that
games, as rule-based systems, are excellent vehicles for knowledge, learning and
entertainment, irrespective to whether these games have been created for fun
or education. Even if they are not created with photo-realistic assets, detailed
physics simulations and multi-million dollar budgets.
Games constitute a new form of rule-based representation that is fundamentally different from static representation through non-interactive text, images
and sounds. According to Rune Klevjer, simulation is a form of procedural representation; simulation represents rules instead of events (2002). Gonzalo Frasca
classifies simulation as an alternative to narrative or representation (2003, 223).
Ian Bogost picks up on Frasca’s work when he defines simulation as follows: “A
simulation is a representation of a source system via a less complex system that
informs the user’s understanding of the source system” (2006, 98).
This link between games, rules and simulation is especially clear in the con1 This chapter also appeared in a slightly altered form as a journal article for Simulation &
Gaming (Dormans, 2011a).
26
Joris Dormans | Engineering Emergence
temporary focus on realism found in many modern games. Over the years games
have grown increasingly more realistic. The power of modern computers allows
us to render nearly photo-realistic images in real time; the visual and auditive
qualities of games quickly approach the quality found in cinema. Games often
refer to elements recognizable from real life: real cars, real environments, real
weapons. Games that look and feel realistic sell well. In some cases the reality
a game refers to is purely fictional. Games set in the Star Wars universe depict
many things that are not real, but still the players have a clear idea what a Star
Wars game should look, sound and feel like. Much industry research is aimed
at making games more realistic. Realism features prominently in the “top ten
hurdles facing game designers today” published on the website of the magazine
Popular Science. All are concerned with the accurate and realistic simulation of
real-life phenomena. Getting water and fire effects right made that list, as did
realistic movement, rendering human faces and artificial intelligence designed to
capture realistic behavior (Ward et al., 2007).
On the other hand, in certain circles of game critics and scholars, it is in
vogue to point out that realism is not what games are about. Steven Poole’s
Trigger Happy deconstructs the supposed realism of games. He argues that most
players play games because they allow them to do things that cannot be done
in reality. A thoroughly realistic race game, for example, would require a player
to undergo thorough training before he can even try to complete a single round
on a racing circuit. A game that is totally realistic ceases to be a game (Poole,
2000, 77). He concludes: “videogames will become more interesting artistically
if they abandon thoughts of recreating something that looks like the ’real’ world
and try instead to invent utterly novel ones that work in amazing but consistent
ways” (Poole, 2000, 240).
The sentiment that games are different from realistic and accurate simulations
can already be found in the early work of Chris Crawford who states:
“accuracy is the sine qua non of simulations; clarity the sine qua non
of games. A simulation bears the same relationship to a game that a
technical drawing bears to a painting. A game is not merely a small
simulation lacking the degree of detail that a simulation possesses; a
game deliberately suppresses detail to accentuate the broader message that the designer wishes to present. Where a simulation is
detailed a game is stylised” (1984, 9).2
Jesper Juul also points out that: “games are often stylized simulations; developed not just for fidelity to their source domain, but for aesthetic purposes.
These are adaptations of elements of the real world. The simulation is oriented
toward the perceived interesting aspects of soccer, tennis or being a criminal in
a contemporary city” (2005, 172). Games allow us to do things not available
to us in real life, and it is rules that grant us this power, as long as the player
follows them. However, rules create both limitations and affordances. Without
2 The paintings that are most stylized are modern paintings that strive to capture the essence
of that which they depict through non-realistic means. I assume that Crawford is referring to
this type of paintings.
Chapter 2 | Rules, Representation and Realism
27
rules, games would have little structure and actions would have little meaning
(Juul, 2005, 58). It is for similar reasons that I linked rules to agency in a study
on pen-and-paper role-playing games, even though many avid role-players tend
to downplay the importance of rules in favor for interactive play-acting. In a
pen-and-paper role-playing game the rules form an interface with the fictional
world, and it is through rules that players can affect that world; game rules
create agency (Dormans, 2006b). When looking at a game it is more important
to look at what rules allow, instead of how they limit the player (cf. WardripFruin et al., 2009). Pressing the jump button in a Super Mario Bros. game
has the satisfactory effect of making the on-screen avatar jump way beyond the
capabilities of any human being. Game rules amplify our own abilities and allow
us to explore strategies or tactics in artificial conflict that would be dangerous,
destructive, impractical or impossible in real life.
2.1
The Iconic Fallacy
Ian Bogost’s definition of a simulation as: “a representation of a source system via a less complex system that informs the user’s understanding of the
source system” (2006, 98) closely resembles the semiotic triparte model of the
sign drafted by Charles S. Peirce. This resemblance provides an opportunity to
investigate realism in games from a different, semiotic perspective. This perspective reveals that the current trend towards realism games is preoccupied with
only a small subset of possible forms: iconic forms. At the same time, Peirce’s
semiotic theory also suggest where to look in order to explore games and rules
beyond realism.
In Peirce’s triparte model a sign is connected to an object: that what the
sign represents, and an interpretant: the mental concept the sign invokes; which
“is neither an interpreter nor a user of signs” (Kim, 1996, 12). This model of the
sign is best known for the classification of signs into icons, indexes, and symbols.
This classification is based on the nature of the relation between the sign and
its object: when a sign resembles its object it is an icon, when the sign has
an existential connection to its object it is an index, and when the connection
is arbitrary it is a symbol (Kim, 1996, 19-21). Figure 2.1 combines Bogost’s
definition of simulation with Peirce’s model of the sign.
If games and simulation are forms of representation, then the same categories
of relations between their form (simulation) and that what they represent (source
system) apply to games. This perspective dictates that games, like any form of
representation, always signify something outside the game. This is true even
for games that are created for the purpose of entertainment only. No matter
Figure 2.1: Triparte model of signs and simulation.
28
Joris Dormans | Engineering Emergence
how much the “poetic function” (Jakobson, 1960) of an entertainment game
calls attention to its representational form, it still is a form of communication
that does refer to many meaningful and recognizable elements outside the game.
Cultural interpretations of relative simple and abstract entertainment games like
Pac-Man (Poole, 2000, 178-183) or Tetris (Murray, 1997, 143-144) have been
made, and even though these interpretations are sometimes quite far fetched, the
point is that, like any form of art, no game exists within a social and cultural
vacuum. As David Myers points out: “human play as a cognitive and symbolic
act that is fundamental to the human representational process” (1999a, 486).
In this semiotic model of games and simulation realism and iconicity are
linked. We call a simulation realistic when the simulation (as a system) closely
resembles the source system; we call a simulation realistic when it is iconic.
From this analogy two other forms of simulation suggest themselves: indexical
and symbolic simulation. If games are ultimately not realistic, then indexical and
symbolic simulation might be interesting notions to help us understand games
better. As we will see in the next two sections, constructions that we could call
indexical or symbolic have been used in games to great effect.
Before exploring indexical and symbolic simulations I would like to push the
analogy between linguistic signs and simulation one step further to make apparent an interesting discrepancy between the current focus on iconic games and
the highly symbolic nature of language. Natural language is by its nature very
abstract, not realistic; most words do not resemble what they stand for. And it
is the abstract nature of language that contributes to language’s great expressiveness. This notion can be traced back a long time. It was already apparent
in the works of seventeenth century philosopher John Locke who observed:
“Men making abstract Ideas, and settling them in their Minds with
names annexed to them, do thereby enable themselves to consider
Things, and discourse them, as it were in bundles, for the easier and
readier improvement, and communication of their Knowledge, which
would advance but slowly were their words and thoughts confined
only to Particulars” (Locke, 1975, 420).
It is on similar grounds that, roughly a century later, the philosopher Edmund
Burke attaches greater aesthetic power to poetry than to the realistic paintings
of his age. Poets use words to “obscure” the image they try to get across.
Paradoxically, this leads to a mental image that is more vivid and evocative
than painting a complete and detailed picture of the same thing(Burke, 1990, 55).
These days the development of abstract art has changed all this and has increased
the expressive power of the image dramatically, as is exemplified by the names
used by art history to identify particular genres: Impressionism, Expressionism,
Abstract Expressionism, etcetera.
Ferdinand de Saussure identifies the arbitrary character of the linguistic signs
as their principal characteristic. Although he does not rule out the possibility of
non-arbitrary signs, he argues that in human languages most signs are arbitrarily
linked to their meaning. There are usually no characteristics of what we are
referring to that are connected to the words we use (de Saussure, 1983, 67-69). In
Chapter 2 | Rules, Representation and Realism
29
other words, language consists mostly of symbols; there are only a few linguistic
icons and indexes. For Saussure too, it is the human faculty to construct a
“system of distinct signs corresponding to distinct ideas” that makes language
possible (de Saussure, 1983, 10). Through the human capability to take abstract
meanings and handle them in bundles, human expression and understanding
is taken beyond the level of particular things and into the realm of general
knowledge. In other words, abstract, non-iconic presentations contain more
expressive and representational power than realistic or iconic representations.
Ian Bogost’s definition of simulation quoted above is not complete. Bogost
emphasizes that subjectivity is inherent to simulation: “A simulation is a representation of a source system via a less complex system that informs the user’s
understanding of the source system in a subjective way” (Bogost, 2006, 98). In
a simulation a system is represented through another system and the choices
made in the construction of the second system reflect the values of its creator:
“no simulation can escape some ideological context” (Bogost, 2006, 99). As
Bogost insists, this subjectivity can be partly attributed to the fact that with
simulation the simulating system is by necessity less complex than its source
system. A simulating system always deviates from its source system and the
choices made in that deviation reflect the understanding and/or ideology of the
person or group that created the simulation. What Bogost exactly means with
‘less complex’ is not made explicit. Here, I interpret ‘less complex’ as ‘consisting
of fewer parts’. The number of parts in a simulation is usually lower than the
number of their counterparts in the source system. This also means that in most
cases those parts are abstractions of more complex subsystems in the source system. For example the parts that make up a simulated weather system bundle
many actual air-molecules that make up real weather. This makes the simulation more convenient to handle, or to paraphrase Locke: it enables us to consider
the multitude of parts of a simulated system in bundles for easier and readier
understanding, and for easier and readier communication and improvement of
that understanding.
Thus, there always exists a gap between a simulated system and its simulation, and that gap always renders the simulation subjective to a lesser or greater
extent. However, this subjectivity is the price we pay for the convenience and enhanced understanding that subjective simulations allow. In most cases the gain
in expressive power outweighs the loss in resemblance to particular instances.
When one considers a simulation as essentially subjective, it is worth noting
that any claim to realism becomes an ideological maneuver in itself. For example,
the high level of verisimilitude in America’s Army can be read as the rhetoric
claim that its apparent realism and correctness in visual representation can be
extended into the ideological domain: ‘the game got its physics right, so its
ethical claims must be realistic, too’ (Bogost, 2007, 78). On the other hand, in
commercial entertainment games realism is often rendered as a special effect. In
these games realism and authenticity becomes a spectacle designed to impress
and to be appreciated by the audience. Realism, with its high poly-count, plasma
effects and particle engines, is foregrounded and hyperreal. Or to use the words
with which Geoff King described a very similar phenomenon in blockbuster films:
30
Joris Dormans | Engineering Emergence
Figure 2.2: A schematic sketch of Diablo’s inventory screen.
it is “the hyperrealistic spectacle-of-authenticity rather than authenticity itself”
(King, 2000, 136).
2.2
Indexical Simulation
To start looking beyond iconic simulation, the notions of indexical and symbolic simulation are obvious points of departure. In this section I will discuss the
first notion, in the next section I will discuss the second. The ‘inventory system’
that first occurred in Diablo and that has since featured in many other games
can be seen as an example of the first. It inspired Warren Spector, developer
of Deus Ex, into saying that: “Diablo got Inventory right. There’s no sense
messing with something that works...”.3 For quite some years now, many computer games have included an ‘inventory’: the game allows the main character
to pick up objects and carry them around. The player can manage these objects
in the game’s inventory screen. Most games restrict the number of objects the
character can carry in some way. There might be a fixed number of objects
the character can pick up, or all the game objects might have a weight value
attached to it and the character can only carry objects up to a particular load.
Diablo’s inventory system takes object size as it main restricting factor (see
figure 2.2). Each item takes up a number of inventory ‘slots’, the available slots
are limited and organized in a grid. An item may take up 1x1, 2x2 or 1x4 slots
for example. Depending on the available room in the inventory an object can
be picked up or not. The upper half of the screen is dedicated to the objects the
main character currently has equipped.
I argue that this is an example of indexical representation in games. The
main restricting factors for somebody to carry objects in real life (shape, size and
weight) are represented by easily understandable two-dimensional shapes. These
3 Quoted on S. T. Lavavej’s Deus Ex webpage. URL: http://nuwen.net/dx.html (last
visited June 23, 2011). However, it must be said that opinions on this system differ; not
everyone is as enthusiastic as Warren Spector (see for example Adams & Rollings, 2007, 516518). Whether it is a good design decision to burden players with the upkeep of their inventory
depends on the type of game and intended gameplay.
Chapter 2 | Rules, Representation and Realism
31
shapes and their relative size can be said to be existentially connected to the size
and weight of their simulated counterparts. Therefore the simulation qualifies as
an indexical construction as it is parallel to indexical signs in which the relation
between the sign and its object is also based on an existential connection (rather
than resemblance or arbitrary convention).
The number of games that have copied this system in one form or another is a
testimony to the quality of this construction. The internal rules and constraints
are immediately apparent (not in the least because they are tailored towards
visual representation on a screen). The management problems the system gives
rise to are very much like those problems in real life. The system even allows
players to make an inefficient mess of their inventory, teaching them something
about the need to organize themselves.
What the Diablo inventory system does very effectively is to take many
related and similar functioning game rules and replace them all by a single
mechanism that is well suited to the medium of the video game. Obviously
some accuracy of simulation is lost (an item cannot be large and light at the
same time), but the overall behavior is retained (the players are limited in what
they can carry). The cleverness of the Diablo inventory is that it collapses
all the nuances of managing an inventory into a problem of size, which is easily
represented by a computer screen, instead of weight which was the more common
choice before, but which translate to the visual medium of the computer less well.
Another example of indexical simulation is the way most games handle ‘health’.
Health of characters and units is often represented by a simple metric, be it a
percentage or a number of ‘hit points’. Obviously in real life the physical health
of a person or the structural condition of a vehicle is a complex matter to which
many different aspects contribute. By using a generic health for a single character games bundle all these aspects into one convenient mechanism. Both players
and computers can easily work and understand the numerical metric to represent
the bundle.
2.3
Symbolic Simulation
Symbolic simulation goes one step further in breaking away from modeling
a system with rules that closely resemble the mechanisms of the source system.
The use of dice in many board games tends to be symbolic. For example, the
roll of a few dice can stand for a complete battle in a game of Risk. In this
case, the relation between rolling dice and fighting is arbitrary, and one simple
action well-known from other games is used to simulate a multitude of actions
for which most players would lack expertise. Dice can replace these battles
because, for the purpose of the game, as the player should have little influence
over the outcome of these battles. Risk is about global strategy, not about
tactical maneuvers on the field of battle. A player cannot control the result of
dice (not without cheating anyway) just as a supreme army commander cannot
win every battle personally. Yet, the player needs some sort of influence and
the rules tie in with the dice rolling: committing more armies to a battle allows
to player to roll more dice and improve chances for success. Something similar
32
Joris Dormans | Engineering Emergence
Figure 2.3: Kriegsspiel is played by maneuvering units around a real map, combat
is resolved using dice.
Figure 2.4: Jumping to avoid or defeat enemies in Super Mario Bros.
goes for Kriegsspiel (figure 2.3) and many successive war games. In contrast
to Risk, these games are all about tactical maneuvering on a battlefield. So
the rules for these maneuvers are quite elaborate. But the rules covering actual
fighting are left to dice and attrition tables. Again, these games are to train
tactical skills, not how to use a gun.
Dice are wonderful devices to create a nondeterministic effect without the
need of detailed rules. From a suitable high level of abstraction, a complex and
nondeterministic system, such as fighting, has a similar effects as rolling a few
dice. Especially when the player is not supposed to have much influence over this
system, dice mechanics can be used to replace the more complex system. The
characteristic randomness of different dice mechanics can be used to match many
superficial, nondeterministic patterns created by more complex systems. Penand-paper role-playing games have come up with many clever and interesting
ways of using dice, allowing more or less influence by the player. In fact, dice
mechanics related to a set of characteristics representing skills and attributes
forms the core of most pen-and-paper role-playing systems. Often the same
mechanism is used to represent a wide variety of actions.
Other examples, such as jumping on top of enemies in order to dispose them
in the classic video game Super Mario Bros. fall somewhere in between
symbolic and indexical forms of simulation (see figure 2.4). Although the precise
implementation differs from enemy to enemy, and certainly does not work against
all enemies, it is a frequent feature throughout the game and the series it belongs
to. It is unlikely that I am the first to point out that this method is a little odd,
to say the least. However, it has become a convention within platform games
that is instantly recognizable to gamers, and ties in with that genre’s defining
action of jumping from platform to platform.
The connection between jumping on top of something and defeating something in real life is not completely arbitrary, but its use in platform games has
become so conventional it parallels the definition of a symbolic sign in language.
In the real world, there are creatures that can be squashed by jumping on top of
Chapter 2 | Rules, Representation and Realism
33
them. However, there is no creature that I know of that is lethal when bumped
into, but not when stepped upon, which makes this exact mechanism somewhat
arbitrary.4 What is more, this method of fighting in Super Mario Bros. is
motivated more by the use of the genre’s most prominent action of jumping,
than it is motivated by any claim to realism. The link between the simulation
and what is simulated is both arbitrary and conventional. Especially in the
multitude of platform games that followed the example set by Super Mario
Bros.
There is, however, an affinity between the skills needed to defeat enemies
in Super Mario Bros. and in real life. In the game, it requires timing and
accuracy, which are among the skills involved in real fighting. The point is, the
simple representation in the game allows us to do more than to hone and train
those skills. The simple metaphor of jumping on top of enemies is easy to grasp
by the player, but the game then goes on by inviting the player to experiment
and develop strategies. In most platform games each level ends with a ‘boss’
enemy which is typically designed to test the effectiveness of players’ strategy.
It is the ultimate test for the players to demonstrate they understand and have
mastered the simulation, and are able to combine different moves.5
What the jumping on enemies mechanism accomplishes is a very clever way
of adding combat rules to a jumping game; it introduces no new actions for the
player. It manages to do this by replacing actions it tries to represent by other,
arbitrary rules already implemented in the game. This reduces the number of
actions players need to learn, allowing players to quickly move on to a deeper,
more tactical or strategical interaction with the game instead of fussing around
with its interface. As is argued below, symbolic simulation effectively reduces the
system to a simpler construction with more or less equivalent dynamic behavior.
2.4
Less Is More
Indexical and symbolic simulation tend to create simpler game systems than
iconic simulation. The reduction in rules these forms of simulation allow is
in general benevolent. Simpler games are easier to learn, yet they still can
be quite difficult to master. Games are not the only medium for which the
expression ‘less is more’ rings true. In almost any form of representational art
saying more with less means is appreciated, especially by critics and connoisseurs.
Christopher Alexander, drawing inspiration from poetry for his pattern language
for architecture and design puts it like this:
“This language, like English, can be a medium for prose, or a medium
for poetry. The difference between prose and poetry is not that different languages are used, but that the same language is used differently.
4 Even
if such creatures do exist, they are certainly are not tortoises.
lessons carry over to situations beyond the game. The mentality of the players
that have learned these lessons is excellently described by John Beck and Mitchell Wade: they
know that solutions will eventually present themselves, and they have mastered a trial and
error approach to many problems in life (2004, 11-14).
5 These
Joris Dormans | Engineering Emergence
34
In an ordinary English sentence, each word has one meaning, and the
sentence too, has one simple meaning. In a poem, the meaning is far
more dense. Each word carries several meanings; and the sentence as
a whole carries an enormous density of interlocking meanings, which
together illuminate the whole.” (Alexander et al., 1977, xli)
And:
“It is essential then, once you have learned to use the language,
that you pay attention to the possibility of compressing the many
patterns which you put together, in the smallest possible space. You
may think of this process of compressing patterns, as a way to make
the cheapest possible building which has the necessary patterns in it.
It is, also, the only way of using a pattern language to make buildings
which are poems.” (Alexander et al., 1977, xliv)
For poetic language, or rather for any form of representation art, this quality
is very important and does not stem from the use of abstract signs only. The
combination and structure of these signs, or to use the linguistic term, syntactical
relations between these signs also play an important role. In this light, Noam
Chomsky observed that language allows speakers to make infinite use of finite
means: the number of words we have may be limited (and is vastly outnumbered
by particular things in reality), the number of combinations we can make with
them is infinite (Chomsky, 1972, 17). This characteristic of language is often
called discrete infinity.
It is impossible to exactly quantify how many rules a game should have; it
is impossible to quantify how much less is how much more. Each individual
design has its own balance. A particular number of rules could be too few for
one game and too many for another. The balance a game should seek to strike
is between the number of gameplay options the rules create on the one hand
and the cognitive burden it requires to understand or operate those rules on the
other. Antoine de Saint-Exupéry’s famous quote “it seems that perfection is
reached not when there is nothing left to add, but when there is nothing left to
take away” (1939) applies extremely very well to games.
In general, games are very good at creating endless possibilities with only
a few rules. It is estimated that there are more possible game states in games
like Chess and Go than there are atoms on earth (see Shannon, 1950). It is
the rules of the game that determine the number possible states, but it is not
necessarily true that more rules will lead to more possible states. In addition,
when a game can create a large number of possible states without using many
rules, the game will be more accessible.
Possible game states and trajectories through a games state space are emergent properties of the game rule system. The elusive notion of gameplay is
related to these properties. Games that allow many interesting trajectories arguably have more gameplay than games that generate fewer trajectories or less
interesting ones. However, determining the type and quality of the gameplay
is hard, if not impossible, by simply looking at the rules. Comparing the rules
Chapter 2 | Rules, Representation and Realism
35
Figure 2.5: In Connect Four gravity makes sure players can only occupy the bottom
most, unoccupied square in each column.
of Tic-Tac-Toe and Connect Four serves as a good illustration of these
difficulties. The rules for Tic-Tac-Toe are:
1. The game is played on a three by three grid.
2. The players take turns to occupy a square.
3. A square can only be occupied once.
4. The first player to occupy three squares in a row (orthogonally or diagonally) wins.
The rules for Connect Four are (with the differences emphasized):
1. The game is played on a seven by six grid.
2. The players take turns to occupy a square.
3. A square can only be occupied once.
4. Only the bottom most unoccupied square in a given column can be occupied.
5. The first player to occupy four squares in a row (orthogonally or diagonally) wins.
While the differences in rules for these two games are only a few, the differences in gameplay are immense. Far larger than the difference in cognitive
effort needed to understand the rules. In the commercially available version of
Connect Four, the most complicated rule (number 4) is enforced by gravity: a
player’s token will automatically fall to the lowest available space in the upright
playing area (see figure 2.5). This relieves players from manually enforcing this
rule and allows them to focus on the rules effects instead. Despite the small
difference in the complexity of the rules, Tic-Tac-Toe is suited only for small
children, whereas Connect Four can also be enjoyed by adults. The latter
36
Joris Dormans | Engineering Emergence
game allows many different strategies and it takes a considerable longer time to
master the game. When two experienced players play the game, it will be an
exciting match, instead of a certain draw as is the case with Tic-Tac-Toe. It
is hard to explain these differences just by looking at the differences in the rules.
These days, emergence of complex behavior from relatively simple elements
is an important aspect of many fields of research in the domains of mathematics,
physics and social sciences. In the research of games, too, emergence is becoming
an increasingly important notion. From the computational side, emergence is an
important technique used in anything from development of artificial intelligence
to the realistic rendering of water and fire. For Penelope Sweetser the disadvantageous loss of creative control in a system that is set up for emergence is
outweighed by the more consistent and intuitive player interactions such systems
allow (Sweetser, 2006, 14). Likewise, game designer Harvey Smith argues that
attempting to design a totally controlled game environment that allows rich interaction is no longer economically viable, as the sheer amount of detail cannot
be efficiently produced manually (Smith, 2001).
One major advantage of games that feature emergent gameplay is that a
rule-system allows, and often even invites, players to experiment with the game,
instead of merely repeating the moves a game designer intended. Ultimately
emergent games allow the transformation of the game rules itself (Myers, 1999b).
This has severe consequences when building an educational game, but also when
the game designer has a particular story or message in mind. For Jan Klabbers it
is the responsibility of the game designer to shape the whole of the game system
in such a way that behavior that conforms the design specifications emerges from
its components. At the same time, the system should leave enough freedom for
players to act according to their own strategies, goals and incentives, in order to
elevate the position of the player into that of a reflexive actor. This is “one of
the major bottle necks in the design” (Klabbers, 2006, 102).
However, in some ways, computer games seem to be moving against the trend
of emergence. Jesper Juul differentiates “games of progression” from “games of
emergence” as a historical newer category associated with computer games. The
rise of computer games, and adventure games in particular, has made games of
progression possible, as without a computer the amount of data and the number
of special case rules facilitating the progression through a multitude of game
spaces would have become unwieldy (Juul, 2005, 5).
Chris Crawford’s notions of data intensity and process intensity (Crawford,
2003a, 89-92) can be pitted against Juul’s observation that games of progression
are a younger form and the implication that progression is the result of a natural
evolution of the medium. Crawford argues that computers are both suited to
handle large amounts of data and crunch vast quantities of numbers, but it is the
latter ability that sets computers apart from most other media. Handling data
is something that all media are good at. The computer often allows faster access
to remote locations within the data, an ability put to good use within hypertext
(Lister et al., 2003, 23-30). However, it is the ability to create new content on the
fly where the computer really shines. Like no other medium before, the computer
has the capacity to surprise players and designers alike (see also Smith, 2001).
Chapter 2 | Rules, Representation and Realism
37
For Chris Crawford, games should capitalize on this ability of the computer,
games should be process-intensive, rather than data-intensive. In other words,
games should be games of emergence rather than games of progression.
2.5
Designing Emergence
Designing emergence is a notoriously hard, somewhat paradoxical, task. Emergent properties of a system only surface when a system is put into motion. Even
when a system behaves in a certain way during all test, there is no guarantee it will do so all the time. In this light the realistic fallacy seems to be a
fairly conservative strategy to avoid the difficulties of designing truly emergent
games: that have relative simple systems and display interesting, complex behavior. Simply adding more, and more detailed rules is only a poor substitute
for creating complex gameplay through a lean and elegant rule system.
Emergence can be the result of relatively simple rules, therefore games do not
need to rely on complex rule systems in order to create interesting gameplay. On
the contrary, using simple means to generate complex gameplay has many advantages. The design becomes easier to manage for the designer, and the game
becomes easier to learn for the player. In the examples of non-iconic simulation above (Diablo’s inventory, the use of hit points, dice in Kriegsspiel and
jumping in Super Mario Bros.), the use of indexical and symbolic simulation
resulted in a simpler rule system than an iconic simulation would have. This is
not a characteristic of the examples discussed above, rather it is the advantage
of using non-iconic rules in games. Compared to a completely detailed, realistic
system that tries to simulate through accurate detail, indexical and symbolic
simulation aims to capture the essence of the source system with fewer means.
When done correctly, the result is a leaner, more elegant system that minimizes
on parts and maximizes on expressiveness.
In essence, indexical simulation bundles a number of related and more or less
isomorphic rules into one game mechanism. Symbolic simulation goes one step
further, it connects rules in the game where they would not be connected directly
in the source system. As in the use of symbols in language, there are symbols
that work better than others. The symbols that work best seem to connect two
unrelated rules that still have some affinity between them. In the case of Super
Mario Bros. there is a natural affinity between the physical skill and timing
involved in both jumping and fighting (also see Lakoff, 1987, 448).
The development of the serious board game Get H2O, in which I took part,
is very good example of the application of non-iconic reduction. In this game,
produced as part of an educational program for adolescents in East-Africa, the
players struggle to survive in the poor residential areas of an African metropolis
(see figure 2.6). The vital resources are scarce, players need to balance carefully
between personal gain and community efforts. The players only have indirect
influence over bad events that might happen, but sometimes players can benefit
from these events, sowing the seeds for conflict. The game simulates life in an
African metropolis, and is designed to give the players a top-down view of their
own lives. It is designed to function as a vehicle for exploration, discussion and
38
Joris Dormans | Engineering Emergence
Figure 2.6: A prototype of Get H2O being played in Nairobi (photo by Butterfly
Works).
reflection.
Instead of trying to simulate the East-African urban life in detail, the game
reduces the number of resources and rules to a relative simple set. The game uses
three main resources: money, houses and clean water. The latter two determine
how many actions a player can take each turn while the money is used to build
more houses. These resources are under constant threat. Money and water
might get stolen, houses might be burned down. In reality, an African family has
many more needs, but these three resources are enough to simulate an economy
of scarcity that behaves not unlike the real economy in urban African areas.
The indexical nature of this simulation makes it possible to create a relatively
simple system that is still recognizable for people who grow up in those areas.
In fact, the game became more recognizable because it lacked detail: fewer
details creates more room for personal interpretation and less chance that the
game does not match the individual experience. What is more, this economy
creates the particular balance between short term personal gain and long term
community interests that causes social instability. Creating this instability was
the prime design goal for the game. The game is supposed to train people in
dealing which such a situation in the first place. Simply put, more resources and
a more complex economy were not needed to replicate the volatile social system
of an East-African urban sprawl.
The game also uses symbolic simulation. After every player has taken a
turn, all players discard one playing card without revealing it. These cards are
normally used for player actions. The discarded cards are then shuffled and
revealed. Every card has a symbol representing bad things that might happen,
from corruption and pollution to arson and drought. If the same symbol is played
Chapter 2 | Rules, Representation and Realism
39
twice the effects are aggravated: one drought symbol does nothing, but two
drought symbols indicate that a drought strikes, often with devastating effects.
Obviously playing cards have nothing to do with the occurrence of real droughts.
The cards are a way of simulating bad events that are mostly beyond the control
of the people living in an African metropolis. One that also conveniently ties in
with other mechanics of the game: all players will also get a secret role which
allows them to benefit from bad events such as corruption, scapegoating and
arson.
When used correctly, indexical and symbolic reduction reduces the number
of elements in a system without affecting its structural complexity and emergent
properties too much. In the Get H2O example many similar resources that
are needed on a daily basis are replaced by just one: water. The feedback
structure that entails having access to these resources is pretty much unchanged.
The number of feedback loops, for example, is not affected. In fact, the game
emphasizes these structural features, by taking away unnecessary detail. By
reducing the number of elements in a game system the cognitive burden of the
player in keeping track of all these elements is also reduced, allowing the player
to focus more on these features and the strategic interaction that they allow, or
in the case of Get H2O, on the social implications they have.
There are more advantages. A system that uses indexical and symbolic simulation can concentrate the experience, allowing a complete session of play to
run much quicker than what the play represents in real time. The player is
confronted with the results of his actions fast and efficiently. It allows players
to ‘handle the rules in bundles for the easier and readier improvement of their
understanding of the system’. On the one hand this allows players to go through
the process more often and on the other hand it will contribute to the pleasurable experience of agency and power that drives many commercial entertainment
games. In the Get H2O game this was certainly one of the design goals. The
game can be played in roughly forty-five minutes, allowing players to experiment
with different strategies efficiently while reducing the costs of failure.
For the designer of games there are advantages, too. A game system that is
reduced to its essence becomes better manageable and easier to balance. Without many parts, the designer can focus on those elements and structures that
contribute directly to the game’s emergent behavior and more easily tweak that
behavior into the desired shape. Games would do well to strive for non-iconic,
discrete infinity rather than detailed realism. Not only is this economically more
feasible, it is also more interesting artistically and it allows for more effective
communication.
The Legend of Zelda series is a great example of gameplay design in
which only a handful of game objects and associated rules are combined in many
interesting challenges. The value of each of these objects and their rules does
not stem from its power to represent some sort of realistic aspect of adventuring
through a dungeon, but form a potential combination with other objects and
rules. The exploration challenges, which the series is famous for, are almost
always the result of combinations of simple, reusable gameplay mechanics that
are often quite indexical or symbolic. For example, in the The Legend of
Joris Dormans | Engineering Emergence
40
Zelda: Twilight Princess the player can find the ‘gale boomerang’ in the
‘Forest Temple’ level, which creates a gust of air strong enough to activate windoperated switches and carry small items to the player. This boomerang can
be used to carry ‘bomblings’, little creatures that explode a few seconds after
the player grabs him by hand or with the gale boomerang. Effective use of
this combination is required to defeat the final boss in that level. The same
boomerang is used in The Legend of Zelda games for the Nintendo DS, but
in this case the player can use the stylus to draw the path of the boomerang
quite freely, directing it around obstacles unrealistically. Players appreciate this
sort of structures as they have the advantage of being inheritably coherent.
And as has been pointed out before, coherence is a strong contributing factor to
gameplay (Poole, 2000, 64-66). One can even argue that the appreciation of such
structures is in its essence an aesthetic appreciation (Huizinga, 1997, 25). It is
the appreciation of the craftsmanship of the game designer in building systems
with interesting structural qualities from which interesting behavior emerges.
It forces to pay attention to the way the game was constructed and how it is
structured (cf. Ryan, 2001, 176).
The meaning that emerges from these games is not necessarily less detailed
or less valuable than games that aim for detailed and realistic simulation. On
the contrary, as the challenges of exploration in The Legend of Zelda are
more abstract, the skills and knowledge the game addresses are more generic;
the message of a game that is less iconic is much better applicable outside the
particular settings of the game. This is especially useful when one wants to
express something through a game that has value beyond the game and its
immediate premise.
2.6
Conclusions
Games and simulations share a representational form: representation of source
system through a system of rules. Even games that are played as a pure form
of entertainment simulate something. Rule-based representation can take many
different forms. Traditional simulation has accurate modeling as its main goal,
whereas most games are designed to entertain. Games that aim to educate can
be said to fall somewhere in between: they seek a certain level of accuracy, but
generally enjoy more design freedom than simulations do. However, many games
still conform to the norm of simulations; they aim to represent a source system
by creating rules that resemble the rules of the source as closely and accurately
as possible. This type of rule-based representation is called iconic simulation, in
analogy to general semiotics.
Also in analogy with semiotics, the notions of indexical and symbolic simulation were explored as possible avenues to differentiate games from simulations.
As pointed out above, the goal of a game is not the same as the goal of a simulation. Indexical simulation, where the rules of the game have some sort of causal
relation with the rules of the source, and symbolic simulation, where the rules
of the game are linked to the source by convention, allow for much simpler game
systems. Although these systems consist of fewer rules and parts, their behavior
Chapter 2 | Rules, Representation and Realism
41
or meaning need not be less complex. The power of non-iconic simulation, such
as games, lies not in its power to accurately model a source system or in the creation of a vast, realistic game world, rather in its efficient use of expressive game
mechanics. I do not wish to claim that the potential of iconic representation
has been fully explored. However, to me it is clear that much more progress can
be made by developing indexical and symbolic building blocks for simulation,
and, more importantly, investigate the effectiveness of particular configurations
of such building blocks. Emergent behavior is more likely to originate from the
interrelations of game parts than simply the parts themselves. It is the craft of
the game designer to create complex systems from appropriate and simple elements. There is little art in creating complex simulation with equally complex
(or worse, more complex) means. Yet this seems to be what many developers
aim for.
Indexical and symbolic simulation as discussed in this chapter are suggestions
to go beyond iconic simulation. They are theoretical notions that help reduce a
game system to its bare minimum without affecting the structure from which the
gameplay emerges too much. This allows the designer to focus on balancing the
emergent behavior and provides the player with a better opportunity to explore
the ludic significance, or generic knowledge, codified by the game.
Yet, to design games with emergent gameplay is by no means an easy task,
even when the rules are kept simple. Emergent behavior is by definition unpredictable. A game designer’s best bet is to create many prototypes and keep
testing them. But even with frequent tests, game designers have to rely on their
experience and intuition to create their games. This practice can be improved
by creating better tools for the job, especially design tools that acknowledge
the emergent aspects of games and focus on those structural qualities that drive
them. In this light, Chapter 3 will discuss previous efforts and existing design
tools while Chapter 4 will present the Machinations framework as a new, alternative scheme to deal with game mechanics and structures of emergence in
games.
42
Joris Dormans | Engineering Emergence
He that would perfect his work must first
sharpen his tools.
Confucius
3
Game Design Theory
The previous chapters focused on the nature of games. This chapter discusses
available tools and theory to assist developers designing games. A number of design guidelines, methods, theories and tools have been developed over the past
years. Some of these were developed specifically to assist the design process,
while others were developed as analytical tools, work methods, or documentation techniques. The main approaches and attempts to assist game designers
that have been developed up until now and that are discussed in this chapter
are: design documents, the MDA framework, play-centric design methodologies,
game vocabularies, design patterns, finite state machine diagrams, Petri nets,
and finally different types of game diagrams. Most of these methods were not
developed as design tools, yet all of them can be used as such or might have
an impact on the development of new design methods and tools. This chapter
discusses the merits of all these methods for the purpose of developing design
methods and tools.
The reader should note that the common discourse about these methods is
quite diffuse. Within the game industry, and to a lesser extent within game
research too, there is no fixed vocabulary. Many concepts are used quite informally, and terminology frequently overlaps or even conflicts. For example, the
term ‘game design document’ captures a wide variety of different documents that
are created for as many different reasons. Furthermore, there seems to be little
distinction between analytical methods and design methods, and the two terms
are sometimes used interchangeable. I have tried to use the original terminology
as much as possible. Hopefully I have done so without creating confusion.
In addition, not everybody in the game industry sees the benefits of methodological approaches to game design, such as the ones discussed in this chapter
and presented in the following chapters. The two most common arguments
against design methodologies are that they have little practical value for game
design and that they cannot replace the creative process of designing games. I
will address these arguments in the last section before drawing a conclusion.
44
3.1
Joris Dormans | Engineering Emergence
Design Documents
Almost every game company creates design documents. On the Internet
many different templates can be found that are used by different companies, and
virtually every book that discusses game design has its own template. The notion
and practice of design documents is as diffuse as it is divers. There are many
different reasons to write these documents, and there are many different moments
in the design process in which companies do so. Game design documents are
sometimes used to record designs before they are build, and sometimes they are
used to record designs after the games have been built. They typically contain
descriptions of a game’s core mechanics, level designs, notes on art direction,
characters and their backgrounds, etcetera. Some advocate lengthy detailed
descriptions covering every detail of a game, while others favor brief documents
that capture design targets and design philosophy.
Over the years, writing game design documents has become a common industry practice, although no standard emerged that describes how, when or to
what purpose these documents should be written. It is not uncommon to produce an entire set of documents, each focusing on a different part of the design
or facilitating a different stage of the design process (see for example Adams &
Rollings, 2007; Rogers, 2010). Without a widely accepted template for design
documents, they do not carry over from company to company or from university
to professional career. Without a widely accepted template, design documents
cannot grow into a standard methodology. The fact that most design documents
have their own style and use their own unique concepts to describe games does
not help to create a generic body of knowledge beyond the scope of each individual project or company. With no industry wide standard in sight, it is unlikely
that design documents are going to be effective in the near future.
What is more, design documents might not be the right tool to deal with the
dynamic, emergent behavior of games. Game design documents that are written
before any prototype is made are the equivalent of requirements documents in
software engineering. A requirements document lists the requirements and functionality of a new, custom-built software application. Its creation is one of the
first steps in the ‘waterfall method’ of developing software, in which each step is
completed before proceeding to the next step. This document is typically written before the software is built and frequently is part of the agreement between
contractor and client. The waterfall method assumes that all requirements are
known and can be recorded before the software is built. Within software engineering creating and documenting functional designs is a time-tested practice,
although, with the recent popularity of agile development methods, the practice of writing complete functional designs as a blueprint for a new software
application has lost its appeal.
There are three important differences between designing games and business applications using a waterfall method that make it difficult or inefficient
to transfer the practice from general, custom software development to game
development:
Chapter 3 | Game Design Theory
45
1. Game design is a highly iterative process; no matter how experienced a
designer is, chances are that the design of a game is going to change as it
is being built. Due to the emergent nature of games (see chapter 1) it is
often impossible to accurately predict the behavior of a game before it is
implemented. In games changes to the implementation are to be expected.1
2. Not all games are created within a contractor-client context. This is especially true for entertainment games, which development shares more
with commercial, off-the-shelf software development. Without this context there is less need to document the design before the game is built.
Although publishers and funders for entertainment games set goals and
milestones, they do not function in the same way.
3. Games are rarely built with upkeep or future development in mind, reducing the necessity to create documentation that aids future developers. Despite the fact that many sequels are produced in the game industry, many
sequels are built from scratch, surprisingly little code is being reused. From
the perspective of software engineering this is a bad practice. However, as
the development techniques are still evolving fast and a new generation of
hardware becomes available roughly every five or six years this practice
makes more sense from the perspective of the game industry.2
Many designers regard the game design document as a necessary evil, and
some have dismissed the practice entirely (Kreimeier, 2003). Everyone agrees
that designs need to be documented and communicated to the team, but in
practice people hardly look at design documents (Keith, 2010, 85-87). Stone
Librande, creative director at Electronic Arts, experimented with a technique
he calls one-page designs to circumvent some of these problems (2010). His
approach is to create design documents that are more like data visualizations
instead of multi-page, written texts. One-page design documents are posters
that capture the essence of a game visually (see figure 3.1). These documents
have four advantages over the traditional design documents:
1. Most designers find them more interesting to create, making the task of
creating a design document less tedious.
2. Because there is a spatial constraint (although the size of the page is left
to the whims of the designer), the designer is forced to focus on the essence
of the game. This makes the document a better match for the agile development process often found in games.
3. As people tend to like the way these documents look, they tend to stick
them to walls, increasing their exposure and impact.
1 Although,
it must be noted that this is also increasingly true for business software.
developers of reusable game components that are sold to game studies (‘middleware’)
are probably the exception to this rule. They do maintain and reuse their code, but as their
core business is not developing games, this practice does not carry over to game design.
2 The
Joris Dormans | Engineering Emergence
46
Figure 3.1: A sample one-page design document from Librande (2010).
4. Stone Librande suggests leaving plenty of whitespace on the documents in
order to invite team members to scribble notes on them. This keeps the
documents up to date.
One-page design documents solve some of the problems associated with design
documents, but not all. There is still no standard for documenting gameplay,
mechanics and rules. Each one-page design document is created for a particular
game, and although the product should be understandable and communicative,
it cannot set a standard. In addition, the lack of detail, which makes a one-page
design document more flexible and therefore is one of its strengths, makes it less
suited to record designs; one-page design documents are very good at capturing
the design vision, goals and direction, but they cannot function as a technical
blueprint at the same time.
3.2
The MDA Framework
The MDA framework, where MDA stands for mechanics, dynamics and aesthetics, has been used to structure the game design workshops at the Game
Developers Conference (GDC) for at least eleven years running (from 2001 to
2011).3 In contrast to the practice of game design documents, the MDA framework quite consciously tries to present a generic approach to the difficulties
3 Unfortunately in software engineering the same acronym is widely used to denote Model
Driven Architecture, this might lead to some confusion. In this dissertation MDA always
stands for the mechanics, dynamics and aesthetics framework.
Chapter 3 | Game Design Theory
47
Figure 3.2: The MDA framework (after Hunicke et al., 2004).
involved in designing games. It has been quite influential and it seems to be one
of the most frequently recurrent frameworks found in university game design
programs all over the world. It probably is the closest thing the industry has to
a standardized game design method.
The MDA framework breaks down a game into three components: mechanics,
dynamics and aesthetics, which correspond to the game’s rules, its system, and
the fun it brings (Hunicke et al., 2004). The MDA framework teaches that
designers and consumers of games have different perspectives on games. Where a
consumer notices the aesthetics first and the dynamic and mechanics afterwards,
a designer works the other way round. A designer creates mechanics first and
builds dynamics and aesthetics on top them (see figure 3.2).
The MDA framework is designed to support an iterative design process and
to help designers to assess how changes in each layer might affect the game as a
whole. Each layer has its own design goals and effects on the game. Mechanics
determine the actions a game allows and it affects the game’s dynamic behavior.
The dynamics layer addresses concepts such as randomness and complexity to
explain a game’s behavior. Finally, the aesthetics layer is concerned with the
game’s emotional target: the effect it has on the player. The MDA framework
describes eight types of fun as prime aesthetic targets. The eight types of fun
are: sensation, fantasy, narrative, challenge, fellowship, discovery, expression
and submission (Hunicke et al., 2004; LeBlanc, 2004).
Despite the influence of the MDA framework and the long running GDC game
design workshop, as a conceptual framework the MDA never seems to have
outgrown its preliminary phase. The distinction between the mechanics and
dynamics layers is not always clear, even to the original authors (see LeBlanc,
2004). The mechanics are clearly game rules. But the dynamics emerge from
the same rules. Yet, the original MDA paper places game devices such as dice
or other random number generators in the layer of the dynamics. To me, those
devices would seem more at home in the layer of the mechanics. Likewise,
the aesthetics layer seems to contain only the player’s emotional responses. The
visuals and story that cue these responses, which would commonly be understood
as being part of an aesthetics, seem absent from the framework. The eight kinds
of fun comprise a rather arbitrary list of emotional targets, which is hardly
explored with any depth. Apart from short one-sentence descriptions, Hunicke
et al. do not provide exact descriptions of what the types of fun entail. They do
state their list is not complete, but they do not justify why they describe these
eight, or even hint at how many more types of fun they expect to find. What
48
Joris Dormans | Engineering Emergence
is more, the whole concept of ‘fun’ as the main target emotion of games has
been criticized by Ernest Adams and Andrew Rollings (2007, 119), and Steven
Johnson (2005, 25-26), among others. Games can aspire to target a much wider
variety of emotional responses. Some additional MDA articles (such as LeBlanc,
2006) have appeared over the years but they have not taken away these concerns.
3.3
Play-Centric Design
A more thorough method of iterative game design is described by Tracy
Fullerton et al. (2006). Coining the term ‘play-centric design’ to describe their
method, Fullerton et al. advocate putting the players at the heart of a short
design cycle. They advise that game prototypes are built quickly and tested
often. Because of the short design cycle more innovative options can be explored
with a reduced risk measured in effort and time.
Play-centric design distinguishes between two levels in a game: the formal
core and a dramatic shell surrounding it. A game’s formal core consists of
rules, objectives and procedures whereas the dramatic shell consists of premise,
character and story. Combined, these two layers contribute to the dynamic,
emergent behavior that supports play. It is the objective of play-centric design
to tune this behavior into a specific target experience. In this context Fullerton
et. al. restate Katie Salen and Eric Zimmerman’s description of games as a
“second-order design problem” (2004, 168). A designer designs the game, but
the game delivers the experience; the designer does not create the experience
directly.
This trend, to involve the player in the design process, is gaining momentum in both academia and development studios. The human-centered design
or user-centered design, originating from software engineering, has been a big
influence on this trend. It should come as no surprise that Microsoft’s game
studios are front runners in this respect, as Microsoft has much experience with
similar methods used in regular software development. Pagulayan et. al. (2003)
describe the heuristics and structured user tests that have been used to develop
several games within Microsoft. Slowly but surely these methods have become
an integral part of designing games, and more and more these methods rely on
a combination of qualitative methods such as heuristic evaluations (Sweetser
& Wyeth, 2005; Schaffer, 2008) and quantitative metrics such as plotting the
locations of player death’s onto a level map (Swain, 2008).
Play-centric design focuses on the process of designing games. By structuring
the design process, involving the player, gathering data from prototypes, and
iterating many, many times everything is done to ensure that the end product,
the finished game, is as good as the design team can make it. For a professional
game designer these methods are (or at least should be) regular tools of the
trade. They do not make the process of designing games less hard, but they do
help the designer to stay on track, break the task down into a series of smaller
subtasks, and steadily progress towards a high quality end product.
With the proper methods and tools this process can be refined. The play-
Chapter 3 | Game Design Theory
49
centric approach would benefit from methods that can speed up each iteration
or increase the improvements that are made in each one. There are several types
of methods that seem applicable. Certainly formal models of game design are
widely accepted amongst them, but also techniques to gather and process data
collected during play-tests.
3.4
Game Vocabularies
Not only designing games is a hard task, talking about them is already difficult: there is no common language to describe their inner workings. There are
plenty of books and articles that discuss games as rule-based systems, but almost
all of these choose their own words. More often than not, these vocabularies are
very good at describing particular games, but they rarely transcend into a more
generic vocabulary.
In a 1999 Gamasutra article designer Doug Church sets out to create a framework for a common vocabulary for game design (Church, 1999). According to
this framework, a game design vocabulary should consist of “formal abstract
design tools”, where “formal” indicates that the vocabulary needs to be precise,
and “abstract” indicates the vocabulary must transcend the particularities of a
single game. For Church the vocabulary should function as a set of tools, where
different tools are suited for different tasks, and not all tools are applicable for
a particular game.
Doug Church describes three formal abstract design tools in his article:
1. Intention: Players should be able to make an implementable plan of their
own creation in response to the current situation in the game world and
their understanding of the game play options.
2. Perceivable Consequence: Game worlds need to react clearly to player
actions; the consequences of a player’s action should be clear.
3. Story: Games might have a narrative thread, whether designer-driven or
player-driven, that binds events together and drives the player forward
toward completion of the game.
These three tools form a list that is by no means complete or exhaustive, nor
did Doug Church intend it to be. Between 1999 and 2002 the Gamasutra website
hosted a forum where people could discuss and expand the framework. The term
‘design tool’ was quickly replaced by the term ‘design lexicon’ indicating that
the formal abstract design tools seem to be more successful as an analytical
tool than a design tool. Bernd Kreimeier reports that “at least 25 terms where
submitted by almost as many contributors” (Kreimeier, 2003). As a project
the formal abstract design tools have been abandoned; however, Doug Church’s
article is often credited as one of the earliest attempts to deal with the lack of a
vocabulary for game design, even though his framework never caught on.
There are several researchers that carry on the torch that Doug Church lit.
The “400 Project” initiated by Hal Barwood and Noah Falstein is one example
(Falstein, 2002). Barwood and Falstein set the goal of finding and describing
50
Joris Dormans | Engineering Emergence
Figure 3.3: One of Craig Lindley’s taxonomies (after Lindley, 2003).
400 rules of game design that should lead to better games. The project website
lists 112 rules.4 However, the project seems to be abandoned as well; the last
update to the website was in 2006.
Craig Lindley (2003) uses orthogonal taxonomies to map games onto a hypothetical space defined by dominant game forms such as narratology (focus on
story), ludology (focus on gameplay) and simulation (focus on realism). Individual games can be mapped to the space depending on their relative closeness to
each of these extremes (see figure 3.3). Lindley describes a few possible, complementary taxonomies using one-, two- and three-dimensional spaces. He designed
these taxonomies as a high level road map to inform the design team of the intended design target of a particular game project. The taxonomy also suggests
suitable tools and techniques borrowed from other fields; a game that veers towards the narrative side will benefit more from traditional storytelling techniques
than a game that is a simulation first and foremost. Lindley’s game taxonomies
provide a systematic framework in which many of the formal abstract design
tools can be embedded, providing structure to what otherwise would remain a
loose collection of labels.
The game ontology project takes the notion of a common vocabulary for
games into yet another direction. This project attempts to order snippets of
game design wisdom into one large ontology. An ontology is a large classification scheme that has a hierarchical organization. Each entry in the ontology
describes a common structure found in games. It lists strong and weak examples of the structure and lists parent and children categories. For example, the
ontology entry “to own” is used to describe the game structure in which game
entities can own other game entities. An example would be the game entity
‘Mario’ that collects ‘mushrooms’ and ‘stars’, etc. “To own” is a child of the
4 http://www.theinspiracy.com/400_project.htm
(last visited June 23, 2011).
Chapter 3 | Game Design Theory
51
“entity manipulation” entry which, in turn, has three children: “to capture”,
“to possess” and “to exchange” (Zagal et al., 2005).5
The game ontology project aims to explore the design space of games without
prescribing how to create good games. More than Doug Church’s formal abstract
design tools, it primarily is an analytical tool; it aims at understanding games
rather than building them. This is a general characteristic of this and other
game vocabularies. Their success as an analytical tool does not translate easily
to being successful as a design tool. Obviously, the development of a high level,
consistent language to describe common game structures will help designers in
the long run, and, as all the vocabulary builders point out, can be a great help in
mapping the relatively unexplored areas of the game design. In fact, all authors
describe how their vocabularies can be used as a brainstorming tool, simply
by selecting and exploring random combinations of notions describing common
aspects of games. However, no matter how useful this practice can be, it can
usually only help with generating ideas. This is only a small part of the entire
process of building a game, yet it requires considerable investment on the part
of designers who must familiarize themselves with many new concepts to learn
the vocabulary. The game ontology project, for example, consists of over one
hundred separate entries, each of which ties in with several other entries in the
ontology. For game developers it can be difficult to see what is the actual return
on their investment in learning a vocabulary of that size.
The many different approaches towards a common vocabulary for games aggravate this problem.6 Every vocabulary has its own unique approach and terminology. Simply determining where and how all these approaches overlap or
collide makes an extensive academic research project in itself. Even when a game
designer invested the time and effort to learn one of these vocabularies, effectively working together or sharing knowledge with somebody who has learned
a different vocabulary is still going to be a problem. The only thing all these
vocabularies seem to share is their rejection by game designers. In the words
of Daniel Cook: “Academic definitions of game design contain too many words
and not enough obvious practical applications where people can actually use the
proposed terminology” (2006).
3.5
Design Patterns
Staffan Björk and Jussi Holopainen’s work on game design patterns also seeks
to address the lack of vocabulary for game design (2005, 4). However, their
approach is slightly different as they drew inspiration from the design patterns
found in architecture and urban design as explored in the works of Christopher
Alexander. According to Alexander: “There is a central quality which is the
root criterion of life and spirit in a man, a town, a building, or a wilderness.
5 The project has an active web page where all entries can be found: http://www.
gameontology.com (last visited July 8, 2011).
6 The discussion here cannot address all vocabularies out there. One approach worth mentioning focuses on tracking gameplay innovations: the Game Innovation Database found at
http://www.gameinnovation.org/ (last visited October 22, 2010).
Joris Dormans | Engineering Emergence
52
This quality is objective and precise but it cannot be named” (1979, ix). His
pattern language is designed to capture this quality. Patterns are presented as
problem and solution pairs, where each pattern presents a solution to a common
design problem. These solutions are described as generically as possible so that
they might be used many times (Alexander et al., 1977, x). The patterns are all
described in the same format. Each pattern also has connections to ‘larger’ and
‘smaller’ patterns within the language, where smaller patterns help complete
larger patterns (Alexander et al., 1977, xii).
This idea has been transfered to the domain of software design by Erich
Gamma, Richard Helm, Ralph Johnson and John Vlissides.7 Within software
engineering the principles of object-oriented programming take the place of
Alexander’s unnamed quality. Software design patterns are a means to record
experience in designing object-oriented software (Gamma et al., 1995, 2). Today,
software design patterns are common tools in teaching and designing software.
A pattern framework for game design following these examples was suggested
by Bernd Kreimeier (2002). However, Björk and Holopainen break away from
existing design patterns. According to them, design patterns as problem-solution
pairs do not suit game design because:
“First, defining patterns from problems creates a risk of viewing patterns as a methodology for only removing unwanted effects of a design
rather than tools to support creative design work. Second, many of
the patterns we identified described characteristics that more or less
automatically guaranteed other characteristics in the game, in other
words, the problem described in a pattern might easily be solved by
applying a related and more specific pattern. Third, the effect of introducing, removing, or modifying a game design pattern in a game
affected many different aspects of the gameplay, making game design
patterns imprecise tools for solving problems mechanically. However,
we believed that game design patterns offer a good model for how
to structure knowledge about gameplay that could be used both for
design and analysis of games.
Based on these conclusions, we have chosen to define game design
patterns in the following fashion: game design patterns are semiformal interdependent descriptions of commonly reoccurring parts of
the design of a game that concern gameplay.” (Björk & Holopainen,
2005, 34)
This decision makes their pattern approach indistinguishable from the game
vocabularies discussed above, and subjects it to all the associated problems.
Their book contains hundreds of patterns, and their website has hundreds more.
This is indicative of Björk and Holopainen’s dedication to their framework, but
also of the fact that their patterns are not built on a strong theoretical notion of
what games are and how gameplay emerges from game parts. Their mention of
games as state machines (Björk & Holopainen, 2005, 8) is not enough to carry
7 Also
know as the ‘Gang of Four’.
Chapter 3 | Game Design Theory
53
the weight of the whole framework. The number of patterns used by software
engineering, by contrast, is much lower: a typical introduction has about twenty
patterns. I doubt that the diversity of problems and solutions encountered in
games is one order of magnitude larger than those encountered in software engineering. The real difference, is that software design patterns are based on the
principles of object-oriented software design. This gives the patterns focus and
provides leverage on the problems they need to deal with, leading to patterns
that are further abstracted from typical applications or implementations. Without a clear theoretical vision on games, drafting patterns becomes an exercise
in cataloging reoccurring parts of games, without ever questioning why they reoccur or whether these and related patterns might be the result of some deeper
mechanism at work within games. Where Christopher Alexander starts from
the notion that his design patterns ultimately allow us to approach some quality that cannot be named, but which is objective nonetheless, the game design
patterns lack a similar theoretical focal point.8
Design patterns work well for architecture and software engineering because
they codify a particular quality in their respective domain. In order to replicate
their success for game design, a similar notion of quality within games should
serve as its foundation. Unfortunately, Björk and Holopainen do not formulate
such a quality for games. Without such a quality no set of game design patterns
can be anything more than a vocabulary of games. The notion of games as state
machines as mentioned by Björk and Holopainen could be the starting point to
develop a notion of quality within games, an opportunity which is missed by
Björk and Holopainen, but which I will explore further.
3.6
Mapping Game States
Games can be, and often are, understood as state machines: there is an
initial state or condition and actions of the player (and often the game, too) can
bring about new states until an end state is reached (see for example Järvinen,
2003; Grünvogel, 2005). In the case of many single-player video games either
the player wins or the game ends prematurely. The game’s state usually reflects
the player’s location, the location of other players, allies and enemies, and the
current distribution of vital game resources. From a game’s state the player’s
progress towards a goal can be read.
There are several techniques to represent state machines. Finite state machine diagrams, for example, represent state machines with a finite set of states
and a finite set of transitions between states. One state is the initial state and
there might be any number of end states. To represent a game as a finite state
machine, all the states the game can be in need to be identified. Next all possible
transitions from state to state need to be identified. For certain simple games
this works. For example, figure 3.4 represents a finite state machine describing
8 Although it must be noted that Alexander’s pattern language also includes several hundred
described patterns. In that sense game design patterns are not very dissimilar. However,
Alexander’s pattern language describes a fairly large number of domains: buildings, towns,
etc. The sets that describe each individual domain are much smaller.
54
Joris Dormans | Engineering Emergence
Figure 3.4: A finite state machine representing an adventure game.
Figure 3.5: A more complex finite state machine, but one that still produces a finite
set of trajectories.
a fairly straightforward and relatively simple, generic adventure game. This diagram utilizes a simplified version of the more elaborate state machines diagrams
specified by Unified Modeling Language (UML) (Fowler, 2004, 107-116).
Many things have been omitted from this finite state machine. For example,
the way the player moves through the world has been left out, which is no
trivial aspect in an action-adventure game with a strong emphasis on exploring
(as is the case with most action-adventure games). Still, movement is easily
abstracted away from this diagram as movement does not seem to bring any
relevant changes to the game state (other than the requirement of being in a
certain location to be able to execute a particular action).
The question is whether or not this type of representation of a game is of
any use. Looking at the diagram this game does not look complex at all. The
possible set of different trajectories through the finite state machine is very
limited. The only possibilities are abcde and abdce. This game is a machine
that cannot produce any other result. It is, to use Jesper Juul’s categories, a
game of progression, and not a game of emergence (Juul, 2005, 5). To be fair,
most adventure games have a much larger set of states and player actions that
trigger state transitions. There might be side quests for the player to follow, or
even optional paths that lack the symmetry of the two branches in figure 3.4. A
game like this might grow in complexity very fast (see for example figure 3.5),
but still the number of possible trajectories remains ultimately finite (unless one
introduces loops, see below). Yet this is what many games have done in the
past.
One way to create infinite possible trajectories is to introduce loops. Noam
Chomsky has shown that by including loops in a state machine the set of possible
results becomes infinite (Chomsky, 1957, 18-25). For example we could allow
the player to go back after opening the door and fight another monster (see
figure 3.6). The possible set of results is now {abcde, abdce, ababcde, ababdce,
Chapter 3 | Game Design Theory
55
Figure 3.6: A looping finite state machine representing an adventure game.
Figure 3.7: A recursive finite state machine representing an adventure game.
abababcde, abababdce, ..., etc.}; the player can fight an infinite host of monsters
before proceeding. Of course, this has little purpose in the context of the game,
unless the player is awarded experience points for defeating each monster and the
number of experience points somehow affects the player’s chance of defeating the
end boss (which might very well be the case in a computer role-playing game).
However, if this were the case, one might argue that each level of experience
would in fact constitute a new state, leading to an infinite number of states in
the diagram.
Finite state machines lack a construct for memory, which would solve the
experience points problem described above. To deal with the memory problem
William A. Woods designed “augmented transition networks” which uses recursion and a stack (Woods, 1970). In an augmented transition network a transition
might invoke a new state in the same or in a separate network of states and transitions. This would cause the old state to be pushed onto the stack and the new
state to be activated. When the machine encounters an end state, it recalls the
last state pushed on to the stack and proceeds from there. It only terminates
when nothing is left on the stack. In an augmented transition network representation of the adventure game above, every time the player fights a monster the
network would call itself recursively, and thus every fight would be pushed to
the stack, and the number of fights on the stack can be checked when the player
fights the boss monster (see figure 3.7).
Game states are usually much better expressed using a mix of variables and
56
Joris Dormans | Engineering Emergence
states. Not only allows such a mixture to model the large number of states
encountered in most games, it also shifts attention towards the transitions between the states, which corresponds to user actions. It is possible to construct
a diagram for Risk with only four states and seven transitions (figure 3.8) in
which each transition affects one or more variable registers representing territories, armies and cards. The diagram shows many loops, and as a result an
infinite number of different paths through the state machine is possible. A diagram that focuses on transitions is clearly more capable to capture the nature
of games and the varied sessions of play. However, the diagram is not very easy
to read as some sort of pseudo code is needed to represent the exact mechanics
that checks and changes the variable registers.
Petri nets are an alternative modeling technique suited for game machines
that are explored by a few researchers (Natkin & Vega, 2003; Brom & Abonyi,
2006; Araújo & Roque, 2009). Petri nets work with a system of nodes and
connections. A particular type of node (places), can hold a number of tokens.
In a Petri net a place can never be connected directly to another place, instead a
place must be connected to a transition, and a transition must be connected to
a place. In a classic Petri net places are represented as open circles, transitions
are represented as squares and tokens are represented as smaller, filled circles
located on a place. In a Petri net tokens flow from place to place; the distribution
of tokens over spaces represents the current state of the Petri net (see figure 3.9).
This way the number of states a Petri net can express is much larger than with
finite state machine diagrams. Petri nets put much more focus on the transitions
and have a natural way of representing integer values through the distribution
of tokens over the places in the network. Indeed, “Petri Nets tend to be, in
general, a more economic representation when state-space complexity increases”
(Araújo & Roque, 2009).
One of the very promising advantages of the use of Petri nets, is that they
have a very solid mathematic foundation. Petri nets can be easily verified and
simulated. They are almost like a visual programming language. But this ad-
Figure 3.8: A state diagram for Risk
Chapter 3 | Game Design Theory
57
Figure 3.9: Four iterations of the same Petri net showing movement of tokens through
the network
vantage often is a double edged sword. Petri nets can model a complete game
with a high level of detail, but this frequently leads to quite complex diagrams
which try to capture a game in its entirety. Petri nets can become the equivalent
of a game’s source code, and just as inaccessible to a non-programmer.
3.7
Game Diagrams
State machine diagrams and Petri nets are not the only diagrammatic approaches to deal with the problem of game design. Over the years, a few other diagrammatic or systematic approaches have been developed that deal with games
exclusively. Game theory as invented by John von Neumann, can be seen as one
of the earliest attempts to deal with game-like systems that feature a similar
state-space explosion as we have seen with finite state machine diagrams. One
could try to map this sort of systems with decision trees, but they would quickly
grow out of control. Instead, game theory uses matrices to chart the gains or
losses of possible moves in relation to the opponents move. From these matrices
rational strategies, or the lack thereof, should become apparent (see Binmore,
2007). Emmanuel Guardiola and Stéphane Natkin (2005) use similar matrices to
represent all possible interactions between a single player and a computer game.
Game theory and its application in computer games focuses on the actions of the
players. It is a very useful technique to balance actions and prevent dominant
strategies to emerge. Game theory works best with relatively simple, two-player
games; it seems to restrict itself mostly to a formal theory of gameplay decisions,
which in itself is a relevant subset of game design. However, it does not scale
very well to the scope of modern computer games, which includes much more
elements (Salen & Zimmerman, 2004, 243).
Raph Koster’s exploration in game diagrams presents yet another approach.
Presented at the Game Developers Conference in 2005, his focus is on atomic
particles that make up the game experience; on what he calls the ‘ludemes’
and devising a graphical language for them (Koster, 2005a). These ‘ludemes’
are essentially the core mechanics of a game. Koster proposes to harvest these
ludemes by reverse engineering existing games. Sadly, as Koster points out
himself, he does not succeed. Figure 3.10 shows his best take on diagramming
Checkers. He believes games can be diagrammed, but he also admits that the
language he came up with is not sufficient for the task.
Inspired by Raph Koster, Stéphane Bura takes the idea of creating game dia-
58
Joris Dormans | Engineering Emergence
Figure 3.10: Raph Koster’s diagram of Checkers (2005a).
grams one step further (2006). Combining Koster’s approach with his experience
with Petri nets, Bura designs game diagrams that try to capture a game’s structure at a higher level of abstraction than simply its core rules. Removing game
diagrams from the burden of modeling rules at the lowest level, allows Bura to
focus more on the emergent properties of games. His diagram models notions
such as ‘skill’ and ‘luck’ as abstract resources that affect other actions in the
diagram, either by enabling or inhibiting them. Figure 3.11 shows the diagram
Bura created to model Blackjack. As should become clear from this diagram,
Bura tries to capture the entire ‘gestalt’ of the game into a single image. In this
diagram the elements that model ‘skill’, ‘luck’ and ‘money’ are similar to places
in a Petri net and can accumulate tokens. The elements ‘gain’ and ‘risk’ act
like transitions. They consume and produce resources according to the arrows
that connect them to other arrows. This diagram also includes two inhibiting
connections (lines that end in a circle) to denote that the ‘luck’ of the house
inhibits the ‘gain’ of the player and that the ‘skill’ of the player inhibits the
money he or she risks. Although Bura is more optimistic than Koster, he also
admits that much work still needs to be done. He suggests a standard library
of ludemes to work with and sub-diagrams to increase the detail. But to my
knowledge, none of these extensions have been published.
There are also a few examples of the use of UML for representing game
systems diagrammatically. Taylor et al. (2006) extend UML use-case diagrams
to describe game-flow: the experience of the player. Their focus is on the play
session and the progression of the player through the game. Perdita Stevens
and Rob Pooley use class diagrams, collaboration diagrams and state machines
diagrams (three different types of UML diagrams) in their educational case study
of modeling the structure of Chess and Tic-Tac-Toe with standard UML
Chapter 3 | Game Design Theory
59
Figure 3.11: Stéphane Bura’s diagram of Blackjack. (2006)
(Stevens & Booley, 1999, 178-189). These attempts suffer from problems similar
to other types of game diagrams. As a specification language, UML can be very
detailed and inaccessible to non-programmers. In a way, UML is too universal
to capture the particular structures that are important in the domain of games.
3.8
Visualizing Game Economies
Working from an extended version of UML collaboration diagrams, as described by Bran Selic and Jim Rumbaugh (1998), I experimented with using
UML notation to capture how different parts of a game interact (Dormans,
2008b). In contrast to the previous attempts to diagram game mechanics, I
started from the idea that these diagrams should map the internal economy that
drive the emergent behavior in many games as described by Adams & Rollings
(2007). This perspective is discussed in further detail in Chapter 4. In these
diagrams I focused on the flow of resources and flow of information between
game elements. Figure 3.12 displays the UML diagram I created for the game
Power Grid. It shows the most important elements and the communication in
the form of flow of resources and information between them. The solid lines and
black shapes indicate flow of resources, and dotted lines and white connections
indicate flow of information.
In the discussion of these diagrams I quickly zoomed in on the feedback
structure that can be read from the diagram. Feedback is realized through a
60
Joris Dormans | Engineering Emergence
Figure 3.12: Power Grid in an adaptation of UML collaboration diagrams.
Figure 3.13: The feedback structure in Power Grid.
closed circuit of communication. Figure 3.13 shows the feedback structure for
Power Grid in a shorthand notation that omits much of the UML. From this
and other analyses I started extracting common patterns found in these feedback
structures (see figure 3.14).
Although this approach produces interesting results and was positively received by game designers and researchers, there are a number of problems with
it. As with many other approaches discussed in this chapter, it is geared towards documenting existing games, although the notation has been used as a
brainstorming tool during design workshops. During brainstorming writing full
UML by hand proved to be impracticable; the shorthand notation was used more
often. Finally, the diagrams are static, while they represent a highly dynamic
system. There are cues in the diagram that suggest the dynamic behavior of the
game it represents, but it leaves too much room for interpretation: one person
might read something completely different from the same diagram than the next
Chapter 3 | Game Design Theory
61
Figure 3.14: The feedback patterns extracted from Power Grid.
person.
Independent from my work, Morgan McGuire and Odest Chadwicke Jenkins
use a similar approach for their “commodity flow graphs” (see figure 3.15). These
graphs also represent a game’s economy and flow of commodities (2009). Their
approach is comparable to my attempts to map a game’s economy, but is different
on three important points:
1. McGuire and Jenkins interpret the notion of economy quite literally and
restrict themselves to games that include trading and tangible resources
(“commodities”), whereas the notion of internal economy taken from Adams
and Rollings ‘commodifies’ many more elements of games that are generally less tangible such as player and enemy health in a shooter game.
2. Although feedback does appear in one of the two examples, their graphs
are not designed to foreground feedback mechanisms.
3. Their graphs are quite informal, even though some elements are explained
in the accompanying text, many important details are not.
3.9
Industry Skepticism
Not everybody in the game industry thinks that developing design theory
or methodology is a very good idea. In an interview with Brandon Sheffield,
Raph Koster recalls that his presentation on game diagrams split his audience
at the Game Developers Conference in 2005. Some thought it was good, some
thought it was a complete waste of time (Sheffield, 2007). I have come across
similar sentiments in discussions about my work with people working in the game
industry. Usually those who dislike the premise of design methodology argue that
they are academic toys with little relevance for real, applied game design; that
they, indeed, are a waste of time. Another common argument against design
methodology is that they can never capture the creative essence that is the
heart of successful games. In this argument, design methodologies represent an
attempt to destroy the very soul of the art of game design; no method can replace
the creative genius of the individual designer (see Guttenberg, 2006). Starting
with the first, I will address both arguments below.
62
Joris Dormans | Engineering Emergence
Figure 3.15: Commodity flow graph for Settlers of Catan, after McGuire &
Jenkins (2009, 243).
The current vocabularies, aids and frameworks have a poor track record. As
should have become clear from the discussion above, there are many of these
out there, many of them designed by academics, not all of whom have actual,
hands-on game design experience. The return value of using them, set against
the often considerable investment required to learn them, is not particular high,
especially for those tools that excel in analyzing games, which is done more
often within universities than outside. The same goes for those frameworks
that allow designers to explore the design-space. The design programs within
universities allow for such exploration, whereas outside there is little time for
such theoretical exploration. The argument that the industry too would benefit
from such exploration is rather hollow if money needs to be made and one cannot
afford to take chances on a new innovative concept.
The sentiment is valid, but does not cripple the effort to create game design
methodologies. It simply suggests criteria for evaluating design methods: design
methods should help design games, not just analyze them. This seems obvious,
but many methods that have been developed over the years are analytical methods, even when they sometimes are presented as design tools. These methods
help us understand existing games or explore the hypothetical design space, but
offer little practical guidance on how to build them. What is more, design methods should return the investment required to learn them. The latter criterium
can be met in two ways: make sure that the required investment is low, or make
sure that the return is high. Obviously a design method should aim to do both
in order to maximize the return on investment.
The second argument, that no design method can replace the creative genius
Chapter 3 | Game Design Theory
63
of the individual designer, is more problematic. People who subscribe to this
opinion dismiss the whole idea of design methodology. However, this opinion is
often informed by a rather naive conception of art. Art is, and always has been,
the combination of creative talent, practiced technique and hard work. There
is no point in denying that one artist has more talent than another, but pure
talent rarely makes up for the other two aspects. Especially within an industry
where much money rides on the success of each project, investors simply cannot
afford to gamble on creative talent to deliver all the time.
The image of the artist as the creative genius is a romantic vision that rarely
fits reality. To create art, one must learn the techniques of the trade and work
hard. This has always been the case for all forms of art. There is no reason to
assume that games are any different. The artist’s techniques are many. They
range from the practical to the theoretical. Painters learn how to use a brush
with different types of paint, on the one hand, but learn about the mathematical principles of perspective and the psychological principles of cognition on the
other. The development of abstract art throughout the nineteenth and early
twentieth century has been a gradual and deliberate intellectual process (Rosenblum, 1975). The scientific invention of the perspective revolutionized Renaissance painting (Panofsky, 1960). The foundation of literary theory that Aristotle
laid over two thousand years ago is still taught today (Vogler, 2007). What has
changed over the years is the widening gap between artist and academic communities. During the Middle Ages art prevailed where academia hardly survived,
as a result the artist and the academic frequently were the same person. These
days, with thriving universities, being an academic has become a profession of its
own, but that does not mean that the ties between art and academia have been
severed. There are still many artists that contribute to the academic debate and
there are still many academics that contribute to the evolution of art. Games
are no different.
In contrast to what skeptics of design methodologies fear, design methods help
shape games but they cannot replace the creative genius. No matter how good a
method or tool is, it can never replace the vision of the designer, nor can it replace
the hard work involved in designing a game. At best it can ease the burden and
refine one’s techniques. Sometimes methods and tools seem restrictive; when
holding a hammer everything starts to look like a nail. But the best methods do
not restrict a designers vision. Rather, they should enhance it, enabling them
to work faster and create better results. Ideally, design methods also facilitate
teamwork and collaboration. For example a design tool that allows accurate
representation of game elements, would reduce the chance that individual team
members end up working toward different visions.
Game designers that take no interest in design methodology are either foolish
or lazy. However, designers have all rights to be critical of design methods, and
I do hope they remain so in the future. After all, they are the final judges that
decide whether or not a given method is worth their time and ultimately expand
their expressive power with the medium of games.
Joris Dormans | Engineering Emergence
64
3.10
Conclusions
To summarize, over the years a number of frameworks, vocabularies and work
methods have been created to assist game designers, with varying success. Game
design documents are generally considered cumbersome and inefficient; they are
seldomly put to good use. Everybody uses game design documents in their own
way. For some designers, these documents capture the creative direction early in
the development process, while for others they are a tedious requirement of the
job of game designer. For the purpose of the discussion here, no generic wisdom
to aid the development of design tools can be extracted from the diffuse practice
of writing design documents.
The MDA framework provides a useful lens on the different aspects of game
design, and can help designers to understand where to start the huge task of
designing a game. It breaks down games into three understandable and useful
layers, and teaches inexperienced designers to look through the outer layers of
a game into the mechanics core. However, the framework has evolved little
over the years, and close examination of its core concepts is likely to raise more
questions than it answers. To serve as a design tool that goes beyond the very
basics of game design; the MDA framework lacks scrutiny and accuracy.
Player-centric design practices, where short iterations and frequent playtesting are the key, are more successful in structuring the hard and laborious
process of designing games. However, there is room for improvement. With the
proper methods and tools every iteration can be made more effective, and new
ways of gathering qualitative and quantitative data might present themselves.
The theory and tools presented in this dissertation are best embedded within in
play-centric design process: they can help designers to improve every iteration,
but they cannot take away the necessity to build prototypes and test them with
real players.
There have been a number of attempts to create game vocabularies and pattern libraries that allow us to talk about games in better, more accurate terms.
However, none of these vocabularies has really gained enough momentum to become something resembling a standard that spans both industry and academia.
From a pragmatic point of view, these vocabularies require a considerable effort
to learn while they are most successful in the analysis of existing games; they
seem to be more useful for academics than they are for developers. In addition,
they usually lack a clear theoretical vision on the artifacts they intend to describe. The result is that these vocabularies hardly scratch the surface of games
and fail to contribute much to what most designers already knew intuitively.
The use of finite state machine diagrams or Petri nets to map games as state
machines are both valuable techniques with a proven track record in their respective domains. However, their respective difficulties in capturing the essence
of games indicates that simply framing games as state machines is not good
enough. The number of game states usually is not finite, and their complexity
quickly becomes problematic if one tries to model a game in every detail. A theoretic perspective on games first needs to develop a concise and objective notion
of quality in games before it can help us understand their inner machinations
Chapter 3 | Game Design Theory
65
from a more generic scope. Once this notion has been developed, a diagrammatic language can be devised to represent these machinations. Petri nets are
more promising but are less accessible to designers.
Game specific diagrams are a relatively unexplored approach towards the
development of game design tools. Apart from some preliminary attempts by
Koster, Bura, McGuire & Jenkins, and myself, little is done in this area. None of
these attempts can claim to be successful and accepted by the game development
community. Yet, the results are interesting, especially if they focus on a more
abstract gestalt of games. A more abstract and generic scope to represent game
designs seems to come quite natural to diagrams. At the same time, they are
fairly easy and intuitive to learn: most diagrammatic languages utilize only a few
core, reusable elements. When these elements express a generic and objective
notion of quality in games, these diagrams could become quite powerful.
Game design tools are needed; they can be used to improve the process of
game design, but the poor track record of current academic approaches created
some resistance within the game industry against the whole notion of design
methodology. Part of this resistance is understandable, as methods frequently
fail to return the investment required to use them. This means that we need
to rethink how design methods and tools should be used: they should not only
facilitate analysis or theoretic exploration of game concepts, rather they should
really help the designer to design. We should also take note of the fact that
game design methods cannot replace the creative talent of the individual game
designer. Game design methods should refine a designer’s technique and increase
the designer’s expressive power, any game design method should ultimately be
a tool, but it remains up to designer to make those tools work.
66
Joris Dormans | Engineering Emergence
The object in constructing a dynamic model
is to find unchanging laws that generate the
changing configurations. These laws correspond roughly to the rules of a game.
John H. Holland (1998, 45)
4
Machinations
In the previous chapter we have seen that there are no consistent and widely
accepted methodologies available for the development of games. Yet, the number
of attempts and calls for such endeavors, indicate that a more formal approach
to game design is warranted. In this chapter I propose a formal framework that
focuses on discrete game mechanics called ‘Machinations’. Initial steps towards
the current framework were presented at the Meaningful Play Conference in
2008 and the GAME ON-NA Conference in 2009 (Dormans, 2008b, 2009). The
last sections appeared earlier in a paper presented at the Workshop on Artificial
Intelligence in the Design Process at the AIIDE Conference in 2011 (Dormans,
2011d).
Taking into account the discussion of existing methodologies and the nature
of games in the previous chapters, the Machinations framework presented here
adheres to the following requirements:
1. A formal framework for game mechanics should formulate a clear theoretical vision on the structure of game mechanics and their relation to quality
in games.
2. A formal framework for game mechanics should not only be an analytical
tool for existing games; rather it should have a direct application for the
development of new games.
3. A formal framework for game mechanics should provide a good return on
the investment to learn it, either by keeping the investment low, or by
making the return high, but preferably both.
The first section presents an overview of the Machinations framework and
its various components. Sections 4.2 through 4.5 describe the Machinations
diagrams that are part of the framework and how they can be used to represent
game mechanics. The sections 4.6 to 4.8 describe how, using Machinations
diagrams, feedback structures, and recurrent patterns in these structures, can
be analyzed and related to a game’s dynamic behavior. In sections 4.9 and
4.10, I will discuss how the Machinations software tool, developed as part of
this research, implements Machinations diagrams, and how it can be used to
simulate, and generate quantitative data in order to balance games. The chapter
68
Joris Dormans | Engineering Emergence
concludes with a case study. The Machinations framework is used to explore the
theoretical game SimWar described by Will Wright, which, to my knowledge,
has never been implemented. With this example I hope to illustrate that the
Machinations framework can be used to simulate and analyze games even before
they are built.
A word of caution: the Machinations framework is a lot to take in at once.
The framework comprises many interrelated concepts that are best understood
in unison. This means that there is no real natural starting point to explain all
these concepts. I advise the reader to occasionally refer back to earlier concepts.
I also like to point out that appendix A presents a single page overview of
the most important concepts. In addition, many of the diagram examples can
be found online in an interactive version at the Machinations wiki page: www.
jorisdormans.nl/machinations/wiki.
4.1
The Machinations Framework
The Machinations framework formalizes a particular view on games as rulebased, dynamic systems. It focuses on game mechanics and the relation of these
mechanics and the dynamic gameplay that emerges from them. It is based on the
theoretical notion that structural features of game mechanics are for a large part
responsible for the dynamic gameplay of the game as a whole. Game mechanics
and their structural features are not immediately visible in most games. Some
mechanics might be close to a game’s surface, but many are obscured by the
game’s system. Only a detailed study of a game’s mechanics can shed a light
on the game’s structure. Unfortunately, the models that are used to represent
game mechanics, such as representations in code, finite state diagrams or Petri
nets, are complex and not really accessible for designers. What is more, they
are ill-suited to represent games at a sufficient level of abstraction, on which
structural features, such as feedback loops, become immediately apparent. To
this end, the Machinations framework includes Machinations diagrams which
are designed to represent game mechanics in a way that is accessible, yet retains
the structural features and dynamic behavior of the games they represent.
The theoretical vision that drives the Machinations framework is that gameplay is ultimately determined by the flow of tangible and abstract resources
through the game system. Machinations diagrams represent these flows and
foreground the feedback structures that might exist within the game system. It
is these feedback structures that for a large part determine the dynamic behavior of games. This is consistent with findings in the science of complexity that
studies dynamic and emergent behavior in a wide variety of complex systems
(see section 1.5). By using Machinations diagrams a designer can give game
systems a shape which would normally remain invisible. The main premise is
that through Machinations diagrams, structure and quality in game mechanics
become tangible.
Machinations diagrams are designed to capture game mechanics. As such,
they are not only a design tool; they are also useful as an analytical tool to
compare and analyze existing games. The Machinations framework allows us to
Chapter 4 | Machinations
69
Figure 4.1: A map of the Machinations framework.
observe recurrent patterns across many different games. Machinations diagrams
are a medium to express and reason about these patterns.
Machinations diagrams can be drawn with any tool. The language was designed to draw easily on a computer, or on paper. At the same time, the syntax
of the language is exact. It describes unambiguously how different elements interact. This allowed the development of a Machinations software tool, which
can be used to simulate and experiment with game systems. With this tool, a
user can run and interact with a Machinations diagram. To a certain extent, a
user can play a game represented as a Machinations diagram. In addition, the
Machinations tool allows users to define ‘artificial players’ that interact with a
diagram automatically. This means that games can be simulated without any
actual interaction of real users. Such a simulation can run very quickly, allowing
a user to experiment and gather quantitative data from thousands of simulated
play sessions quickly and efficiently. To support this, the tool features automatic
charts to collect data from each simulated play session.
Figure 4.1 is an overview of the Machinations framework and its most important components. It summarizes the discussion above.
Machinations diagrams create an abstracted perspective on game mechanics
and are often used to focus on certain aspects of a game. The framework does not
dictate how much detail a diagram should capture or what aspects of the game
economy one should represent. Using Machinations diagrams many different
aspects can be captured at many different levels of detail. The best perspective
and level of abstraction is largely determined by context and purpose. Often
it is sufficient to model games from the perspective of a single player, even if
the game is actually played by multiple players. In these cases, it is often fairly
easy to imagine how a diagram might be duplicated and combined to represent
the multiplayer situation. In other cases it is useful to model one player at a
70
Joris Dormans | Engineering Emergence
higher level of detail than other players. Likewise, particular aspects of games
such as taking turns might be abstracted away. At a high level of abstraction,
there often is little difference in the effects of players acting simultaneously and
asynchronously, or in alternating turns. I have tried to keep the level of detail
low and the level of abstraction high in the examples used in this chapter. This
way the structural features of the internal economy are best foregrounded, which
best serves the purpose of examining the effects of these structures on emergent
gameplay. For this reason the natural scope for Machinations diagram is that
of a single player and his or her individual perspective on the game system.
Although it is certainly possible to model multiplayer systems, the framework,
as it currently stands, does not include features designed to support multiplayer
games in particular. For example, the main input device for interaction with
a Machinations diagram is the mouse; there is no support for multiple input
devices. Likewise, turn taking is geared towards a single player only: the system
responds to every turn in a similar way. There is no support for alternating
turns for a number of players; and it does not prevent the players to take actions
outside ‘their turn’.1
The Machinations framework focuses on rules and mechanics and does not
take into account all elements of game design. Most importantly, the Machinations framework excludes all elements of level design. As such, the framework
is more effective for games that do not rely on level design that much. This
includes most board games, strategy games and management simulation games.
While the framework will still be applicable to games that rely more on level
design, for these games the framework can only describe a part of the picture.
Level design will be the subject of the next chapter.
Moreover, The Machinations framework does not involve the player or any
cultural dimension of representation through games. The framework treats
games as complex state machines: interactive devices that can be in many different states, and whose current state determines the transition to a new state.
This is an approach that can be and has been critiqued: without players games
would be, quite literally, meaningless. The formal rule systems of games are
subject to constant change and reinterpretation by players. A formal approach
always runs the risk of turning a blind eye towards this dynamic and important
dimension of games (see Malaby, 2007). However, building game systems is an
important task of a game designer. It is this system that codifies the player’s
possible interaction and generates individual game experiences. The aim of the
Machinations framework is to understand the elementary structures that contribute to emergent gameplay and that ultimately facilitates the expressive and
dynamic nature of games.
Finally, one more word of caution: when one sets out to model anything as
complex as games, a model can never do justice to the true complexity of the
reality of gameplay. The best models succeed in stripping down the complexity
of the original by leaving out, or abstracting away, many important details.
This is certainly the case with the framework and diagrams I present here.
1 This type of mechanics can be modeled using the current framework, but it does require
fairly elaborate constructions. There is no ‘native’ support for this type of features.
Chapter 4 | Machinations
71
However, any model is a tool that can help us understand and work with complex
systems. To be able to use the model to the best effect, understanding the
concepts that informed the creation of the model is required. As any model, the
Machinations framework and diagrams only facilitate understanding; they are
never a substitute for it.
4.2
Flow of Resources
The Machinations framework utilizes the idea of ‘internal economy’ (Adams
& Rollings, 2007) to model activity, interaction and communication between
game parts within the game system.2 A game’s economic system is dominated
by the flow of resources. In games resources can be anything: from money
and property in Monopoly, via ammo and health in a first person shooter
game, to experience points and equipment in a role-playing game. More abstract
aspects of games, such as player skill level and strategic position, can also be
modeled through the use of resources. A game’s internal economy consists of
these resources as well as the entities or actions that cause them to be produced,
consumed and exchanged. In the case of Risk, territories, armies and cards are
the main resources; the player’s option to build will produce more armies, while
with an attack the player risks armies to gain territories and cards.
In order to model a game’s internal economy Machinations diagrams uses
several types of nodes, that pull, push, gather and distribute resources. Resource
connections determine how resources move between elements and while state
connections determine how the current distribution of resources modifies other
elements in the diagram. Together, these elements form the essential core of
Machinations diagrams. This section explains the basics of resource flow. The
next section discusses different types of state connections. Sections 4.4 and 4.5
discuss more specialized nodes used to represent common operations in the game
economy. An two-page visual overview of these core elements can be found in
appendix A.
Machinations diagrams represent the flow of resources between different game
elements. In this respect, Machinations diagrams are dynamic: just like tokens
in a Petri net can move between places, resources in a Machinations diagram can
move between nodes. The flow of resources is dictated by resource connections
2 The history of the notion of internal economy within games is somewhat fragmented.
It appears in a chapter title in Rollings & Adams (2003) which is a precursor to Adams &
Rollings (2007), but the notion itself is not really discussed in that book. There are a few
examples of the use of the term or the synonym ‘in-game economy’ in the context of games,
such as in Simpson (2000), Burke (2005) and McGuire & Jenkins (2009). In these three cases
the term is reserved for trade and inventory systems (as opposed to, for example, combat
systems). McQuire and Jenkins even include ‘commodity flow graphs’ that visualize the flow
of resources through a game, but they do not foreground feedback structures as Machinations
graphs do (see discussion of their work in section 3.7). It is not until the notion was discussed in
more detail in Adams & Rollings (2007) that it started to encompass more types of resources
(such as health) than a strict interpretation of ‘economic’ would allow, and that a game’s
internal economy could actually include combat systems, leveling systems, etc., as well. Since
then, the term appears in lectures and syllabuses that follow Adams and Rollings book. It is
in this wider sense that the term is used here.
72
Joris Dormans | Engineering Emergence
Figure 4.2: In a Machinations diagram the distribution of resources over nodes represents the state of a game. Resource connections indicate how resources might be
redistributed.
represented as solid arrows connecting the nodes of the diagram. The state
of the game corresponds to the distribution of resources over the nodes (see
figure 4.2). Thus on one hand, a Machinations diagram represents a single state
a game can be in, but on the other it determines subsequent states that are its
direct result. The nodes and connections of the diagram visualize the structure
that ultimately shapes the probability space of game states. The static version
of Machinations diagrams accompanying this text obviously cannot display the
state changes over time. The distribution of resources in a static Machinations
diagrams always represent a game’s initial state, unless stated otherwise.
In a Machinations diagram the nodes are active: they can fire and by firing
they cause resources to be redistributed. A node in a Machinations diagram can
be in one of three different activation modes:
1. A node can fire automatically: it fires at intervals determined by the diagram’s time mode (see below). All automatic nodes fire simultaneously.
2. A node can be interactive, which means that it represents a player action
and fires in response to that action. In the dynamic version of a Machination diagram, interactive nodes fire after users click on them.
3. A node can be passive, which means it can only fire in response to a trigger
generated by another element (see section 4.3 below). Passive nodes still
accumulate resources.
As Machinations diagrams are dynamic it is important to understand how
they handle time. There are three different time modes for Machinations diagrams:
1. In synchronous time mode all automatic nodes are activated at regular
intervals of arbitrary length specified by the user. All interactive nodes
clicked by players fire at the next time step, at the same time when automatic nodes fire. In this mode all actions in one time step take place
simultaneously. It is possible to for a user to activate multiple interactive
nodes during a time step, but during a time step, each interactive node
can only be activated once.
Chapter 4 | Machinations
73
2. In asynchronous time mode automatic nodes in the diagram are still activated at regular intervals of arbitrary length specified by the user. However, players can activate interactive nodes at any time within the intervals
and the resulting actions are executed immediately. In this case an interactive node can be activated multiple times during a time step.
3. Alternatively, a Machinations diagram can be in turn-based mode. In this
mode time steps do not occur at regular intervals. Instead a new time step
occurs after the player has executed a specified number of actions. This
is implemented by assigning a number of action points to each interactive
node and allotting players a fixed budget of action points each turn. After
all action points are used, all automatic nodes fire and a new turn starts.3
The flow of resources from node to node is instantaneous: resources simply
disappear and reappear at their new location. However, in order to help designers
understand how the internal economy works, the Machinations tools can be set to
visualize the resource flow by animating the movement of the resources along the
resource connections. This is a visual aid only and does not affect the diagram
in any way.
Resource connections, that dictate how resources are distributed when nodes
fire, have a label that determines the flow rate. Many labels are numbers indicating a fixed flow rate, but the Machinations framework also uses simple
expressions or icons to represent flow rates based on randomness, skill or other
uncertain factors outside the game mechanics. When the label is omitted, the
flow rate of a resource connection is considered to be 1.
The most basic nodes connected by resource connections are pools. They are
represented as open circles. A pool collects resources that flow into it. Pools
can be used to represent any collection of resources in a game. For example,
the money possessed by a player of Monopoly can be represented as a pool,
the money in the ‘bank’ can be represented as another pool. The resources
themselves are represented as small dots. Different colors can be used to denote
different types of resources. Alternatively, to avoid visual clutter, the number of
resources on a pool can be represented as a number. Resource connections that
lead into a node are considered to be its inputs while resource connection that
lead out of a node are considered to be its outputs.
The activation mode of a pool is indicated by its visual appearance. Passive
pools have a single outline, interactive pools have a double outline, automatic
pools have a single outline and are marked with a star (‘*’, see figure 4.4).
When a pool fires it will try to pull resources through its inputs. The number
of resources pulled is determined by the rate of the individual input resource
connection. Alternatively, a pool can be set in ‘push mode’. In this mode, a pool
will push resources along its outputs when fired. Again, the number of resources
pushed is determined by the flow rate of the output resource connection. A pool
3 It is possible to create interactive nodes that cost no action points to activate. When all
interactive nodes cost no action points, except a single ‘end turn’ action (that has no other
effect), this can be used to create a game where players can take any number of actions until
they are done.
74
Figure 4.3: resource connections in a
Machinations diagram.
Joris Dormans | Engineering Emergence
Figure 4.4:
Different activation
modes of a pool.
in push mode is marked with a ‘p’ (see figure 4.4). A pool that has only outputs
is always considered to be in push mode, in which case the marker is omitted.
It might happen that two pools try to pull resources from the same source
simultaneously, while there are not enough resources to serve both pools. For
example, in figure 4.5 every time step pool B automatically pulls one resource
from A and both C and D attempt to pull one resource from B. This means
that after one time step, B will have one resource and C and D will both try
to pull it. How this is resolved depends on the the time mode. In synchronous
time mode, neither C nor D can pull the resource. After two iterations when
B has pulled a second resource, both C and D will pull one resource from B.
While the diagram runs, C and D will both pull a resource once every two time
steps simultaneously. As A starts with nine resources, after nine time steps C
and D will have four resources and one resource will remain on B. The state of
the diagram will then no longer change.
In asynchronous or turn-based mode, either C or D will pull a resource one.
Which pool has priority is initially random; subsequently, the priority alternates
every time step. This means that C and D will both pull one resource from B
on alternating time steps and eventually there will be four resources on C and
five on D, or vice versa.
Figure 4.5: How simultaneous pulls are handled in a Machinations diagram depends
on the diagram’s time mode.
Chapter 4 | Machinations
4.3
75
Flow of Information
Machinations diagrams are dynamic beyond the flow of resources. Resources
flow rates can be modified by changes in resource distribution. In addition, nodes
can be triggered, activated or deactivated in response to changes in resource
distribution. These modifications are indicated by a second class of edges called
state connections. State connections indicate how the current state of a node,
the number of resources on it, affects target elements in the diagram. State
connections are represented as dotted arrows. Labels and the type of elements
they connect specify the type of state connection. There are four types of state
connections that are characterized by the type of elements they connect and
their labels:
1. Label modifiers connect a source node to a target label (L) of a resource
connection or a state connection. It indicates how state changes of the
source node (∆S) modify the value of the target label as indicated by the
modifier’s label (M ). The new value takes effect on the subsequent time
step. Thus, the next value of label that is the target of a number (n) of
label modifiers is given by the following formula:
n
X
(Mi ∆Si )
Lt+1 = Lt +
i=1
For example, in figure 4.6 every resource added to pool A adds 2 to value
of the resource flow between pools B and C. Thus the first time B is
activated, 1 resource flows to A and 3 resources flow to C, the second
time, 1 resource still flows to A, but now 5 resources flow to C. The label
of a label modifier always starts with a plus or minus symbol indicating
incrementation or decrementation.
Label modifiers are frequently used to model different aspects of game
behavior. For example, a pool might be used to represent a player’s accumulated property in a game of Monopoly. The more property a player
has, the more likely it is that player will collect money from other players.
This can be represented by the diagram in figure 4.7.4 Note that in this
4 Note
that many details are omitted in this diagram, in particular the diagram does not
Figure 4.6: A label modifier affecting the flow rate between two pools. In
effect: F lowBC = 3 + 2A
Figure 4.7: In Monopoly the state
of your property positively affects the
chance other player’s money flows to
you.
Joris Dormans | Engineering Emergence
76
+3
+2
-2
-0.4
C
A
B
Figure 4.8: Node modifiers affect the
number of resources on a pool. In effect: C = 3A − 2B
+1
points
villages
+2
cities
Figure 4.9: In Settlers of Catan
your score is determined by the number of villages and cities you possess
(amongst other things).
case the exact value of the label modifier is unspecified, it only indicates
that the effect on the random flow rate is positive.
2. Node modifiers connect two nodes. They indicate how state changes of
the source node (∆S) modify the number of resources on the target node
(N ) as indicated by the modifier’s label (M ). The state changes to the
target node are further processed during the subsequent time step. Thus,
the new number of resources on a node that is the target of a number (n)
of node modifiers is given by the following formula:
n
X
(Mi ∆Si )
Nt+1 = Nt +
i=1
Figure 4.8 illustrates a node with two modifiers. By using negative node
modifiers or redistributing resources from a node that has positive input
node modifiers it becomes possible that the number of resources on a node
becomes negative. In this case, the negative number of resources indicates
a shortage. No resources can be pulled from a node that has a shortage,
and resources that flow into a node with a shortage are used to compensate
for the shortage first.
Node modifier can have labels that are fractions, for example ‘+1/3’ or
‘-2/4’. In this case the number of resources of a target node is modified
by the value indicated by the fraction’s numerator every time there is a
change to the number resources on the source divided by the fraction’s
denominator and rounded down. Thus when a source node changes from 7
to 8, the number of resources on the target is lowered by 2 if the modifier
is -2/4, but if the modifier is +1/3 the number of resources on the target
node does not change.
A simple, real-life example of the use of node modifiers can be found in
Settlers of Catan where players gain 1 point for every village in their
possession and 2 points for every city in their possession (see figure 4.9).
show how a player might acquire property. Diagrams representing Monopoly that paint a
more complete picture are included later in this chapter.
Chapter 4 | Machinations
Figure 4.10: A trigger in Monopoly
enables the acquisition of property by
spending money.
77
Figure 4.11: In Caylus the presence
of a laborer at a goldmine activates the
option to collect gold.
Node modifiers change the number of resources on other nodes. In effect,
the use of node modifiers causes production or consumption of resources.
These effects can also be achieved with a slightly more elaborate, but
also somewhat cumbersome construction. This construction includes other
elements that are introduced later. However, as the use of node modifiers is
quite common, they are useful ‘syntactic sugar’: simpler notations of more
elaborate constructions. For this reason node modifiers are here treated as
a particular type of state modifiers.
3. Triggers are state connections that connect two nodes or connects a source
node to the label of a resource connection. Triggers fire when all the inputs
of its source node become satisfied : when each input passed the number
of resources to the node as indicated by its flow rate.5 A firing trigger will
in turn fire its target. When the target is a resource connection, it will
pull resources as indicated by its flow rate. A node that has no inputs will
fire outgoing triggers whenever it fires (either automatically, or in response
to a player action, or to another trigger). Triggers are identified by their
label which is a star (‘*’).
Triggers are commonly used in games to react to redistribution of resources.
For example, in Monopoly players might transfer money to the bank in
order to ‘trigger’ the transfer of property from the bank into their possession. This can be represented as the diagram in figure 4.10.
4. Activators connect two nodes. They activate or inhibit their target node
based on the state of their source node and a specific condition. The activator’s label specifies this condition. Conditions are written as an arithmetic
expression (for example ‘==0’, ‘<3’, ‘>=4’ or ‘!=2’) or a range of values
(for example ‘3-6’). If the state of the source node meets this condition
then the target node is activated (it can fire). When the condition is not
met the target node is inhibited (it cannot fire).
5 This means that resource connections have a memory, they keep track of how many resources were requested and how many were delivered to pulling resources.
78
Joris Dormans | Engineering Emergence
Figure 4.12: Different types of gates in a Machinations diagram.
Activators are used to model many different game mechanics. For example, in the board game Caylus players place their laborers (a resource)
at particular buildings on the board to enable them to execute special actions associated with that building. For example, a player might place a
laborer at a goldmine in order to collect gold (see figure 4.11). However,
as indicated by the trigger in figure 4.11, in Caylus every time a player
exercise this option, the laborer is returned to the player’s laborer pool.
4.4
Controlling Resource Flow
Pools are not the only possible nodes in a Machinations diagram. Gates are
another type of node. In contrast to a pool a gate does not collect resources,
instead it immediately redistributes them. Gates are represented as diamond
shapes that often have have multiple outputs (see figure 4.12). Instead of a flow
rate, each output is labeled with a probability or a condition. The first type
of outputs are referred to as probable outputs while the other are referred to as
conditional outputs. All outputs of a single gate must be of the same type: when
one output is probable, all must be probable and when one output is conditional,
all must be conditional.
Chapter 4 | Machinations
79
Probabilities can be represented as percentages (for example ‘20%’) or weights
indicated by single numbers (for example ‘1’ or ‘3’). In the first case a resource
flowing into a gate will have a probability equal to the percentage indicated by
each output, the sum of these probabilities should not add up to more than 100
percent. If the total is less than 100 percent there is a chance that the resource
will not be sent along any output and is destroyed. In the latter case the chance
that a resource will flow through a particular output is equal to the weight of
that output divided by the sum of the weights of all outputs of the gate. Gates
with probable outputs can be used to represent chances and risks. For example,
in Risk players risk armies in order to gain territories. This type of risk can
easily be represented by a gate with probable outputs indicating the rates for
success or failure.
An output is conditional when it is labeled with a condition (such as ‘>3’ or
‘==0’ or ‘3-5’). In this case, all conditions are checked every time a resource
arrives at the gate and one resource is sent along every output whose condition
is met. As the conditions might overlap this can lead to duplication of resources,
or, when no condition is met, to the destruction of the resource.
Like pools, gates have three activation modes: gates can be passive, interactive or automatic. Interactive gates also have a double outline and automatic
gates are also marked with a star. When a gate has no inputs, it triggers every
time it fires, this way gates can be used to produce triggers either automatically
or in response to player actions.
Furthermore, gates have one of two different distribution modes: deterministic distribution and random distribution. A deterministic gate will distribute
resources evenly according to the distribution probabilities indicated by percentages or weights if it has probable outputs. When it has conditional outputs it
will count the number of resources that have passed through it every time step
and uses that number to check the conditions of its outputs.6 A deterministic
gate has no special symbol and is represented as a small open diamond.
A random gate generates a random value to determine where it will distribute
incoming resources. When it has probable outputs it will generate a suitable
number (either a value between 0 and 100 percent, or a number below the total
weights of the outputs). When its outputs are conditional it will produce a value
between 1 and 6 to check against the conditions, just as if the diagram rolled a
normal six-sided die.7 Random gates are marked with a dice symbol.
Gates might have only one output. Gates with one output act exactly the
same way as gates with multiple outputs. The gates on the middle row of
figure 4.12 will (from left to right) randomly let 30 percent of all the resources
pass, immediately pass the resource to the output regardless of the output’s flow
rate, and let only the first two resources pass.
All output state connections from a gate are triggers; gates do not accumulate
resources and therefore label modifiers, node modifiers and activators originating
from a gate have no function. The triggers are activated instead of redirecting
6 It can be convenient to think of a deterministic gate with conditional outputs as ‘counting
gate’.
7 This value can be set represent other types of random distribution if needed.
80
Joris Dormans | Engineering Emergence
Figure 4.13: An automatic, random gate controlling the flow of resources between
two pools. In this case there is a 30% chance three resource will flow from A to B every
time step.
resources. These triggers can also be conditional or probabilistic. In this way
gates can be used to control the flow of resources (see figure 4.13).
4.5
Four Economic Functions
In their discussion of a game’s internal economy represented by the flow of
resources, Adams and Rollings identify four basic economic functions: sources,
drains, converters and traders (Adams & Rollings, 2007, 331-340). Sources create
resources, drains destroy resources. Converters replace one type of resource for
another, while traders allow the exchange of resources between players or game
elements. In theory, a pool or combination of pools and gates can fulfill all these
functions, but for clarity it is useful to introduce special nodes to represent
sources, drains, converters and traders.
Sources are nodes that produce resources. In Risk, the building action is a
source: it produces armies. Likewise passing ‘Go’ in Monopoly also is a source:
it generates money. Health packs are sources of health in shooter games. As any
node in a Machinations diagram, sources have an activation mode that is either
passive, interactive or automatic. An example of an automatic source is the
steady regeneration of the protective shields of the player’s star fighter in Star
Wars: X-Wing Alliance. The rate at which a source produces resources is a
fundamental property of a source. The building action in Risk and the passing
of ‘Go’ in Monopoly are examples of sources activated by user actions and
game events respectively. Adams and Rollings distinguish between ‘limited’ and
‘unlimited’ sources (Adams & Rollings, 2007, 333). A limited source can easily
be represented as a pool without inputs that starts with a number of resources
on it. An unlimited source can be represented as a pool without inputs but with
a sufficiently large number of resources on it (or rather an infinite number of
resources). To represent unlimited sources, the Machinations framework includes
a special source node represented as a triangle pointing upwards (see figure 4.14).
Note that Adam and Rollings’ notion of a ‘limited source’ is still represented as
as a pool with no inputs.
Drains are nodes that consume resources. In an adventure game where you
can cross hot lava at the cost of loss of health points, the lava acts as a drain.
Being underwater in most games causes a resource representing breath to be
drained away. The rate of a drain is determined by the flow rate of its input
resource connection. Some drains consume resources at a steady rate while oth-
Chapter 4 | Machinations
81
Figure 4.14: The four economic functions in a Machination diagram, with their
equivalent constructions.
ers consume resources at random rates or intervals. Drains could be represented
as a pool with no outputs, but the Machinations framework includes a special
drain node represented as a triangle pointing downwards (see figure 4.14).
Converters convert one resource into another. In Monopoly the option to
buy property acts as a converter: the resource money is converted into another
resource: property. In a shooter game, killing enemies might also invoke a
converter. In this case ammunition is used in an attempt to kill, which in turn,
when the enemy is put down, might be converted in new ammunition or health
packs dropped by the enemy. Converters act exactly as a drain that triggers a
source, consuming one resource to produce another. As with sources and drains,
converters can have different types of rates to consume and produce resources.
The Machinations framework represents a converter as a vertical line over a
triangle pointing to the right.
Traders are nodes that cause resources to change ownership when fired: two
players could use a trader to exchange resources. The board game Settlers of
Catan is built around a trading mechanism allowing players to trade the five
types of resource cards among each other against exchange rates they establish
among themselves. A player that has many of timber cards, might for example
decide to exchange three timber cards for two wool cards of another player. Compared to converters, traders are relatively rare; in many games, what appears
to be a trader is often implemented as a converter. For example, depending on
the implementation, the merchants in many computer role-playing games where
players can barter for goods and equipment are converters, not traders. These
merchants will happily buy whatever the player has to offer and seem to have
an unlimited supply of handy items and money to buy the player’s unwanted
loot (Castronova, 2005, 198-199). Fallout 3, where all traders’ supplies are
limited, is an exception. A trading mechanism can be constructed by two gates
connected by a trigger ensuring that when one resource is received the other
is returned in exchange. The Machinations framework represents a trader as a
vertical line over two triangles pointing left and right.
The difference between converters and traders is not always immediately
Joris Dormans | Engineering Emergence
82
Figure 4.15: A Machinations diagram for Monopoly.
clear. Especially from the perspective of an individual player, converters and
traders will almost have the same function: pass a number of resources to it, and
get a number of other resources in return. Yet, there is an important difference.
As pointed out above: a converter can be seen as a combination of a drain and
a source. Using a converter resources are actually consumed and produced, and
therefore the total number of resources in the game might change. Whereas
with a trader the number of resources always stays the same (also see Adams &
Rollings, 2007, 334).8
The economic function of a particular game element can change according to
the perspective of the diagram. For example, figure 4.15 represents Monopoly
from the perspective of an individual player. Only in the perspective of an
individual player rent income is a source. From the perspective of the entire
game rent acts as trader: money simply exchanges ownership. In this case,
the bank should be modeled as an individual pool with which money is only
traded. After all, the boxed game comes with a finite supply of play money.
All of these would constitute valid models of the same game. As mentioned
in the introduction, the Machinations framework allows one to model a game
with different levels of detail and abstraction. The important question is: what
does one try to model? For a basic understanding of behavior of Monopoly a
simple, limited perspective as in figure 4.15 might suffice. Especially when one
can imagine how this same structure is repeated for every player. But for a more
thorough analysis more detail might be required.9
8 When
a converter is replaced by two pools as the equivalents of a source and a drain
suggest the number of resources also stays the same. However, if one considers resources that
are trapped on a pool without outputs and which state no longer affect the game to be inactive
then the number active resources might change.
9 For an example of such an analysis see http://www.jorisdormans.nl/machinations/wiki/
index.php?title=Tutorial_1. In this example the model for Monopoly includes two players.
The only reason I can imagine why anyone would want to model the bank is in a detailed
study to find out how much money should be included in the published game.
Chapter 4 | Machinations
4.6
83
Feedback Structures in Games
The structure of a game’s internal economy plays an important role in a
game’s dynamic behavior and gameplay. Just as feedback plays an important
role in any dynamic system (see section 1.5), it plays an equally important role
in a game’s internal economy. The idea of applying the concept of feedback to
games is not new. During his 1999 lecture at The Game Developers Conference
Marc LeBlanc introduced feedback loops to the game design world (1999). Since
then, feedback loops have been discussed by a number of influential designers,
including Salen & Zimmerman (2004), Adams & Rollings (2007) and Fullerton
(2008). A classic example of feedback in games can be found in Monopoly
where the money spent to buy property is returned with a profit because more
property will generate more income. This feedback loop can be easily read
from the Machinations diagram of Monopoly (figure 4.15): it is formed by the
closed circuit of resource and state connections between the money and property
pools. More specifically, for feedback to exist, a close circuit of connections is
required that consists of at least one state connection that is not an activator.
A closed circuit of resource connections can only create a loop of resources. To
change the rate of the flow, at least one label modifier, trigger or activator that
changes or triggers the production or consumption of the resources must be part
of the loop. Note that for this reason a closed circuit of resource connections
that includes a converter or trader also constitutes a feedback loop, as their
equivalent constructions do include a state connection (see figure 4.14).
As is the case in classic control theory (DiStefano III et al., 1967), Marc
LeBlanc distinguishes between two types of feedback: positive and negative feedback (also see section 1.5). Positive feedback strengthens itself and destabilizes
a system. Positive feedback occurs when an effect fuels itself. In the Monopoly
example above, the feedback is positive because investing money will generate
more money. Positive feedback amplifies small differences between players: a
player that has a lucky break early in the game will find this luck amplified over
time: a player that by chance gets the option to get more money or good property early in a game of Monopoly is very likely to win.10 Positive feedback can
be applied to ‘positive’ game effects but also to ‘negative’ game effects, as is the
case with loosing pieces in Chess, which increases the chances of loosing more
pieces, and which will eventually make you lose. LeBlanc suggests that positive
feedback drives the game to a conclusion and magnifies early successes (see also
Salen & Zimmerman, 2004, 224–225).
Negative feedback is the opposite of positive feedback. It stabilizes a game
by diminishing differences between players, by applying a penalty to the player
who has done something that takes him closer to his goal and winning the game,
or by giving advantages to the trailing players. Many racing games use negative
feedback to keep a race close, either by giving trailing players more advantages or
by hindering leading players. This effect is often described as ‘rubber-banding’.
10 Which, incidentally, is exactly the point the original designers of Monopoly’s ancestor
The Landlord Game were trying to make: it was a critique of capitalism that favors those
that possess capital and works to widen the gap between the rich and the poor (Fron et al.,
2007).
84
Joris Dormans | Engineering Emergence
Figure 4.16: The core feedback loop involving armies and territories in Risk.
It can be implemented by blatantly giving trailing players better acceleration
and more grip, or more subtely as is the case in Super Mario Kart by having
the most effective weapons in the game affect cars in front of the player that
uses them. LeBlanc points out that in most multiplayer games that allow direct
interaction, some sort of negative feedback is already in place: rational players
will target the leader more than any other player. As one might expect, negative
feedback can prolong a game and magnifies late successes.
Control theory, in almost all cases, strives for negative feedback while avoiding positive feedback, as it aims to create stable systems. A large part of control
theory concerns itself with determining and optimizing the stability of the system. For games the situation is, of course, very different. Positive feedback
loops are much more frequent in games because, in general, players do not want
to play a game that is stable and drags on forever. Negative feedback does have
an application within games: most games with only positive feedback will seem
too random to many players as they will be unable to catch the player who took
an early lead; negative feedback is often used to balance out early successes.
Just as one feedback loop will only create weak emergence in Jochen Fromm’s
typology (see section 1.5), most games need multiple feedback loops to display
truly interesting emergent behavior. The game Risk is an excellent illustration
of this as in this game four feedback loops interact.
The core feedback loop in Risk involves the resources armies and territories.
Figure 4.16 depicts this core feedback loop. The label ‘+1/3’ of the label modifier
that sets the output flow rate of the source ‘build’ indicates that the output of
the source is improved by one for every three territories the player has.
The second feedback loop in Risk is formed by the ‘cards’ that are gained
from a successful attack (see figure 4.17). Only one card can be gained every
turn, thus the flow of cards passes through a limiter gate first. Collecting a
set of three cards can be used to generate new armies. Not every set generates
armies, and some sets generate more than others. In figure 4.17 these effects
are indicated by the random symbol labeling the output of the converter that
converts cards into armies.
The third feedback loop is activated when a player manages to capture a
continent, which will give the player bonus armies every turn (see figure 4.18). In
Risk predefined groups of territories form continents as indicated by the design
of the game board. In the diagram this level of detail is not possible, instead
Chapter 4 | Machinations
85
the construction is represented as a pool connected to another pool with a node
modifier. In this particular case, seven territories will count as one continent
which will in turn activate the bonus source.
Finally, the fourth feedback loop is activated by the loss of territories due to
the actions of other players (figure 4.18). Which player is going to attack which
other player is dependent on many factors, including those players strategies and
preferences. Sometimes it is opportune to prey on weaker players in order to
gain territories or cards, sometimes it is important to oppose stronger players
to keep them from winning. In the diagram this is indicated by the multiplayer
dynamic label (an icon depicting two pawns) affecting the resource flow to the
drain on the right. The number of continents a player captured has a strong
influence on this. As in general, players do recognize that other players that have
a continent have a big advantage and will usually act against that more vigor.
The player might also lose armies to actions of other players. I have chosen to
omit them from the diagram to avoid too much clutter. It should not affect the
argument too much. The important thing is, that in Risk there is some form of
friction caused by other players, and the strength of this friction increases when
the player has captured continents. This type of friction is a good example of the
negative feedback that can almost always be found in multiplayer games where
players can act against each other as pointed out by LeBlanc (see above).
The first three feedback loops in Risk all are positive: more territories or
cards will lead to more armies which will lead to more territories and cards. Yet
they are not the same. The feedback of cards is much slower that the feedback
of territories as a player can get only one card each turn, but at the same
time the feedback of the cards is also much stronger. Feedback from capturing
continents operates faster and even stronger. These properties are important
characteristics of the feedback loops that have a big impact on the dynamics
of the game. Players are more willing to risk an attack when it is likely that
the next card they will get completes a valuable set: it does not improve their
chances of winning a battle but it will increase the reward if they do. Likewise
the chance of capturing a continent can inspire a player to take more risk than the
Figure 4.17: The second feedback in Risk involving cards and armies.
86
Joris Dormans | Engineering Emergence
Figure 4.18: The third feedback loop in Risk: bonus armies through the possession
of continents.
Figure 4.19: The fourth feedback loop in Risk: negative feedback caused by capturing
continents. Note that in this figure the label modifier affecting the resource connection
from the territories pool to the opposition drain does create a closed circuit as both
nodes at the end and the start might be affected by the changing flow rate.
Chapter 4 | Machinations
87
player should. In Risk the player’s risks and rewards constantly shift, making
the ability to understand these dynamics and to read the game a decisive skill
in this game. These three positive feedback loops play an important role but
simply classifying them as positive does not do justice to the subtlety of the
mechanics.
4.7
Feedback Profiles
To really appreciate the feedback structure of Risk the differences between
the three positive feedback loops must be explored in more detail. The feedback
of capturing territories to be able to build more armies is straightforward, fairly
slow and involves a considerable investment of armies. Often players lose more
armies in an attempt to conquer territories than they will regain with one build.
This leads to the common strategy to build during multiple, consecutive turns.11
The feedback involving the cards requires a considerably larger effort on the part
of the player. Players can only gain one card during a single turn, no matter
how many attacks were successful. However, depending on the cards the player
draws, the return might be much higher. The feedback involving continents is
very fast and with a high return. Players will receive bonus armies every turn no
matter whether they choose to build or to attack. The feedback is so strong and
obvious that it will typically inspire fierce counter measures from other players.
Table 4.1 lists seven characteristics that, together with the determinability
characteristic discussed below, form a more detailed profile of a a feedback loop.
At a first glance some of these characteristics might seem overlapping, but they
are not. It is easy to confuse positive feedback with constructive feedback and
negative feedback with destructive feedback. However, positive destructive feedback does exist. For example, loosing pieces in a game of Chess will increase
the chance of loosing more pieces and loosing the game. Likewise, the board
game Power Grid employs a mechanism in which the game leaders have to invest more resources to build up and fuel their network of power plants: negative
feedback on a constructive effect.
The strength of a feedback loop is an informal indication of its impact on the
game. Strength cannot be attributed to a single characteristic: it is the result of
several. For example, permanent feedback with a little return can have a strong
effect on the game. The effects of a feedback loops on a game can drastically
change with these characteristics. Feedback that is indirect, slow but with a lot
of return and not durable has a strong destabilizing effect. In this way even
negative feedback can be used to destabilize a system if it is applied erratically
or when its effects are strong, but slow and indirect.
In many games the profile of a feedback loop is also affected by factors such
as chance, player skill and social interaction. Table 4.2 lists the different types
on determinability used in the Machinations framework and the icons used to
11 In Risk players can choose to build or attack during a given turn; when they choose to
build they will gain a fixed number of armies, when choosing to attack they can attack as
many times as they choose during a turn. In Risk there is a rule that forces a player to attack
after three consecutive turns of building.
88
Joris Dormans | Engineering Emergence
Table 4.1: Feedback Characteristics
Chapter 4 | Machinations
89
Table 4.2: Determinability
denote them. These icons can be used to annotate resource connections and
gates in a diagram. A single feedback loop can be affected by multiple and
different types of nondeterministic resource connections or gates. For example,
the feedback through cards in Risk is affected by a random gate and a random
flow, increasing its unpredicatablity.
The profile of multiplayer feedback in a game that allows direct player interaction, like Risk, can change over time. As LeBlanc already pointed out, it
often is negative feedback as players act stronger, or even conspire against the
leader. At the same time, it can also be positive as in certain circumstances, as
mentioned above, it can be beneficial to prey on the weaker player.
The skill of players in performing a particular task can also be a decisive
factor in the nature of feedback, as is the case for many computer games. For
example, Tetris gets more difficult as the blocks pile up, the rate at which
players can get rid of the blocks is determined by their skill. Skillful players will
be able to keep up with the game much longer than players with less skill. Here
player skill is a factor on the operational or tactical level of the game. In games
of chance, tactics, or games that involve only deterministic feedback, a whole
set of strategic skills can be quite decisive for the outcome. However, that is a
result of a player’s understanding of the game’s feedback structures as a whole,
and as such it is not an element that can or needs to be modeled within the
structure. This feedback loop in Tetris is also affected by randomness. The
shape of the block is randomly determined by the game. Although, the skill is
90
Joris Dormans | Engineering Emergence
generally more decisive in Tetris, the player just might get lucky.
Games that feature only deterministic feedback can still show surprising unpredictable outcomes, as emergence itself can be a source of unexpected and
hard to predict behavior. In fact, it is my conviction that a well-designed game
is built on only a handful feedback loops and relies on chance, multiplayer dynamic, and skill only when it needs to and refrains from using randomness as an
easy source of uncertainty.
A feedback loop’s characteristics and determinability form the feedback’s
profile. While a profile like this can be very helpful in identifying the nature
of feedback in a game, it does little to reveal the interaction between different
feedback loops. This is where diagrams, such as figure 4.20, excel. Many of the
characteristics of feedback loops described above can be read from the diagrams.
The effect of the feedback is directly related to the constructive or destructive
nature of the feedback loop, whereas return and investment depends on the
number of resources involved. A feedback loop that consists of almost only
state connections and triggers, and few interactive nodes, is likely to have a high
speed. Range can be read from the number of elements involved in the feedback
loop, speed from the number of iterations required to activate the feedback. The
return of a feedback loop must be read from the modifiers of the arrows that
create the closed circuit, as some of these modifiers might be nondeterministic
the return is more difficult to assess or actually becomes uncertain. The type
of feedback (positive or negative) is perhaps the most difficult to read from a
static representation, and requires careful inspection of the diagram, but this
is possible, too. Note that the plus symbols in the diagrams do not indicate
positive feedback, only that there is positive correlation between the number of
resources in the pool and the label it is affecting. A positive correlation can
induce negative or positive feedback.
4.8
Feedback Analysis and Recurrent Patterns
An analysis of a game’s feedback loops can be used to identify structural
strengths and flaws in its design. To create interesting and varied gameplay
feedback is an important device, and most successful games incorporate two or
more, but not that many more, feedback loops in its main structure. Structural
flaws, or ‘bad smells’ in analogy to software engineering, are constructions that
are best avoided. If we take Risk again as our example, we can identify one of
its problems from play experience: building as often as you can is an effective,
almost dominant, strategy. To counter this strategy the game includes a a
special rule preventing the players from building on more than three subsequent
turns. Inspection of the feedback structure of the game suggests other ways of
resolving the problem. Attacking feeds into a triple positive feedback structure
(through territories, cards and continents), which is a strength of it its design,
but apparently the feedback is not effective enough. Strengthening the feedback
of territories will help only a little as building is part of the same feedback
loop and will probably encourage the unwanted behavior. Either the feedback
through cards or the feedback through continents needs to be improved. The
Chapter 4 | Machinations
91
Figure 4.20: A Machinations diagram for Settlers of Catan, which is won by
collecting ten points. The game’s five resources are collapsed into one for this diagram.
Normally, a player has to pay with a particular set of resources to perform an action.
The relative value of these different resources varies as production of each individual
resource is subjected to chance. Settlers of Catan has three main feedback loops:
1) The slow, expensive but durable increase of production through the investment in
roads, villages and cities. 2) Buying cards, which is fast, unpredictable and has no
durability. And 3) through trade with other players which is subject to multiplayerdynamics.
92
Joris Dormans | Engineering Emergence
Figure 4.21: The dynamic engine pattern.
card feedback loop involves two random factors: success of attack and the blind
draw of the card itself. This makes the feedback unpredictable and very hard
for the player to assess. In general, involving too much randomness in the same
loop is best avoided, especially when this randomness affects different steps in
the loop. It is very hard to balance and predict the feedback of such a loop,
so reducing the randomness, for example by allowing the winner a pick of three
open cards, will help a lot.
Alternatively the feedback through the capture of continents can be improved. The problem with this feedback is that it has a high return, is permanent, direct and fast: it is very obvious and will inspire strong reaction by
opposing players, in other words it acts as a red flag. Combined with a relative
high investment, it constitutes an effective but risky strategy. The strength and
the obviousness of the feedback invites a strong negative feedback. This creates
a feedback loop that is too crude: it is either on and going strong or it is off.
Either the player succeeds in taking and keeping a continent and has a very good
shot at winning, or the player is hit hard and loses what usually is a considerable investment. By making the feedback less strong, and perhaps increase the
number of continents (or rather regions) for players to conquer, a more subtle
feedback loop is created that will pay-out more often without unbalancing the
game too much.12
Looking at feedback structures in games, many recurrent patterns emerge.
For example, both Monopoly and Risk share a similar structured, positive
feedback loop that can be found in many other games as well. This pattern,
which I call a dynamic engine, revolves around a source producing one type of
resource, which can be converted to improve the source. Figure 4.21 depicts this
elemental feedback pattern, using the generic names energy and upgrades for the
two resources involved. In Monopoly these resources are money and property
respectively, whereas in Risk these are armies and territories. The pattern can
be found in many more games. In StarCraft the player invests minerals to
build SUV units to mine more minerals.
Settlers of Catan (see figure 4.20) has a more complicated implementation of this pattern, one where a player needs to build roads before that player
can build villages, and where villages can be upgraded to cities. In this case the
12 On
the other hand, the game is called ‘risk’ for a reason, risk taking is an intended part
of the game play. How much risk is suitable for this game is also a matter of taste.
Chapter 4 | Machinations
93
dynamic engine is also part of a engine building pattern (see appendix B).
A dynamic engine has a very typical gameplay signature. When play begins
players will invest most of their energy in upgrades for a while. At a certain
point, players start to use the energy elsewhere or, when that is the set goal for
the game, simply collect it. When one plots the output of energy over time in a
graph, this leads to a sharply cornered line (see figure 4.22). This signature is
recurrent in all games that use a dynamic engine, although it might be obscured
by the randomness or nondeterministic behavior caused by other feedback structures. For example, with one strategy in StarCraft called “turtling”, players
invest a lot in building their base, before starting to build an offensive force to
attack the enemy. When these players start attacking their offensive output is
usually quite large. In the opposite strategy, called the “Zerg rush” after one
of the game’s playable factions, the players invest little in their base, instead
they start building offensive units as fast as they can in a bid to overpower their
opponents before they have built up adequate defense (see figure 4.23). The
effectiveness of the latter strategy depends on the speed of the attacking player,
but also on the balance between offensive capabilities, defensive capabilities and
the building costs of the units involved. Section 4.11 will discuss the balance
between these two strategies in more detail.
Twelve more recurrent feedback patterns are described and discussed in appendix B.
4.9
Implementing Machinations Diagrams
The online Machinations tool does not only allow users to draw Machinations
diagrams, it can also run diagrams.13 While running, the resources in a diagram
flow from node to node and flow rates change according to their distribution.
The digital version of the diagrams introduces an extra activation mode, two
different modes the nodes can use to push or pull resources, three new types of
nodes, and the concept of color-coded resources. All of these are discussed in
this section.
The new activation mode digital Machinations diagrams introduce is the
13 See
the tool’s web page: http://www.jorisdormans.nl/machinations
Figure 4.22: The gameplay signature
of a dynamic engine.
Figure 4.23: The turtle and rush
strategies in StarCraft are the result
of a dynamic engine.
94
Joris Dormans | Engineering Emergence
Figure 4.24: Extra modes and nodes in automated Machinations diagrams.
‘starting action’ mode. Nodes in this mode fire once when the diagram is started
and are marked with an ‘s’ instead of the star used to mark automatic nodes
(see figure 4.24).
The pull modes of a node specify in more detail how a node pulls resources
from another node. There are two different pull modes:
1. By default, a node pulls as much resources as it can, up to the flow rates
of its inputs. If not all resources are available, it still pulls those that are.
2. Alternatively, a node can be set to pull all or none resources. In this mode,
when not all resources are available, none are pulled. Nodes that are in
‘all or none’ pull mode are marked with an ‘&’ sign (see figure 4.24).
These modes also apply to pushing modes: by default a pushing node sends
as many resources out along its output resource connection up to outputs’ flow
rate. A pushing node in ‘all or none’ mode only sends resources when it can
supply all of its outputs. This means that nodes in push mode might be marked
with both a ‘p’ and a ‘&’.
The three new nodes are: end conditions, charts, and artificial players. End
conditions specify when a diagram has reached an end state and can stop further
execution. Usually such a state is reached when a specified number of resources
is collected or when a particular resource is completely drained (see figure 4.25).
End conditions need to be activated through an activator, they do not have a
activation mode as other nodes do. End conditions can be used to set goals or
build simple timers to limit the game’s length. Diagrams that have end conditions are suited to ‘quick run’: instead of displaying the dynamic behavior as it
develops over time, it runs the game to its completion immediately. Diagrams
can also be run several times in succession, in this case the tool will show which
end condition was reached how many times.
Charts can be used to plot the state of pools into a graph. Pools and graphs
are connected using node modifiers, but to avoid visual clutter the tool represents
these state connections as two small arrows, one leading out of the pool and one
leading into the graph. The color of these arrows corresponds with the color of
Chapter 4 | Machinations
95
Figure 4.25: Two different end conditions.
Figure 4.26: An example diagram that produces a chart and includes an artificial
player.
the lines in the graph (see figure 4.26). The data collected by these graphs can
be exported as simple comma separated values to be analyzed further by other
tools.
Artificial players allow the use of the Machinations tool to simulate players interacting with the diagram. This introduces the possibility of automated
multiple tests runs. The implementation of artificial players is rudimentary, but
effective. Basically the artificial player has a list of options to activate a specified
node and either goes through these options in sequence, or works down the list
testing a specified probability for each option until it finds one node to activate.
These options might be affected by the state of a pool. For example, the artificial
player script for figure 4.26 reads:
invest = 100 - upgrades * 30
run = 100
This script will initially cause the artificial player to invest, but with every
upgrade the chances it will invest are decreased by thirty percent. If it does not
invest, there is a one-hundred percent chance it will run instead.
In the digital Machinations tool the color of resources is meaningful. If a
resource connection has a different color than the color of the pool then only those
96
Joris Dormans | Engineering Emergence
Figure 4.27: Automated Machinations diagrams allow the use of color-coded resources.
resources which color matches the color of the resource connection can be pulled
through that resource connections. This allows different types of resources to be
stored on the same pool. Likewise sources and converters producing resources,
produce resources in the respective colors of their outputs, when these outputs
have a different color than the source or converter. I use the term color-coding to
refer to this use of colored resources and resource connections in a Machinations
diagram. Label modifiers, node modifiers, and activators which color is different
from the color of the node they originate from, act according to the number of
resources of that color on the pool. Figure 4.27 illustrates how this can be used in
a diagram. In this figure source A produces a random number of orange and blue
resources every time it is activated. Both are collected at pool B, the number of
orange resources on B increases the number of blue resources produced and vice
versa. The user can only activate drain C when there are at least 20 red and
20 blue resources on pool C. Color-coded resources are used in the case-study of
SimWar later in this chapter.
Digital Machinations diagrams offer the opportunity to collect data on the
behavior of a game system before the game is built. It allows designers to test
typical playing strategies. The artificial players do not have very advanced artificial intelligence, but they can still easily be programmed to follow certain
strategies, and will happily do so over thousands of runs. As will become clear
in the discussion of SimWar in section 4.11, this can be a valuable tool in identifying dominant strategies and testing the balance in a game. Artifical players
can be activated and deactivated individually, allowing the user to define different artificial players set up to represent and experiment with different strategies
within a single diagram.
4.10
Randomness and Nondeterministic Behavior
In many games, complexity is not the only source of nondeterministic behavior. As was argued in chapter 2, dice (or other random generators) are a good
way to create nondeterministic behavior for those mechanics that are not the
core of the gameplay. In this way, dice can be used to simulate the outcome
of battles in Risk or Kriegsspiel on the one hand, or to simulate the conditions that affect the rate of production in a game like Settlers of Catan on
the other. From the perspective of the Machinations framework, randomness
is a good tool to inspire particular behavior from the players but it might also
be used to obscure dominant gameplay signatures that originate from certain
feedback structures, such as the dynamic engine pattern.
Chapter 4 | Machinations
97
An account on how randomness can affect the behavior of the player is given
by John Hopson (2001). He argues that consistent with the findings in behavioral
psychology experiments, player behavior is affected by chance and the interval
the player is awarded for actions. When a player has a chance to be awarded at
regular intervals, the player’s attention and activity will spike at those intervals,
where as when those intervals have random lengths, the player will be active
most of the time. After all, the next action might lead to a new reward. As
such, it is vital that in a Machinations diagram you can set up random values
as well as random intervals.
The effect of randomness on the dynamic engine signature is illustrated with
the following experiment. A simple racing game for two players utilizes a simple
dynamic engine. The goal of the game is to collect thirty ‘distance resources’ by
running. But the player can also choose to invest three energy to produce an
upgrade which will increase the rate of production of energy (see figure 4.28).
The energy production starts at a rate of 0.1 which means 1 resource is produced
every ten seconds. Every upgrade improves this rate by 0.1. The artificial players
controlling the black and gray diagrams are set up slightly differently. The black
player will first buy four upgrades before it starts running. The gray player will
buy only two upgrades. Obviously, black’s strategy is superior: black wins every
time. The chart in figure 4.28 shows this trend: the black line reaches 30 before
the gray line does.
When the energy sources are changed to have a probable output of 10 percent
every second, and each upgrade will increase this probability by another 10
percent (see figure 4.29), the pattern is broken. Figures 4.30 and 4.31 show
sample data generated by different runs. Gray now has a chance of about 23
percent to win.14 In effect, the randomness can counter the effect of the feedback
loop as was also suggested by Ernest Adams and Andrew Rollings (2007, 387).
The Machinations tool allows the designer control over random values produced. As we have already seen, the percent sign (‘%’) is used to denote a
probability. A source that has probable output is labeled ‘20%’ will have a
twenty percent chance to produce a resource every time step. The tool can also
simulate dice rolls by using a similar notation for dice rolls and calculations that
is commonly used in pen-and-paper role-playing games. In these games ‘D6’
stands for a random number produced by a roll of a single six-sided die, where
as ‘D6+3’ adds three to the same dice roll, and ‘2D6’ adds the results of two sixsided dice and thus will produce a number somewhere between two and twelve.
Other types of dice can be used as well: ‘2D4+D8+D12’ indicates the result of
two four-sided dice added with the results of an eight- and twelve-sided die. Unlike pen-and-paper role-playing games, the Machinations tool is not restricted to
dice that are commercially available. It can use five-, seven or thirty-five-sided
dice.
Intervals are created by using a slash (‘/’) for inputs and outputs. For example, a source that produces ‘D6/3’ resources will produce between one and six
resources every three seconds. Intervals can also be random: a drain that has an
input with a modifier that states ‘3/D6’ will drain 3 resources with an interval
14 I
ran the diagram 1000 times which resulted in 768 wins for black and 232 wins for gray.
98
Joris Dormans | Engineering Emergence
Figure 4.28: Two players racing for distance in a game with deterministic behavior.
Chapter 4 | Machinations
99
Figure 4.29: Two players racing for distance in a game with random behavior.
Figure 4.30: Sample result of a race
with random behavior.
Figure 4.31: Another sample result
of a race with random behavior.
Joris Dormans | Engineering Emergence
100
between one and six seconds.
4.11
Case study: S IM WAR
The Machinations framework can be used to study existing games and support the design of new games. To illustrate the use of the framework I choose
to discuss a game of some renown within the design community, yet has never
been built. SimWar was presented during the Game Developers Conference in
2003 by game designer Will Wright, who is well-known for his published simulation games: SimCity, The Sims, etc. SimWar is a hypothetical, minimalistic
war game that features only three units: factories, defensive units, and offensive units. These units can be built by spending an unspecified resource that
is produced by factories. The more factories a player has the more resources
come available to build new units. Only offensive units can move around the
map. When an offensive unit meets an enemy defensive unit there is a fifty percent chance that one destroys the other and vice versa. Figure 4.32 can be seen
as a visual summary of the game and includes the respective building costs of
the three units. During his presentation Will Wright argued that this minimal
real-time strategy game still presents the player with some interesting choices,
and displays dynamic behavior that is not unlike the behavior found in other
games within the same genre. Most notably Wrights argued that a ‘rock-paperscissors’ mechanism affects the three units: building factories trumps building
defenses, building defenses trumps building offensive units, whereas building offensive units trumps building factories. Wright describes a short-term versus
long-term trade-off and a high-risk/high-reward strategy that recalls the ‘rush’
and ‘turtle’ strategies found in many real-time strategies (see section 4.8).
Building up a model SimWar using Machinations diagrams is best done in
few steps. Starting with the production mechanism, a pool is used to represent
a player’s resources. The pool is filled by an automatic source. The source’s
production rate is initially zero, but is increased by 0.25 for every factory the
player builds. Factories are built by clicking the interactive converter labeled
‘BuyF’, which will pull resources only when at least five are available. Figure 4.33
contains this diagram. The structure is a typical implementation of a dynamic
Figure 4.32: SimWar summary, after Wright (2003)
Chapter 4 | Machinations
101
Figure 4.33: The production mechanism of SimWar.
engine pattern that we have seen before. As all dynamic engines, it creates a
positive feedback loop: the more factories a player builds the quicker resources
are produced which in turn can be use to build even more factories. Notice that,
in this case, the structure requires players to start with at least 5 resources or 1
factory otherwise players can never start producing.
Resources are also used to buy offensive and defensive units. The mechanics
for this are represented by figure 4.34. This diagram makes use of color coded
resources. The resources produced by the converter labeled ‘BuyD’ are black
while the resources produced by ‘BuyO’ are green as indicated by the color of
their respective outputs. This means that black resources (representing defensive
units) and green resources (representing offensive units) are both gathered on
the ‘Defending’ pool. However, by clicking the ‘Attack’ gate, all green resources
are pulled towards the ‘Attacking’ pool. Thus only offensive units can be used
to launch an attack.
Figure 4.35 illustrates how combat between two players is modeled. Each
attacking unit of one player (red on the left) increases the chance a defending unit
of another player (blue on the right) is destroyed, and vice versa. In addition,
attacking units increase the chance a factory is destroyed, but that drain is only
active when the defending player has no defending units left.
Combining the structures of each step, a model can be created for a two
player version of SimWar (see figure 4.36). One player controls the red and
orange elements on the left side of the diagram, while the other player controls
the blue and green elements on the right side of the diagram. Both sides are
symmetrical. Note that, in contrast to figure 4.33, the supply of resources is
Figure 4.34: Offensive and defensive units in SimWar.
102
Joris Dormans | Engineering Emergence
Figure 4.35: Combat in SimWar.
ultimately limited (as it is in most RTS games). This is to prevent the game
from potentially dragging on for ever. If both players run out of resources before
they managed to destroy the other, the game ends in a draw.
Figure 4.37 displays the relative strength of each player as it developed over
time during a simulated session. The strength was measured by adding five for
every factory the player owns plus one for each offensive and defensive units.
The chart displays what might be called the fingerprint of an interesting match.
This particular session was played by two artificial players set up to follow
what might be called a ‘turtling’ strategy, favoring factories and defensive units
over offensive units. The script these players followed was:
Chapter 4 | Machinations
103
Figure 4.36: A Machinations diagram for SimWar. It features two players. One
player controls the red and orange nodes on the left, the other the blue and green
nodes on the right.
Figure 4.37: A chart showing the relative strength of each player as it developed over
time during a simulates session eventually won by the red player.
Joris Dormans | Engineering Emergence
104
Attack
BuyF =
BuyD =
BuyO =
= Defense*5-30
100-Factories*30
100-Defense*15
Factories*20+Resources*2
Another type of artificial player was created by setting up the script to follow
a ‘rushing’ strategy, by building one factory before directing all resources towards
building offensive units:
Attack
BuyF =
BuyD =
BuyO =
= Defense*10-70
200-Factories*100
100-Defense*50
100
The ‘rushing’ strategy proved to be very unsuccessful. Figure 4.38 plots the
strengths of both players over a typical session and also indicates when attacks
where launched. The ‘rushing’ player (red) builds up a large attack, but does
not recover once its units are lost. After that attack, it is fairly easy for the
‘turtling’ player (blue) to defeat red with a series of smaller attacks. Out of
one thousand simulated session, the ‘rushing’ strategy managed to win only
twice. Most published real-time-strategy games are balanced towards rushing
strategies, as the latter tend to be harder to execute, and mastered later by
players. In order to balance the game a number of ‘tweaks’ were tested: I
increased the costs for factories and defensive units, and decreased the cost for
offensive units and run the simulation one-thousand time for every modification
(see table 4.3). Surprisingly, increasing the cost for defensive units seem to
have little effect. Even when a defensive unit costs more than an offensive unit,
making it a really poor choice, the turtle strategy remained dominant. This
leads to the conclusion that the balance between rushing and turtling strategy is
mostly affected by the balance between production and offensive units, and little
by the balance between offensive and defensive units. Also note that increasing
the factory costs initially increases the average game length, but decreases it for
costs above eight. This can be explained by the fact that increasing factory cost
slows the game as it takes more time to build up production capacity. At the
same time a very high factory cost favors the rushing strategy, which tends to
win faster than the turtling strategy. At higher factory costs the second effect
dominates the first effect.
4.12
Conclusions
The Machinations framework formulates a clear theoretical vision on the
structure of game mechanics and quality in games. Within the framework,
gameplay is an emergent property of the system of rules. Machinations diagrams visualize those structures that directly contribute to emergent gameplay.
Quality is attributed to the structure of the game mechanics, or rather to the
feedback structures that are present within game mechanics. Up until now, feedback loops in games have been characterized as being either positive or negative.
Chapter 4 | Machinations
105
Figure 4.38: A chart showing a rushing player (red) against a turtling player (blue).
The orange spikes indicate attacks waves launched by red, the green spikes indicate
attack waves launched by blue.
Table 4.3: Tweaks to SimWar’s economy and the effects for ‘turtling’ versus ‘rushing’
strategies.
106
Joris Dormans | Engineering Emergence
However, in order to explain gameplay we need more characteristics of feedback.
The Machinations framework proposes to characterize feedback loops by their
type, effect, speed, range, investment, return and durability. In addition, several types of nondeterministic flows can affect a feedback loop. The delicate
interaction between multiple feedback loops must be taken into account to get a
complete picture of the dynamics of games as complex machines that generate
an internal economy. Patterns such as the ones discussed in this chapter codify
structures that have proven to be successful in the past.
Although the Machinations framework utilizes a number of key concepts and
terms, it is not a design vocabulary that needs to be expanded in order to include
previously unencountered structures. Machinations diagrams work with only a
handful of elements which can be combined in infinite constructions to capture
just about any game. When these basic elements are understood, a designer
should be able to read any Machinations diagram. The framework offers a high
range of expressiveness for little investment on the part of the designer.
Machinations diagrams have an exact and consistent syntax. This means
that the diagrams can be interpreted by a computer. In fact, the Machinations
software tool implements diagrams. In other words, the diagrams can be run:
they are interactive and dynamic, just like the games they are modeling. It
allows models of games to be tested and explored quickly and efficiently. The
tool even allows the designer to quickly gather quantitative data from simulated
play sessions, offering a high and concrete return. Unfortunately this dynamic,
interactive property of the software tool does not translate to paper; the interactive tool, and many of the examples discussed in this chapter, can be found
on the Machinations web page: http://www.jorisdormans.nl/machinations.
The Machinations framework is a design tool first and foremost, but it can
be used to record existing and non-existing games equally well. For pragmatic
reasons many of the examples discussed in this chapter are models of well-known,
existing games: it is hard to show the relation between rule structures and
emergent gameplay when the reader is unfamiliar with the latter. In practice,
using the Machinations tool allows a designer to simulate and run a design many
times before building a prototype. It can also be used to track down flaws and
suggest improvements for prototypes and published games. This applications of
the Machinations framework should help the designer to get more out of each
iteration in a play-centric design process.
The list of feedback patterns presented in appendix B and on the accompanying website is neither definitive nor complete. It is best to consider this
framework as a set of building blocks that can be used to build an infinite number of different structures, some of which are recurrent patterns that can be used
to analyze existing games and explore new concepts alike.
There are some limitations to the use of Machination diagrams. The idea of
internal economy suits some games better than others. In particular, it works
very well for board games, strategy games and management simulation games.
Games that rely more on level design and mechanics of progression are addressed
in the following chapter. In games where economy is more abstract it can be
difficult to determine the best level of abstraction and scope for the model, as
Chapter 4 | Machinations
107
is the case for games such as Chess and Go. Those games seem to derive their
emergent behavior less from an internal economy and more from the mechanics
that govern tactical maneuvering, that fall outside the scope of this dissertation.
Many games can be diagrammed in multiple ways depending on the designer’s
focus and the diagram’s particular perspective. Still, feedback loops can go a
long way in explaining the flow of a game, and should be consistent features
even with different levels of abstraction and different perspectives.
108
Joris Dormans | Engineering Emergence
Storytellers in all media and all cultures are,
at least partially, in the business of creating
worlds.
Scott McCloud (2000, 211)
5
Mission/Space
All games have rules and mechanics, but different types of games utilize different
types of mechanics. As was discussed in Chapter 1 mechanics to govern physics,
economy or levels are of a different nature and therefore must be treated differently. The previous chapter discussed the Machinations framework, diagrams
and tool to represent discrete game mechanics that govern a game’s internal economy, and to correlate their structure to emergent gameplay. It is a framework
that works well for certain types of games. As was pointed out, the Machinations
framework does not incorporate level design. Yet for level-driven games, such as
most action-adventure games and most first-person shooter games, quality level
design is critical. Unfortunately, level design has been studied less extensively
than game mechanics dealing with physics or economy.
Level design is one of the aspects of game design where the different disciplines
of art, design and technology converge. Level design combines all of these aspects
as it works with core mechanics and art to create a spatial and temporal experience within the technological boundaries of the game’s software-architecture.
Level designers are usually responsible for turning gameplay elements and art
assets into game spaces that create a concise and compelling experience. Often
this requires scripting of simple behavior to create mechanics to control player
progress. The specialized role of the level designer was one of the last roles to
emerge from the growing game industry (Byrne, 2005; Fullerton, 2008). As a
result there is no definite set of principles guiding level design yet, although a
few common patterns and strategies can be found in different sources.
In this chapter I will present the Mission/Space framework. This framework
provides a structural perspective on levels and players’ progression through them.
However, this chapter starts with investigating a few existing structural models
for level design that are suggested by others in the game development community.
The first section presents an overview and a discussion of common typologies
for level layouts. The second section explores the models that focus on levels
as a set or series of challenges to be overcome by players. The Mission/Space
framework builds on these two different perspectives and includes separate graph
representations for both the mission and space that comprise a level. With a
detailed analysis of a game level from The Legend of Zelda: Twilight
Princess I will illustrate and discuss the relation between the missions and
Joris Dormans | Engineering Emergence
110
spaces. This discussion extends into the final sections of this chapter where I
will show how the Mission/Space framework can be leveraged and expanded to
get a better grip on mission, spaces and their relation. In chapters 6 and 7 I
will combine the Mission/Space framework with the Machinations framework
in order to integrate structures of emergence and progression more closely, and
to unify both in an formal approach to game design that is based on model
transformations, respectively.
5.1
Level Layouts
Level design has not been studied extensively. Yet, it is generally acknowledged that levels benefit from having a relatively simple gestalt or purpose,
especially when this gestalt matches the intended gameplay, rhythm and pacing. To this end, a number of scholars of games and interactive storytelling
categorized spatial structures frequently found in games. An overview and comparison of four of these typologies is presented in figure 5.1. This figure lists
all the categories in each typology horizontally. Their relative position and size
on the horizontal axis suggests overlap between the categories of each typology. Figure 5.2 illustrates each category with a schematic representation of their
structure.
Of these typologies, Marie-Laure Ryan’s specifically concerns itself with interactive story structure, while the others concern themselves with game levels.
Still, the similarities between many of the structures they describe is striking.
In fact, it is a common observation that in games, stories are, at least partially,
structured spatially instead of temporarily (Jenkins, 2004). This causes some
confusion whether these categories concern themselves with level geography or
topology. As Ryan focuses on interactive storytelling, her categories are clearly
topological, but the other three typologies are much more geographical in nature. From these typologies it appears that in level design topological structures
and geographical structures are frequently isomorphic or at least often treated
as such.
The different categories in the typologies in figure 5.1 can be grouped in to
three super-categories. The categories on the left all concern tightly controlled
structures that are fairly linear and directed; players usually cannot go back in
these structures. Categories that describe railroading and branching techniques
fall in this group. The categories in the center are all networks of some sort.
These categories offer players more freedom to explore the world but still restrict
players’ positions to a restricted set. The categories on the right feature more
open world that impose few restrictions on player movement.
One observation that can be made from figure 5.1 is that, although there
definitely is some overlap between the different typologies, their differences are
also quite prevalent. What is a main category in one typology might be a
special case of a category in another. For example, ‘bottlenecking’ with Byrne
can be regarded as a special instance of ‘parallel structures’ with Adams and
Rollings. Categories present in most typologies are omitted in another: Adams
and Rollings have no category that covers the ‘tree/branching structures’ with
Chapter 5 | Mission/Space
111
Figure 5.1: A comparison of four level typologies, based on (Ryan, 2001; Byrne, 2005;
Adams & Rollings, 2007; Schell, 2008)
Ryan and Byrne. Other categories, such as ‘dynamic levels’ with Byrne and
‘grids” with Schell, probably have no place in the typology at all, as they are
separated by their mode of creation and not their topological structure.
The categories within some typologies do not appear to be on the same level
as some other categories within the same typology. For example the ‘hub-andspoke’ structure can be regarded as a special case of ‘network layout’ in Adam
and Rollings typology. It might be a quite common structure in games, but it
does not capture nearly as many different structures as the network layout does.
At the same time, other specialized structures can be found that do not feature in
any of these typologies. For example, a variation of the ‘hub-and-spokes’ layout:
the ‘hub-with-loops’ layout, in which the player progresses through longer linear
and directed level sections before returning to the hub can also frequently be
encountered in games (see figure 5.3). Three of the nine main dungeons in The
Legend of Zelda: Twilight Princess1 have this type of layout: Link, the
game’s protagonist, can often quickly return to a central location after activating
the trigger that unlocks a new section. The other six main dungeons have a
normal hub-and-spoke layout. The point is, that from the typologies presented
in figure 5.1, one cannot be certain to classify the hub-and-loops structure as a
special combination of hub-and-spokes and linear structures or as a category of
its own. The typologies do not provide enough guidance to safely argue one way
or the other.
These problems indicate that understanding levels in games is not as straightforward as it may seem. The reason for this may be, as is one of the main points
of this chapter, that levels do not constitute a single structure: they contain
1 This includes the ‘Hyrule Castle’ level that builds up to the final conflict with the game’s
main antagonist.
112
Joris Dormans | Engineering Emergence
Figure 5.2: Illustrations of the level typologies found in (Ryan, 2001; Byrne, 2005;
Adams & Rollings, 2007; Schell, 2008)
Figure 5.3: A hub with loops structure
Chapter 5 | Mission/Space
113
Figure 5.4: Cell diagram representing level structure, after Byrne (2005, 69).
at least a geography and a topological structure. Separating a level’s topology
from its geography will help to create much clearer perspective on game levels.
In this respect, I consider the layouts discussed above to be geographies. A
level’s topology would focus more on the logical structure of the player’s tasks.
5.2
Tasks and Challenges
A level’s topology concerns itself with the tasks and challenges players must
perform in order to complete a level. These tasks or challenges are usually
fairly straightforward tests of the player’s abilities. They might take the form
of puzzles, fights, traps or hidden objects that need to be collected. Ed Byrne
suggests structuring these tasks in cell-diagrams outlining the game’s flow and
highlighting the different player tasks (see figure 5.4). These cell diagrams are
simple informal structures that read almost like a storybook that help design a
level’s layout or rhythm. Cell diagrams focus on a game’s logical and temporal
structure instead of its spatial layout (Byrne, 2005).
The temporal dimension of games and the players experience is also emphasized by the approach of Ben Cousins (2004). Cousins introduces the idea of a
hierarchy to order the gameplay experience. In a detailed analysis of the game
Super Mario Sunshine he divides the experience in five layers, where the top
layer constitutes the whole game. The subsequent layers describe the individual
missions, mission elements, input elements, and primary elements. The middle
layers intuitively correspond to player plans and intentions. In cognitive science
they would have been called basic level categories for the games actions; these
actions are most accessible to players and would typically be the actions players would use when describing the gameplay. The lowest layer represents the
actions made possible by the game mechanics and interface; they correspond to
the buttons pressed by a player and the resulting actions of the avatar, such as
jumps or simple moves. Cousins stresses the importance of the quality of these
114
Joris Dormans | Engineering Emergence
Figure 5.5: Cell diagram representing level structure after Adams & Rollings (2007,
282).
low level actions as the player spends much time performing them. Using simple
metrics Cousins shows that 55 percent of the primary elements in a four minute
session of Super Mario Sunshine consisted of running forward and changing
direction.
The ‘hierarchy of challenges’ described by Ernest Adams and Andrew Rollings was directly inspired by Cousins’ analysis. However, where Cousins focuses
on individual sessions, with the hierarchy of challenges Adams and Rollings abstract away from individual sessions and represent all possible trajectories in a
single image. In this hierarchy all the game’s challenges are ordered into a layered structure representing what a player needs to accomplish to complete the
game (see figure 5.5). To build a hierarchy of challenges, inside knowledge of the
level’s design is required. Some challenges can be performed in different order
or even simultaneously. At the lowest level in the hierarchy are the atomic challenges: the micro actions the player needs to perform to get ahead. At higher
levels in the hierarchy there are goals of individual sections and missions. At
the highest level the game’s ultimate goal resides. Adams and Rollings discuss
the different needs of visibility among the levels: games need to be very clear
in their high level and atomic level challenges while they can be less clear for
intermediate challenges. They also stress that presenting a player with simultaneous atomic challenges considerably increases the game’s difficulty (Adams &
Rollings, 2007).
Creating simultaneous or alternative options in the hierarchy of challenges
leads to a potentially more varied gameplay. However, this is more difficult in the
higher levels of the hierarchy than it is in the lower levels (see also Arinbjarnar &
Kudenko, 2009). Deus Ex is a good illustration of this. In this game the player
has to get through a series of connected missions. At most low level challenges
the player is presented with options: to deal with a guard the player might use
stealth or violence. As players progress they choose how the game’s protagonist
develops his skills: players can choose to specialize in different strategies to
Chapter 5 | Mission/Space
115
overcome common challenges, such as the the use of force or stealth. Yet, at the
high end of the challenge structure there are only few options and branches. It
is only at the very end of the game that the player gets to choose between one
of three alternative endings.
5.3
The Mission/Space Framework
The two different approaches, focusing on the geographic layout of a level on
the one hand and on the sequence of tasks on the other, suggest that in a level
both structures exist at the same time. These structures are superimposed onto
each other and as a result it is all too easy to confuse one with the other or
to take their interrelation for granted. The Mission/Space framework formalizes
both perspectives and foregrounds that spaces and missions coexist within levels;
a level has a particular geometric layout and a series of tasks that need to be
performed in that space. At a first glance this observation seems obvious, and
in fact, the level layouts and the hierarchy of challenges as discussed above
seem to acknowledge a similar distinction. At the same time, many designers
seem to forget about this distinction: for many games the mapping between the
mission and the space is quite direct and their structures often are quite similar,
even isomorphic. This section formalizes the structures of missions and spaces
in games into a single framework. The advantages of this framework, where
missions and spaces are treated as separate but related structures, is that their
individual structures, but also their relations, can be investigated with much
more clarity. The relation between mission and space is more sophisticated
than a superficial survey would suggest. Games might reuse the same space for
different missions, as is the case in System Shock 2 where the player traverses
the same areas of a spaceship multiple times. System Shock 2 shows that the
same space can accommodate multiple missions (assuming that the individual
mission structures do not resemble each other too closely). Reuse of game space
in this way is often economic: the developer does not have to create a new space
for every mission in the game. It has gameplay benefits as well. For example,
players can use previous knowledge of the space to their advantage, adding to
the their sense of agency and the depth of the gameplay.2
A designer of a level works with both missions and spaces, just novelist wotk
with plots (the sequence of events) and their narration (the way these events are
told).3 During the design process, the designer is likely to loop between mission
constraints and spatial affordances while setting up a delicate balance between
2 Agency is commonly used to describe a player’s power to affect and influence a game world.
Agency is a core concept in Janet Murray’s book Hamlet on the Holodeck where she defines it
as follows: “the satisfying power to take meaningful action and see the results of our decisions
and choices” (1997, 126).
3 With plot and narration I refer to the similar notions of “story” and “narration” used by
Gérard Genette (cf. 1980, 29). However to avoid confusion between the common sense use of
the word “story” and Genette’s more technical use I use replaced it with “plot”. To add to the
general confusion, Foster uses the terms “plot” and “story” to indicate a similar distinction,
but here “story” means narration, and is completely opposite to Genette’s use of the word
(Foster, 1962).
116
Joris Dormans | Engineering Emergence
the two that ultimately facilitates the target gameplay. For players, the space is
the access point: in space, through play, they realize a mission.
Mikhail Bakhtin’s notion of the ‘chronotope’ provides another interesting
parallel for the combined mission and spaces in games. The chronotope refers to
the artistic relation between time and space in literature. Bahktin observes that
in literature time and space are “fused into one carefully thought-out, concrete
whole” (Bahktin, 1981, 84), and that their particular artistic intersection is
important to the literary form. He also observes that particular chronotopes
correlate to particular literary genres. The first genre Bahktin describes is that
of Greek Romance. The plot typically involves two lovers who have to overcome
many obstacles in order to marry. This requires many adventures that take place
in several far-away countries. The intersection of plot-structure and the journey
through hostile environments is surprisingly uniform across many stories from
this genre. For games it is the artistic intersection between mission and space
that has a similar, large impact on the experience. It also seems that particular
configurations of missions and spaces correspond to particular game genres. For
example, the mission/space configurations found in action-adventure games are
usually far more linear and restricted than the configuration of an open-world
with multiple quests that is typically found in computer role-playing games.
Often the complete, combined mission/space structure of a level is set up to
generate a narrative experience. For example, the Forest Temple level in The
Legend of Zelda: Twilight Princess is set up to generate play trajectories
that resemble elements of the hero’s journey as described by Vogler (2007). As
Link enters the temple, his mission is set up as he finds the first monkey of eight
monkeys he needs to liberate. Shortly after this he encounters a large spider
guarding the first hub in the level. Defeating this spider grants access to many
locations in the first part of the level; the spider acts as a threshold guardian
and marks the transition into the dungeon realm of adventure. What follows are
many tests and obstacles, during which our hero Link meets many new friends
and enemies. Halfway through the level, Link’s confrontation with the monkey
king fulfills the same role as “approach to the inmost cave” where the hero
confronts and fights the adversary or an important henchman for the first time.
The hero escapes with the “elixir” that ultimately helps him to defeat the main
adversary in a final climactic confrontation. In this case that elixir is the ‘gale
boomerang’: a magic device that acts as a weapon and a key at the same time.
With the gale boomerang Link can activate triggers that unlock the second part
of the dungeon that hides the final level-boss. Just as the same structure never
seems to grow stale for fairy tales and adventure films, it is a recipe that can be
found in many of Link’s adventures in this game or any game within the series.
It is also found in many Mario games, Nintendo games, and many other games.
More importantly, the Forest Temple level illustrates how both mission and
space are used to shape the play experience. The relation between the mission
and the space is less straightforward than it may seem. In section 5.6 I will
investigate the structure of the Forest Temple in more detail. First, I will discuss
how the Mission/Space framework uses two kinds of graphs to represent missions
and spaces respectively. These graphs have their own language and structures
Chapter 5 | Mission/Space
117
that are discussed in the next two sections.
5.4
Mission Graphs
Mission graphs represent the players’ progress towards a goal not by tracking
their physical location, but by tracking the tasks they must perform to finish a
level. A mission graph is a directed graph that represents a sort of to-do list with
each node representing a task that might or must be executed by the players.
Nodes can be in one of three different states; the current state of a mission is
determined by the respective states of its tasks:
1. A task can be available. When a task is available the player can execute
it.
2. A task can be unavailable. When a task is unavailable the player cannot
execute it.
3. A task is completed when the player has successfully executed the task.
Edges in the mission graph determine how changes in tasks’ states affect the
state of other, subsequent tasks. There are three types of edges between tasks
(also see figure 5.6):
1. A strong requirement indicates that the completion of the source task
(called a strong prerequisite) is a necessary condition for the availability of the target task. A strong requirement is represented as a double
arrow directed towards the target task.
2. A weak requirement indicates that the state of a target task is changed from
unavailable to available when the source task (called a weak prerequisite)
is completed or available, unless this is barred by the incompleted state of
a strong prerequisite of the target task, or when this is barred by explicit
inhibition (see below). A weak requirement is represented as a single arrow
directed towards the target task.
3. Inhibition indicates that the state of a target task is set to unavailable
when at least one the source task (called an inhibitor) is completed, unless
the state of the target task is completed. Inhibition is represented as a
solid line that ends in an open circle at the side of the target task.
118
Joris Dormans | Engineering Emergence
Figure 5.6: Basic elements of mission graphs
When a task has multiple predecessors, its state is determined as follows:
If it is completed, then its state no longer changes.4
Else, if it has inhibitors, then its state is set to unavailable when
at least one of the inhibitors is completed.
Else, if it has strong prerequisites, then its state is set to available
when all strong prerequisites are completed.
Else, if it has weak prerequisites, then its state is set to available
when at least one of the weak prerequisites is completed.
Else, its state is set to unavailable.
There are three classes of tasks in a mission graph (also see figure 5.6):
1. Regular tasks are represented as circles. Different colors and letters can be
used to represent specific (types of) tasks.
2. An entrance is a task that serves as an entry point for a player. A mission
graph should have at least one entrance, but might have several entrances
representing different starting points for the mission. Entrances cannot
have prerequisites and one entrance starts in the available state to represent
the player’s point of entry.5 Entrances are represented as circles that are
marked with an arrow head pointing towards it.
3. A goal is a task that finishes the mission when it is completed. A goal
cannot be a prerequisite or an inhibitor for another tasks. A mission can
have multiple goals in which case it is finished when one of these goals is
completed. Goals are represented as circles that have a double outline.
4 Curiously, in games it is fairly uncommon to be able to undo a task, which might make
sense when that task inhibits other tasks. Allowing a special undo edge that could revert the
completed state of a task would allow mission graphs to incorporate the option of allowing
players to overcome problems they created for themselves. This notion, tantalizing as it may
be, is left unexplored in this dissertation.
5 In the case of multiple entrances, what entrance is activated depends on the player’s or
the game’s actions prior to that mission.
Chapter 5 | Mission/Space
119
Figure 5.7: Mission graph displaying state to represent progress.
Static mission graphs cannot display changes to the states of the nodes.
However, in a digital version of mission graphs this can be done. For example,
figure 5.7 displays a mission graph that does represent the state of the tasks:
unavailable tasks have a gray outline and fill, available task have a solid outline
and completed tasks are checked. Figure 5.7 represents a mission state where a
player has progressed through half the entire mission. A dynamic representation
of mission graphs displaying states in this way can be developed into a useful
design and analytics tool. This notion is explored further in section 5.8.
At a first glance, when a task has only one predecessor, it seems to make little
difference for a mission whether that predecessor is a weak or strong prerequisite. However, as there is no need to execute and complete a task that is a weak
prerequisite in order to proceed, players might simply ignore weak requirements.
In the case when a fight is a weak prerequisite to finding a key, players might
ignore the fight and simply pick up the key. Would the fight be a strong prerequisite for finding the key, for example because the enemy is actually carrying it,
players are forced to fight.
Weak prerequisites are quite common in games. For example, the twodimensional, vertical scrolling shooter game Star Defender launches a sequence of enemies against the player (see figure 5.8). In this type of game,
players are usually not required to actually kill those enemies; they can simply
ignore them and proceed to the end of the level without firing a single shot.6 The
structure of a game like that might be represented as figure 5.9. In a way, this
structure can be replaced by a wide branching mission (see figure 5.10). After
all, none of the tasks in the mission is a strong prerequisite for another; the tasks
might be executed in any order. The player might even choose to complete the
goal immediately. Once players start the mission, all other tasks become available. However, in this case, the order in which players will encounter tasks is
lost. This order might be important, as increasing difficulty, pacing and task
variation are all considerations that impact mission design. Therefore, the mission structures in figures 5.9 and 5.10 are not the same and specifying the order
in which players will (likely) encounter weak prerequisites does matter.
6 Although this would make the game very, very difficult. In addition, the final enemy of a
level usually is a level-boss that must be defeated to finish the mission.
120
Joris Dormans | Engineering Emergence
Figure 5.8: Many of the tasks in Star Defender are not required to finish a level.
Figure 5.9: A linear mission with
weak requirements.
Figure 5.10: A branching mission
with strong requirements.
Figure 5.11: A linear mission with
one solution.
Figure 5.12: A linear mission with six
solutions.
Chapter 5 | Mission/Space
121
In addition, there is an important difference in the way strong and weak
requirements map to game spaces. It makes sense to place all tasks with a
weak prerequisite closely after that prerequisite. That way the order in which
the player encounters them is preserved. On the other hand, because strong
requirements dictate the order in which players can execute tasks more strictly,
a task can be placed much further from a strong prerequisite. One might think
of a weak prerequisite as a task players must encounter in order to get to the
task that requires it. This difference will be explored in more detail later in this
chapter and in chapter 7.7
Games with a linear structure of many strong requirements restrict the number of play trajectories considerably. For example, the mission in figure 5.11
only has one solution: players must simply execute all tasks in order to finish it.
Although structures like this can be found in games, they offer little variation
and replay value. Many games would do well to include weak requirements and
branches of strong requirements in order to create more interesting missions. For
example, the mission in figure 5.12 does not dictate in what order the player
must perform tasks A and B, but both must be completed before the player can
proceed. In addition, the player needs to perform at least C or D to be able to
complete the mission. Even old and linear games such as Super Mario Bros.
often offered alternative paths with different challenges to get to the end of a
level. Frequently, their mission structure is more like the one in figure 5.12 than
the one in figure 5.11.
5.5
Space Graphs
In the Mission/Space framework space is also represented as graphs consisting
of nodes and edges. In contrast to mission graphs, space graphs represent space
directly, most nodes represent places the player can be in. A space graph has
three different types of nodes. Any node in a space graph can be further specified
by using colors and letters to indicate different types. The three types of space
nodes are (also see figure 5.13):
1. A place is the elemental unit of space. What constitutes a place can
vary from game to game. In a game with discrete spaces such as a text
adventure or a tile-based board game, places correspond with the positions
the player can occupy. In games with continuous spaces a place might be
a room, a platform or a zone. Usually mechanics help define places in
these games: if a particular platform gives the player different options to
travel onwards than another platform in a gravity based platform game,
then both platforms are considered to be individual places. Places are
represented as open circles, whose size can be used to indicate the place’s
7 It might also suggest a second type of strong requirement, one that suggests a prerequisite must be completed and must be immediately precede the task that requires it in space.
However, the precise nature of the tasks can be used to determine how far apart a task and its
prerequisite can be placed: if the prerequisite is finding a key, they can be placed far apart;
when the prerequisite is jumping across a wide gap, it must be placed immediately before the
task.
Joris Dormans | Engineering Emergence
122
Figure 5.13: The basic nodes and edges of a Space Graph
relative dimension.
2. A lock is a special type of place that is not accessible until the player has
activated or acquired the correct key or keys. A lock can be in two states:
locked and unlocked. When a lock is locked it is represented as a black
circle with a white lock symbol, when it is unlocked it is represented as a
white circle with a black outline and a black opened lock symbol.
3. A game element is an item, character, creature or feature that can be
found in a particular place. Game elements are represented as small colored
shapes that are embedded within the circle of the place they are located in.
Game elements often have letters indicating their particular type. Game
elements can only be located in places; locks cannot contain game elements.
Typical game elements that are frequently found in the examples that
follow are entrances, goals, keys and triggers (marked with ‘e’, ‘g’, ‘k’ and
‘t’, respectively). These elements represent game objects that implement
mission logic.
Space graphs can have many different relations represented by different types
of edges. Some of these depend on a game’s particular implementation. The
more common relations include:
1. A path indicates that players can move freely between two places. A path
can be traversed in two directions, but as it is usually important to know
which place the player is likely to encounter first, a path is represented as
an arrow pointing towards the space that is further away from the player’s
point of entrance. 8
2. A valve connects two places or a game element and a place. It can be
traversed only in one direction. A valve is represented as a solid arrow
8 In addition, this difference will play an important role when performing transformations
on game spaces, certain transformations will allow game spaces to be reorganized in such way
that game element or connections are moved towards the starting location, but not the other
way round.
Chapter 5 | Mission/Space
123
Figure 5.14: A window in Prince of Persia showing a potion that cannot be
reached from this direction, prompting the player to paths that are not yet discovered.
with a different arrow head ( ) indicating the direction of travel. A
valve between two places is only traversed when a player chooses to do so.
A valve that starts from a game element automatically transports a player
when the player activates the element (voluntarily or involuntarily). This
can be used to represent teleporters or traps.
3. A window indicates that the player can see a place from another place, but
cannot directly travel between these places. A window is not necessarily
literally a window: any place the player can see but not immediately reach
is considered to be connected by a window. Windows can prepare the
player for challenges to come. They can also point the player at areas that
are otherwise hidden or difficult to reach and the rewards that getting there
might yield (see figure 5.14). In general, players assume that a visible yet
inaccessible place does have a path leading to it. Windows are represented
as dotted arrows connecting two places. The arrow points towards the
place that can be observed.
4. An unlock relation connects a game element with a lock, or it connects two
game elements. When connected to a lock, the source game element acts
as a key to open it. Locks can have multiple keys in which case all keys
must be acquired or activated before the player gains access to the lock.9
An unlock between two game elements indicates that the first element is
required to use, obtain or activate the second element, or it might indicate
that an element is protected by an another element. Unlock relations are
represented as dotted arrows pointing towards the lock or the element
being activated.
5. A lock relation connects a game element with a lock. The game element
acts as a trigger to close the lock. Lock relations are represented as dotted
lines that end in an open circle at the side of the lock.
As with mission graphs, the basic elements of space graphs can be used
to create a number of elementary constructions. These include, but are not
9 Although it is imaginable that for a particular game this is implemented differently and
any one key provides access to a lock.
124
Joris Dormans | Engineering Emergence
Figure 5.15: A pathway and an open space.
restricted to:
• A graph that connects two places with paths and valves is called a pathway.
The pathway includes the two places it connects, and any places in between
(see figure 5.15). The shortest possible pathway simply connects two places
directly with a single path.
• A pathway that does not contain any closed locks or valves is called an
open space (see figure 5.15). How the open space is constructed does not
really matter; even if the place is constructed from a set of connected
rooms in a textual adventure it is still considered to be an open space,
as long as the player has the ability to travel to all places in the open
space freely without having to activate game elements. A linear pathway
through which the player can freely travel back and forth is also considered
an open space.
• A locked short-cut can be used as an alternative for a valve. It blocks
player’s access from one direction but can be easily unlocked from another
direction. Once it is unlocked it will stay open. Locked short-cuts are
often implemented with doors that open only from one side (see example
a in figure 5.16).
• A hub is a central place frequently visited by the player from which several separate (usually four or more) pathways start (see example b in figure 5.16). Although some pathways might be initially closed and some
pathways might lead back to the hub through a valve. A hub is a good
place to locate entrances and ‘save points’ as it can minimize the amount
of backtracking after a set-back.
• A set-back forces players back to a previously visited place as a result of a
failed challenge or through the use of traps (see example c in figure 5.16).
It will take time and perhaps resources to return to an earlier place. Games
with ‘save points’, and in which the player frequently dies, automatically
Chapter 5 | Mission/Space
125
Figure 5.16: Example illustrating a locked short-cut (a), a hub (b) and a set-back
(c).
have set-backs. Set-backs are also frequent in platform games where failing
crucial jumps can force the players to lose much altitude that might not
be easily regained (also see Compton & Mateas, 2006).
• In contrast to the set-back, a force-ahead pushes players into previously
unexplored areas for which they might be ill-prepared or lack the proper
equipment. In general, chances of player death and failure increase, but the
player should be allowed a chance to return to the previous stage, leading
to a survival challenge. The old dungeon crawler Eye of the Beholder
II includes trapdoors that drops the player into lower, more difficult part of
the dungeon from which the player would struggle to return. A force-ahead
can be realized with a valve to a new and very dangerous place.
5.6
Level Analysis: The Forest Temple
The differences between mission and space, and their relation, become more
clear from the analysis of the ‘Forest Temple’ level in The Legend of Zelda:
Twilight Princess (see figure 5.17). The The Legend of Zelda series of
games are well known for their quality game and level design and this game is
no exception. In this level, the player, controlling the game’s main character
Link, sets out to rescue monkeys from an evil presence that has infested an old
temple in the forest. The mission consists of the player freeing a total of eight
monkeys, the defeat of the mini-boss (the misguided monkey king Ook), finding
and mastering of the ‘gale boomerang’ before finally defeating the level-boss (the
‘Twilit Parasite Diabara’). Figure 5.18 displays the forest temple level map as it
is published in the official game guide. Figure 5.19 presents a mission graph and
a space graph of the level. In order to reach the goal, Link needs to confront the
level-boss in a final fight. In order to get to that fight, Link needs to find a key
and he needs to rescue four monkeys, for which he needs the gale boomerang, for
which he needs to defeat the monkey king, etcetera. Some tasks can be executed
in different order: it does not really matter in what order Link liberates the
126
Joris Dormans | Engineering Emergence
Figure 5.17: Confronting the misguided monkey king Ook to obtain the gale
boomerang in The Legend of Zelda: Twilight Princess.
Figure 5.18: The map of the Forest Temple from the official game guide (Hodgson
& Stratton, 2006, 68).
Chapter 5 | Mission/Space
127
Figure 5.19: Mission graph and space graph for the “Forest Temple” level from The
Legend of Zelda: Twilight Princess. The gray lines connect the corresponding
fights in both graphs.
128
Joris Dormans | Engineering Emergence
monkeys two, three four. Other tasks are optional, but lead to useful rewards.
The mission structure for the Forest Temple level has a few striking features.
One is the bottleneck formed by the fight with the mini-boss and the retrieval
of the boomerang halfway through the structure and the two sections of relative
nonlinearity before and after the bottleneck. This structure corresponds directly
to the learning curve of the level in which the player needs to train with the new
weapon before using it to defeat the level-boss. It is a structure that is frequently
found in Nintendo games.
The game space features a hub-and-spoke layout (see section 5.1) that supports the parallel tasks of the mission structure. From the central hall (where
the first guardian is fought) the player can go into three directions. The pathway
that leads to the right, quickly branches into three more pathways. Three pathways lead to captured monkeys and one to the mini-boss. The last pathway is
only open to Link after he has freed the first four monkeys. After the player has
retrieved the boomerang, additional spaces in the first hub-and-spoke structure
and a new hub can be reached. This structure is not the only possible space
to accommodate the mission. It is not very hard to imagine different types of
spaces that accommodate the same mission structure. Even a linear layout, in
which the rooms are aligned in a long row with different tasks in each of them
would be a possibility.
When playing the game, a trajectory through the space is generated by the
player. This trajectory can be represented as a linear string of visited nodes.
Trajectories that lead to the end goal of the level might be called solutions. The
individual trajectory is constrained by the structure of the map. The player has
to start at the entrance and the only way to progress is to find the first monkey.
From there on an infinite set of trajectories becomes possible (infinite, because
the space allows the player to travel back to earlier locations). Yet, this does not
mean that all combinations are possible. Players must first defeat the mini-boss
before retrieving the boomerang, and they must have retrieved the boomerang
before they can reach the master key. The layout of space constrains the possible
play-trajectories and solutions.
From this analysis it should become clear that both mission and space have
their own individual structure. There is some natural affinity between the huband-spoke space and the required parallel branches in the mission, but the two
are not directly linked. As mentioned above the same mission structure could
be mapped to a linear structure in space, whereas a linear mission could also
be mapped to a hub-and-spoke layout (one where the order of the spokes to be
visited is fixed).
The gale boomerang itself is a good example of the lock and key mechanisms
typical for the series and indeed this is used in many action-adventure games (cf.
Ashmore & Nietsche, 2007). Lock and key mechanisms are one way to translate
strong prerequisites in a mission into spatial constructions that enforce that
relation between tasks. The boomerang is both a weapon and a key that can
be used in different ways. It has the capability to activate switches operated by
wind. Link needs to operate these switches to control a few turning bridges to
give him access to new areas. In order to get to the master key that unlocks
Chapter 5 | Mission/Space
129
the door to the final room with the level-boss, he needs to use the boomerang
to activate four switches in the correct order. At the same time the boomerang
can be used to collect distant objects (it has the power to pick up small items
and creatures), and can be used as a weapon. This allows the designer to place
elements of the second half of the mission (after the mini-boss has been defeated)
in the same space that is used for the first half of the mission. This means players
will initially run into obstacles they cannot overcome until they have found the
right ‘key’.
5.7
Mission-Space Morphology
The structuralist tradition in narratology understands the part of the art of a
well-told story in terms of the structural relations between the different elements
in plot and narration. In a well-told story plot and narration are rarely the same.
Narration speeds up through, or omits uninteresting events. Narrations that use
flashbacks or flash-forwards present the events in the plot in a different order
than they have occurred. A common trick is to start a narration “in medias
res”, where the narration starts in the middle of or just the main action, only
to step back to relate the events that led up to the starting scene. Differences
between plot and narration are powerful narrative devices that are exploited
by master storytellers in any medium. What is more, a plot is not necessarily
linear, it might contain events that occurred in simultaneously; when the story is
narrated in a linear medium (as most traditional media for storytelling are), the
storyteller must use these devices to be able to tell the story at all. Traditional
structures and templates exist that have proven their value in the past and are
still used in books and films alike. From the structuralist analysis of Russian
fairy tales by Propp (1968) and theory of the monomyth of Campbell (1947)
to more modern interpretations used in contemporary cinema by Vogler (2007).
Many of these structures facilitate anticipation and involvement of the reader
with the narrated events. Differences between plot and narration are employed
to foreshadow, to change perspectives, to shock and to educate. The possibilities
are endless.
Similar, commonly used structures do exist for game spaces and missions. In
fact, the structures in figure 5.1 are examples of this type of structures, even
though mission and space are somewhat tangled up in them. Currently much
level design depends on a direct mapping between mission and space: all too
often, mission and space are isomorphic structures. A strong indicator for this
dependency is the popularity of the quest in many adventure games, a genre
that typically features strongly articulated spaces and missions. In the trope of
the quest, in which the journey of the hero indicates personal growth, mission
and space are isomorphic; they share the same structure. In many ways this is
very convenient. Dungeons & Dragons designer Dave Arneson is attributed
to have once remarked that Dungeons & Dragons levels (or dungeons) are
intentionally designed as flowcharts.10 It could be argued that a certain level of
isomorphism between mission and space guarantees a smooth player experience.
10 From
personal correspondence with David Ethan Kennerly (2009).
130
Joris Dormans | Engineering Emergence
In its most extreme case where a player is both railroaded through space and
mission the play experience can be tailored to that one single trajectory. HalfLife and Half-Life 2 have refined this strategy to perfection, and to great
commercial and critical success (see section 1.6). Strong isomorphism between
space and mission allows the use of many spatial metaphors or other spatial narrative devices. The symmetrical pathways mentioned above, could for example
be used to suggest a similarity between the tasks that must be completed along
those pathways.
On the other hand, in the ‘Forest Temple’ level, mission and space are not
isomorphic (see figure 5.19 above). During the first half of the level, the mission
with its parallel chains maps fairly directly to the hub-and-spoke layout of the
space. But once the gale boomerang is obtained the player has to traverse back
through the dungeon to gain access to previously inaccessible places. Would
mission and space be more isomorphic, Link would have to travel further into
a new section of the dungeon. Of course, as conventional level design wisdom
dictates, there is little point in using a lock and key system where Link encounters
all the keys first. In general, it is better to have the player find the lock before
the key in this way for three reasons.
1. When keys are encountered first, players will simply be forced to collect
everything they encounter without discrimination, which makes rather simplistic gameplay.
2. With obstacles and items that act as locks and keys but are represented
as something else, it is easier to recognize the key if players know what
the lock is; players then usually realize where they can proceed; they will
actively formulate the intention to return to the lock.
3. When players can negotiate obstacles they were unable to get past earlier,
they will experience progress and accomplishment. Thus, forcing Link
go back to previously visited places foregrounds his growth as a character.
Where there were obstacles he could first not overcome, he now can conquer
that space. It sets up a natural, clever contrast between Link’s state before
and after the defeat of the mini-boss and marks his progress along the
hero’s journey.
There is no reason why in a game mission and space should be isomorphic.
In fact, just as stories benefit from different morphologies for plot and narration,
so do games benefit from having different morphologies for mission and space.11
5.8
A Software Tool for Level Design
As is the case for the Machinations framework, the Mission/Space framework
is formal enough to be implemented as an automated design tool. A tool like
11 On a side note, I once experimented with an ‘in medias res’ structure for a tabletop roleplaying session, where I had players play through the events that lead up to the opening scene
they played through earlier. ‘In medias res’ is also used in Prince of Persia: Sands of Time
where most of the game is framed as a long flashback. In both cases the structure translates
well to games and actually contributes positively to the gameplay experience.
Chapter 5 | Mission/Space
131
that, with the name Ludoscope, was created as part of this research. Ludoscope
allows a designer to draw mission and space graphs efficiently. In fact, the mission and space graphs that feature in this chapter were created with Ludoscope.
However, Ludoscope is more than just a drawing aid. It also implements the
logic that governs mission and spaces. As a result, it can be used to simulate
levels at an early stage and help developers identify design flaws.
The implementation of mission graphs in Ludoscope allows designers to hook
up a network of tasks. Designers can indicate which tasks are entrances and
which tasks are goals. Based on that information, it determines which tasks
need never be encountered and marks them as such. By selecting a task, the
designer can immediately see what other tasks can be reached from that point in
the graph, these tasks are given a green outline. When a selected task inhibits
another task it displays the inhibited task with a red outline. Other tasks that
have become unreachable because of this are also displayed with a red outline
(see figure 5.20). This allows designers to select tasks one at a time in simulation
of a player completing tasks in the game and checking if the mission still can
be completed. Ludoscope can easily be extended to include more checks for
inconsistencies or warn the designer for potential deadlocks. In addition many
types of heuristics for games might be implemented to assist designers, such as
space syntax measures described by Paul Martin (2011).
The implementation of space graphs in Ludoscope allows designers to do
something similar with spaces representing missions as well. By selecting a place
in a space graph, all places that can be reached from that location are rendered
in with a green outline. In addition, designers can select game elements to unlock
(or lock) locks. Again, this allows them to simulate a player’s progress through
a level (see figure 5.21).
Ludoscope is still work in progress, and as will be discussed in Chapter 7
has much more functionality than drawing Mission and Space Graphs only. In
particular, Ludoscope initially evolved from prototypes for procedural level generation tools, and it contains support to represent and involve mechanics through
Machination diagrams as well. This latter aspect allows designers to zoom in
on the relation between their game’s mechanics and level design, which, as will
be argued in the next chapter, is a critical aspect of designing games where
Figure 5.20: Selecting task B in Ludoscope reveals that task A is inhibited making
it inpossible to reach goal C. Goal D can still be reached.
132
Joris Dormans | Engineering Emergence
Figure 5.21: Simulating the Forest Temple level. The light blue outline indicates
selected places and activated game elements. By selecting the place containing the
entrance and various keys, player progress can be simulated. In this case, the simulated
player is in the early stages of the level having made some progress in the lower left
and right corners.
Chapter 5 | Mission/Space
133
structures of emergence and progression are tightly integrated.
5.9
Conclusions
As is the case with game mechanics, level design benefits from a more structural approach. A game level is a complicated construction consisting of many
different parts. Different models and perspectives can help level designers to keep
their focus and to structure their labor. To this end, this chapter introduced
the Mission/Space framework. This framework includes specialized graphs to
represent missions and spaces separately.
Mission graphs focus on the tasks and challenges that players need to complete in order to finish a level. There are three types of relations between the
different tasks in a mission: weak requirements, strong requirements and inhibition. Mission structure foregrounds the interdependencies between tasks and
can be used to identify potential deadlocks or other inconsistencies, as well as
a balanced learning curve and training structure. Space graphs represent the
topographical structure of the level. They are indicative of how different places
are connected, a level’s flow, pacing, and what game elements can be used to
unlock new areas.
Mission and space are related, creating spaces to accommodate missions or
creating missions to suit a particular space is an important aspect of level design.
Although, sometimes games have levels where missions and spaces are almost
isomorphic, the structures of mission and space are independent. A detailed
analysis of the Forest Temple level in The Legend of Zelda: Twilight
Princess showed that games benefit from more sophisticated relation between
missions and spaces.
In order to aid level designers, the Mission/Space framework can be used to
create level design tools that help designers create complex but consistent levels.
The Ludoscope prototype implements some of these ideas. It allows designers
to create mission and space graphs and subsequently simulate players progress
through them.
The Mission/Space framework provides a solid structure that can be used
as a stepping stone in the development of further theory for game design. In
the next chapter, the Machinations and Mission/Space framework will be used
to explore how structures of emergence and progression can be integrated. In
Chapter 7 both frameworks are leveraged to create a framework and prototype
that allows the automation of certain task in order to support the creative process
of designing games.
134
Joris Dormans | Engineering Emergence
The challenge for game designers who want to create rich,
open game worlds and tell interesting stories at the same
time, is to move beyond the constraints of unicursal corridors or multicursal hub structures while keeping the
player’s attention on a storyline. And it is no easy task.
Espen J. Aarseth (2005, 11)
6
Integrating Progression and
Emergence1
This chapter focuses on the relation between mechanics and levels, which were
the subjects of the previous two chapters. The Machinations framework and the
Mission/Space framework formalize different perspectives on these two aspects of
games. Machinations diagrams help visualize and understand a game’s internal
economy. It focuses on the mechanics that govern the economy and contribute
to emergence in games. The Mission/Space framework, on the other hand,
focuses on the construction of levels, and the mechanics that control players
progress through a game. These two aspects of game design: mechanics and
emergence, versus level design and progression, are often understood as two
opposing or conflicting ways of generating gameplay. It is the goal of this chapter
to leverage both frameworks in order to integrate structures of emergence and
progression more closely and bridge the gap between game mechanics and level
design. Ultimately, a close integration between emergence and progression would
lead to games where progression through levels, and perhaps stories, emerges
from the rules.
In section 6.1 I will discuss how professional designers typically create mechanics before they start focus on level design. In the two sections that follow, I will explore how the relation between mechanics and levels is typically
approached in games that traditionally rely on structures of progression, and
sandbox games that allow players to build and structure the internal economy
themselves, respectively.
Integrating levels and mechanics is not easy, and the number of games that
have successfully done so in the past is low. In section 6.4 I will discuss the
limitations that have been encountered in designing game levels that display
more dynamic and adaptive behavior. The problem often resides in a conceptual mismatch between relatively crude mechanisms for level and story on the
one hand and quite sophisticated and expressive mechanisms for physics simulations and/or economy on the other. In section 6.5 I will explore how feedback
mechanisms might be applied to mechanics to control progression. This explo1 This chapter is an extended version of a paper presented at the DiGRA Conference in
2011 (Dormans, 2011b).
136
Joris Dormans | Engineering Emergence
ration will lead to two new techniques to create progression through emergence:
1) feedback mechanisms for lock and keys, and 2) modeling progress as an economic resource. These techniques are discussed and illustrated in the final two
sections of this chapter.
6.1
From Toys to Playgrounds
In general, mechanics are designed before levels, and this means that in most
games, level design is dependent on the game mechanics. Particular mechanics
often serve as a starting point for levels in a game: a level might introduce
or focus on particular mechanics. For this reason, it is usually not possible
to finalize a game’s level design before finalizing the mechanics. After all, the
challenge posed by a platform carefully placed at a distance that is just less than
the maximum jump distance completely changes when the jumping distance is
increased. Worse, what was intended as an impossible jump, might change into
a possible jump or the other way round. It is also one of the reasons why the
MDA-framework stresses that game designers look at games starting with rules
and mechanics (see section 3.2).
Kyle Gabbler summarizes the dependency of level design on game mechanics
as one of seven tips in his video keynote for the first Global Game Jam in 2009
(Gabler, 2009).2 He advices game designers to “make the toy first”: he urges the
participants not to create any art assets until the basic mechanics of the game
are in place. These basic mechanics should already deliver an interaction that is
fun, even without goals or well designed levels. For example, if one sets out to
build a racing game, the car should be built first, it is only when the car is fun
to drive around, that one should start work on graphics and on tracks. From
my own experience as a Global Game Jam participant in 2009, 2010 and 2011 I
can testify that this tip is very useful.
Gabler’s tips are not restricted to game jams and games built under extreme
time constraints only: they are valuable for any game development project.
Games are developed by creating many prototypes during short iterative steps.
The earlier prototypes tend to resemble Gabler’s ‘toys’: they often implement
the basic mechanics and interactions, without spending much attention to goals
or levels. In most cases, the levels that exist at this stage, serve as testing
grounds to illustrate the basic game mechanics. Designers typically throw in as
many game objects as they can for play-testers to play around with. The goal
of these prototypes usually is to explore the limits of the mechanics, and to look
for opportunities to create interesting interactions, or explore how game objects
might be combined into interesting puzzles and challenges. It is only when the
designers are convinced that the basic mechanics are solid that the virtual space
is organized into a level set up to deliver a concise and well-structured experience;
when the individual toys are organized into a playground.
The emergent simulation game Spore was developed using a large number
of small prototypes, and serves as an excellent illustration of this process. What
2 The Global Game Jam is an annual event in which teams at hundreds of locations all over
the world enter a competition to build a game based on a set theme within 48 hours.
Chapter 6 | Integrating Progression and Emergence
Figure 6.1: SPUG prototype for the
creature stage in Spore
137
Figure 6.2: Screenshot from the creature stage of Spore.
is unique for a large commercial game, such as Spore, is that many of these
prototypes are available for download.3 These prototypes were built for various
reasons. Some were experiments with the procedural techniques used to create
terrains, worlds and even entire galaxies. While others are gameplay experiments designed to try out various rules and behaviors. One of these, SPUG
(figure 6.1), is a prototype for the “creature stage” in the published game (figure 6.2). According to accompanying text on the website in this prototype “no
limitations are placed on leveling up or cheating stats. This tool was intended
to give designers the opportunity to explore different economies for the creature
game, so limitations on power-ups and level ups are self-imposed.”4 Thus it was
used to experiment with the same type of game mechanics that is the focus of
the Machinations framework.
The design strategy that gives mechanics dominance over level design has
a number of implications. It provides certain types of games with a natural
structure for progression, as will be explored in the next section. Other games,
that focus more on dynamic gameplay created by sophisticated mechanics (as
explored in section 6.3), use levels only to set some rough boundaries that provide
some color and perhaps some constraints on the emergent gamepay.
6.2
Progression through Structured Learning Curve
One of the many considerations of designing levels, is to train the player in
the required gameplay skills necessary to complete the game. In general, players
do not want to read manuals in order to play a game; they expect that learning
the mechanics is a natural result of playing. This means that levels need to be
structured in such ways that players actually learn the game while playing it.
This type of structure fits well with games of progression that frequently model
their structure after the hero’s journey. A sense of growth and accomplishment
drives these games and sets up a particular relation between game mechanics
3 See
4 See
http://eu.spore.com/extra/?id=10488 (last visited July 17, 2011).
http://eu.spore.com/extra/?id=10626&lang=en (last visited July 17, 2011).
138
Joris Dormans | Engineering Emergence
and level design that is common among action-adventure games, first-person
shooters and genres that typically include structures of progression.
Daniel Cook’s skill atoms constitute one of the most concrete theoretical perspectives on this aspect of level design (Cook, 2007). He analyses the individual
steps a player goes through in learning a new game skill, and the way individual
skills are hooked up into chains. Once the design team has decided on the final
mechanics to be included in the game, levels can be structured in such way that
the player is taught these mechanics.5 The most straightforward approach is to
spread out the chain of skills over the level and to organize the level accordingly.
In this case, the chain of skills is integrated into the level’s mission structure,
which is then related to the level’s spatial layout in a similar way as described
in the previous chapter. However, levels are not there to teach the player the
required skills only; there is usually more to a level than just a tutorial. Levels
are also structured to facilitate exploration. Once the player has learned the
basics of playing a particular game, levels provide the player with opportunities to display their mastery. During this stage, the mechanics become a means
towards the goal of exploring the level or completing an interactive story.
The way the player is taught to use new items and strategies through level
design also resembles the learning stages found in martial arts training (Kohler,
2005; Dormans, 2005). This structure plays an important role in the level’s mission and is a common feature in many action games, in particular action games
produced by Nintendo. It is an excellent technique to learn the player the required actions to finish the game. It is an easily recognizable pattern of a game’s
mission, although to use it to the best effects, the space needs to be designed
to support this structure. In both action games and martial arts, students need
to master a whole vocabulary of different moves and combinations of moves.
In martial arts learning progresses through four stages in which techniques are
learned, practiced, combined and tested. These stages, called kihon, kihon-kata,
kata and kumite, are all present in the Forest Temple level of The Legend of
Zelda: Twilight Princess (see section 5.6). The kihon stage, in which a
player learns an individual technique in isolation and relative safety, is present
in both the earlier tasks involving the ‘bomblings’ and the tasks for which the
player needs the ‘gale boomerang’. These tasks are repeated a few times (kihonkata) before the player is taught how to use the boomerang to pick up the
bomblings to deliver them over greater distances before he can reach the final
two monkeys (kata). Finally the player needs this combined technique in order
to defeat the level-boss (kumite). The order and logic of these stages is dictated
by the mission structure. The player cannot encounter the level-boss before she
successfully demonstrated the techniques needed to defeat it (see figure 6.3).
This structure of introducing game mechanics gradually and having them act
as locks and keys can be found in a very pure form in certain smaller independent
5 Although I like to point out that the development process for games is highly iterative,
and that work on level design is often started before all mechanics are complete. It is not
uncommon that mechanics are changed because of insights gained during the design of levels.
However, the design process usually starts with mechanics, and mechanics are usually made
definitive before all levels are designed.
Chapter 6 | Integrating Progression and Emergence
139
Figure 6.3: A simplified version of Forest Temple mission highlighting the kihon,
kihon-kata, kata, kumite structure.
games. Knytt Stories and Robot Wants Kitty are good examples of such
games. Both of these games are platform games where the player’s goal is to
reach a particular location (even though these games’ story might frame it a little
bit differently). Both basically consist of one large level where the player gathers
a number of power-ups that act as locks and keys. But also the challenges the
player meets get progressively more difficult. Where, for example, the double
jumping ability allows the player to jump longer distances in both games, the
gaps the player needs to jump across do get wider, and the penalty for failing
a jump increases from, having to replay a little part of the level in order to try
again, to dying and/or replaying longer parts of the level.6 Table 6.1 lists the
power-ups in Knytt Stories in the same order they are encountered. Note
that some of these power-ups combine in even more powerful combinations. For
example, once players have found both the double jumping and wall climbing
abilities, they can use a double-jump after jumping from a wall. Likewise when
players have the double jump and the high jump, they can use both to jump
even further.
Larger games, such as The Legend of Zelda: Twilight Princess,
take more time to introduce their mechanics. As we have seen in the detailed
analyses of this game’s Forest Temple level in section 5.6, only two important
mechanics/power-ups were introduced in this level: the ‘gale boomerang’ and
the use of ‘bombling’ creatures. The Legend of Zelda: Twilight Princess
typically takes more time to train the player in the use of these mechanics, and
explores more possible combinations of the two, allowing it to design an entire
level around these two mechanics. Although I estimate the average time a player
6 A double jump is a common mechanism found in many platform games. A double jump
allows player controlled characters to jump once while in mid-air, effectively allowing the player
character to jump twice. It can be used to jump higher or further and emphasizes the timing
involved in performing jumps.
140
Joris Dormans | Engineering Emergence
Table 6.1: The power-ups of Knytt Stories in the order they are encountered and
their associated locks.
Chapter 6 | Integrating Progression and Emergence
141
Figure 6.4: Game elements as they are introduced in The Legend of Zelda: Twilight Princess. The data for this figure was collected from playing the game and
consulting a printed game guide (Hodgson & Stratton, 2006).
142
Joris Dormans | Engineering Emergence
will spend in the forest temple level is roughly half of the time spent to complete
Knytt Stories, The Legend of Zelda: Twilight Princess still takes
more time to introduce each power-up and explore its potential. As players explore more levels in The Legend of Zelda: Twilight Princess more and
more mechanics are introduced, although every level focuses on only a few new
mechanics and a few reused mechanics. Figure 6.4 provides an overview of this
game’s most important mechanics, when they are introduced and in what levels
they are reused. Mechanics that only act as locks and keys (such as ‘fused shadows’ and ‘mirror shards’) without being weapons or affecting the gameplay are
omitted in this figure. Many of the mechanics combine in interesting ways. For
example, wearing the ‘Zora armor’ allows the player to breath under water, when
also wearing the ‘iron boots’ will make the player sink allowing the exploration
of deep water caverns. In addition, the enemies and traps players encounter
require an increasing mastery of the game’s mechanics and their combinations
to defeat or avoid them.
What both Knytt Stories and The Legend of Zelda: Twilight
Princess do well, is to integrate the mechanics that control the progression
through the game world with the emergent mechanics that determine the core
gameplay. This is an effective and proven way to combine emergence and progression in a game. This is testified by the success of the Zelda games and the
frequency with which mechanics or power-ups are used in this way. However,
whether the two modes are truly integrated remains doubtful. Progression in
these games still relies on a fairly low number of game state changes, and needs
to be planned in detail to work. Over the past few decades, game designers have
learned to forge the two into pretty smooth game experiences. But it has not
led to a true synthesis of emergent and progressive gameplay.
6.3
Economy Building Games
Strategy games and management simulation games usually follow a different
design strategy to integrate their mechanics into level design. In stead of creating levels to dictate a structure of progression, these games allow players to build
and shape an emergent, internal economy within the constraints set up by the
level’s space or mission. This design strategy can be found in computer games as
Civilization, Caesar III or StarCraft, but also for modern, management
simulation board games such as Puerto Rico, Caylus or Agricola. In these
games an internal economy is built during play. Game spaces constrain the way
the economy might be built, but rarely accommodate a mission in the sense that
was discussed earlier. Instead of traveling towards a particular goal, the goal of
the game is to build up an effective game space. Missions, insofar they exist in
these games, often constitute of the single task to build an economy according to
some sort of specification. These economies usually start out quite simple with
the production of basic resources, but tend to get more complicated quickly. For
example in Civilization players build cities to produce food, wealth, knowledge, buildings and units. The location of the city affects the production rates:
building a city on fertile grasslands will increase the food production, rivers will
Chapter 6 | Integrating Progression and Emergence
143
Figure 6.5: In the game Caesar III players build cities in the Roman era.
increase trade and wealth, while hills and mountains offer the opportunity to
build mines to increase production of buildings and units. Players need to find
locations that best suit their needs. A player that is going for a strong military
will need more production, while building close to rivers can speed up trade,
wealth and scientific advancements. There are both long term and short term
considerations: cities that grow fast will eventually output more resources than
cities that are located close to interesting resources but far from fertile soil. The
default mode of playing Civilization generates new maps for every game the
player starts. Players will thus have to make the most out of the land they have
discovered.
Modeling a game like Civilization with the use of the Machinations framework presents us with a problem. Although many of the individual mechanics
can be easily captured using machinations diagrams, different sessions require
different diagrams as players effectively hook up game elements differently every
time. By building and changing elements in the game’s space, players also set up
their own game economy. These economies usually start out quite simple with
the production of basic resources, but they tend to quickly get more complicated.
For example, in the Roman city simulation game Caesar III (see figure 6.5)
players build cities in the Roman era. Players need to build infrastructure for
traffic and water, buildings to produce food and other basic resources. To build
up the city’s economy the player needs residences, workshops, markets and warehouses. Citizens will demand temples, schools and theaters, and at the same time
the player must provide security against different types of threats by building
prefectures, city walls and hospitals. Finally the player must train soldiers to
protect the city from invading barbarians. The cities economy is dominated by a
144
Joris Dormans | Engineering Emergence
Figure 6.6: The basic economic relations between the elements in Caesar III.
multitude of resources. Farms produce wheat, fruits, or olives, clay pits produce
clay that can be converted into pottery in special workshops. Other workshops
convert olives into oil, or metal into arms. The residences the players build are in
constant need of these and other goods. The better the player can supply these
residences the wealthier their inhabitants become, and that will in turn improve
tax income needed to build more farms, workshops and residences or pay for
other needs such as prefectures to fight crime and fire or military structure to
protect the city from invading barbarians. In the meantime players will need to
build granaries, markets and warehouses to distribute all these needs effectively
over the growing city. Figure 6.6 represents the basic economic relations between
some of these elements. The consumption of trade goods in residences triggers
the production of wealth. More wealth has a positive effect on the amount of
money generated through taxes. At the same time, wealth drains quickly creating an ever increasing need to supply residences with high quality trade goods.
In the game the actual connections between all these elements are flexible: a
farm might deliver its crops to a granary, warehouse or workshop depending on
the needs and the distances to these locations (see figure 6.7). The challenge of
Caesar III is to utilize space effectively and build a smooth running economy.
Players gradually build this economy as they see fit, but it will invariably be
dominated by the positive feedback loop that involves production, consumption
by citizens and tax income. This positive feedback is balanced by the negative
feedback provided by the dynamic friction built into the citizens mechanism (see
figure 6.8). The more effective players are in utilizing space and building up their
city, the more effective their economic engine will run.7
Games where players build the game economy in this way, very clearly fall into
7 Caesar III includes a few more elements that I have left out for the sake of discussion.
In the real game players also need to manage a number of hazards such as crime and fire by
building special buildings to counter these effects. They serve to complicate the production
mechanisms further. Apart from nutrition and wealth the citizens also demand entertainment,
culture, education and religion which are produced and consumed in similar ways. Finally, in
most levels of Caesar III players also need to deal with demands of the Roman emperor and
fight invading barbarians, they act as extra (periodic) friction to the economic engine.
Chapter 6 | Integrating Progression and Emergence
145
Figure 6.7: A possible spatial configuration for some elements of Caesar III.
Figure 6.8: The dominant economic structure of Caesar III.
146
Joris Dormans | Engineering Emergence
Jesper Juul’s category of games of emergence: it is almost impossible to write a
walkthrough for Civilization or Caesar III, but there are plenty of strategy
guides available online. Still, playing these games does offer something of an
experience of progression. Caesar III, for example, has a series of scenarios,
each with its own particular challenges and goals, and within each scenario there
are a number of scripted events. But even without these events, the process of
building a city goes through a number of stages, from the initial planning stage,
when players still have much money to build anything they might need, to the
managing of a crisis or the city’s defense, to the tweaking and fine-tuning of the
city’s economy in order to reach very hard economic goals during the end stages
of the later levels. Caesar III, as many other games of emergence, has its own
rhythm and progression that partly emerges from its dynamic game economy,
and partly from the scripted events that are unique for every scenario.
To design mechanics and to plan scenarios for these games is not easy. It
requires a thorough understanding of the game’s potential economy. Most of the
game elements the player can add to the game will somehow expand the game’s
economy, but there are only so many ways these elements can combine. Most
of these games will have a dominant economic structure as the one presented in
figure 6.8: these structures are the most effective way of setting up the economy
and players will eventually veer towards building similar structures. In some
cases, a game might contain a few alternative dominant structures. Most of
the time players will be able to familiarize themselves with only a few elements
at a time, learning the particularities of the game’s potential economy over
multiple sessions. This requires that players should be able to combine individual
elements in different ways, and preferably in such ways that these combinations
offer different strategic options.
At the same time, to keep players on their toes, a few scripted events, or a
rhythm that is dictated by a mechanism that is not very emergent at all, can be
used to structure scenarios and can give the game clear goals or closure. Caesar
III’s scripted events are one example of these, but similar mechanics can be found
in many other games, including board games (where one perhaps least expected
such mechanics). In the board game Caylus, for example, players compete for
the favor of king Charlemagne by building him a castle and surrounding town.
The game progresses through three stages after which a part of the castle is
finished and the players are rewarded for their contribution (or punished for
the lack thereof). Every round, the timeline that keeps track of the progress is
advanced one or two steps based on the players’ actions. Once the timeline is
advanced at least six steps, the first stage is completed; after at least twelve steps
the second stage is completed and after at least eighteen steps the third stage
is completed and the game is finished. This timeline drives the game towards a
conclusion, but does not constitute what I would call a dynamic mechanism.
6.4
A Mismatch Between Mission and Space
Although the design strategies to integrate mechanics and progression as
described in the previous two sections are successful in their own way, it seems
Chapter 6 | Integrating Progression and Emergence
147
that the development of design strategies that integrate the two more closely has
hit a barrier. This barrier seems to come from an imbalance in the advances that
have been made with creating sophisticated mechanics and game spaces on the
one hand, and the lack thereof in the development of missions and interactive
stories (Wardrip-Fruin, 2009). This mismatch between the mechanics of mission
and space has held back further integration of emergence and progression.
Over the past few decades the game development community has accumulated much experience with creating compelling game spaces with interesting
rules. From the early limited spaces from the seventies and eighties to the vast
virtual areas found in modern games, game spaces have grown into highly detailed constructions with near analogue qualities. Traversing the space of a
contemporary game is no trivial task, especially for those games that involve
movement in their core mechanics, as is the case with most action games. But
also for strategic games this evolution has been fast. One needs to simply compare the open free world of StarCraft II to the tile based combat found in
Civilization or indeed classic boardgames such as Chess, to appreciate the
strategic depth allowed by freely positioned units and more continuous terrain
features. Seeing these huge strides in the development of game spaces towards
structures with a high granularity, I agree with Noah Wardrip-Fruin who observes that it is curious that game stories and quests have not grown as much;
game missions usually work with a very limited set of possible states, all of which
are known before play (Wardrip-Fruin, 2009, 59).
One cause for the mismatch in granularity between mission and space, Noah
Wardrip-Fruin points out, is the popular quest logic and dialog trees common
to most games. The player’s progression through a mission is simply tracked by
setting up a few bottlenecks or gates to act as milestones in a story. Once the
player fulfilled the task associated with a milestone, the story is advanced. The
implementation is as simple as keeping track of a few simple Boolean story flags
that control the visible entries for the in-game journal that records the game
story (Wardrip-Fruin, 2009).
For example, the story-driven first-person shooter game Deus Ex has detailed rules to simulate combat in urban environments, where players can use
different strategies to overcome obstacles. An important aspect of the gameplay
is that players can choose between various upgrades of their character, allowing
them to specialize in stealth or direct combat, hacking computers or bypassing
locks. Most of the levels are set up to accommodate different playing styles. It is
even possible to complete the game without using lethal force against human opponents. Deus Ex accommodates a fairly continuous space of viable strategies
and tactics to overcome obstacles. On the other hand, the options presented to
players during narrative cutscenes and dialogs are always presented as discrete
choices between a handful of options (see figure 6.9), which have little effect on
the development of the narrative events. The progression of the levels is always
the same, sometimes dialogs and cutscenes are different, but the overall outcome
is usually the same. Only during the very final level, the actions of the player
will determine which one of three endings is selected.
The common implementation of dialog trees can further serve as an example
148
Joris Dormans | Engineering Emergence
Figure 6.9: A dialog in Deus Ex where the player can choose between four options.
of the problems faced by designers of interactive missions and stories in games.
Dialogs feature in many games, and while certain games do not even bother to
make their dialog interactive, those that do often resort to using dialog trees. A
dialog tree offers players a few optional lines to advance the dialog. Reaching
certain leaves in a dialog tree might change the story state. Many dialog trees
are not really trees, but are more akin to directed graphs as often different paths
through the tree will take players to the same node in the dialog. One problem
commonly associated with these tree-like structures is that they quickly become
inefficient and overly complex; the number of options that need to be created
is much larger than the average player will ever see, and without proper editing
tools the writer of a dialog tree might quickly lose track of all the options.
Worse, dialog trees do a rather poor job of really creating the illusion of freedom
or agency. With only a few options available at a time, chances are that players
will feel constrained in their options (Wardrip-Fruin, 2009, 56). In all likelihood
players will recognize the tree-like structure, and it is not uncommon for them to
traverse the entire tree in order to explore all possible gameplay options, which
mostly is a trivial yet tedious task. In short, at the micro-level of the dialog, these
tree-like structures often constitute poor gameplay (Dormans, 2006c). Still, at
the macro-level of mission or story, they are quite common.
It is probably one reason why many story-driven games do not offer any
variation in the way the story unfolds. Games such as Half-Life 2, The
Legend of Zelda: Twilight Princess, or StarCraft, all use their levels
and challenges as simple bottlenecks that players must pass in order to advance
the story. Failure to overcome an obstacle simply means that players will have
to try again. Sometimes there is little room for variation as is the case with
StarCraft II, but that rarely goes beyond the option to choose in which order
levels are completed and the occasional choice between two alternative options.
In the ten years that lie between the release of Deus Ex and StarCraft II
there seems to be little changes.
Chapter 6 | Integrating Progression and Emergence
149
For Noah Wardrip-Fruin the problem ultimately lies in the shape of the underlying processes: the processes that underly both the dialog tree and larger
interactive story/mission implementations are rather uninteresting. He suggests
that a new approach to game fiction is warranted and that this approach should
be fundamentally different from the quest flags and dialog trees that govern most
missions in games. (Wardrip-Fruin, 2009, 76). I propose that a closer inspection
of the mechanics that control game missions offers plenty opportunities to arrive
at a better shape for interactive missions.
6.5
Mechanics to Control Progression
The mechanics that govern progression through a level can be represented as
Machinations diagrams. These diagrams can be added to a space graph in order
to provide more detail. For example figure 6.10 represents a rough representation
of the Forest Temple space graph as was discussed in the previous chapter. In
this representation fights and tests of skills are omitted, the focus is on the
items that must be collected to finish the level: on the mechanics that control
progression. However, where before keys were connected directly to a lock,
keys now activate mechanics represented by Machinations pools and resources,
which in turn operate locks in the space graph. Extending space graphs with
Machinations diagrams allows the level to be represented in more detail.
The diagram might seem a bit overwhelming at first. The nodes that represent places in the level’s space all have a black outline, whereas the nodes
representing mechanics all have a colored outline. The translation of a lock and
key mechanism into a Machinations diagram is done with two pools and one
resource representing the key. This lock and key mechanism is isolated in figure 6.11. In this diagram the key element in the space graph triggers the pool
below it causing the resource representing the key to be transfered to the pool
on the right and there by activating (unlocking) the lock. Similar mechanics
are used to represent those locks that are opened with multiple keys, as is the
case with the monkeys in the Forest Temple level. In this case multiple pools
and resources represent those keys which must be collected on a single pool to
unlock new areas (see figure 6.12). In the forest temple level there are more lock
and key mechanisms that are omitted from figure 6.10 because they operate
only on a local scale. One of these mechanisms involves the bombling creatures
that spawn at particular locations and which explode a few seconds after they
have been picked up. The creatures can be used to destroy certain walls. This
mechanism is different from the other lock and key mechanics presented thus
far. However, it can easily be captured with a mixed graph representing both
space and mechanics (see figure 6.13). In this diagram, the bombling element
activates a source that produces a bombling resource that the player can use to
trigger a drain on a destructible wall, another drain is activated once per five
seconds representing the time the player has to use the bombling before it blows
automatically. This also means that player might actually fail to perform the
task to open the lock. In The Legend of Zelda: Twilight Princess this
failure poses no problem, as the bomblings respawn quickly after they have been
150
Joris Dormans | Engineering Emergence
Figure 6.10: A simplified space graph of the Forest Temple level extended with
Machinations to represent mechanics. The space nodes and edges have a black outline,
the Machinations nodes and edges are all colored. In this figure: b = boomerang, e =
entrance, g = goal, k = master key, m = monkey.
used, allowing the player to try again. While the player has a bombling, the
source producing new once is deactivated to make sure there is only one active
bombling in play. In addition, the time constraint (bomblings explode a few
seconds after the player has picked them up) constrains this mechanism to locks
that can be reached within a few seconds after grabbing a bombling. Hence the
use of a source and the additional drain in figure 6.13.8
In the remainder of this chapter I will focus on the theoretical possibilities
and difficulties of relating levels and mechanics using Machinations diagrams to
extend mission and space graphs. From the discussion in chapter 4 it became
clear that game mechanics benefit from having feedback loops. The lock and key
mechanics discussed above do not involve any feedback. Feedback needs a closed
circuit that consists of at least one state connection that is not an activator or
event; none of the mechanics above fulfill those requirements. There are two
design strategies to include more feedback in the game mechanics that control
progression through a level: 1) designers could develop lock and key mechanics
that involve feedback directly, or 2) progress itself could function more like an
abstract resource that can be gained (and might be lost) through mechanics that
operate on a fairly large scale. Roughly speaking the first strategy boils down
8 Although the diagram does not show that players might loose health when they fail to use
the bombling in time.
Chapter 6 | Integrating Progression and Emergence
151
Figure 6.11: The lock and key mechanism in isolation. In this figure: e = entrance,
g = goal, k = key.
Figure 6.12: The lock and multiple keys mechanism. In this figure: e = entrance, g
= goal, k = key.
Figure 6.13: The bombling lock and key mechanism. In this figure: b = bombling,
e = entrance, g = goal.
152
Joris Dormans | Engineering Emergence
to adding interesting feedback mechanics to control locks to a space graph, and
the second strategy boils down to replacing missions graphs with more sophisticated mechanics. Both strategies are explored in the following two sections,
respectively.
Feedback in the game mechanics that interact with level structures constitute an important and interesting category. As was mentioned in the general
discussion of the relation between feedback and emergence in section 1.5, feedback that traverses between different levels of organization within the system
contributes to stronger types of emergent behavior. Internal economy and level
design constitute such levels of organization. Games that are set up to so that
their internal economy and level design affect one and other during and as a
result of play, have a greater potential for emergent behavior.
6.6
Feedback Mechanisms for Locks and Keys
To create lock and key mechanisms that involve feedback, a good starting
point is treating the keys more as a resource that can be produced and consumed.
For example, figure 6.14 represents a mechanism where players need to “harvest”
ten keys before they can open the lock. Feedback is implemented through the
application of dynamic friction on the number of keys players have collected.
The more keys that are collected, the quicker the keys are drained. This makes
it somewhat harder to estimate how many keys need to be harvested to get
past the lock. Obviously, this gets even more difficult as the distance between
the location where keys can be harvested and the lock increases. However, the
mechanism is not very interesting in itself: it boils down to harvesting enough
keys and than make a run for the door, there is little strategy involved.
In an attempt to create a more interesting mechanism, we can apply the
dynamic engine pattern (see chapter 4 and appendix B) to a lock and key mechanism. Figure 6.15 represents such a mechanism. This time players need to
collect more than 25 keys in order to proceed, but now they have the option
to invest 7 keys to increase the harvest rate by 0.5. However, this mechanism
is probably too simple, too. It is not very difficult to find out what number of
upgrades is ideal for for this scenario.9 But even that is not necessary, play9 As
it turns out, the ideal number is actually one or two upgrades, both arrive at 26 keys
Figure 6.14: Simple feedback applied to a lock and key mechanism. In this figure: e
= entrance, g = goal, k = key.
Chapter 6 | Integrating Progression and Emergence
153
Figure 6.15: The dynamic engine pattern applied to a lock and key mechanism. In
this figure: e = entrance, g = goal, k = key, u = upgrade.
ers can achieve the goal without needing to upgrade at all. These weaknesses
should not come as a surprise: as I argued in chapter 4 one feedback loop is
generally not enough to create an interesting dynamic mechanism. The particular strategies are the direct result of the use of the dynamic engine pattern.
Games that do mostly rely on just a dynamic engine as their sole, or single-most
important, feedback loop, such as Monopoly, usually include random factors
to make it more interesting and unpredictable, but that is not the direction I
want to explore here.
To create a more interesting lock and key mechanism, we can complement the
dynamic engine pattern by some form of dynamic friction (see figure 6.16). In
this case enemies spawn that will consume the harvested keys. Now players have
to balance between three tasks: harvesting, upgrading and fighting the enemies
to keep their numbers down. This is not a trivial task, playing the interactive
version of the Machinations diagram10 is already a fairly interesting challenge.
Simply harvesting will probably not bring the player very far, and although it
is possible to achieve the goal by switching between harvesting and fighting,
this requires players to maintain a delicate rhythm of switching between the two
for a long time; it is very hard to accomplish. Players need to find a balance
between the three actions in order to reach the goal. When the fighting is made
skill-based, then the most effective balance can actually vary between individual
players. In its essence the mechanism is very similar to the basic gameplay
mechanism of the real-time strategy game I have discussed earlier: players need
to balance between harvesting raw materials, fighting, and upgrading.
at exactly the same time, while taking no upgrades or more upgrades turns out to be slower.
10 Also available at http://www.jorisdormans.nl/machinations/wiki/index.php?title=
Mission_Mechanics.
154
Joris Dormans | Engineering Emergence
Figure 6.16: The dynamic engine pattern and dynamic friction applied to a lock and
key mechanism. In this figure: e = entrance, f = fight enemies, g = goal, k = key, u =
upgrade.
6.7
Progress as a Resource
In many games that integrate emergent gameplay with progressive level design, the goal is to reach a certain location. This goal can be represented abstractly with a very simple diagram (see figure 6.17). In its essence, the ‘core
mechanics’ of The Game of Goose are similar. The main elaborations this
game implements are the use of dice to determine how much progress the players
are making each turn and the chance that a player might gain extra progress,
lose turns or lose all progress. More advanced games elaborate more: the most
common strategy for action-adventure type games is to make the production of
progress non-trivial and interesting in itself. To a certain extent, the experiments
with lock and key mechanisms that involve feedback, discussed in the previous
section, fall into the same category. This section seeks to go one step further,
it explores the possibility to involve progress itself in a mechanism to make the
gameplay more dynamic. Put differently, this section explores how feedback
mechanics might replace typical missions found in most games of progression.
An interesting and fairly abstract implementation of a progress mechanism
can be found in the latest edition of Warhammer Fantasy Roleplay. The
rules of this tabletop role-playing game include the concept of a ‘progress tracker’
as a generic tool to manage the players’ progress towards a single goal, competition for conflicting goals by multiple parties, or even the players’ party’s internal
tension and friction. The progress tracker takes the form of a track that can be
built from individual track pieces (see figure 6.18). This allows the ‘game master’ to build tracks with lengths that suit the current situation.11 Markers on
11 In
a tabletop role-playing game one player takes the role of the game master (also known as
Chapter 6 | Integrating Progression and Emergence
Figure 6.17: A simple progression mechanism
155
Figure 6.18: Pieces that are use to build a
progress track in Warhammer Fantasy Roleplay
the track indicate the progress of individual parties. The rules suggest a number
of ways a progress tracker can be used to facilitate scenes that involve races,
chases, investigations. The tracker can also be used to represent a time limit
by forcing the players to complete a certain task before a marker on a progress
track reaches the end, or to create tension by using it to track the build-up of
some “looming danger” unknown to the players.
Crucially, progress tracks in Warhammer Fantasy Roleplay do not only
track progress towards some goal (or danger), they might affect gameplay as
well. For example, in the scenario that is published with the rules, the progress
track to represent the players’ investigation into some secret cult includes a
special position. Once the players’ marker reaches this position, the game master
should provide the players with an extra hint in order to speed up their progress.
This occurrence creates a one-off, positive feedback loop. Similar events on the
party tension meter, a progress track that is part of the core rules, can cause
the players’ characters to suffer additional stress, fatigue or wounds, causing
destructive feedback.
As Warhammer Fantasy Roleplay illustrates, progress mechanics can be
used to cause feedback, and this is an excellent way to involve progress in the
dynamic behavior of the game. However, more sophisticated forms of feedback
can be used to evolve this further. A suitable pattern to accomplish this is
the escalating complications pattern (see figure 6.19). With this pattern, which
is found in simple, emergent games like Pac-Man or Space Invaders, the
player’s goal is to complete a number of tasks. In Pac-Man this task is to
eat all available dots; in Space Invaders it is to destroy all alien invaders.
The task is getting progressively harder as the player is making progress. The
dots get harder to reach in Pac-Man while the alien invaders start to move
faster and faster as their numbers decrease. A slightly more complex variation
of the pattern, called escalating complexity can be found in Tetris. In this
variation some form of complexity is created and it is the player’s task to keep
this complexity under control. However as complexity increases this task gets
progressively more difficult, usually another mechanism ensures that complexity
is produced at an increasing rate (see figure 6.20). In Tetris the blocks cause
dungeon master or storyteller). Where normal players usually control one character, it is this
players responsibility to set the scenes and take care of all other characters in the game. The
game master and the players are not opponents; they collaborate in creating an entertaining
game experience.
156
Joris Dormans | Engineering Emergence
task
loose
>n
*
+-
n%
n
+
complexity
*
progress mechanism
Figure 6.19: Escalating complications: the player’s task gets more difficult as there are fewer targets.
Figure 6.20: Escalating complexity: the
player’s task gets more difficult as complexity increases.
the the complexity, and the game speeds up every time the player reaches a new
level.12
Applied to progression, it is possible to model the progress towards a certain
goal and have that progress affect the mechanics. In a sense, games of progression have always mimicked this effect by ordering a fixed sequence of challenges
roughly from easiest to the most difficult. Nonlinear missions with alternative
branches can be built using a similar principle in order to create dynamic levels with more replay-value, but as has been argued before, this strategy is not
very effective. Most of the time many more branches and challenges need to
be created than an average player is ever going to see. By creating a system
where story-like progression emerges directly from the game mechanics, endless
possibilities can be created efficiently. When the mechanics are set up to produce enough variety, this could lead to games where interactive experiences, and
perhaps stories gain a whole new dimension.
In order to illustrate how a game like this might work, I will discuss an independent Flash game I designed and developed myself.13 In this game, Seasons
(see figure 6.21), the player controls a few characters in a 2D platform game
constructed as a frame story. In the frame story itself, the player controls a girl
that lives in a land that lost all color and is covered in perpetual winter. She
travels to her grandfather who tells her stories of the past, when the world still
12 For a more detailed discussion of escalating complications and escalating complexity see
these patterns’ description in appendix B.
13 The reality is that I could not find a published game that could serve as an example. Games
where progression truly emerges from mechanics are quite rare, and the examples I know of
do not lean towards storytelling. Seasons has the advantage of being a game that follows
the tradition of games of progression, while it does treat progression a little different than
most games in this tradition. While developing Seasons I focused more on the story structure
of the game than I explored the emergent properties of the game’s ecosystem. Originally it
was an experiment with a frame story structure for games first and foremost. At the same
time, the contrast between the highly linear stories and the open sandbox-like frame story was
deliberate. As was the reflection of the player’s progress through the state of the environment
in the main quest.
Chapter 6 | Integrating Progression and Emergence
157
Figure 6.21: Restoring the environment in Seasons.
had colors and seasons. These stories are also playable levels during which the
player controls other characters. By listening to her grandfather’s stories the
girl learns how the land lost its color and why the cycle of the seasons stopped.
The player is encouraged to use that knowledge to slowly restore the land back
to life. The girl needs color magic to do so, utilizing color magic, she can restore
elements back to their original state. Certain elements, such as trees and living
creatures, in return produce new color magic for the player to use. She also needs
color magic, to perform double jumps, to fly, or to control the wind, amongst
other things, all of which are necessary actions to reach the places where she can
restore crucial elements or creatures and to ultimately locate the evil villain. In
a way, her progress is reflected in the current state of the environment, the more
color is restored the closer the girl is to reaching her goal; in order to reach the
villain, the girl needs to restore all colors to the land, and it is through this
restoration that the villain is defeated in the end.
When I originally designed Seasons I was aware of the powerful dynamic
engine that is created by its main mechanism: by restoring plants and creatures
the player can easily get more color magic. I countered this by adding static
friction: restoring plants only works for a limited amount of time. Figure 6.22
represents this mechanism in a Machinations diagram. Apart from the dynamic
engine, the most important sources of color magic in Seasons are reaching
certain locations which would unlock and supply the player with a new type of
color magic and the stories themselves, as the secret stars the player can find in
these stories supply the player with extra color magic in the frame stories.
In hindsight, it would have been more interesting to use a mechanism that
would have implemented the escalating complications pattern. For example, by
158
Joris Dormans | Engineering Emergence
Figure 6.22: The economy of color magic in Seasons.
having restored elements also feed the aggression of those elements and creatures
that have not been restored, requiring the player to spend more color magic in
order to avoid them while traveling the world or even to actively oppose them
(see figure 6.23). This would in effect aggravate the contrast between those
parts of the game world that have been restored and those that have been not
restored and cause a gradual increase in difficulty that is an emergent result of
the mechanics, and not of careful, progressive level design.
A game set up like this, still relies on triggers to advance the storyline. In
Seasons reaching certain locations in the frame story will unlock new stories
to be told by the girl’s grandfather and a few other characters in the game.
Completing these stories will sometimes unlock a new type of color magic or
direct the player to those locations in the game’s frame story where they can be
unlocked. This set-up works, but much progression is still designed in advance.
There are only a few alternative routes the player can take within this structure.
This could also be changed. In Seasons there is a rudimentary ecosystem that
produces color magic. This economy could be elaborated so that instead of only
one or two restored elements providing the girl with all color magic she needs,
the player needs to structure many more elements. By relocating creatures and
activating plants the player could build and tune an economy not unlike the citybuilding in Caesar III (see above). By making a few changes the ecosystem
in Seasons could be rewired by the player to produce the color magic that is
needed for the present task. Reaching certain objectives in this economy could
still trigger story events, but the control of these events is far less linear, as it
would be fairly easy to design the system in such a way that the player has
many options in respect to what objectives to go for first. As with Caesar
III or SimCity the game would be able to accommodate a multitude of player
strategies, and the game’s world would reflect a player’s preferences. It would
require a multi-part ecosystem that could function in roughly the same way as
the basic economy of Caesar III (see figure 6.6).
Chapter 6 | Integrating Progression and Emergence
159
Figure 6.23: A redesign of color magic economy in Seasons that implements escalation.
6.8
Conclusions
In games, mechanics and levels cannot be separated. Games use level structures to teach the mechanics to players, and use special mechanics to control
players’ progress through a level. In strategy and management simulation games,
players typically have much control over the game economy that is built from the
individual mechanics. However, games that find a good balance between the two
are rare. One would expect missions to be the place where levels and mechanics
converge. But as games have evolved over the past few decades, and ways have
been found to create detailed spaces and articulate mechanics, the evolution of
missions and mechanics to control progression has mostly stood still. As a result
it is rare to find a game that truly integrates emergent gameplay and progressive
level design.
In this chapter I have explored the relation between mechanics and levels in
order to find a better balance between these two elements of game design. Leveraging the formalisms developed in the previous two chapters, I have explored
how mechanics that control level progress could benefit from implementing feedback loops, but also how progress itself can be integrated better in the general
economy of a game. These suggestions are preliminary; apart from some promising prototypes they have not been implemented and have not been thoroughly
tested. My intention in this chapter was to illustrate how games might integrate
these two important elements of game design. The possibilities for the convergence of progression and emergence in games are far greater than the illustrations
in this chapter.
The main point, however, is that emergence and progression are not two
separate dimensions of game design. Using the right tools designers can shape
160
Joris Dormans | Engineering Emergence
emergent mechanics to produce progressive experiences, and by having a clear
perspective on a game’s internal economy and mechanics designers can structure
levels that go beyond a structured learning curve. The internal economy of a
game and its level designs constitute different levels of organization. General
theory on emergence suggests that feedback that traverses these different levels
creates a greater potential dynamic behavior than feedback that operates only
within one of these levels. In this way mechanics that drive emergence and
progression can be combined to create compelling game experiences that offer
great freedom to the player.
Procedural content generation algorithms are
hard. Not hard as in you need to get a smarter
programmer. Hard as in the travelling salesman ate my NP-complete halting state hard.
Andrew Doull (2008)
7
Generating Games
The last three chapters presented frameworks to deal with game mechanics, and
game levels, and discussed their interrelation. Both frameworks are different
perspectives on the same object: games. Each perspective foregrounds different
aspects of games and their design. Where the Machinations framework foregrounds emergent behavior that stems directly from the structure of the rules,
the Mission/Space framework foregrounds progression. Both perspectives are
complementary ways of looking at games. When designing a game, designers move back and forth between these perspectives, and in all likelihood also
other perspectives such as perspectives that foreground theme, art, interaction,
market-value, or suitability for a target audience.1 These different views on the
same game do affect each other. As was already mentioned, what mechanics
operate in a game has an effect on how levels might be structured. When designers change the mechanics, this might also lead to changes in the levels and
vice versa. This relation suggests that there also might be a relation between
the formal models used in this dissertation to represent these perspectives; it
suggests formal relationships between Machinations graphs, mission graphs and
space graphs that go beyond the possibility of mixing graphs that we saw in the
previous chapter. This chapter explores this relationship and discusses how it
can be exploited in the creation of powerful, procedural game design tools.
In the first three sections I will discuss model transformation, formal grammars and rewrite systems. This will provide the theoretical background for a
formal approach to game design leveraging the multiple perspectives and models.
In section 7.4 I will discuss how formal grammars and rewrite systems can be
generalized to work with graphs in order to make them applicable to the models
that have been discussed in previous chapters. In addition, I will do the same
for shapes that are relevant to create actual game geometry in section 7.5. Next,
I will illustrate the use of model transformation, grammars and rewrite systems
by discussing one important transformation in the process of level design: that
of adding locks and keys that allows the designer to transform missions into
topological and topographical representations of spaces. In the sections that
follow, I will discuss how similar techniques can be used to generate game spaces
and game mechanics. In the final three sections of the chapter I will discuss the
1 Needless
to say these perspectives do not fall within the scope of this dissertation.
162
Joris Dormans | Engineering Emergence
application of these techniques to procedural content generation, which generally refers to the automatic creation of content using clever algorithms. I will
discuss three different applications of procedural content generation: the automatic generation of game content, adapting games to player performance and
creation of “mixed-initiative” automated design tools. To support this research
I have created a number of prototypes that implement the rewrite systems discussed in this chapter. On occasion, I will explain how the theories and ideas
have been implemented in these prototypes. These prototypes focused on level
generation for action-adventure games first, but their scope widened as the research progressed to include more game content and to support a wider variety
of games.
7.1
Game Design as Model Transformation
Model transformation is a notion taken from the practice of ‘model driven engineering’ or ‘model driven architecture’ within computer science. Model driven
engineering describes the process of creating software as a series of model transformations where, for example, a business model is transformed into a software
architecture, which in turn can be transformed into software code. Model driven
engineering is intended to deal with the complexities of designing enterprise-scale
software solutions. It depends strongly on a formalized conceptual framework,
expressed through different models, which can be used to design systems and
communicate about system architecture. It also plays an important role in automatic software generation. One of the main premises is that it relieves programmers from many tedious, manual tasks and elevates the task of programming to
a higher level of abstraction, where the most is made of their creativity and ingenuity. Through model driven engineering, the quality and efficiency of software
production are to be improved (Brown, 2004). Model driven engineering works
with many different models, some of which are specific for a certain domain,
while others are more generic. There is a strong push to use Unified Modeling
Language (UML) as a standard modeling language independent of platform and
implementation (Selic, 2003).
This is not the first time that a model driven approach is explored in the
context of game design and development. In a short paper Emanuel Montero Reyno and José Á Carsı́ Cubel (2008) explore the use of standard UML
techniques and tools for the rapid, and mostly automated, creation of game
prototypes. They conclude that the UML approach caters better to software
engineers than to game designers. In a later paper the same authors sketch a
platform-independent modeling language for gameplay specification (Reyno &
Carsı́ Cubel, 2009a). This modeling language still relies heavily on the use of
UML, although they do add game structure diagrams and rule set diagrams to
the palette of models offered by UML in order to deal with specifics of the domain of games. Using this language, they were able to generate code for game
prototypes quickly (Reyno & Carsı́ Cubel, 2009b). The main difference between
their approach and the approach taken here is that they aim to generate games
and prototypes of a particular kind (platform games). Their domain is quite
Chapter 7 | Generating Games
163
narrowly defined, and there is only little room for design. Their language formalizes a particular subset of games, but not the process of designing them.
For the approach presented here, I use model transformations to formalize the
design process in a more generic fashion. Automation is only a secondary goal.
As a result, the models I am using are not based on UML. They are models
specific for the domain of game design rather than for the domain of software
production.
The game design process includes many steps during which designers focus
on particular aspects of games. Models of some sort have always been an important outcome of these steps. Design documents capturing design vision, early
prototypes as a sketch for the intended gameplay, lists of tasks outlining a mission, or hand drawn maps detailing game spaces, all of these are examples of
models of (some aspects of) the final game. The steps and products are not
separate. Decisions made early on have consequences for later steps. One might
say that the labor of a game designer is to take a design vision and to transform
it into a working set of game mechanics. Later, these mechanics are transformed
into a mission designed to teach players the game. That mission could be transformed into a game space to accommodate it, etcetera. Framed as a series of
model transformations, and using clearly defined models that are specific to the
domain, this process can be formalized. As it happens, the Machinations framework and the Mission/Space framework provide us with suitable, clearly defined
models specific to the domain.
In practice there are many different ways of designing games. The iterations
and incrementations towards a finished product are not set in stone. Some
designers might start with a premise, design rules to go with it and then proceed
to levels and detailed stories. Others might start with a map that constrains the
design of game mechanics. Even within the process of designing a game, steps to
create particular parts of the game might differ. A tutorial level requires that the
game mechanics are clear and finished, but other level content might be dictated
by a storyline rather than game options. With applying model transformations
to game design, I acknowledge that the process is flexible and different types
of games call for different approaches and different types of transformations.
However, in order to discuss and illustrate game design as a series of model
transformations I have chosen to focus on a particular order of design steps and
transformations that in my experience is a sensible way of designing games. In
this case, I propose that mechanics are designed before levels and when creating
levels, missions are created before spaces. In this process a level designer first
creates a mission by generating a list of tasks the player must perform to finish
the level, next the designer transforms this mission into a space by arranging
these tasks onto a map of the level. The designer then adds detail to the map
until it is sufficiently detailed and populated to function as a game level (see
figure 7.1). Usually, the later models are more complex and more detailed.
One can assume that in this case the original mission is embedded within the
representation of space, but not the other way round.
In an alternative approach, not discussed in detail in this dissertation, a
designer might begin by designing a space first, and then design a mission that
Joris Dormans | Engineering Emergence
164
Figure 7.1: Level design as series of model transformations.
create
space
distill
mission
adjust
space
refine
level
Figure 7.2: An alternative series of transformations.
matches the space, and maybe make some adjustments in order to facilitate that
mission before adding detail (see figure 7.2). This approach is better suited to
generate levels where spaces conform to some rational or architectural principle
outside the mission; a level might be a mine first, furnished with all the elements
that one expects from such an environment and then transformed into a space
to accommodate interesting gameplay. This way a single space might also host
multiple missions, as is the case in System Shock 2 where the player traverses
the decks of a space ship, and returns to previously explored decks during later
stages in the game.
In order for model transformations to work, the models the transformations
operate on need to be defined clearly. Model transformation typically uses formal
grammars to describe the models. For this reason, I will discuss formal grammars
in the next section, and show how they can be used to describe graphs and maps.
The transformations themselves are described with rewrite systems, which use
a similar type of rewrite rules to describe how a model can be transformed into
another model. Rewrite systems are discussed in section 7.3.
7.2
Formal Grammars
Formal grammars originate in linguistics where they are used to describe
sets of linguistic phrases that constitute natural languages (Chomsky, 1972).
In formal language theory, a language is a, possibly infinite, set of strings. A
grammar is a finite characterization of a language. To be more exact, a formal
grammar for a language that consists of strings of letters has four elements:
1. A finite set of terminals that are elements of the language the grammar
is to produce. This set is called the alphabet. For example: {a, b} is a set
consisting of two terminals a and b. It is conventional to use lowercase
letters to represent terminals.
2. A finite set of nonterminals that are not elements of the strings of the language the grammar is to produce. The grammar’s rules will replace these
nonterminals with terminals. For example: {A, B, S} is a set consisting of
three nonterminals A, B and S. It is conventional to use uppercase letters
to represent nonterminals.
Chapter 7 | Generating Games
165
3. A symbol from the set of nonterminals that is the start symbol. It is
conventional to use the symbol S as the start symbol.
4. A finite set of rewrite rules that take the form of: left-hand → right-hand.
Both the left-hand and the right-hand of the rule are strings that consist of
terminals and/or nonterminals. Every rule specifies a subsequence of symbols (the left-hand) that can be replaced by another sequence of symbols
(the right-hand).
A formal grammar specifies the set of possible strings of terminals that can
be generated by a finite number of applications of the rewrite rules starting from
the start symbol. For this reason they are also known as generative grammars.
Below is an example grammar that describes a simple ‘mission language’.
Note that this example is overly simplistic; it is intended to illustrate the use of
formal grammars for strings, it does not make any claim about the structure of
missions or the formal grammars that might describe them.
• The alphabet consists of two symbols: {g, t}, where g stands for goal and
t for task.
• The set of nonterminals is {S, T }.
• S is the start symbol.
• The following rewrite rules apply:
S → Tg
T → tT
T →t
Following this grammar the strings {tg, ttg, tttg, ttttg, ...} are all elements
of a set containing all possible missions; they are all part of the language the
grammar generates. The grammar describes a mission ‘language’ where a level
must have a singular goal preceded by at least one, but possibly more occurrences
of tasks. Note that the grammar above is recursive: the second rule creates a
string on which the same rule is applicable again. The language generated by
this grammar thus contains all strings consisting of a g preceded by at least one,
and possibly more t’s.
The Chomsky hierarchy classifies formal grammars based on the form their
rewrite rules (Chomsky, 1959):
• Type 0, or unrestricted grammars, have rewrite rules of the form α →
β where α and β may be arbitrary strings containing terminals and/or
nonterminals.
• Type 1, or context-sensitive grammars, have rewrite rules of the form
αBγ → αβγ where B is a nonterminal and α, β and γ may be arbitrary
strings containing terminals and/or nonterminals. In addition, α and β
may be empty strings.
Joris Dormans | Engineering Emergence
166
• Type 2, or context-free grammars, have rewrite rules of the form A → α
where A is a nonterminal and α may be an arbitrary string containing
terminals and/or nonterminals.
• Type 3, or regular grammars, have rewrite rules of the form of A → a or
A → aB where A and B are nonterminals and a is a terminal.
The mission grammar in the example above is a context-free grammar, because all the rewrite rules are of the form that one nonterminal is replaced by
a string of terminals and nonterminals. We might specify another mission language, using an unrestricted grammar as follows:
• The alphabet consists of three symbols: {g, r, t}, where g stands for goal,
r for reward and t for task.
• The set of nonterminals is {S}.
• S is the start symbol.
• The following rules apply:
S → tg
tg → ttg
tt → trt
This grammar generates a language that includes all missions that end with
a goal preceded by at least one occurrence of a task, and where rewards can be
included if the reward is preceded and followed by a task: {tg, ttg, trtg, ttrtrtg,
...}. For generating missions, using an unrestricted grammar has advantages:
in this case, generated missions might still be expanded allowing designers, or
games, to adjust a mission when needed. When a grammar is context-sensitive,
context-free, or regular, a new mission can only be generated from scratch.
7.3
Rewrite Systems
The process through which one model is transformed into another model can
be captured using rewrite systems. Rewrite systems share many similarities with
formal grammars. Most importantly, rewrite systems also make use of rewrite
rules. However, where formal grammars describe and define languages, rewrite
systems define particular classes of transformations. These transformations can
be used to translate an expression from one language to its equivalent in another,
or to elaborate and refine expressions within one language. For example, rewrite
systems can be used to transform a mission into a space, or to add detail to
existing missions.
In its most generic form, a rewrite system for strings, sometimes called an
abstract reduction system or abstract rewrite system, consists of two elements
(Klop, 1992):
Chapter 7 | Generating Games
167
1. A finite set of elemental symbols A.
2. A finite set of rewrite rules that take the form of: left-hand → righthand where both the left-hand and the right-hand are strings consisting of
symbols from A.
The difference between a formal grammar and a rewrite system is that a
rewrite system is not used to generate a language. It has no starting symbol and
does not distinguish between terminals and nonterminals. Instead it can operate
on any string of symbols; this means that it can operate on an input string
specified by a formal grammar. This also means that a transformation described
by a rewrite system does not terminate in the same way as the generation process
described by a formal grammar does. Where generation terminates when the
string it is generating only contains terminals, any application of a rewrite rule
in a rewrite system results in a string of symbols that is meaningful and might
be further transformed. In contrast, the transformation described by a reqrite
system terminates when the string is in normal form. A string of symbols is in
normal form if there is no rewrite rule that allows the string to be rewritten to
another string. In addition, a string has a normal form when for that string there
exists a sequence of rewrite operations that generates a normal form. A rewrite
system is called weakly normalizing when all symbols have a normal form, and it
is called strongly normalizing, or terminating, when all symbols have a normal
form and when infinite sequences of rewrite rules are impossible (Klop, 1992,
5-6).
To illustrate the use of rewrite systems using an overly simplistic example,
consider the following rewrite system:
• The set of symbols is: {f, g, r, t}, where f stands for fight, g stands for goal,
r for reward and t for task.
• The following rules apply:
tg → trg
tt → trt
t→f
In this case f would be the normal form of t. The set tg has two normal forms:
f g and f rg. Likewise, tt also has two normal forms: f f and f rf . The symbols
f , g and r are normal forms. As there are no sequences of rules that could
go on indefinitely, this rewrite system is terminal. When this rewrite system is
applied to the mission tttg, the application of a randomly selected rewrite rule
generates one of the following results: {tttrg, trttg, ttrtg, f ttg, tf tg, ttf g}. The
set of missions that are generated after the transformation terminates is: {f f f g,
f rf f g, f f rf g, f f f rg, f rf rf g, f rf f rg, f f rf rg, f rf rf rf }.
When all strings and symbols have a single normal form, the rewrite system
is called confluent, which this rewrite system is not. For a rewrite system that is
terminal and confluent every starting set of symbols has a unique terminal set.
For certain transformations this is a useful property, for others it is not. In the
Joris Dormans | Engineering Emergence
168
case of generating games, many transformations are not confluent, and might not
even be terminal. As was argued before, a single mission might map to several
game spaces and several game spaces might accommodate several missions, this
means that the rewrite systems that allow us to transform missions into space,
or vice versa, must not be confluent. On the other hand, a transformation
describing how a mission can be translated to an isomorphic space must be
confluent.
A rewrite system does not define a language or a model; it cannot be used
to describe or analyze a game or a level. It can, however, codify design principles: a rewrite system specifies the operations a designer might perform on a
model in order to transform one model to the next. When implemented as an
automatic transformation, these rewrite systems are very strict; they allow only
the operations that are represented by their rules and nothing more. Real-life
designers are more flexible, yet they also obey certain restrictions. If the aim is
to create a level that is solvable, no designer would place a crucial key behind a
lock opened by that same key, as this would create a deadlock.
The advantage of using a rewrite system is that transformations are defined
consistently. If the definition is correct, there is no room for mistakes.2 Obviously, this depends on the ‘correctness’ of the rewrite system. It requires
considerable effort to design rules that never generate inconsistencies, and to
verify that this is indeed the case. But it is possible, and as the set of rules is
small in comparison to the set transformations they define it often is worth the
trouble. Rewrite systems and the transformations they define are recurrent between projects and games. I imagine that a game company or design community
develops and refines sets of transformations over the course of many projects.
These sets would then represent the accumulated design lore of that company
or design community.
As an additional advantage, rewrite systems allow the design process to be
(partly) automated. Automated transformations have two more advantages: 1)
they can produce different games or levels quickly, enhancing the output of the
level designer. And, 2) this output can be used to validate the correctness of the
rewrite systems and the underlying principles by quickly generating games and
levels and by verifying them either manually or procedurally.
7.4
Graph Grammars
The models used to describe games in this dissertation are graphs, not strings.
In order to to use formal grammars and rewrite systems to generate and transform these models, formal grammars and rewrite systems need to be generalized
to work with graphs as well. Graph grammars that generate graphs consisting
of edges and nodes have been described by Rekers and Schürr (1995). Rewrite
systems designed to work with graphs grammars have been discussed by Reiko
Heckel (2006). In a graph grammar a structure containing one or several nodes
and interconnecting edges can be replaced by a new structure of nodes and edges.
Figures 7.3 and 7.4 illustrate this process. After a group of nodes has been se2 Although
what might be considered to be correct will vary between games.
Chapter 7 | Generating Games
169
2:B
2:A
4:B
1:A
1:a
3:b
Figure 7.3: A graph grammar rule. Square nodes denote nonterminals and circular
nodes denote terminals.
Figure 7.4: The process of applying a rule to a graph.
lected for replacement as described by a particular rule, the selected nodes are
numbered according to the left-hand side of the rule (step 2 in figure 7.4). Next,
all edges between the selected nodes are removed (step 3). The numbered nodes
are then replaced by their equivalents (nodes with the same number) on the
right-hand side of the rule (step 4). Then any nodes on the right-hand side that
do not have an equivalent on the left-hand side are added to the graph (step 5).
Finally, the edges connecting the new nodes are put into the graph as specified
by the right-hand side of the rule (step 6) and the numbers are removed (step
7).
Nodes might be deleted as a result of the application of a graph rewrite
rule. However, in this case, it is important that the nodes to be deleted are not
connected to any other edges other than the ones specified in the left-hand side
of the rule.
The graph grammars and rewrite systems used in this dissertation have a
form that is analogous to the unrestricted grammars in the Chomsky hierarchy:
they have rewrite rules of the form α → β where α and β may be arbitrary
graphs containing terminals and/or nonterminals nodes. This form offers the
most flexibility for generating and transforming models representing games in
various stages of design.
In order to generate game levels, graph rewrite systems can be used to transform graphs representing missions into graphs representing game spaces. Figure 7.5 depicts mission and space graphs and grammars in relation with each
other and a rewrite system. A transformation from mission to space should ter-
Joris Dormans | Engineering Emergence
170
Figure 7.5: Level design as a model transformation.
minate after all mission elements have been replaced with a space element. In
other words, in the rewrite system describing this transformation all elements
representing mission nodes and edges should all have a normal form that is part
of the space graph language. After this transformation the space graph can
be further elaborated and refined using different rewrite systems defining the
relevant transformations.
7.5
Shape Grammars
Graph grammars can be used to generate Machinations diagrams, mission
graphs, and topological models of space. In order to generate game geometry
another form of formal grammars, called shape grammars, is required. Shape
grammars have been around since the early 1970s when they were first described
by George Stiny and James Gips (1972). In shape grammars, shapes are replaced
by new shapes following rewrite rules similar to those of formal grammars. Special markers are used to identify starting elements and to help orientate (and
sometimes scale) the new shapes.
The implementation of shape grammars in the software prototypes to support this research works with three geometric primitives: points, line-segments
and quadrilaterals (quads). In the implementation all shape grammar rules are
context-free: the left-hand of any shape grammar consists of one single element,
while the right-hand can consist of multiple elements.3 Rewriting works differently for each geometric primitive: the operations for rules that have a point,
line-segment or quad as the left-hand element are explained below.
Points only have a location and an orientation. They are represented as
3 Creating the shape grammar equivalent of context-sensitive or unrestricted grammars
would require a way of detecting the context of a shape in order to identify sets of shapes
to be replaced. Where graphs have edges to specify these relations explicitly, shapes lack such
an element. For the purpose of generating game geometry context-free shape grammars are
sufficient.
Chapter 7 | Generating Games
171
Figure 7.6: Points and point rules in a shape grammar.
A
a
A
a
a
a
a
A
A
a
Figure 7.7: A series of transformations based on iteration of rule 2 in figure 7.6.
circles with a triangle indicating their orientation. Colors and letters inside
the circle identify the type of point. Capitals indicates nonterminal points,
while lowercase indicates terminal points. Shape grammar rules that have a
nonterminal point as their left-hand must have at least one point in their right
hand, but the right-hand can also consist of multiple points. One point on the
right-hand is the starting point. The starting point has a black triangle and
replaces the left-hand point and matches its orientation (see figure 7.6). If there
are multiple points in the right-hand the extra points are placed relative to the
starting point’s location and orientation (see figure 7.7).
Line-segments have a location, orientation and a particular length. They
are represented as a black line with a triangle indicating their orientation. Colors and letters next to line identify the type. Capitals indicate nonterminal
line-segments, while lowercase indicates terminal line-segments. For terminal
line-segments the letters and triangles can be omitted (see figure 7.8). Shape
grammar rules that have a nonterminal line-segment as their left-hand element
must have at least one starting line-segment in their right-hand. The starting
line-segment is identified with a black triangle indicating its orientation. In addition, points, other line-segments, and quads can be added to the right-hand.
The left-hand line-segment is of a set unit length. To replace a line-segment
of arbitrary size and orientation, all right-hand elements are rotated and scaled
so that the starting line-segment matches the original line-segment in size and
orientation. If the right-hand starting line-segment is gray and dashed it is not
placed but only used as a reference to determine the scaling and rotation; in
effect, the original line-segment is simply removed (see figure 7.9).
Quads have a location, orientation, and shape. They are represented as a
quadrilateral shape with a small square indicating their orientation by marking
one side. Colors and letters identify the type. Capitals indicate nonterminal
quads, while lowercase indicates terminal quads. For terminal quads the letters
Figure 7.8: Points and point rules in a shape grammar.
Joris Dormans | Engineering Emergence
172
A
A
A
A
Figure 7.9: A series of transformations based on iteration of rule 2 in figure 7.8.
Figure 7.10: A quad based transformation rule.
and squares can be omitted. Quads represent areas in a game, their sides are not
barriers. Barriers must be represented with explicitly as line-segments. Shape
grammar rules that have a nonterminal quad as their left-hand can have any
set of elements as their right-hand. The left-hand nonterminal quad is always
shaped as a unit-sized square. In order to replace a quad a complex matrix
transformation is used to map the unit-sized square to the shape of the original
quad (see Arvo & Novins, 2007). Figure 7.10 gives an example of a rule for quad
based replacement. Figure 7.11 depicts a series of transformations based on this
rule starting with a different shaped quad.
Combined, these rules can be used to generate quite sophisticated game geometry. Figure 7.12 is an example of a relatively simple grammar. A possible
structure that is the result of random application of rules from this grammar is
found in figure 7.13. In this case, recursion in the rules creates self-similarity
in the structure. In order to prevent such recursion from generating infinitely
detailed structures, constraints can be placed on the transformations and replacements. For example, constraints specify how far the right hand is allowed
to scale up and/or down in order to match the line-segment or quad to be replaced. Disallowing new line-segments and quads to overlap existing elements in
the structure, or be placed outside certain bounds, are other useful constraints
to prevent illogical and unwanted results.
There are more possible implementations of shape grammars than the suggestions above. For certain games it will make more sense to use triangles instead
of quads as the primitive shape for two-dimensional objects. For other games
it will be useful to support square tiles, or to extend these primitives into the
third dimension. The choice for quads and the use of only two dimensions in the
prototypes for this research was made for convenience in implementation and
ease of designing relevant grammars.
Figure 7.11: A series of transformations based on iteration of the rule in figure 7.10.
Chapter 7 | Generating Games
173
Figure 7.12: Rules for a shape grammar.
Figure 7.13: An example transformation based on the random application of the
rules in figure 7.12.
7.6
Example Transformation: Locks and Keys
Rewrite systems can be used to describe useful, recurrent transformations in
game design. This section describes how rewrite systems can be used to add locks
and keys to a mission graph and use these locks and keys to transform the mission
graph into a non-isomorphic space graph. As was mentioned in section 5.6,
this transformation consititues one of the most important design principles in
action-adventure games such as The Legend of Zelda. Locks and keys control
players’ movement through the level and foreground their progression. Although
the locks and keys in The Legend of Zelda have many different guises, their
basic functionality remains the same. Model transformations from mission to
space can be leveraged to explain this design principle in more detail.
Essentially, what locks and keys allow a designer to do is to take a linear series
of tasks, which by itself would make for an equally linear level, and transform
it into a branching structure (see figure 7.14) which lends itself much better for
the creation of nonlinear spaces. This transformation can be captured with only
two mission graph rewrite rules (see rules 1 and 2 in figure 7.15).
There are plenty of rules that could be added to this basic set in order to
generate more interesting levels. For example, a rule can be designed that moves
a lock backwards, towards the goal (see rule 3 in figure 7.15). However, this rule
breaks with the level design wisdom that is generally better to have the player
encounter the lock before the key. Another rule can be designed that allows
tasks that are placed after a lock to be placed in front of a key associated with
Joris Dormans | Engineering Emergence
174
that lock (rule 4). This will in effect hide the key, making sure that the player
needs to accomplish more tasks before finding it. Other options include using
multiple keys on a single lock (rule 5) or creating keys that are used multiple
times (see rule 6).
The technique of using rewrite systems is highly controllable. If you consider
a lock and key combination to be a single task, then none of these rules changes
the number of tasks in the level. This way the size of a level is dictated by the
length of the initial mission. In addition, these rules also make sure that a lock
will always be followed by another element. This can be verified by inspecting
the rules: there is no rule that allows the removal of the last node after a lock,
and all additional branches that are created end with a key node that is required
to proceed elsewhere. This means that all tasks must be completed in order to
finish the level. This restriction explains why in rule 6 the second lock uses a
different symbol than the first lock. It means that the second lock cannot be
moved by rule 2. Should this lock be allowed to be moved by applying rule 2,
the first lock ends up leading to a string of tasks not ending in a goal or a key,
or even no tasks at all. This situation is undesired as it might cause the number
of tasks the player must perform to complete the mission to be reduced (see
figure 7.16).
Figure 7.17 shows a few examples of level structures that were generated with
rewrite rules depicted in figure 7.15.
7.7
Generating Space
Once a mission structure is generated that consists of multiple tasks with
locks and keys, there are several strategies to build spaces to accommodate the
mission. In an earlier paper, I described a method that uses shape grammars to
define spatial parts which are used to build up a space not unlike a jig-saw puzzle
a
K
L
b
K
c
L
Figure 7.14: Addition of a lock and key transforms a linear mission (a) into a branching structure (b) in which the lock can be moved forward (c).
Chapter 7 | Generating Games
175
4:K
2:?
1
2:?
1:K
2:?
1:L
3:?
3:L
4
Rule 1: add lock
3:?
4
1:K
1:L
1:L
2:?
3:?
4:K
1:K
2:?
4:?
3:?
3:L
5:?
3
1
2:L
2:L
3:K
2:L
4:?
1:K
Rule 5: duplicate key
3:K
2
5:?
Rule 4: pull task to hide key
Rule 2: move lock towards entrance
1:L
2:?
3:?
5:K
3:?
1:L
2:?
Rule 3: move lock towards goal
5:K
3
4:?
1:L
2:?
3:Ld
4:?
Rule 6: duplicate lock
Legend:
?
any
K
key
L
lock
Ld
lock duplicated
task
strong requirement
weak requirement
Figure 7.15: Rewrite rules governing the transformations enabled by the use of locks
and keys. In these rules, the nodes marked with a question mark can be any node: the
question mark acts as a wild card.
K
K
L
L
L
L
Figure 7.16: Undesired transformation that is the result of applying rule 2 from
figure 7.15 to a duplicated lock that is not marked as such.
Joris Dormans | Engineering Emergence
176
K
K
K
K
Ld
K
Ld
L
L
Ld
L
K
K
L
K
K
L
K
K
K
K
Ld
L
K
L
L
L
K
K
K
L
K
K
L
K
Figure 7.17: A sample of the levels generated by randomly applying the rules 1, 2,
4, 5 and 6 of figure 7.15 on a mission of twenty-one tasks.
(Dormans, 2010).4 Although this approach works, it has difficulty generating
spaces for missions which allow multiple paths to converge at the same goal. To
deal with this problem I take advantage of the spatial nature of a two-dimensional
representation of a graph, which can be translated into a shape easily. This
approach is also outlined in (Bakkes & Dormans, 2010). The research prototype
can generate an organic layout for mission graphs by simulating all nodes in
the network as nodes with connections functioning as springs with some basic
algorithms to reduce the number of overlapping connections. This algorithm
was used to generate the graphs in figure 7.17.
After this, a fairly simple and generic rewrite system is used to replace tasks
with places of various sizes, to place keys inside them and, to create locks to
connect the places (see figure 7.18).5 This rewrite system is both confluent and
terminal: it will always generate the same structure for the same input (although
in this case the room sizes are set randomly). The result of this transformation
is a space graph similar to the one in figure 7.19. From this a spatial structure
can be generated that follows the same outline but consists of quads, and line
segments (see figure 7.20). This step does not use a rewrite system, as the
other steps do. This is because it would require a rewrite system that can mix
graph grammars with shape grammars. There is no off-the-shelf solution for
rewrite systems to deal with such a mix of two completely different types of
grammars. This is complicated further by the fact that all elements of a space
4 Incidentally, this process is very similar to the “dynamic” level layout of Schell (2008) in
figure 5.2 in the previous chapter.
5 This grammar assumes that tasks that have a direct game element equivalent, such as
keys, switches and enemies, are replaced before this grammar is run. For example a task “key”
is automatically replaced if there also exists a game element “key”.
Chapter 7 | Generating Games
177
Mission:
3:?
2
Space:
followed by
path
e
entrance
required by
unlocks
g
goal
L
lock
?
any
lock
Ld
lock duplicated
?
any game element
lock (not movable)
task
?
any lock
place
1:?
2:?
1:?
create unlocks
1:g
2
3:?
1
2:?
add goal after game element
2:?
2
2:?
1:?
3
place game element
1:g
2
1
1:?
2:?
3
add goal after place
2:?
3
1:?
1:?
protect game element
2:?
1:Ld
2:?
create duplicated lock
1
2:?
1
2:?
1
replace task 1
2:e
1
1
2:?
1
2:?
1
create entrance
replace task 2
3:?
2
3:?
2
1:L
1:?
2
3:?
1:?
2
3:?
1
create lock
resolve path from game element
Figure 7.18: A generic rewrite system to transform a mission graph to a space graph.
graph probably need to be transformed into a game geometry simultaneously.
Instead, it uses an algorithm designed especially for this step, which is tailored
to generate a two dimensional dungeon map. Other implementations might be
designed to generate different types of spaces: such as three dimensional terrains,
or cities.
The next step involves using a shape rewrite system to flesh out this basic
shape and generate more detail. Rules like those in figure 7.12 have been used
to add detail to the spatial construction to transform figure 7.20 into figure 7.21.
The shape rewrite rules used in this transformation result in a natural looking
cave. This need not be the case. With different rules the transformation yields
rooms that look much more like artificial constructions (see figure 7.22). In this
case the effect would have benefited from aligning the mission structure to a
grid before translating it into a spatial construction. After this step, additional
178
Joris Dormans | Engineering Emergence
Figure 7.19: Tasks replaced with rooms of various sizes
Figure 7.20: The mission structure from figure 7.19 translated into a spatial construction.
Chapter 7 | Generating Games
179
Figure 7.21: Space transformed with shape rewrite system to produce organic dungeon walls.
Figure 7.22: Space transformed with shape rewrite system to produce straight dungeon walls.
Joris Dormans | Engineering Emergence
180
1:S
1
4
collect
1
1:S
1:S
destroy
7:>n
3
4
3:win
5
6
prevent
5:>n
2
1
1
1:S
4:win
2
3:win
5
6
survive
5:==0
2
7:==0
3
4:win
2
Figure 7.23: Transformation rules to create a goal from an arbitrary starting point
(the nonterminal “S”).
transformations can be used to populate the level with treasure, creatures, traps
and decorations.6
7.8
Generating Mechanics
Just as it is possible to generate missions and spaces using rewrite systems, it
is possible to design rewrite systems that generate mechanics from an arbitrary
starting point, add mechanics to existing missions and spaces, or generate missions and spaces that match specific game mechanics. Machinations diagrams
are an ideal point of departure for such an endeavor as these diagrams are graphs,
that can be embedded within missions and space graphs, and can be subjected
to the same type of rewrite rules as missions and topographic representations of
space.
Rewrite rules can be used to codify recurrent constructions found in games.
These constructions include typical game goals, not unlike those described by
Björk & Holopainen (2005), Nelson & Mateas (2007), and Djaouti et al. (2008).
Figure 7.23 features a number of rewrite rules that might be constructed to
include a number of these goals in games. It is not difficult to see that from
these starting constructions the mechanics can be expanded by replacing simple
mechanics with more sophisticated ones. Examples of rules that describe such
transformations can be found in figure 7.24.
Notice that graph rewrite rules for Machinations diagrams require that both
nodes and edges have a unique number identifying them for transformation (see
section 7.4). Edges in Machinations diagrams behave as nodes in certain respects: they can have textual modifiers and other edges might connect to an
edge instead of a node. This means that in many cases simply removing all
edges between nodes prior to transformation and adding them after a transformation is not going to work.
It is also possible to include rules that in fact make the mechanics simpler.
6 In the end, the implementation of shape grammars to generate spaces will turn out to be
quite specific to a particular game. It might even be that other types of grammars are used.
For example, two-dimensional or three-dimensional tile based grammars will be very applicable
to particular games.
Chapter 7 | Generating Games
181
Figure 7.24: Rules to transform mechanics.
As was argued in chapter 2, this can improve the game. Simplifying a game
is a stage that almost every design goes through (or should go through), as
many elements are usually added during the initial brainstorm. The trick of
simplifying, following the discussion in chapter 4, is to cut elements that do not
contribute to the structure of the feedback loops present in the mechanics. In
this way interesting emergent behavior the game might display is not destroyed
during this stage. Rewrite systems are a good tool to guide such a process: for
example, they can be used to identify overly complex mechanisms and replace
them with simpler ones with equivalent behavior.
Rewrite systems allow mechanics to be generated in close relation with levels
and vice versa. For example, it is possible to transform lock and key mechanisms
in a space diagram using Machinations and then to use them to elaborate on the
mechanics (see figure 7.25). In a similar vein, more elaborate mechanics to deal
with other aspects of the game can also be added to mission graphs or space
graphs.
As with the transformations used to describe the process of designing a
level, there are many different sequences of transformations possible. The most
straightforward point of departure is a space graph, and refine it by adding
mechanics using rules as illustrated in figure 7.25. However, it might be more
interesting to start with mechanics and transform them into a mission that utilizes these mechanics, after which they can be transformed into a space as shown
before. In order to transform mechanics into a mission a good starting point
would be to associate tasks with all the elements in the Machinations diagram
that are interactive (see figure 7.26). Next, these tasks need to be connected in
some logical order. This can be done by ‘mirroring’ resource connections with
followed by or required by connections in the mission graph, as is suggested by
the rewrite rules in figure 7.26. A number of typical Machinations constructions
182
Joris Dormans | Engineering Emergence
Figure 7.25: Rewrite rules to add and elaborate mechanics for locks and keys in space
graphs.
Chapter 7 | Generating Games
183
1
1
2
3:*
1
create task for converter
3:>n
1
2:win
4:>n
5:L
6
create goal
2
3:*
1
create task for drain
1
2:n
2
1
1:resources==0
3:?
4:*
3:*
1
create task for source
2:n
3:?
4:*
7:>n
1
5
6:L
5
create proto lock
5:?
4
3:?
6
2
7:?
1
5:?
4
3:?
5:?
2
6
7:?
1
3:?
5:?
6
2
6
2
7:?
1:L
7:?
1:L
4
connect task 1
5:?
4
7:?
5:?
2
6
1
7:?
connect task 2
3:?
connect lock 1
3:?
6
4
4
3:?
2
1
5:?
6
7:?
4
3:?
2
1:L
5:?
6
7:?
4
3:?
2
1:L
connect lock 2
Figure 7.26: Rewrite rules to generate mission structures (black) from Machinations
diagrams (blue).
lend themselves for the creation of locks. Missions generated in this way tend
to branch pretty wide and might have several starting points. This is only to
be expected as missions generated in this way will be more open (or nonlinear)
than missions typically found in games of progression.
Figure 7.27 shows the result of a rewrite system as suggested in figure 7.26
applied to a Machinations diagram. The mission generated in this example is
relatively simple. However, after the initial generation, missions can be elaborated further using more traditional lock and key mechanisms in order to create
levels that mix elements of emergence and progression.
Structuring the learning curve is still important in levels generated in this
way. One solution to structure the learning curve can be found in the generation
process of the mechanics itself. Assuming this process started out with fairly
simple mechanics, like the simple goals presented in figure 7.23, or perhaps with
one or two elaborations, a first level, or the first challenges of a level, could
Joris Dormans | Engineering Emergence
184
*
*
*
*
n
*
n
+
*
n
n
+
n
*
+
+
n
*
+
n
*
>=n
>n
*
*
win
*
+
n
L
>=n
>n
L
L
>=n
L
Figure 7.27: Transformation of a mechanics (blue) to mission (black). The mechanics
are taken from the Seasons example from the previous chapter (see figure 6.23).
Figure 7.28: Steps in the generation of game mechanics could inform the generation
of progressive levels.
be generated from these mechanics. Subsequent levels could be generated from
further transformations (see figure 7.28). The transformation history that led
up to the complete design of the mechanics could be used as a basis for such a
structure of level progression. This way levels might be created that are coherent
and where earlier challenges prepare the player for the challenges that are still
to come. At later stages of the game, parts of the mechanics might be removed
in order to be replaced with new mechanics in order to create variation in the
gameplay.
In a way, these transformations describe how games might be designed on a
formal level. The framing of game development as model transformation might
assist game designers because it helps to structure the way we think about these
processes and helps to codify design wisdom in the form of formal grammars
and rewrite systems. The additional advantage in describing these processes
on a formal level is that this is an important step towards developing tools that
Chapter 7 | Generating Games
185
help designers by automating parts of this process; formal grammars and rewrite
systems are instrumental in developing tools for procedural content generation.
7.9
Procedural Content in Games
There are multiple ways to look at procedurally generated content in games.
The most common application of procedurally generated content is games that
generate all or some of their levels automatically with complex algorithms. But
these days procedural content generation is also gaining ground as an aid for
design teams during the development process or sometimes to assist players to
create content while playing.
Games with procedurally generated content have been around for some time.
The classic example of this type of game is Rogue, an old Dungeons & Dragons style, ASCII, dungeon-crawling game whose levels are generated every time
the player starts a new game.7 Newer games that use procedural techniques include Diablo, Torchlight, Spore and MineCraft. The typical approach of
these games can be classified as a brute-force random algorithm that is tailored
to the purpose of generating level structures that function for the type of game.
Often these algorithms generate a large sample and rely on evaluation functions
to select the level that is the most fit (Togelius et al., 2010). Others evaluate
the level in order to remove areas that turn out to be unreachable (Johnson
et al., 2010). One strategy is to generate a tile map that is filled with tiles
representing solid rock and to ‘drill’ tunnels and rooms into the map starting
from an entrance. Multiple paths can be created by drilling into new directions
from previously created locations. The dungeon is then populated with creatures, traps and treasures.8 Another strategy involves zoning the dungeon into
large tiles, generate dungeon rooms in some of these zones in the next step, and
finally connecting the rooms with a network of corridors.9 To create game space
to represent wilderness areas, cellular automata can be used to generate more
organic structures.10
Although these algorithms have a proven track-record for the creation of
roguelike games, the gameplay their output supports does not necessarily translate to the generation of other types of games.11 For this research the pro7 Rogue is so influential that other games that follow in the same procedural tradition are
often referred to as ‘roguelikes’.
8 See Mike Anderson’s dungeon building algorithm on RogueBasin http://roguebasin.
roguelikedevelopment.org/index.php?title=Dungeon-Building_Algorithm (last visitid July
23, 2011).
9 See the “Grid Based Dungeon Generator” algorithm on RogueBasin http://roguebasin.
roguelikedevelopment.org/index.php?title=Grid_Based_Dungeon_Generator (last visitid
July 23, 2011).
10 See Jim Babcock’s “Cellular Automata Method for Generating Random Cave-Like Levels”
on RogueBasin http://roguebasin.roguelikedevelopment.org/index.php?title=Cellular_
Automata_Method_for_Generating_Random_Cave-Like_Levels (last visitid July 23, 2011).
11 A typical major component of the gameplay of roguelike games is character building. This
type of gameplay, which stems directly from a rather mechanistic interpretation of pen-andpaper role-playing, resolves for a large part around gathering experience points and magical
equipment to improve the main character. As game designer Ernest Adams points out in his
satirical “letter from a dungeon”, there seems to be little purpose behind these mechanics,
186
Joris Dormans | Engineering Emergence
totypes focused on generating content for action-adventure games, which are
story-driven games where exploration, puzzle-solving, conceptual and physical
challenges make up the majority of the gameplay (Adams & Rollings, 2007).
Compared to simulation games and role-playing games, action-adventures typically have a relatively simple set of simulation rules and only a few available
power-ups. These games usually do not have an elaborate leveling system where
character development, expressed in terms of skills and attributes, is an essential
part of the gameplay. Lacking these, action-adventure games must rely more on
level design as their prime source of gameplay. As a result, a structured learning
curve, clever pacing of action, challenges and puzzles play a more prominent role
for the levels in an action-adventure game. A procedure to generate levels for
this genre must include a way to incorporate these elements. It is in a similar
light that Gillian Smith et al. point out that generating levels for an action
platform game requires different techniques, as level design is also a far more
critical aspect of that type of game (Smith et al., 2009).
As it turns out, level design principles, like flow, pacing and structured learning curves, are difficult to implement with the algorithms commonly used for
roguelike games. These algorithms generally cannot express these principles as
these principles mostly operate on larger structures than the individual dungeon
rooms and corridors the algorithms work with. The solution of Smith et al. is
to create a “rhythm-based approach” to generate levels with “a strong sense of
pacing and flow” (2009). The perspective of game design as a series of model
transformations provides us with a formal framework that allows us to take yet
another approach. Rewrite systems allow us to codify and implement game
design knowledge at many different scales. It allows us to start from design,
rather than from algorithm, and generate different types of game content (see
figure 7.29). What is more, it is applicable beyond automated level generation,
which is the focus of most academic efforts. As we have seen in the previous
section, it can bridge the gap between levels and mechanics; it is also applicable
to procedural generation of game mechanics, which is a relatively unexplored
area of procedural content generation (Nelson & Mateas, 2007; Reyno & Carsı́
Cubel, 2009b). The process is also very flexible, it supports many different sequences of generation. Mechanics might be generated first, or levels might be
generated first, or the procedure might switch back and forth between these two.
A detailed example of these techniques to generate level structures in the style
of The Legend of Zelda games can be found in appendix C.
Procedural content generation is still a growing research field within game
studies and computer science. It is attracting more and more attention as many
designers realize that games might benefit from procedurally generated content.
resulting in a shallow representation of character growth as a faint echo of the mythical quest
(2000). Gameplay of this type, although forming a viable niche of its own, is well suited for a
random dungeon layout. It does not require the same standard of level-design quality as, for
example, an action-adventure game from the Zelda series. In action-adventure games this style
of character development plays only a little part, as is mentioned in an interview by Shigeru
Miyamoto, the Zelda series main designer (quoted in DeMaria & Wilson, 2004, 240). Just as
the random encounter table is an appreciated tool to facilitate a particular style, but not all
styles, of role-playing in Dungeons & Dragons (Dormans, 2006b).
Chapter 7 | Generating Games
187
Figure 7.29: One of the prototype games built in support of this research. This game
is a 2D action-adventure game with procedurally generated levels, where the player,
the light circle equipped with a triangular ‘sword’ and a rectangular ‘shield’, collects
items, fights enemies and solves puzzles.
These days triple-A titles require so much content that companies are actively
pursuing methods that could lead to shorter development times. The economic
benefits aside, there is also an increased interest in the area because procedural
content generation could lead to better game experiences. For example Jesper
Juul argues that games that generate new levels each time players have to restart
again because they failed during the game, reduce the costs of this failure. In
general, players dislike having to perform exactly the same actions over and over
again (Juul, 2009, 2010). Used in this way, procedural content could lead to
games with more varied gameplay.
7.10
Adaptable Games
The generation techniques discussed in this chapter can also be employed
to generate content during play, allowing for the opportunity to let the actual
performance of the player impact this generation and create games with highly
adaptable gameplay. There are several strategies to accomplish this. A straightforward strategy would be to transform certain elements according to rewrite
systems as the player plays. The selection of transformation rules could be
based on the player’s performance. In this case, the whole level or even the
whole game will come to reflect the player’s unique performance.
An interesting example of this technique is discussed by Julian Togelius et.
al. (2011) in relation to a Super Mario Bros. clone that features procedural
content by the name of Infinite Mario Bros. In this game the levels are
adapted to players’ actions directly. In one version, whenever the player presses
the jump button a platform would be created at that position in the next level.
188
Joris Dormans | Engineering Emergence
Similarly when the jump button is released the ground is changed, and enemies
are added to the next level in response to presses of the fire button. In another
version of the same game, the transformations are applied to the same level but
just outside the view of the player (or sometimes even in plain view). Extra
transformations were added: new enemies are spawned for every coin the player
collects and new coins are created for every enemy the player destroys.
In response to the article by Julian Togelius et. al. I created a small experimental version of the classic game Boulder Dash. In this game, with
the uninspired name Infinite Boulder Dash, the player collects diamonds
from a two dimensional mine that also features boulders and patches of dirt.
In contrast to the original game, the game automatically scrolls to the left and
the player must keep up with the scrolling to prevent getting killed. The mine
wraps around: what disappears on the left reappears on the right. However,
as the player collects diamonds new elements are added to the level, including
more boulders, diamonds or moving enemies. What element is created depends
on the number of diamonds the player collects on each subsequent move, and
on the rules of the current level. Figure 7.30 shows the rules for one such level.
It indicates that after the player collects the first diamond a boulder is created,
after the player collects a second diamond a new diamond is created, two boulders are created after the player collects three diamonds in a row, etcetera. In
addition, every time the player collected enough diamonds and leaves the level
through the exit, these creation rules change. In general the new level will be
more difficult, as more boulders are created, enemies might be spawned, or useful bonuses might become harder to obtain. The player’s performance is a factor
in this. The game keeps track of the number of diamonds collected, the number
of uncollected diamonds, the number of lives lost, among other things and these
statistics affect how rules might change. The actual changes are implemented
through simple rewrite rules. For example in order to make the game more difficult the rule ‘3rd diamond = 2 boulders’ (collecting three diamonds in a row
results in the creation of two boulders) might be replaced with ‘3rd diamond = 3
boulders’, thereby increasing the number of boulders that is likely to be created
and thus also the difficulty of the game.12
Infinite Boulder Dash implements a fairly simple variant of adaptation
techniques. The use of rewrite systems to create adaptable games can be taken
much further. Imagine a simple vertical space shooter where the behavior of
enemies is described by a simple graph (see figure 7.31). Transformations could
be used to evolve these graphs to create enemies with different behavior. The
game can easily modify this behavior based on the performance of the player: if a
player destroys an enemy quickly that enemy is probably very weak, on the other
hand if the enemy manages to damage the player it is probably more successful.
Particular player actions such as destroying entire enemy waves, the collection
of upgrades, or the completion of levels might trigger these transformations.
Each trigger might differently affect the probability of transformation rules being
selected. At the same time the game could use the same data to trigger similar
transformations on the graph describing a level-boss, creating a final adversary
12 Infinite
Boulder Dash can be played at www.jorisdormans.nl/InfiniteBoulderDash.
Chapter 7 | Generating Games
189
Figure 7.30: In Infinite Boulder Dash players collect diamonds in an endless
series of generated levels that adapt to the players performance.
that is appropriate for the level. This way the boss would use similar weapons
and maneuvers the player already encountered during the level; it would allow
the game to prepare the player for the final boss fight.
The game could use a similar technique to build a model of the player. Such
a model would start simple, but certain actions of the player would trigger
transformations that in turn affect the way the game reacts to the player. This
technique could easily be applied in games that feature role-playing elements
where the player frequently can choose between options and solutions that represent different ethical attitudes and where non-player characters in the game
react to the choices made by the player. Examples of these games are Deus Ex
and Fable. Where most of these games use (relatively) simple variables to track
to what extent a certain non-player character likes or trusts the player character
(see for example Crawford, 2005), transformations would allow the AI for these
non-player characters to build up a far more complex model of the player and
act accordingly.
Something similar can be accomplished for game stories, too. When player
actions trigger transformations on the general plot, the possible number of generated plots quickly expands beyond the potential of the commonly used branching
story trees. Instead of the simple boolean logic that plagues many interactive
plots (cf. Wardrip-Fruin, 2009), a game constructed in this way would apply certain transformations on the current plot based on the performance of the player.
This could lead to an interaction of much finer granularity between player and
game. It could also quite literally lead to an implementation of an interactive
structure that Marie-Laure Ryan calls a fractal story where a story keeps offering
more and more detail as the player turns her attention to certain parts of the
story (Ryan, 2001, 337). Marie-Laure Ryan describes this structure following
190
Joris Dormans | Engineering Emergence
Figure 7.31: Three enemies for a vertical space shooter represented by graphs. The
two enemies below might have evolved from the enemy above.
some ideas on interactive storytelling that feature in Neal Stephenson’s novel
The Diamond Age (1995). In this novel a young, lower-class girl called Nell
acquires a state-of-the-art interactive storytelling book that teaches her important skills, prepares her for later life and responds to the girl’s actual real-life
situation. It does so by telling stories of Princess Nell, which take the form of
a classic fairytale. The basic premise of the fairytale is always the same, but
the details are expanded every time the girl reads further. The story adapts
and reflects Nell’s life outside the book, helping her to overcome the problems
she faces in real life. An important difference between the fractal stories and
branching stories is that where branching stories build towards different, predesigned endings along pre-designed paths, the fractal story transforms itself to
accommodate many different paths that essentially lead towards the same goal;
the general outline of the fractal story in Stephenson’s book is known from the
start, when Nell is reading the book the story is not so much advanced as it is
expanded (also see Dormans, 2006a).
In order for these techniques to work, the game needs to be able to assess
players’ actions. Luckily games are typically pretty good at this task; they
already reward and penalize players for many actions, for example by rewarding
points or taking away lives. The completion of tasks or (sub)quests could easily
trigger transformations. Likewise failure to complete tasks or (sub)quests could
trigger other transformations. What transformation is selected can be affected
further by the model of the game or story as it has been created thus far.
Transformation rules automatically become inapplicable when what they need
Chapter 7 | Generating Games
191
to replace (the left-hand part) no longer can be found in the current model.
What elements can be replaced with new elements can depend on many
things. Rewrite systems can replace any thing in a current representation of
a game. However, once a player has encountered certain elements they might
need to be excluded from further rewrite operations.13 Elements that have not
been encountered yet could always be transformed into something else, should
the need arise. Even elements that have been encountered, but have not yet
been fully explored, might change function or behavior, and if the designer of
the game is prepared to risk consistency, anything might be subjected to further
transformation.
7.11
Automated Design Tools
Procedural techniques can also be leveraged to automate game design tools.
Such tools assist designers to create quality games or game content by automating some tasks of the design process. This approach has been called a “mixedinitiative approach” (Smith et al., 2010; Smelik et al., 2010) and is contrasted
with procedural content generation tools that build games and game content
without interference of human designers. Although the latter is interesting in
itself, there are relatively few games that actually consist of fully generated content. Interest in tools that focus on assisting designers is growing as more and
more game companies acknowledge that such tools can increase the effective
output of their staff; it allows level designers to focus on the creative aspects
of their job and delegate more of the manual tasks to the computer. There are
even opportunities for those games that allow players to become the co-creators
during play, as is the case with Little Big Planet.
Model transformations and rewrite systems are an excellent match for the
mixed-initiative approach. They provide the designer with many opportunities
to control the process of level generation at many different levels of abstraction.
At the top most level of abstraction, designers might specify the sequence of
transformations, selecting different rewrite systems for each step. In effect this
would allow designers to specify whether the level is designed with a particular
mission as its starting point (as outlined in figure 7.1) or whether a particular
space guides the design of the level (as outlined in figure 7.2). There could
even be alternative modules to generate different types of spaces: one rewrite
system might generate a ‘dwarf fortress’ while another might generate an ‘orc
lair’. Additional transformations might change the ‘dwarf fortress’ into a ‘dwarf
fortress overrun by orcs’, etc.
The prototypes that support this research initially all focused on level generation. The rewrite systems that operated in these prototypes were designed to
formalize design knowledge such as lock and key structures and learning curves
(Dormans, 2010; Bakkes & Dormans, 2010). These prototypes were successful,
in the sense that they were able to generate levels quickly and with some interesting gameplay and progression. Where the first prototype had some difficulty
in generating levels where different, alternative routes converged, these problems
13 The
version of Infinite Mario Bros. that does this is very strange, almost unplayable.
192
Joris Dormans | Engineering Emergence
were more or less solved in the second prototype. Currently, the implementation
of the transformation from a graph that represents a level space to a map is
very specific for top down 2D action-adventure style games. The shape rewrite
system to refine the space is also implemented in only two dimensions, but, when
needed, the same type of grammars and rewrite systems can be made to work
in three dimensions.
There are four points that could lead to further improvements:
1. What transformations are involved in the generation process and their
order is more or less fixed; in order to make the most of the model transformation approach this should be implemented with more flexibility.
2. Designers could be given more control over the generation approach in
order to make the prototype more suitable to a mixed-initiative approach
to procedural content generation.
3. Structuring the learning curve can be improved by inclusion of the generation of mechanics as outlined above.
4. The transformation from space graph to a geographical map of the level is a
step that is implemented without the use of a rewrite system. Although the
step is quite small and rewrite systems control the process before and after,
it is something that should be improved for a truly generic application of
these techniques. Currently there is no off-the-shelf rewrite system that
can deal with topographical graphs and geographic spaces at the same
time.
For the third and final prototype (Ludoscope, see section 5.8) I focused on
dealing with the first three improvements. Flexibility was created by implementing ‘recipes’ (see figure 7.32). A recipe consists of a series of instructions that
can be specified by the user. These instructions include, among others, opening rewrite systems, clearing graphs, applying rewrite rules, and changing the
automatic layout settings. A recipe can specify a specific number of times that
a rule should be applied, a range from which the tool will randomly select, or
it can specify that the rule must be applied as long as there are suitable nodes
to apply it to. In this way the user is able to specify the steps involved in the
generation process. The interface allows the user to iterate through the steps, to
skip certain steps if need be, or to complete the entire process at once. Applying
a recipe leads to similar, but different levels.
Manual control over the generation process was also implemented. This feature makes Ludoscope suited for a mixed-initiative approach to content production. Designers can manually select nodes in the graph and then apply any
applicable rule to it, or when no node is selected the tool finds out which rules
are applicable to any node and offers designers a choice between them (see figure 7.33). When designers choose to apply a rule, a suitable node is selected
randomly, unless a specific node was selected. Ludoscope implements an automatic layout system to handle the changing graph representations, but designers
can manually change the layout by dragging individual nodes around. In addition, Ludoscope allows designers to directly manipulate individual nodes and
Chapter 7 | Generating Games
193
Figure 7.32: Ludoscope, a prototype for an automated level design tool, implements
a recipe (on the left) to allow users to specify a multi-step, automatic generation.
edges ans well. Designers can add, change or delete nodes as they see fit. To
support manual editing, a number of simulation features are implemented that
help designers check levels and mechanics for consistency. These features were
already discussed in Chapter 5. This way Ludoscope becomes a more general
design tool that assists the designer with planning and creating consistent game
experiences. What is more, in the current implementation, designers can easily
move back and forth between automatic and manual modes of producing game
content.
Finally, Ludoscope works with space graphs, mission graphs and Machinations diagrams. With the tool it is possible to create graphs that include elements
of all of these different representations of games and use them to represent, and
generate, these different aspects of games in unison. It provides designers with
a powerful tool, allowing them to experiment and simulate game designs in an
early stage of development.
In developing Ludoscope much effort was put into the editors that allow the
creation of the various formal grammars and rewrite systems that are involved
in the process (see figure 7.34). Creating them is a critical step in automating the design of games. I assume that specific transformations are needed for
particular games. Although generic rewrite systems, such as the lock and key
grammar discussed above, can serve as useful starting points, I expect that all
rewrite systems need to be adapted to a particular implementation. This makes
the process of setting up such tools difficult and time-consuming. However, I
do believe, that, with some experience, the benefits are far greater than these
investments.
7.12
Conclusions
Game design framed as a series of model transformations and the use of
rewrite systems allows us to capture and experiment with design principles at
194
Joris Dormans | Engineering Emergence
Figure 7.33: Ludoscope. The highlighted rule on the left indicate which rule currently
is applicable to selected node in the mission graph (with the white outline).
Figure 7.34: Editing graph rewrite rules in Ludoscope.
Chapter 7 | Generating Games
195
a formal level. This is applicable in the automatic generation of game content,
but also functions as a useful tool in the process of designing levels by hand,
or assisting designers by automating parts of the design process. In all cases, it
allows level designers to approach their task on a high level of abstraction. At
this level of abstraction level designers can focus on the truly creative aspects of
their task. This increases their effectiveness in designing levels, and reduces the
chance of flaws in the design.
As an example the role of using locks and keys, in various different guises,
was discussed in relation to the transformation of linear missions into nonlinear
spaces. These transformations can be formalized using rewrite systems, providing us with a structured method to express and experiment with these level
design techniques. Adding locks and keys to a mission structure is but one step
in the process of designing a level or a game. Many more steps can be defined
in a similar way including the generation of game mechanics. The process of designing a game involves many specialized transformations, some of these are only
applicable within the context of a single game or genre. By specifying many different steps and individual rewrite systems for each step, a highly flexible body of
game transformations can be created. For each individual game, designers might
select the transformations that are the most applicable and create a sequence of
transformations that yield the best results.
In order to validate the effectiveness of this approach to game development
a series of prototypes were created. Initially these prototypes focused on level
design and generation, but similar techniques can be used to generate mechanics,
adapt gameplay to player performance, and to shed new light on interactive storytelling. What is more, model transformations, formal grammars and rewrite
systems suit an approach to content generation where the computer assists human designers by automating certain aspects but leaving the most critical and
creative aspects under control of the designers. I expect this approach to gain
momentum in the near future as it will help game companies to improve efficiency and quality of their production.
The prototypes outlined in this chapter are still preliminary. There is plenty
of room for improvement. As was already pointed out, the transformation from
graph to space would benefit from a more generic implementation. Also, to
increase the quality of the output, a fitness function, based on game and level
design heuristics might be implemented to directly or indirectly affect the transformation process. These things are left to be explored in the (near) future.
196
Joris Dormans | Engineering Emergence
Emergence is not the same thing as magic.
Chris Crawford (2005, 138)
8
Conclusions and Validation
This dissertation set out to develop formal tools for the hard task of designing
games. Games, as rule based systems, often display emergent behavior; one of
the core aspects of games, gameplay, is an emergent property of game rules and
mechanics. Many games try to create gameplay with a relative simple set of
rules that offer much variation and freedom to players, although not all games
do so in equal measure. For games that aim to deliver a structured story-like
game experience, level design usually is more critical to the final gameplay than
mechanics that build emergent gameplay are. However, these days, games that
build on a pure structure of level progression are quite rare; while emergence
plays a role in virtually every game.
As rule based systems, games share kinship with simulations. However, where
the purpose of simulation is to accurately model (at least certain parts of) a
source system, games are often considered stylized representations of systems
that might, or might not, exist outside the game. Games can afford many
more liberties in order to entertain, to persuade, or to educate. The relation
between game rules and the systems they represent, goes beyond accurate, iconic
simulation. For games this means that they can use relative simple means to
represent dynamic and complex systems. In fact, there are some arguments
to suggest that this is where the real power of games lies: games allow us to
play with useful shortcuts in understanding complex systems. Yet, as even
relative simple games can display complex behavior, games naturally utilize those
structural features of rule systems that give rise to the complex behavior in the
first place.
Game studies is still a young discipline. Over the years there have been
many attempts to create more generic and formal approaches to games and
the process to design them. So far, none of these approaches has grown into a
standard that spans the game industry and academia. Game design theories face
several problems. Perhaps the most prevalent among them is the little trust the
game industry puts in the approaches that have been developed over the years.
This lack of trust finds its roots in a poor applicability of many of these theories
for actual design, especially when one considers that, in general, the investment
required to master these theories is quite high. Many theories, frameworks,
vocabularies and abstract design tools are more successful as analytical tools
198
Joris Dormans | Engineering Emergence
than they are as tools that might help to engineer games. Another problem is
that designers can be quite skeptical towards the entire enterprise of developing
design methodology; these designers usually dismiss any theory that tries to
formalize (and thereby ‘steal the soul from’) the creative process of designing
games.
With these concerns in mind I have presented two frameworks and outlined
the possibilities to develop automated design tools that are firmly grounded in
design practice and aim to assist designers in creating quality games effectively
and efficiently. The first framework, Machinations, focuses on discrete game
mechanics and formalizes a perspective on game rules and emergent behavior
in games up until the point where games can be simulated using the Machinations tool before they are build. The Mission/Space framework focuses on level
design. Perhaps it is less innovative than the Machinations framework, but it
helps to structure the process of designing game levels, nonetheless. What is
more, combining both frameworks sheds new light on integrating emergence and
progression structures in games that many treat as conflicting ways of creating game challenges. The outline for automated design tools builds on the idea
that game design can be framed as a series of model transformations. By using
formal grammars to describe the models involved in this process and designing
rewrite systems to govern transformations between them, a highly flexible yet
formal approach to game design has been created. This approach has been made
concrete in a series of prototypes that can generate game content automatically
or assist human designers in creating game content.
In this final, concluding chapter I will formulate an answer to this dissertation’s research question, and discuss how all frameworks and prototypes have
been evaluated and validated. Finally I will look ahead what research and developments they might inspire in the near future.
8.1
Structural Qualities of Games
The central research question of this dissertation is: what structural qualities
of game rules and game levels can be used in the creation of applied theory and
game design tools to assist the design of emergent gameplay?
The Machinations framework has been developed to formulate one part of the
answer. For games with discrete mechanics, feedback loops within the internal
economy of the game play a critical role in quality of the emerging gameplay. A
game’s internal economy is formed by the production and consumption cycles
that involve the game’s most important abstract and tangible resources. Depending on the game, these resources might be anything: from weapons, ammunition and health in a shooter game, to food, status and safety in a city building
game. Feedback is created when the accumulation, production or consumption
of these resources directly or indirectly affect its accumulation, production or
consumption in the near future. Feedback can have many different characteristics. Traditionally, positive and negative feedback are distinguished, but for
games it is equally relevant whether or not the effect of the feedback is constructive or destructive, at what speed it operates and how durable it is, among other
Chapter 8 | Conclusions and Validation
199
things.
The Machinations diagrams are designed to represent a game’s internal economy and to foreground feedback structures within. From the analyses of a number of games it became apparent that several feedback structures are recurrent
and can be found in many games. In addition, most games that display interesting, emergent gameplay feature multiple, but not too many, feedback loops.
To explain the structural qualities of games that are level-driven, internal
economy and feedback loops do not help us much. Levels have their own structures that are independent of, yet often also related to, these mechanics. In
general level design has not been studied as extensively as game mechanics have
been. From a review of a number of level design typologies it has become apparent that one of the problems is that in levels two structures operate that
are often only poorly distinguished. To this end, the Mission/Space framework
separates missions from space. In this way the framework helps to create a clear
perspective on level design. A mission is a series of related tasks players must
perform in order to complete a particular level. A space describes the topographic or geographic layout of a level. Missions and spaces are related, but
in contrast to what some design strategies suggest, they need not be isomorph.
Often missions are laid out in space in such a way, that players are assisted in
setting exploration goals, have a fair sense of where they need to go, are rewarded
for formulating intentions and plans, and experience the growth of their own, or
their characters’, abilities.
Lock and key mechanisms play an important role in many games that are
level-driven. Lock and key mechanisms create flexibility in the way missions
might be mapped to game spaces, breaking away from levels in which mission
and space have to be isomorph in order to create a coherent game experience.
Distinguishing between mission and space also helps to identify more clearly
the difficulties in designing levels that allow for a more articulate interactive experience. Space and physics have evolved much faster during computer games’
relative short history than mechanics to control progression have. By applying
the lessons learned from the Machinations framework, more interesting mechanics to control player progression through a mission can be created. Feedback
mechanisms that traverse between internal economy and level design can be created in order to push the emergent behavior of games towards areas that hitherto
have been relatively unexplored: it helps designing games in which progression
is an emergent property of the underlying system.
Although the Machinations and Mission/Space frameworks can be used as
analytical tools, they are set up as design tools first and foremost. They help
to formalize existing design knowledge and experience, they allow designers to
express and discuss designs, but mostly importantly they make tangible certain aspects of game design that normally are quite intangible. Machinations
diagrams give shape to a system that otherwise is largely invisible, the Mission/Space framework untangles two superimposed structures that co-exist in
level design and creates a clear perspective on both. Both frameworks are fairly
easy to learn, they do not rely on extensive vocabularies or pattern catalogs.
Yet, despite their simplicity they are quite expressive, and can capture a large
Joris Dormans | Engineering Emergence
200
variety of different game structures. I hope that the many examples I have used
throughout this dissertation are ample illustration of this.
The concerns about formalizing game design, as expressed by some game
designers, cannot be taken away by these frameworks themselves. In a manner,
this dissertation does exactly what some of them oppose: it seeks to objectify
quality in game design. By pointing out that certain creative aspects within
game design cannot be objectified they dismiss any formal approach to their
craft. However, the game industry must acknowledge that quality is not equally
distributed over all games. We are able to compare games and discriminate
between good and bad games. The frameworks presented are designed to help
designers identify, create and discuss these qualities in games. These frameworks
are not designed to replace creativity, rather, they are designed to support it.
8.2
Validation
In this respect, one important question remains unanswered: how successful
is the applied theory presented in this dissertation; does it really support game
designers in their labor? Clearly, it has not emerged as an industry standard,
but nobody can expect that anything could have been developed into an industry standard within the scope of writing this dissertation. The theories have
been used by people in the industry and by students, but I have not gathered
quantitative data to evaluate its effectiveness. How, then, can the frameworks
and the prototypes be validated? How can I be sure that what I have presented
is of any real value to a game designer?
I have tried to validate the applied theory of this dissertation in four ways.
These ways are described below briefly. In the sections that follow they are
discussed in more detail.
1. Implementing many of the theories in software tools to create diagrams,
simulate games and generate content, has made these theories much more
concrete and has validated them to some extent. On many occasions implementing the theory has led to changes, and vice-versa. In the end I have
managed to implement (nearly) all of the theories somehow. The work
on visualizing game mechanics eventually has led to the implementation
of the Machinations tool, the work on level design has triggered a series
of level generation prototypes for action-adventure games that later has
grown into a more generic tool for generation of game content: Ludoscope.
2. During the period of this research I have actively sought opportunities to
present results to the game development community. In practice this led
to a number of talks and workshops. The responses from these workshops
were usually positive. Although, I did also meet people who were less
enthusiastic for reasons I have discussed in Chapter 3. A number of companies and professional designers have used some of the tools here, and
continue to do so.
3. I have presented much of the material as peer-reviewed conference papers
Chapter 8 | Conclusions and Validation
201
and two journal papers. Many of the chapters in this dissertation are
the result of several iterations. For example the Machinations framework
started out as a diagrammatic notation based on UML and was initially
never intended to be interactive. Many fellow academics, but also people
from the industry, have commented on earlier work. Their suggestions led
to many improvements and shaped the frameworks into the form presented
here.
4. Last, but not least, as I was in the fortunate position to be involved in
the development of several game design courses for the Hogeschool van
Amsterdam (the Amsterdam University of Applied Sciences) I have had
the opportunity to use many of these theories to structure courses and
train students.
8.3
Teaching Game Design
Over the past five years I have been involved with the development of several
game design courses at the Hogeschool van Amsterdam. In February 2007,
my colleagues and I started with a half-year minor program on game design
for third year students in the departments of computer science and interactive
media. This minor was to complement the already existing game technology
minor that, at the time, had been running for two years. This was followed by a
serious games course that started in 2008, and a full, four-year game development
program starting from September 2009.
My responsibility in these courses have been many, but most of them resolved
around setting up and teaching design courses where the students learn how to
build rule systems in order create interesting, or meaningful gameplay experiences. Common industry practices have always informed these courses. This
includes the MDA framework (mechanics, dynamics and aesthetics), play-centric
design, physical prototyping techniques, heuristic evaluations and play-testing
strategies (see chapter 3 for a discussion of some of these). But much of the
material developed as part of this research also quickly found its way into the
course material, starting with the Machinations framework and its precursors.
I have always been a strong advocate of designing games ‘inside-out’. By this
I mean that it makes sense to start with designing mechanics, and build the game
up from there. The MDA framework suggests something similar. Unfortunately,
mechanics and rule systems are not the easiest points of departure for novice designers. It takes experience to appreciate the full effects of small changes to the
rules. Lacking experience and an accurate vocabulary to express these nuances,
the Machinations diagrams proved to be a useful educational tool. A recurrent assignment was to have students create Machinations diagrams for existing
games but also for prototypes they were working on. The first assignment was
designed for students to become familiar with the framework, the second assignment to help them get a clear perspective on their own work. These assignments
are not easy, for many student it takes time to understand the subtleties of the
Machinations diagrams. However, even when students made diagrams that were
202
Joris Dormans | Engineering Emergence
Figure 8.1: A student Machinations diagram by Gerard Meier for the game CrimeBloc
(made for Machinations version 2.0)
poor representations of their games, I found that these diagrams were always a
good starting point to discuss their designs with great precision and accuracy.
Some students became very skilled in creating these diagrams and kept using
them for other assignments as well. These diagrams sometimes turned out to
be very detailed and complex (see figures 8.1 and 8.2). But it illustrates that
students were able to use them and that at least some of them saw the benefits
of keeping to use them. This way, the framework has had an impact beyond
the individual workshops and was instrumental in helping students to create a
professional perspective on the design of game mechanics.
The workshop that introduces the Machinations framework has evolved into a
more or less fixed format. I have used the workshop or variations on it at several
occasions. Obviously, it was used during regular classes at the Hogeschool van
Amsterdam, but I also used it for workshops I hosted at the Willem de Koning
art academy in Rotterdam, at the Game in the City industry event in Amersfoort, workshops at the T-Xchange lab in Enschede, and at Paladin Studios in
Leiden (all of these locations are in the Netherlands). During these workshops,
participants start with designing a simple board game mechanism that allows
them to move pieces on a board, but that also involves some sort of resource.
Chapter 8 | Conclusions and Validation
203
Figure 8.2: A student Machinations diagram by Rik Ruiten. This diagram (made
for Machinations version 2.0) represents a fully functional prototype for a real time
strategy game.
204
Joris Dormans | Engineering Emergence
Next, I explain the basic idea of a game’s internal economy and introduce the
Machinations diagrams to model this economy. After the participants created
diagrams to represent their mechanics in this way, I introduce the notion of feedback and stimulate the participants to explore the concept in their own design.
Finally I challenge them to redesign their mechanics in such way that it includes
multiple feedback loops. This format is an effective introduction to the framework, and often led to quite interesting designs. Most of the participants were
able to create mechanics that were interesting, even when the point of departure
was rather simple.
Another, more advanced workshop focuses on the use of recurrent design
patterns. This workshop started with a short lecture to explain the framework
and then challenged participants to create designs based on a random selection
of patterns. Workshops like this were held at a meeting of the Dutch chapter of
the Digital Games Research Association in Utrecht in 2008, at the INTETAIN
conference in Amsterdam in 2009, and at the DIGRA conference in Hilversum
in 2011 (all of these locations are in the Netherlands). This workshop is more
advanced and suitable for participants with more experience and expertise.
I was not the only teacher involved in these workshops. My colleagues taught
similar classes and workshops. They have similar experiences and point out
the advantages of having an online, dynamic implementation visualizing these
dynamics. In the near future, we plan to capitalize on this advantage more be
expanding our workshop repertoire utilizing this property of Machinations.
The Mission/Space framework was developed later than the Machinations
framework. As a result it has not been used as extensively in courses as the
Machinations diagrams have been. Yet it has also been instrumental in structuring design courses. The game design course for the second year students in
the game development program, was heavily influenced by the ideas developed
as part of this framework. The focus of this course was on building prototypes.
Every two weeks students needed to finish a new prototype. During each of these
two weeks there was a central subject. Mission was one of these subjects, and
space was another. During lectures the individual perspectives of missions and
spaces were introduced and discussed. For students the identification of lock and
key mechanism proved to be a practical and highly applicable perspective that
helped structure their designs. Over the years I have received multiple reports
that clearly indicate that students are able to work with these concepts, both in
analyses and in designs (see figure 8.3).
Framing the game design process as a series of model transformations and
procedural content generation has not been the subjects of courses. These subjects are quite advanced, probably too advanced for most of the bachelor students
I have worked with. At the moment of writing, the most advanced students are
at the end of their second year, and the courses these students will take in their
final two years are still under development. Despite the fact that some of these
students have expressed a direct interest in content generation, it is unlikely
that much of this material will make it in to the program. At best, it might
be the subject of some of these students bachelor theses, or play a role in the
research program of the game lab that has been set up at the Hogeschool van
Amsterdam.
Chapter 8 | Conclusions and Validation
ROOM (1280x720)
LOCK
ICE
KEY
RUBBLE
205
SUBWAY PART 2
This map is not scaled properly.
The puzzles and platforms are bigger to give an
indication how they work.
The map should be created according to scale
when the visuals are made.
SWITCH DOOR
PLATFORM
SWITCH
EMERGENCY DOOR
BUTTON DOOR
SPRINKLER
BUTTON
COMPUTER TO
TURN POWER ON
SAVE
START
BIG FIRE
METRO
UPGRADE GEM
NEXT PART
FROST GEM
SAVE
WATER
Figure 8.3: One of many level designs by Robin Brouwer and Dennis Kempe for the
student game Enchanted Hands focusing on structure and progress. Many of the
upgrades and game elements in this level act as locks and keys.
8.4
Building Prototypes
The working examples of the prototypes (many of which are available online)
should validate the theory to some extent. Validation always was one of the
reasons for the creation of these prototypes. The initial level generation prototypes, in particular, were created with the idea that if a formalized perspective
of level design principles would lead to an interesting implementation, then the
perspective must be of some value. This remains hard to test, and the success
of the levels generated by these prototypes were never subjected to an objective
measure. However, general reception of demonstrations of several prototypes
has always been quite positive.
Initially, I did not intend Machinations diagrams to be interactive, but I am
very happy that I did take that step early. This has led to many improvements
in the theory, and it also has led to the creation of a very useful design tool.
The fact that it can be used to simulate games is very important in this respect.
Representations of games in the Machinations tools can, to a certain extent, be
played. The fact that in this form the representation retains many of the same
dynamic behaviors, is a strong indication that the structures these diagrams
foreground indeed play a critical role in creating emergent gameplay.
The development of Ludoscope also proved to be an important test bed for the
design theory presented in this dissertation. The theory and the tool evolved
in parallel. There were many iterations for both and with each iteration one
kept improving the development of the other. This process went on into the
very last stages of writing this dissertation, and possibly beyond. Over time,
the Mission/Space framework grew more formal, and every time it did, the
implementation of the framework made the implementation of Ludoscope leaner
206
Joris Dormans | Engineering Emergence
and more powerful. At the same time, the implementation of the framework
as an automated design tool kept opening up new avenues for exploration. For
example, with mission graphs implemented in Ludoscope it became very easy to
add features to check for mission and space inconsistencies. This process is not
entirely finished. Ludoscope is still under development. Without doubt, this
development will lead to new theoretical insights that will in turn strengthen
the development process.
In the end, developing theory and developing software almost became one
continuous process. I feel very fortunate that I was in the position to combine
these two methods of developing theoretical and applied tools. One might assume that writing and coding was a tough combination but the truth is that
the synergy between the two gave me more energy to develop both to a much
higher level than I would have been able to do when working on each project
individually.
The research also led to the development of a few experimental games. Most
of which are never released as they were only rough prototypes. Some of these
evolved further, such as the Infinite Boulder Dash example discussed in
Chapter 7. In other cases, the game projects I have been involved in served as
an excellent testbed for the theory and tools presented here. For example, for
Get H2O I created Machinations diagrams to help design the game’s economy.
Games such as Seasons benefited from my level design research. These examples are discussed in this dissertation. Other projects include Ascent, a small
platform game that was an experiment with abstract resources; Dungeon Run,
Flix and Flix 2 were (at least partly) experiments in level design; Campagne,
a satirical card game that benefited from applying non-iconic reduction and creating a symbolic simulation of Dutch politics; and Bewbees that benefited from
the Machinations framework in order to create emergent gameplay.
Another experimental game has been built by three students at the Hogeschool
van Amsterdam under my direction. For this game, LKE (figure 8.4), I asked
the students to experiment with a lock and key mechanism that utilizes a consumable resource as the key and implements multiple feedback mechanisms. In
this game the player collects ‘key energy’ that is consumed in different measures
by different doors. In addition, the more energy players have, the more they can
see of their environment. At the same time, enemies become more aggressive
when the player has much energy. This means that players need to balance their
energy level carefully in order to progress through a level safely. From this experiment we learned that this type of construction works well, especially when the
‘key energy’ also taps into mechanisms, such as the visibility of the environment.
The game is still under development and we are still experimenting with more
mechanism that consume or are affected by ‘key energy’. For this experiment
the students built a Machinations diagram in order to grasp the game system,
and they found it very useful to balance the system but also to guide further
development (see figure 8.5).
Chapter 8 | Conclusions and Validation
207
Figure 8.4: The student game LKE experiments with feedback mechanisms for locks
and key. Game was built by Tim van Densen, Jerry Gomez and Martijn de Jong.
Figure 8.5: The machinations diagram for LKE, made by Tim van Densen.
208
8.5
Joris Dormans | Engineering Emergence
Academic and Industry Reception
The theory presented in this dissertation has grown over the last three years
and during that period I reported my progress through the publication and
presentation of ten peer-reviewed academic papers. In particular, Chapter 2
is based on a conference paper that I later turned into a journal paper (Dormans, 2008a, 2011a). The Machinations framework and its predecessors were
discussed in three conference papers (Dormans, 2008b, 2009, 2011d). The Mission/Space framework evolved while I was working on procedural content and
was introduced in a paper on that work (Dormans, 2010). The strategy for
integrating structures of emergence and progression outlined in Chapter 6 has
been accepted as a conference paper (Dormans, 2011b). The material presented
in Chapter 7 was earlier presented in three conference papers and one journal
article (Dormans, 2010; Bakkes & Dormans, 2010; Dormans & Bakkes, 2011;
Dormans, 2011c).
I also actively sought out people in the game industry and academia to test
my ideas through other channels. In some cases these people approached me
because they found my work on the Internet or were pointed in my direction
by others already familiar with this work. An important outlet for my theories
were Machinations design workshops I hosted at conferences and at different
companies. These workshops were not very different from workshops I hosted
as part of the game development courses at the Hogeschool van Amsterdam,
although the pace in these workshops was much higher. The participants of
these workshops needed some time to grasp the diagrams, but most of them
quickly understood what it was these diagrams try to convey and did see the
value of them. Many participants reported that they enjoyed the workshop and
that it brought into focus an aspect of game design that normally is not very
articulated. Derk de Geus, CEO of Paladin Studios states: “The Machinations
framework has been an invaluable tool to visualize our game’s mechanics. Using
Machinations’ systems thinking, we have been able to outline complex game
systems and identify the strong and weak points in the mechanics. I highly
recommend Machinations for designing games and other interactive systems. As
an example, I’m seriously considering to use it as a tool for business process
modeling” (2011, from personal correspondence).
The person that most strongly influenced the development of the Machinations tool is Stéphane Bura. He contacted me in early 2009 about my first
attempts to visualize game mechanics. As Bura himself has an interest in the
matter and contributed to the debate himself (2006, also see Chapter 3), he was
immediately enthusiastic about the premise of creating visual diagrams to represent internal economies of games. We met shortly afterwards to discuss our work
and he immediately encouraged me to develop a tool to express and to execute
the diagrams that later would become the Machinations diagrams. After each
iteration of the tool and framework Bura was one of the first people to comment
and to push me to develop features that would actually help designers to test
out designs. This resulted in the features that allow designers to gather data
from many simulated runs of which Bura stated: “This is exactly the kind of
Chapter 8 | Conclusions and Validation
209
tool I need to balance the economies of the games I’m working on” (2010, from
personal correspondence).
Experienced developers frequently see the educational value of the framework as its strongest point. As is made clear by Isaac Jeppsen, a system design
engineer at Aptima, Inc. who wrote: “I was ecstatic to find your paper on
Machinations. I’ve spent the past year at Aptima (a human factors engineering
company) trying to explain to them exactly what you illustrated so clearly in
your paper. After being a producer at a game development shop, I was continually frustrated by the lack of good tools to explain to my new academic
colleagues the essence of a game as the underlying systems that drive it” (2011,
from personal correspondence). This would indicate that the framework is more
useful for students and novice designers. In a way I agree with this sentiment.
Experienced designers need not rely on aids like these frameworks as much as
inexperienced designers do, just as programming paradigms and design patterns
seem second nature to experienced software developers. Although, I do like to
point out that adding the option to simulate games using these framework is a
more recent and deliberate attempt to increase the value of the frameworks for
more experienced designers.
Finally, during the 2010 G-Ameland game jam I had the opportunity to
show and discuss my work with game design veteran Ernest Adams. His books
on game design have had a strong influence on the Machinations framework,
as the notion of internal economy that lies at the heart of the Machinations
framework was conceptualized by Adams in the first place. Adams immediately
saw the potential of the diagrams: “Machinations is the best practical design and
testing tool for game mechanics that I’ve ever seen. It’s much more convenient
and intuitive than using spreadsheets or writing code” (2011, from personal
correspondence).
8.6
Omissions
The Machinations framework and the Mission/Space framework do not cover
all aspects of games. There is more to games than just mechanics. Art, artificial
intelligence, and interactive control schemes are some of the areas that have not
been discussed here, but that are relevant for game design. This dissertation
focused on mechanics, but not even all mechanics have been discussed in detail.
As indicated in chapter 1, this dissertation has zoomed in on the mechanics that
build a game’s internal economy and control level progression. Mechanics that
deal with physics, maneuvering, and social interaction are beyond the scope of
this dissertation. However, it is important to consider these omissions at this
point, as all of these type of mechanics can be the cause of interesting and
emergent gameplay.
The mechanics that govern a game’s physics are of a different nature than
those governing a game’s economy. Because physics mostly deal with continuous rules and simulation, the discrete representation of rules and mechanics
in the Machinations framework are a poor match. Likewise the graphs of the
Mission/Space framework do not match the continuous space made possible by
Joris Dormans | Engineering Emergence
210
the physics. To involve a game’s physics on the same level of abstraction as the
Machinations and Mission/Space framework, a new framework should be found
or created that can deal with that sort of rules easily and effectively. What
structural qualities such a framework should foreground I cannot tell. I suspect feedback structures will also affect physics. After all, continuous, emergent
behavior such as the flocking of birds also involves feedback. A formal framework for game physics is a tantalizing subject in and by itself, that requires
considerable research.
Something similar can be said of mechanics that govern maneuvering and
social interaction. Both comprise large fields of research, yet I do think that
in these areas great strides can be made by trying to identify those structural
qualities that lead to emergent gameplay. As it stands now, the theory presented
in this dissertation does little to explain the quality of a game like Go that seems
to derive its quality from maneuvering almost exclusively. I have tried, several
times, to create Machinations diagrams for Go but was never successful. Games
that rely a lot on maneuvering probably benefit from an approach based on
cellular automata. Most of these games have many similar functioning units
that do respond to their immediate surroundings.
There are also are some blind spots within the frameworks themselves. For
example, failure is something neither really takes into account. However, failure is a very common occurrence in games, even though it is often neglected in
many contemporary games (Juul, 2010). The frameworks, and especially mission graphs, can be changed so that failure to perform certain task becomes
an explicit option and integral part of the Mission/Space framework. As was
briefly mentioned in a footnote, relations that specify how failure, or inhibition
of crucial tasks, might be undone could be a great addition to the framework.
This way design strategies for games that better accommodate failure might be
developed.
8.7
Future Research
So, what is next? The development of the Machinations and Mission/Space
work with their associated tools and prototypes is not finished. I see six opportunities for future research and development:
1. The Machinations diagrams might be quite complete, but the research
into recurrent design patterns utilizing the framework is not. The patterns
described in appendix B are just the beginning. Many more patterns might
be found, developed and described. I would advocate that the total number
of patterns be kept low. This means that existing patterns in the library
should be reexamined from time to time. I already dismissed and merged
some patterns.
2. The Machinations tool might be expanded to allow users to embed diagrams in user defined elements to be reused in other diagrams. This would
expand the expressive power of the tool extensively. However, there are
some risks involved in exploring this direction. One of the strengths of
Chapter 8 | Conclusions and Validation
211
the current framework is the level of abstraction it enforces. By allowing
embedded diagrams, the focus on the larger structure might be lost.1 On
the other hand, it would enhance the capacity of the current diagrams
into something closer to a visual programming language specific for the
design of games. This might present opportunities for developing prototyping tools and programming environments dedicated to the generation
of games.
3. I already mentioned that I have used the frameworks and theory in the
development of game prototypes. Unfortunately, none of those games is
finished at the moment of writing. But that is because of a lack of time,
not a lack of applicability of these frameworks. I know I will continue to
work on these projects, but I also hope to inspire others to make use of
these techniques to create games with adaptable, dynamic gameplay from
which story-like progression emerges naturally.
4. Ludoscope, the automated design tool outlined in chapter 7, remains a prototype. I do think that Ludoscope can be developed into a fully-functional,
generic game design tool to assist game designers. This would require much
work, and plenty of research, but I do foresee that such a tool can be very
beneficial to the effectiveness and enjoyability of game design.
5. As suggested in the previous section, formal frameworks might be extended
with the development of frameworks for the mechanics of physics, maneuvering, and social interaction. This would allow us to approach more types
of games and tap into more sources to engineer emergent gameplay.
6. More work could be made of validating and refining the results of this
research. This can be done by further exploring how the applied theory presented here can be used to develop more advanced game design
courses. The Mission/Space framework has not seen as much action as
the Machinations framework. I do not expect that these frameworks have
been exploited entirely. In addition, more could be done to present these
theories to a wider, industry audience. That way, these frameworks might
gather more support and might evolve further to better reflect the industry’s needs.
8.8
Final Conclusions
Designing games remains a hard task, but gameplay can be engineered. The
applied theory presented in this dissertation can help designers to get a better
understanding of those structures in games that contribute to the creation of
emergent gameplay. It can assist them in their labor to design games, but it
can never relieve them of the responsibility that comes from the creative effort
required. I do not wish to prescribe the way designers should use this material.
1 This
was pointed out to me by Chris Lewis in a personal conversation.
212
Joris Dormans | Engineering Emergence
Since theories, frameworks, and automated design tools are only as effective as
their users, the real challenge lies still ahead.
The applied theory on game mechanics and level design aims to elevate game
development to a higher level of abstraction using clearly defined, formal perspectives. Assisted by the frameworks and the tools outlined in this dissertation,
designers can focus more on those aspects of the process that require their creativity and ingenuity. In the end, I like to think that I have supplied designers
with new, more powerful tools, with which their work might become more effective, but also more enjoyable.
Chapter 8 | Conclusions and Validation
213
214
Joris Dormans | Engineering Emergence
A
Machinations Overview
216
Joris Dormans | Engineering Emergence
B
Machinations Design Patterns
B.1
Static Engine
Type
Engine
Intent
Produce a steady flow of resources over time for players to consume or to collect
while playing the game.
Motivation
A static engine creates a steady flow of resources that never dries up.
Applicability
Use a static engine when you want to limit players’ actions without complicating
the design. A static engine forces players to think how they are going to spend
their resources without much need for long-term planning.
Structure
A static engine must provide players with some options to spend the resources
on. A static engine with only one option to spend the resources on is of little
use.
218
Joris Dormans | Engineering Emergence
Participants
• Energy - the resource produced by the static engine.
• Actions - a number of game actions the player can spend energy on.
Consequences
The production rate of a static engine does not change. Therefore the effects of
the engine on game balance are very predictable. A static engine can only be
the cause of imbalance when the production rate of a static engine is not the
same for all the players.
A static engine generally does not inspire long term strategies: collecting
resources from a static engine, if possible at all, is going to be quite obvious.
Implementation
Normally it is very simple to implement a static engine: a simple source that
produces the energy will suffice. It is possible to add multiple steps in the energy
production, but in general this will add little to the game.
A static engine can be made unpredictable by using some form of randomness
in the production. An unpredictable static engine will force the player to prepare
for periods of fewer resources and reward players that make plans that can
withstand bad luck. The easiest way to create an unpredictable static engine is
to use randomness to vary the output of resources or the time between moments
of production, but skill or multiplayer dynamics could work as well.
The outcome of random production rates can be, but does not need to be, the
same for every player. By using an unpredictable static engine that generates
the same resources for all players the luck factor is evened out without affecting
the unpredictability. This puts more emphasis on the planning and timing the
pattern introduces. An example would be a game where all players secretly
decide how many resources all players can get. The lowest number will be the
number of resources to enter play for everyone, while the players who offered the
lowest can act first. This would automatically set up some feedback from the
games current state to this mechanism.
Examples
In many turn based games the limited number of actions a player can perform
each turn can be considered a static engine. In this case the focus of the game
is the choice of actions and generally players cannot save actions for later turns.
The fantasy board game Descent: Journeys in the Dark uses this mechanism. Players can choose between one of four actions for their hero every turn:
move twice, attack twice, move once and attack once, or, move or attack and
prepare a special action (see figure B.1).
The energy produced by the spacecrafts in the Star Wars: X-Wing Alliance is an example of a static engine. The energy can be diverted to boost
Appendix B | Machinations Design Patterns
219
Figure B.1: Distribution of action points in the board game Descent: Journeys
in the Dark.
the player’s shields, speed and lasers. This is a vital strategic decision in the
game and the energy allocation can be changed at any moment. The amount
of energy generated every second is the same for all spacecraft of the same type
(see figure B.2).
Fearsome Floors employs an interesting variation of the static engine.
In this game the player controls a team of three or four characters that need to
escape the game board. Each turn each character has a limited number of action
points which he can use to move and push object around. The number of action
points oscillates between two values (that for all character add up to seven).
This makes the game surprisingly less predictable, and requires the player to
plan ahead at least one turn.
In Up the River players roll a die each turn and can move one of their ships
as many places closer to the harbor. Because the player controls three ships and
ships that are not moved eventually fall of the board, the decision of what ship
to move is not as straightforward as it might seem at first glance. Elements on
the board add extra complications that further emphasize planning and looking
ahead.
In Backgammon dice determine how far the player can move up to two
of his playing pieces, or up to four if he rolled a double. As the player has
fifteen playing pieces in total there are many options. In important strategy in
Backgammon is to move the playing pieces in such way that they are safe from
the opponent and give the player as many possible options for the next turn.
The Game of Goose or Snakes and Ladders are poor implementations
of an unpredictable static engine as both games lack multiple actions: the player
Figure B.2: Distribution of energy in Star Wars: X-Wing Alliance.
220
Joris Dormans | Engineering Emergence
does not have any choice and is left to the whims of chance.
Related Patterns
A (weak) static engine can prevent deadlocks in a converter engine.
B.2 Dynamic Engine
Type
Engine
Intent
A source produces an adjustable flow of resources. Players can invest resources
to improve the flow.
Motivation
A dynamic engine produces a steady flow of resources and opens the possibility
for long term investment by allowing the player to spend resources to improve
the production. The core of a dynamic engine is a positive constructive feedback
loop.
Applicability
Use a dynamic engine when you want to introduce a trade-off between long term
investment and short-term gains.
Structure
Participants
• Energy - the resource produced by the dynamic engine.
• Upgrades - a resource that affects the production rate of energy.
• Actions - a number of game actions the player can spend on, including:
Appendix B | Machinations Design Patterns
221
• Invest - an action that creates upgrades.
Collaborations
The dynamic engine produces energy that is consumed by a number of actions.
One action (invest) produces upgrades that improves the energy output of the
dynamic engine. A dynamic engine allows two different types of upgrades a
player can invest to improve:
• The interval at which energy is produced.
• The number of energy tokens generated each time.
The differences between the two are subtle, with the first producing a more
steady flow than the second, which will lead to bursts of energy.
Consequences
A dynamic engine creates a powerful positive constructive feedback loop that
probably needs to be balanced by some pattern implementing negative feedback,
such as any form of friction, or use escalation to create challenges of increasing
difficulties.
When using a dynamic engine one must be careful not to create a dominant
strategy either by favoring the long term strategy, or by making the costs for
the long term strategy to high.
A dynamic engine generates a distinct gameplay signature. A game that
consist of little more than a dynamic engine will cause the players to invest at
first, appearing to make little progress. After a certain point, the player will start
to make progress and needs to try and do so at the quickest possible pace. The
charts these patterns typically generate quite clearly show this behavior. This
effect of play is clearly visible in Monopoly (see figure B.3, see also section 4.8).
Implementation
The chance of building a dominant strategy that favors either long-term or shortterm investment is lessened when some sort of randomness is introduced in the
Figure B.3: The gameplay signature of a dynamic engine.
Joris Dormans | Engineering Emergence
222
Figure B.4: The harvesting of minerals in StarCraft.
dynamic engine. However, the positive feedback loop that exists in an unpredictable dynamic engine will amplify the luck a player has in the beginning of
the game, which might result in too much randomness quickly.
The outcome of random production rates can be, but does not need to be, the
same for every player. By using an unpredictable dynamic engine that generates
the same resources for all players the luck factor is lessened without affecting
the unpredictability. This puts more emphasis on the player’s chosen strategy.
Some dynamic engines allow the player to convert upgrades back into energy,
usually against a lower rate than the original investment. When upgrades are
expensive, and the player frequently needs large amounts of energy this becomes
a viable option.
Examples
In StarCraft one of the abilities of SUV units is to harvest minerals which can
be spend on creating more SUV units to increase the mineral harvest (see figure B.4). In its essence this is a dynamic engine that propels the game (although
in StarCraft the number of minerals is limited, and SUV units can be killed
by enemies). It immediately offers the player a long term option (investing in
many SUV units) and a short term option (investing in military units to attack
enemies quickly or respond to immediate threats).
Settlers of Catan has at its core a dynamic engine affected by chance (see
figure B.5). The roll of the dice which game board tiles will produce resources
at the start of each player’s turn. The more villages the player builds the more
chance that the player will receive resources every turn. The player can also
upgrade villages into cities which doubles the resource output of each tile. Settlers of Catan gets around the typical signature a dynamic engine creates
by allowing different types of invest actions, and using a measure of upgrades
instead of energy to determine the winner.
Related Patterns
The dynamic friction and attrition patterns are good ways to balance a dynamic
engine.
Appendix B | Machinations Design Patterns
223
Figure B.5: The production mechanism in Settlers of Catan.
B.3
Converter Engine
Type
Engine
Intent
Two converters set up in a loop create a surplus of resources that can be used
elsewhere in the game.
Motivation
Two resources that can be converted into each other fuel a feedback loop that
produces a surplus of resources. The converter engine is a more complicated
mechanism than most other engines, but also offers more opportunities to improve the engine. As a result a converter engine is nearly always dynamic.
Applicability
• Use the converter engine when you want to create a delicate mechanism to
provide the player with resources. It increases the difficulty of the game
as the strength and the required investment of the feedback loop are more
difficult to asses.
• Use the converter engine when you need multiple options and mechanics
to tune the signature of the feedback loop that drives the engine, and the
thereby the stream of resources that flow into the game.
224
Joris Dormans | Engineering Emergence
Structure
Participants
• Two resources: energy and fuel.
• A converter that changes fuel into energy.
• A converter that changes energy into fuel.
• Actions that consume energy.
Collaborations
The converters change energy into fuel and fuel into energy. Normally the player
ends up with more energy than the player started with.
Consequences
A converter engine introduces the chances of a deadlock, when both resources
dry out, the engine stops working. Players run the risk of creating deadlocks
themselves by forgetting to invest energy to get new fuel. Combine a converter
engine with a weak static engine to prevent this from happening.
A converter engine requires more work from the player, especially when the
converters need to be operated manually.
As with dynamic engines, a positive feedback loop drives a converter engine,
in most cases this feedback loop needs to be balanced by applying some sort of
friction.
Implementation
The number of steps involved in the feedback loop of a converter engine for a
large part determines its difficulty to operate efficiently: more steps increase
the difficulty, fewer steps reduce the difficulty. At the same time the number of
steps also positively affects the number of opportunities for tuning or building
the engine.
With too few steps in the system, the advantages of the converter engine are
limited and one might consider replacing it with a dynamic engine. Too many
Appendix B | Machinations Design Patterns
225
Figure B.6: The production mechanism in Power Grid. The converter engine is
printed bolder.
Figure B.7: Elaboration of a converter used in Power Grid.
steps might result in an engine that is cumbersome to operate and/or maintain,
especially in a board game where the different elements of the engine usually
cannot be automated.
It is possible to create an unpredictable converter engine by introducing randomness, multiplayer dynamics or skill into the feedback loop. This complicates
the converter engine further and often increases the chances of deadlocks.
Many implementations of the converter engine pattern put a limiter gate
somewhere in the cycle in order to keep the positive engine under control and
to keep the engine from producing to much energy. For example, if the number
of fuel resources that can be converted each turn is limited the maximum rate
at which the engine can run is capped.
Examples
A converter engine is at the heart of Power Grid (see figure B.6). Although
one of the converters is replaced by a slightly more complex construction (see
figure B.7). The players spend money to buy fuel from a market, and use that
fuel to generate money in power plants.1 The surplus money is invested in
more efficient power plants and connecting more cities to the player’s power
network. The converter engine is limited: the player can only earn money for
every connected city, which effectively caps the energy output during a turn,
and provides another opportunity for dynamic engine building. Power Grid
also has a weak static engine to prevent deadlocks: the player will collect a little
1 The fiction of the game is that players generate and sell electricity, but as electricity is not
a tangible resource in the game it is left out of this discussion.
226
Joris Dormans | Engineering Emergence
Figure B.8: Travel and trade in Elite. The player’s location on planet A or planet
B activates the converters that implement the trading mechanisms in the center. A
few possible ship upgrades are included on the right.
amount money during a turn even if the player failed to generate money through
power plants. The converter engine of Power Grid is slightly unpredictable as
players can drive up to price of fuel by stockpiling it, which acts as a stopping
mechanism at the same time.
The eighties computer space-trading game Elite features an economy that
occasionally acts as a converter engine. In Elite every planet has its own market
selling and buying various trade goods. Occasionally players will discover a
lucrative trade route where they can buy one trade good at planet A and sell it
at a profit planet B, and return with another good which is in high demand on
planet A again (see figure B.8). Sometimes these routes involve three or more
planets. Essentially such a route is converter engine. It is limited by the cargo
capacity of the player’s ship, which for a price can be extended. Other properties
of the player’s ship might also affect the effectiveness of the converter engine: the
ship’s “hyperspeed” range, and also its capabilities (or cost) to survive a voyage
through hostile territories all affect the profitability of particular trade routes.
Eventually trade routes become less profitable as the player’s efforts reduce the
demand, and thus the price, for certain goods over time (a mechanism that is
omitted from the diagram).
Related Patterns
A converter engine is well suited to be combined with the engine building pattern
as it has many opportunities to change settings of the engine: the conversion
rate of two converters and possibly the setting of a limiter.
A converter engine is best balanced by introducing some sort of friction or
some other form of negative feedback.
Appendix B | Machinations Design Patterns
B.4
227
Engine Building
Intent
A significant portion of gameplay is dedicated to building up and tuning an
engine to create a steady flow of resources.
Motivation
A dynamic engine, converter engine or a combination of different engines form
a complex and dynamic core of the game. The game includes at least one,
but preferably multiple, mechanics to improve the engine. These mechanics can
involve multiple steps. It is important that assessing the current state of the
engine is no trivial task.
Applicability
Use engine building when you want to create a game that:
• has a strong focus on building and construction.
• focuses on long-term strategy and planning.
Structure
The structure of the Core Engine is an example. There is no fixed way of
building the engine. Engine building only requires that several building mechanics operate on the engine and that the engine produces energy.
Participants
• The core engine usually is a complex construction combining multiple
engine types.
• At least one, but usually multiple building mechanisms to improve the
core engine.
• Energy is the main resource produced by the core engine.
228
Joris Dormans | Engineering Emergence
Collaborations
Building mechanisms increase the output of the engine. If energy is required
to activate building mechanisms then a positive, constructive feedback loop is
created.
Consequences
Engine Building increases the difficulty of a game, it is best suited for lower
paced games as it involves planning and strategic decisions.
Implementation
Including some form of unpredictability is a good way to increase difficulty,
generate varied gameplay and avoid dominant strategies. Engine building offers
many opportunities create unpredictability as the core engine tends to consist of
many mechanisms. The complexity of the core engine itself usually also causes
some unpredictability.
When using the engine building pattern with feedback it is important to
make sure the positive, constructive feedback is not too strong and not too fast.
In general, you want to spread out the process of engine building over the entire
game.
An engine building pattern operates without feedback when energy is not
required to activate building mechanisms. This can be viable structure when
the engine produces different types of energy that affect the game differently,
and allows the players to follow strategy that favor particular forms of energy
above others. However, it usually does require that the activation of building
mechanisms is restricted in some way.
Examples
SimCity is a good example of engine building. The energy in SimCity is money
which is used to activate most building mechanisms, which in the game consists of
preparing building sites, zoning, building infrastructure, building special buildings, and demolition. The core engine of SimCity is quite complex with many
internal resources such as people, job vacancies, power, transportation capacity,
and three different types of zones. Feedback loops within the engine cause all
sorts of friction and effectively balance the main positive feedback loop, up to
the point that the engine can more or less collapse if the player is not careful
and manages the engine well.
In the board game Puerto Rico each player builds up a New World colony.
The colony generates different types of resources that can be reinvested into
the colony or converted into victory points. The core engine includes many
elements and resources such as plantations, buildings, colonists, money and a
selection of different crops. Puerto Rico is a multiplayer game in which they
players compete for a limited number of positions which allow different actions
Appendix B | Machinations Design Patterns
229
to improve the engine; they compete for different building mechanics. This way
a strong multiplayer dynamic is created that contributes much of its gameplay.
Most real-time strategy games follow a complex engine building pattern combined with the attrition pattern.
Settlers of Catan follows an engine building structure that uses feedback
and multiple types of energy. In this particular case the energy produced by the
engine is never used outside the engine and its building mechanisms. This works
in Settlers of Catan because the objective of the game is to score a number
of points that are directly derived from the number of upgrades to the engine.
Related Patterns
Applying multiple feedback to the building mechanisms is a good way to increase
the difficulty of the engine building pattern.
All friction patterns are suitable to balance the possible positive feedback
that is created by an implementation of engine building that consumes energy
to activate building mechanisms.
B.5
Static Friction
Type
Friction
Intent
A drain automatically consumes resources produced by the player.
Motivation
The static friction pattern counters a production mechanism by periodically
consuming resources. The rate of consumption can be constant or subjected to
randomness.
Applicability
• Use static friction to create a mechanism that counters production, but
which can eventually be overcome by the players.
• Use static friction to exaggerate the long-term benefits from investing in
upgrades for a dynamic engine.
230
Joris Dormans | Engineering Emergence
Structure
Participants
• A resource: energy.
• A static drain that consumes energy.
• A production mechanism that produces energy.
• Other actions that consume energy.
Collaborations
The production mechanism produces energy players need to use to perform actions. The static drain consumes energy outside players direct control.
Consequences
The static friction pattern is a relative simple way to counter positive feedback
created by engine patterns. Although it tends to accentuate the typical gameplay
signature of the dynamic engine because it makes upgrades of the static engine
more valuable as the friction is not affected by the number of upgrades, current
energy levels, or players’ progress.
Implementation
An important consideration when implementing static friction is whether the
consumption rate is constant or subjected to some sort of randomness. Constant
static friction is the easiest to understand and most predictable, whereas random
static friction can cause more noise in the dynamic behavior in the game. The
latter can be good alternative to using randomness in the production mechanism.
The frequency of the friction is another consideration: when the feedback is
applied at short intervals the overall system will be more stable than when the
feedback is applied at long or irregular intervals, which might cause periodic
behavior in the system. In general, the effects of a gradual loss of energy on the
dynamic behavior of the system is less than the same amount of energy is lost
periodically.
Appendix B | Machinations Design Patterns
231
Figure B.9: Static friction in Monopoly.
Examples
In the Roman city building game Caesar III the player needs to pay tributes
to the emperor at particular moments during each mission. The schedule of the
tributes is fixed for each mission and not affected by the player’s performance.
In effect, they cause uses a very infrequent, and high form of static friction, that
causes a huge tremor in the game’s internal economy. See section 6.3 for a more
detailed discussion of this game.
The dynamic engine in Monopoly is countered by different types of friction,
including static friction (see figure B.9). The main mechanism that implements
static friction is the chance cards through which the player infrequently looses
money. Although some of these cards take into account the player’s property,
most of them do not. One could also argue that paying rent to other players
is also a form of static friction, as the frequency and severity is far beyond the
direct control of the player. But, the rate of the friction does change over time
and players have some indirect effect on it: when a player is doing well, chances
are that the opponents are not, which negatively impacts this friction. The
diagram in figure B.9 does not show this as it is made from the scope of an
individual player. This type of friction is an example of the attrition pattern.
Related Patterns
As static friction exaggerates long-term investments it is best suited to be used
with combination with a static engine, converter engine, or an engine building
pattern.
B.6
Type
Friction
Dynamic Friction
232
Joris Dormans | Engineering Emergence
Intent
A drain automatically consumes resources produced by the player, the consumption rate is affected by the state of other elements in the game.
Motivation
Dynamic friction counteracts production but adapts to the performance of the
player. Dynamic friction is a very typical application of negative feedback in a
game.
Applicability
• Use dynamic friction to balance games where resources are produced too
fast.
• Use dynamic friction to create a mechanism that counters production that
automatically scales with players’ progress or power.
• Use slow dynamic friction to reduce the effectiveness of long term strategies
created by a dynamic engine in favor for short term strategies.
Structure
Participants
• A resource: energy.
• A dynamic drain that consumes energy.
Appendix B | Machinations Design Patterns
233
• A production mechanism that produces energy.
• Other actions that consume energy.
Collaborations
The production mechanism produces energy players need to use to perform actions. The dynamic drain consumes energy outside players direct control, but
affected by the state of at least one other element in the game system.
Consequences
Dynamic friction is a good way to counter positive feedback created by engine
patterns. Dynamic friction adds a negative feedback loop to the game system.
Implementation
There are several ways of implementing dynamic feedback. An important consideration is the choice of the element that causes the consumption rate to change.
In general, this can either be the energy itself, the number of upgrades to a
dynamic engine or a converter engine, or the player’s progress towards a goal.
When energy is the cause for the friction to change, the negative feedback tends
to be fast, when progress or production power is the cause the feedback is more
indirect, and probably slower.
When dynamic friction is used to counter a positive feedback loop it is important to consider the difference in characteristics of the positive feedback loop and
the negative feedback loop implemented through the dynamic friction. When
the characteristics are similar (equally fast, equally durable, etc.), the effect is
far more stable that when the differences are large. For example, when a slow
and durable dynamic friction is acting against a fast but non-durable positive
feedback that initially yields a good return, players will initially make a lot
of progress, but might suffer in the long run. Fast positive feedback and low
negative feedback seems to be the most frequently encountered combination.
Examples
Dynamic friction is used the city production mechanism in Civilization (see
figure B.10). In this game the player builds cities to produce food, shields and
trade. As cities grow they need more and more food for their own population.
Players have some control over how much food is produced compared to other resources, but are limited in their options by the surrounding terrain. By choosing
to produce a lot of food cities initially produce fewer other resources, but grow
faster to great potential. Fast growth creates a problem however, the happiness
rating a city must stay equal to or higher than half the population, else the
production stops because of civil unrest. Initially a city has a happiness value of
2. Players can create more happiness by building special buildings, or by converting trade into culture. Both ways cause more dynamic friction with different
234
Joris Dormans | Engineering Emergence
Figure B.10: The city economy of Civilization. Dynamic friction is printed bolder.
The player can freely adjust the culture and research settings in order to control unrest
and research production. These settings are global and affect all cities equally.
profiles on the production process; the first is slow, requires a high investment,
but is highly durable, and has relative high return, whereas the second is fast,
but has a relative low return for its investment.
Related Patterns
Dynamic friction is a good way to balance any pattern that causes positive
feedback, and often is part of the multiple feedback pattern.
B.7 Attrition
Type
Friction
Intent
Players actively steal or destroy resources of other players that they need for
other actions in the game.
Appendix B | Machinations Design Patterns
235
Also Known As
Multiplayer, destructive feedback
Motivation
By allowing player to directly steal or destroy each others resources, players can
eliminate each other in a struggle for dominance.
Applicability
• Use attrition to allow direct and strategic interaction between multiple
players.
• Use attrition to introduce feedback into a system which nature is determined by the strategic preferences or whims of the players.
Structure
Participants
• Multiple players that have the same (or similar) mechanics and options.
• A resource used by all players: energy to perform actions.
• A special attack action that triggers or activates a drain of other player’s
energy.
Collaborations
By performing attack actions players can drain each other’s energy. Attacking
typically costs some sort of energy too, or at least attacking players can spend less
time on other actions. The balance between the attack costs, how much energy
players loose to an attack, and how beneficial the other actions are, determine
the effectiveness the attack and the pattern.
Consequences
Attrition introduces a lot of dynamism into a system as players directly control
the measure of the destructive force applied to each other. Often this introduces
destructive feedback as the current state of a player will cause reactions by other
236
Joris Dormans | Engineering Emergence
players. Depending on the nature of the winning conditions and the current state
of the game this feedback might be negative when it stimulates players to act
and conspire against the leader, but it also might cause positive feedback when
players are stimulated to attack and eliminate weaker players.
Implementation
For attrition to work it is best that players must invest some sort of resource into
attack that could also be spend otherwise. Without this investment attrition
simply becomes race to destroy each other with little or no strategic choices.
Although this might work when there are multiple players are involved and the
game facilitates social interaction between players. In this case the players need
to chose whom to attack.
It is quite common to implement attrition using two resources to represent
life and energy. In this case players use energy to perform actions and loose when
they run out of life. When using these two resources it is important that they
are somehow related, often players are allowed to spend energy to gain more life.
Sometimes the relation between life and energy is implicit, for example when a
player must choose between spending energy or gaining life, there is an implicit
link between the two as players generally cannot do both at the same time.
In a two player version of attrition the game must include other actions, and
games for more than two players often also allow other actions to be performed
by the player. Most of the time a these actions constitute some sort of production mechanism for energy, increases the effectiveness the players defensive or
offensive capabilities. Most real-time-strategy games include all these options,
often with multiple variants for each.
The winning conditions and effects of eliminating another player have a big
impact on the attrition pattern. The winning condition does not need be the
elimination of players, players might score points, or reach a particular goal
outside the attrition pattern which automatically widens the number of strategies
involved. When there is a bonus to attacking or eliminating players, the pattern
can be made to stimulate weaker players.
Examples
Almost all strategy game implement some sort of attrition as it is often an
important goal to eliminate other players in this type of games. The SimWar
example, discussed in Chapter 4, provides a good example.
The trading card game Magic the Gathering implements a decorated
version of the attrition pattern. Figure figure B.11 presents this implementation,
although it show the details from the perspective of a single player only. In
Magic the Gathering, players can play one card every turn, these cards
allow players to add lands, summon creatures or cast spells to heal, or deal
direct damage to their opponent or their opponent’s creatures. But all actions,
except playing lands costs mana, the more mana players have the more they can
spend each turn and the more powerful actions they can play. Creatures will fight
Appendix B | Machinations Design Patterns
237
Figure B.11: The attrition mechanism in Magic the Gathering.
other creatures and when there are no more enemy creatures they will damage
the opponent directly. Players that loose all their life points are eliminated from
the game. gameMagic the Gathering is an example of a game that implements
attrition using separate resources for life and energy (in this case life and mana).
The different options illustrate how attrition can work differently: direct damage
activates a trigger briefly. As its name implies, it is fast and direct. On the
other hand, summoning creatures activates a permanent drain on the opponent’s
creatures and life. The effects usually are not as powerful as direct damage, but
as they accumulate over time they can be quite devastating. What options are
available to a player and how powerful these options are exactly, is determined
by the cards in the player’s hand. As players build their own decks from a
large collection of cards, deck building is an important aspect of Magic the
Gathering.
The most obvious way to implement attrition is in a symmetrical game. However, many single-player games and even certain types of multiplayer games use
‘a-symmetrical’ attrition. An example of ‘a-symmetrical’ attrition can be found
in the board game Space Hulk where one player, controlling the a handful of
‘space marines’, tries to accomplish a mission while the other player, controlling
an unlimited supply of alien ‘Genestealers’ tries to prevent that. The genestealer
player tries to reduce the number of space marines to stop them from accomplishing their goals, they win when they have destroyed enough space marines.
On the other hand, the space marines usually cannot win by destroying a certain
number of genestealers, but they must keep the number of genestealers under
control in order to survive, as the genestealers become more effective as their
numbers grow. Figure B.12 is a rough illustration of these mechanics in Space
Hulk.
Joris Dormans | Engineering Emergence
238
Figure B.12: A-symmetrical attrition in Space Hulk.
Related Patterns
Attrition works well with any sort of engine pattern. Trade can be used as an alternative form of multiplayer feedback that is constructive instead of destructive,
and nearly always negative.
B.8 Stopping Mechanism
Type
Friction
Intent
Reduce the effectiveness of a mechanism every time it is activated.
Motivation
In order to prevent a player from abusing a powerful mechanism its effectiveness is reduced every time it is used. In some cases the stopping mechanism is
permanent, but usually its not.
Applicability
Use stopping mechanism to:
• prevent players from abusing particular actions.
• counter dominant strategies
• reduce the effectiveness of a positive feedback mechanism.
Appendix B | Machinations Design Patterns
239
Structure
Participants
• An action that might produce some sort of output.
• A resource energy that is required for the action.
• The stopping mechanism that increases the energy cost or reduces the
output of the action.
Collaborations
For a stopping mechanism to work it the action must have an energy cost, an
output, or both. The stopping mechanism reduces the effectiveness of an action
mechanism every time it is activated by increasing the energy costs or reducing
the output.
Consequences
Using a stopping mechanism can reduce the effect a positive feedback loop considerably and even make its return insufficient.
Implementation
When implementing a stopping mechanism is important to consider whether or
not to make the effects permanent or not. When the accumulated output is used
to measure the strength of the stopping mechanism the effects are not permanent, in that case it requires players to alternate frequently between creating the
output and using the output in other actions.
A stopping mechanism can apply to each player individually or can affect
multiple players equally. In the latter case, players will reward players that use
the action before other players do. This means that the stopping mechanism
can create a form of feedback depending on whether leading or trailing players
are likely to act first.
Examples
The price mechanic of the fuel market in Power Grid involves a stopping
mechanism (see figure B.13). In Power Grid players use money to buy fuel
240
Joris Dormans | Engineering Emergence
Figure B.13: The stopping mechanism in Power Grid drives up the price for fuel
and causes negative feedback, especially for leading players.
and burn fuel to generate money. This positive feedback loop is counteracted by
the fact that buying a lot of fuel actually drives up the price for all players. As
in Power Grid the leading player acts last, this stopping mechanism causes
more negative feedback for the leading player.
Related Patterns
A stopping mechanism is often used as a way to implement multiple feedback.
B.9 Multiple Feedback
Intent
A single gameplay mechanism feeds into multiple feedback mechanisms, each
with a different signature.
Motivation
An player action activates multiple feedback loops at the same time. Some
feedback loops will be more obvious than others. This creates a situation where
the exact outcome or success of an action might be predictable on the short
term, but can have unexpected result on the long run.
Applicability
Use multiple feedback to:
• increase a game’s difficulty.
• emphasize the player’s ability to read the current game state.
Appendix B | Machinations Design Patterns
241
Structure
In the example structure above there are two feedback mechanisms. The
action (black) activates one feedback mechanism (red) which is positive, fast
and direct, but it also activates a secondary feedback mechanism (blue) which is
negative, slow and indirect. This illustrates just one way of setting up multiple
feedback loops. There are many more.
Participants
• An action that can be activated by the player.
• Multiple feedback mechanisms that are activated by the action.
Collaborations
The action activates multiple feedback mechanisms that ultimately feedback into
the action.
Consequences
For the player multiple feedback loops are more difficult to understand that single
feedback loops. As a result using this pattern makes a game more difficult.
If the feedback loops the action activates can have dynamic signatures that
change during play (which they often have) it is very important for the player to
be able to read the current signature, as their balance might shift considerably
during the game.
Finding the right balance between the multiple feedback loops is an important
issue in a game that uses this pattern.
Implementation
When creating a game with multiple feedback it is very important to make sure
that the signatures of the different feedback loops are different. In particular
the speed of the feedback needs to vary if this pattern is going to be effective.
Alternatively varying the signature of the feedback over time can also work well.
To this end adding playing style reinforcements and stopping mechanisms to one
or more of the feedback loops is a good design strategy.
The most common combination for multiple feedback seems to be fast, constructive, positive feedback coupled with slow, negative feedback. This creates
a trade-off between short term gains and long term disadvantages.
Joris Dormans | Engineering Emergence
242
Examples
Attacking in Risk feeds into three positive feedback loops of varying speeds and
strengths. The most obviously using armies to capture more lands allows the
player to build more armies. A slower form of feedback is realized by the cards: a
player that successfully attacks gains a card, certain combination of three cards
allow him to gain extra armies. The last type of feedback is created by the
capturing of continents. A continent will give a player a number of bonus armies
each turn, this is very fast and strong feedback loop, but one that requires a
higher investment of the player.
Related Patterns
Playing style reinforcements and stopping mechanisms are good ways to ensure
that the signature of the feedback loops an action feeds into change over time.
B.10
Trade
Intent
Allow trade between players to introduce multiplayer dynamics and negative,
constructive feedback.
Motivation
Players are allowed to trade important resources. Usually this means that leading players will get tougher deal while trailing players can help each other to
catch up. Trade works especially well when the flow of resources is unstable
and/or not equal distributed among players.
Applicability
Use trade to:
• introduce multiplayer dynamics to the game.
• introduce negative, constructive feedback.
• introduce a social mechanic that involves players outside their own turns.
Appendix B | Machinations Design Patterns
243
Structure
Participants
• A trading mechanism that allows trade of resources.
• Multiple tradable resources that can spent in various ways.
• Actions that require the use of tradable resources.
Collaborations
The tradable resources can be exchanged by the players using the trading mechanism.
Consequences
Trade introduces negative feedback that does not really slow down the game but
usually helps trailing players catch up, (as it is not destructive).
Trade favors players with good social and bartering skills.
Implementation
In board games trade is very easy to implement, the game simply needs to
specify how and when players can trade resources. In a multiplayer computer
game trade is also easy. However, creating a trading mechanism that involves
artificial players, is far from trivial.
To implement a successful trading mechanism, multiple tradable resources are
required, and the production rates of these resources must fluctuate or at least
be different among players. Trading only works when there is an imbalance in
the distribution of resources among the trading parties. It also helps to include
many actions that consume the tradable resources and to create actions that
consume resources of mutliple types, as this futhere exaggerates the imbalance
when players choose different courses of action.
Examples
In Settlers of Catan players build up an uncertain dynamic engine: villages
and cities that produce the resources used to build more villages and cities. The
Joris Dormans | Engineering Emergence
244
randomness of these engines is partly countered by allowing all players to trade
resources with the player that is taking his or her turn. The exchange rate is set
by in mutual agreement and usually determined by the availability, accessibility
of the resource and the position of the player. Players that are in the lead can
afford to pay more for their resources. When close to winning players might find
it impossible to make a deal.
In Civilization III players can exchange strategic resources, money, and
knowledge. This mutually benefits both parties and allows weaker civilizations
to catch up rather quickly.
Related Patterns
Attrition can be an alternative source of multiplayer feedback that is not constructive but destructive in nature.
B.11
Escalating Complications
Type
Escalation
Intent
Player progress towards a goal increases the difficulty of further progression.
Motivation
A positive feedback loop on the game’s difficulty makes the game increasingly
harder for players as they get closer to achieving their goals. This way the game
quickly adapts to the player’s skill level, especially when the good performance
allows the player to progress more quickly.
Applicability
Escalating complications is well suited for fast-paced games based on player skill
where the game needs to adjust quickly to the player’s skill level.
Structure
Appendix B | Machinations Design Patterns
245
Participants
• Targets represent unresolved tasks.
• Alternatively, progress represents the player’s progress towards a goal.
• A task that either reduces the number of targets or produces progress.
• A feedback mechanism that makes the game more difficult as the player
progresses towards the goal or reduces the number of targets.
Consequences
Escalating complications is based on a simple positive feedback loop affecting the
difficulty of the game. It is mechanism quickly adjusts the difficulty of the game
to the skill level of the player. If failure of the task ends the game escalating
complication ensures a very quick game.
Implementation
The task in a game that only implements the escalating complication pattern is
typically affected by player skill, especially when the escalating complications is
makes up the most of a game’s core mechanics. When the task is a random or
deterministic mechanic, players will have no control over the game’s progress.
Only when the escalating complication pattern is part of a more complex game
system, where players have some sort of indirect control over the chance of
success, a random or deterministic mechanic becomes viable. Using multiplayer
dynamic mechanisms is an option but probably works better in a more complex
game system as well.
Examples
Space Invaders is a classic example of the escalating complication pattern. In
Space Invaders the player needs to destroy all invading alien before they can
reach the bottom of the screen. Every time the player destroys an alien all other
aliens speed up a little, making it more difficult for the player to shoot them.
Pac-Man is another example. In Pac-Man the task is to eat all the dots
in a level, while the chasing ghosts further increase the difficulty of the task and
adding to strategic decision to eat an ‘energizer’ to get rid of them.
Related Patterns
By combining static friction or dynamic friction with escalating complications
a game can be created that quickly matches its difficulty to the ability of the
player.
Joris Dormans | Engineering Emergence
246
B.12
Escalating Complexity
Type
Escalation
Intent
Players act against growing complexity, trying to keep the game under control
until positive feedback grows to strong and the accumulated complexity makes
them loose.
Motivation
Players are tasked to perform an action that grows more complex if the players
fail and which complexity contributes to the difficulty of the task. As long as
players can keep up with the game they can keep on playing, but once the positive
feedback spins out of control, the game ends quickly. As the game progresses
the mechanism that creates the complexity gains speeds up ensuring that at one
point players can no longer keep up and eventually must loose the game.
Applicability
Use escalating complexity when you aim for an addictive skill based game.
Structure
task
loose
>n
*
+-
n%
n
+
complexity
*
progress mechanism
Participants
• The game produces complexity that must be kept under under a certain
limit by the player.
• The task is a player action that reduces complexity.
• Progress mechanism increases the production of complexity over time.
Appendix B | Machinations Design Patterns
247
Collaborations
Complexity immediately increases the production of more complexity, creating
a strong positive feedback loop that must be kept under control. The player
looses when complexity exceeds a particular limit.
Consequences
Given enough skill players can keep up with the creation of complexity for a long
time, but when players no longer keeps up complexity spins out of control and
the game ends quickly.
Implementation
The task in a game that implements the escalating complexity pattern is typically
affected by player skill, especially when escalating complexity makes up most
of the game’s core mechanics. When the task is a random or deterministic
mechanic, players will have no control over the game’s progress. Only when the
escalating complexity pattern is part of a more complex game system, where
players have some sort of indirect control over the chance of success, a random
or deterministic mechanic becomes viable. Using a multiplayer dynamic task is
an option but probably also works better in a more complex game system.
Randomness in the production of complexity creates a game with a varied
pace, where players might struggle keeping up with a peek in production before
catching some breath when production dies out a little.
There are many ways to implement the progress mechanic, from a simple
time based increase of the production of complexity (as is the case in the sample
structure above) to complicated constructions that rely on other actions of the
player or other player actions. This way it is possible to combine escalating
complexity with escalating complication by introducing positive feedback to the
progress mechanic as a result of the execution of the task.
Escalating complexity lends itself well to combine with a multiple feedback
structure where the complexity feeds into several feedback loops with different
signatures. For example, escalating complexity can be partially balanced by having the task feeding into a much slower negative feedback loop on the production
of complexity.
Examples
In Tetris a steady flow of falling bricks causes complexity. There is a slight
randomness in this production as the different blocks are created all the time.
Player need to place the blocks in such way that they fit together closely. When
a line is completely filled it disappears making room for new blocks. When
players fail to keep up the blocks pile up quickly and they will have less time
to place subsequent blocks, this can quickly increase the complexity of the field
when players are not careful and causes them to loose the game if the pile of
Joris Dormans | Engineering Emergence
248
blocks reach the top of the screen. The game progressively speeds up by making
the blocks fall faster and faster, making it more and more difficult.
Related Patterns
Any type of engine pattern can be used to implement the progress mechanism.
B.13
Playing Style Reinforcement
Intent
By applying slow, positive, constructive feedback on player actions, the game
gradually adapts to the players preferred playing style.
Also Known As
RPG-elements (as in ‘a game with RPG-elements’)
Motivation
Slow, positive, constructive feedback on players’ actions that also have another
game effect causes the players’ avatar or units to develop over time. As the
actions themselves feed back into this mechanism the avatar or units specialize
over time, getting better in a particular task. As long as there are multiple
viable strategies and specializations, the avatar and the units will, over time,
reflect the player’s preferences and style.
Applicability
Use Playing Style Reinforcement when:
• you want players to make a long-term investment in the game that spans
multiple sessions.
• you want to reward players for building, planning ahead, and developing
personal strategies.
• you want players to grow into a specific role or strategy.
Appendix B | Machinations Design Patterns
249
Structure
Participants
• Actions players can perform which success is not completely dependent
on the players skill.
• A resource ability that affects the chance actions succeed and than can
grow over time.
• An optional resource Experience points that can be used to increase an
ability.
Collaborations
Ability affects the success rate of actions.
Preforming actions generates experience points or directly improves abilities.
Some games require the action to be successful, while others do not.
Experience points can be spent to improve abilities.
Consequences
Playing style reinforcement works best in games that are played over multiple
sessions and over a long time.
Playing style reinforcement only works well when multiple strategies and play
styles are viable option in the game. When there is only one, or only a few, every
one will go for those options.
Playing style reinforcement can inspire min-maxing behavior with players.
With this behavior players will seek the best possible options that will allow
them to gain powerful avatars or units as fast as possible. This can happen
250
Joris Dormans | Engineering Emergence
when the strength of the feedback is not distributed evenly over all actions and
strategies.
Playing style reinforcement favors experienced players over inexperienced
players, as the first group will have a clearer view over all the options and
long-term consequences of their actions.
Playing style reinforcement rewards the player who can invest the most time
in playing the game. In this case, time can compensate for playing skill, which
can be a wanted or unwanted side-effect.
It can be very ineffective to change strategies over time in a game with
playing style reinforcement, as previous investments in another play style will
not be used as efficiently.
Implementation
Whether or not to use experience points is an important decision when implementing play style reinforcement. When using experience points there is no
direct coupling between growth and action, allowing the player to harvest experience with one strategy to develop the skills to excel in another strategy.
On the other hand, if you do not use experience points you have to make sure
that the feedback is balanced for the frequency of the actions: actions that are
performed more often should have weaker feedback that actions that can be
practiced infrequently.
Role-playing games are the quintessential games build around the play style
reinforcement pattern. In these games the feedback loops are generally quite
slow, and balanced by dynamic friction or a stopping mechanism to make sure
avatars do not progress to fast. In fact, most of these games are balanced in such
way that progression is initially fast and gradually slows down, usually because
the required investment of experience points increases exponentially.
Whether or not the action needs to be executed successfully to generate the
feedback is another important design decision and can inspire totally different
player behavior. When success is required, the feedback loop gains influence.
In that case it is probably best to have tasks difficulty also affect the success
of an action, and to challenge the player with tasks of different difficulty levels
allowing them to train their avatars. When success is not required, players
have more options to improve neglected abilities during later and more difficult
stages. However, it might also inspire players to perform a particular action at
every conceivable opportunity, which could lead to some unintended unrealistic
or comic results, especially when the action involves little risk.
Examples
In the board game Blood Bowl players coach football teams in a fantasy setting. Individual team members score ‘star player points’ for successful actions:
scoring touch downs, throwing complete passes or injuring opposing players.
When a team member collected enough star player points he gains new or improved abilities. Many of these increase their ability to score, pass or injure
Appendix B | Machinations Design Patterns
251
opponents. Improvements occur only between matches and players build up a
team over a long series of matches. Blood Bowl facilitates a wide variety of
playing styles that generally fall somewhere between two poles: agility play with
a strong focus on ball handling and scoring, and strength play with a strong
focus on taking out the opposition in order to win the game.
In computer role-playing game The Elder Scrolls IV: Oblivion the
player avatar’s progress is directly tied to their actions. The avatar’s ability
corresponds directly to number of times he performed the associated actions.
In Civilization III there are different ways in which a player can win the
game. A player reinforces his chosen strategy of military, economic, cultural or
scientific dominance (or any combination), by building city improvements and
world wonders that favor that strategy. In Civilization III several resources
take the role of experience points: money and production among them. These
resources are not necessarily tied to one particular strategy in the game: money
generated by one city can be spent to improve production in another city in the
game.
In the board game Caylus players occasionally win a privilege. They can
choose to use the privilege to gain money, points, building resources, or the
opportunity to build new structures. Every time they pick one of them the
option is improved (the first time you get 1 point, the second time 2 points,
etc). Rules are in place to prevent a player from spending privileges on the
same option to often. This is an example of playing style reinforcement without
experience points.
Related Patterns
When playing style reinforcement depends on the success of actions, it creates a
powerful feedback. In that case a stopping mechanism is often used to increase
the price of new upgrades to an ability.
252
Joris Dormans | Engineering Emergence
C
A Recipe for Zelda-esque Levels
The following ‘recipe’ describes the formal grammar and rewrite systems to
generate levels in style of The Legend of Zelda.
Step 1: generate basic mission
This step creates a basic mission. It applies the following rules once: entrance and goal; add key item; add level boss; add mini boss; add master key;
add guardian; lock mini boss. Next it applies the rule ‘expand act 2’ between
somewhere between 3 and 5 times. Finally it applies the rule ‘expand act 3’
between somewhere between 4 and 6 times. Sample outcome of this step:
254
Joris Dormans | Engineering Emergence
Step 2: add locks
This step adds a few extra locks to the mission. It applies its single rule
between 1 and 3 times. Sample outcome of this step:
Step 3: complicate locks
This step refines the existing locks by duplicating keys and locks. Rules are
Appendix C | A Recipe for Zelda-esque Levels
255
applied until all options are exhausted. Sample outcome of this step:
Step 4: transform into space
This rewrite system transforms the mission graph into an almost isomorphic
space graph. Rules are applied until all options are exhausted. Sample outcome
of this step:
256
Joris Dormans | Engineering Emergence
Step 5: move locks
This step moves locks towards the entrance. It applies its single rule between
6 and 11 times. Sample outcome of this step:
Step 6: expand space
This steps adds a few rooms to the structure. It applies rules randomly
between 5 and 9 times. Sample outcome of this step:
Appendix C | A Recipe for Zelda-esque Levels
257
Step 7: clean up locks
This rewrite systems makes sure locks are no longer directly connected to
other locks. Rules are applied until all options are exhausted. Sample outcome
of this step:
258
Joris Dormans | Engineering Emergence
Step 8: create short cuts
This step introduces short cuts to allow the player to quickly return to a lock
after finding a key. It applies the rule ‘create short cut’ between 2 and 3 times,
then it applies the rule ‘move short cut’ between 0 and 3 times. Finally it applies
the rule ‘replace unlock-loops’ as many times as possible. Sample outcome of
this step:
Summary
Game design theories aim to assist game developers in creating and understanding games. Applied theory for game design supports creativity, speeds up the
design process, and increases quality of games. This dissertation presents two
theoretical frameworks for designing games and discusses how designers can use
them to refine the game development process.
Currently there are several design theories for games. However, none of these
theories has surfaced as a widely accepted standard within the game industry.
Most existing (often academic) design theories have a poor track record. There
are two reasons for this. One: they require a considerable investment to learn,
and two:, they are more successful as analytical tools than as actual development
tools. It should come as no surprise that this has caused considerable skepticism
towards design theory within the game industry.
With these concerns in mind, the frameworks presented in this dissertation
were designed to be practicable and applicable. The first framework frames
games as rule-based systems. It helps to identify and understand the structure
of game rules in order to facilitate design. The second framework helps designers
to create open, dynamic worlds that to provide players with a smooth gameplay
experience. The two frameworks are not completely separate: in a good game,
game worlds and rules combine into a complex, carefully balanced whole. One
cannot do without the other.
The first framework that focuses on game rules is called Machinations and it
utilizes dynamic, interactive diagrams. These diagrams model a game’s internal
economy that consists of the flow of vital, and sometimes abstract, resources.
Resources can be of many types, for example: points, money or ammunition.
The most important structural features these diagrams capture are feedback
patterns. Feedback is generally recognized to play an important role in emergent
behavior in complex systems in general, and games in particular. Feedback
occurs when state changes in a game element affect other elements and ultimately
feed back to affect the first element. In games, feedback often creates an upward
spiral. For example, in the classic board game Monopoly investing money in
property generates more money to invest.
The Machinations framework distinguishes many different types of feedback:
feedback can be positive or negative, constructive or destructive, fast or slow,
etcetera. For designers it is important to understand the nature of the feedback
loops that operate in their games. What is more, from the study of several games
and their internal economy recurrent patterns emerge. These patterns codify
general solutions to commonly encountered design challenges. Machinations
diagrams are an excellent vehicle to identify and reason about these patterns.
Game designers can create Machinations diagrams with the software developed for this study. The main advantage of this software is that these diagrams
can be run: like the games they model, Machinations diagrams are dynamic
and interactive. This allows designers to simulate games easily and efficiently;
it allows them to balance and tweak a game’s internal economy for particular
effects long before a prototype is built.
260
Joris Dormans | Engineering Emergence
The second framework, Mission/Space, focuses on level design and the mechanics that control player progression through a game. Unlike existing models
of level design, the Mission/Space framework capitalizes on the idea that in a
game level two different structures exists. Mission graphs are used to formalize
the structure of the tasks and challenges laid out for players, while space graphs
are used to formalize the geometric construction of game levels. These two constructions are related, but they are not the same. Rather, the different ways
missions might be mapped onto a particular space play an important role in
the resulting player experience. Mechanism for locks and keys, which are often
used to prevent players from reaching certain areas in a level too quickly, are an
common tool to create mission/space structures.
As with the Machinations framework, the Mission/Space framework was used
to develop software tools to assist designers in creating levels. In this case, the
program Ludoscope allows designers to implement mission graphs and space
graphs and simulate a player’s progress through a level. This way Ludoscope
helps to identify potential deadlock, or other flaws, in their design.
The Machinations framework focuses on structures of emergence in games
and the the Mission/Space framework focuses on structures of progression. Combining the two frameworks provides leverage to investigate how emergence and
progression can be combined. Many games combine elements of both. However,
there also are many games where there is a considerable mismatch between the
sophistication of the mechanics in their economy, space and physics on the one
hand, and the mechanics to control progression on the other. By combining the
Machinations framework with the Mission/Space framework, two new strategies
for integrating emergence and progression surface.
The first strategy zooms in on lock and key mechanisms that rarely include
feedback mechanisms. Machinations diagrams can be used to explore more sophisticated and emergent mechanics to create locks and keys, and to control
player progress. In the second strategy progress itself is modeled as a resource
allowing it to become part of feedback mechanisms that operate on the scale of
the entire level or game.
The Machinations framework and the Mission/Space framework formalize
different perspectives on games. The notion of ‘model driven engineering’, taken
from software engineering, can unify these different perspectives into one formal
approach to game development. This perspective ultimately allows us to automate certain aspects of creating games. Using formal grammars and rewrite
systems it is possible to codify transformations that allow designers to generate
spaces from mission, mechanics from spaces, or missions from mechanics. By
dividing the development process into a series of small transformations, flexible
tools can be created that can (help to) automatically generate entire game levels,
and, to a certain extent, entire games. Another application of these techniques
is the development of games that adapt to player performance.
To implement these techniques, the Ludoscope prototype was extended to
allow designers to define such grammars and rewrite systems. This creates a
potentially very powerful, generic game development tool that allows games to
be designed at a high level of abstraction.
Summary
261
The various tools developed as part of this research are one way to validate
the applicability of the theory developed in this dissertation, but it is not the
only one. Through various presentations and workshop, academy peers and industry veterans were able to comment on and influence the development of the
frameworks and tools. In addition, the frameworks, and Machinations in particular, were used in the game development courses taught at the Hogeschool
van Amsterdam. We found that these frameworks help students to understand
games and improve their designs. Mainly because they formalize important aspects of game design that are not immediately apparent or tangible, and because
they facilitate students to simulate and experiment with these aspects.
In the end, designing games remains a tough challenge. The tools and theory
presented in this dissertation assist game designers by codifying design lore,
simulating games before they are built, and even by automating certain aspects
of the design, but they can never replace the human creativity and ingenuity
that goes into their design. However, I do think that these tools and theory will
assist them and can elevate the art of game design to the next level.
262
Joris Dormans | Engineering Emergence
Nederlandse Samenvatting
Game ontwerptheorieën helpen game designers bij het maken en begrijpen van
hun werk. Zulke toegepaste theorieën stimuleren het creatieve proces en kunnen de ontwikkelkosten van games drukken, het ontwerpproces versnellen en de
kwaliteit van de eindproducten verbeteren. Dit proefschrift presenteert twee theoretische kaders voor het ontwerpen van games en beschrijft hoe game designers
deze kunnen inzetten om het game ontwerpproces te stroomlijnen.
Er bestaan op dit moment meerdere ontwerptheorieën voor games, maar geen
enkele kan rekenen op een breed draagvlak binnen de game industrie. Vooral
academische ontwerptheorieën hebben regelmatig een slechte reputatie. Daar
zijn twee redenen voor. Ten eerste vergen ze een grote tijdsinvestering om te
leren hanteren. Ten tweede zijn ze vaak geschikter als analyse- dan als ontwikkelinstrument. Het zal dan ook niemand verbazen dat elke nieuwe ontwerptheorie
binnen de game industrie op enige scepsis stuit.
De ontwerptheorieën uit dit proefschrift zijn daarom ontwikkeld met het idee
dat zij praktisch en makkelijk toepasbaar moeten zijn voor game designers. Het
eerste theoretische kader biedt handvatten om games te bekijken en begrijpen
als systemen van spelregels. Wie de achterliggende spelregels van een game kan
identificeren en hun werkwijze begrijpt, kan deze spelregels immers gerichter
en effectiever aanpassen en verbeteren. Het tweede ontwerpkader kan game
designers helpen bij het creëren van een open en tegelijkertijd dynamische wereld
die spelers meezuigt in een soepele spelervaring. De twee ontwerptheorieën staan
overigens niet los van elkaar: in een goede game vormen de spelwereld en de
spelregels een complexe, goed uitgebalanceerd eenheid. Het een kan niet zonder
het ander en andersom.
Het eerste kader dat game designers inzicht biedt in spelregels en hun werking
heet Machinations en maakt gebruik van dynamische, interactieve diagrammen.
Deze diagrammen brengen de interne economie van een game in kaart door de
stroom van belangrijke, en soms abstracte, resources te modelleren. Resources
kunnen grondstoffen zijn. Maar ook punten, geld of kogels.
De belangrijkste structurele eigenschap die Machinations-diagrammen voor
het voetlicht brengen zijn feedbackpatronen. Feedback wordt over het algemeen
gezien als een belangrijke oorzaak van dynamische gedrag in complexe systemen
zoals games. Feedback zorgt ervoor dat een statusverandering van één gameelement andere game-elementen beı̈nvloedt waardoor vervolgens het eerste element weer verandert. In games heeft dit vaak een zelfversterkend effect. Denk
aan het klassieke bordspel Monopoly. Wie daar geld investeert in vastgoed verdient dat uiteindelijk ruimschoots terug door huuropbrengsten en kan daardoor
nog meer investeren.
Het Machinations-kader maakt onderscheid tussen verschillende soorten feedback: feedback kan positief of negatief zijn, constructief of destructief, snel of
langzaam, etcetera. Voor game designers is het van belang goed zicht te hebben
op de feedbackwerking in hun games en deze te bestuderen en waar nodig aan
te passen. Zij kunnen daarbij gebruik maken van patronen in interne speleconomieën die steeds weer terugkeren. Deze patronen zijn een weerslag van
264
Joris Dormans | Engineering Emergence
veel voorkomende oplossingen voor frequente ontwerpproblemen. Machinations
diagrammen zijn een uitstekend vehikel om deze patronen te identificeren en ze
verder te bestuderen.
Game designers kunnen Machinations-diagrammen maken met de software
die voor dit onderzoek is ontwikkeld. Het grootste voordeel van deze software is
dat digitale diagrammen net zo dynamisch en interactief zijn als de games die ze
representeren. Dit stelt game designers in staat games gemakkelijk en efficiënt
te simuleren. Ze kunnen de interne speleconomie modelleren en balanceren nog
voordat er een prototype van de game is gemaakt.
Het tweede theoretische kader van dit proefschrift, Mission/Space, richt zich
op level-ontwerp en spelmechanismen die de voortgang van een speler bepalen. In
tegenstelling tot bestaande modellen voor level-ontwerp, bouwt Mission/Space
voort op het idee dat er in een level twee verschillende structuren bestaan.
Mission-diagrammen worden gebruikt om de structuur van taken en uitdagingen
voor de speler te formaliseren, terwijl space-diagrammen de ruimtelijke constructie formaliseren. Beide constructies zijn aan elkaar gerelateerd, maar zijn niet
hetzelfde. De verschillende wijzen waarop missies geprojecteerd kunnen worden
op een bepaalde ruimte speelt uiteindelijk een belangrijke rol in de totstandkoming van de spelervaring. Spelmechanismen voor sloten en sleutels, die vaak
gebruikt worden om te voorkomen dat spelers te snel een bepaald doel bereiken,
spelen hier bij een belangrijke rol.
Zoals in het geval van Machinations leent het Mission/Space-kader zich voor
het ontwikkelen van software die ontwikkelaars kan ondersteunen in het ontwerpen van game levels. Het programma Ludoscope implementeert mission- en
space-diagrammen en stelt ontwikkelaars op die manier in staat om de voortgang van spelers door een level te simuleren. Zo helpt Ludoscope in een vroeg
stadium met het identificeren van potentiële fouten in het ontwerp.
Waar Mission/Space zich richt op level-ontwerp en games of progression,
richt Machinations zich meer op interne speleconomie en games of emergence.
Door beide kaders te combineren ontstaat een nieuw perspectief waarmee de
combinatie van emergence en progression kan worden onderzocht. Veel games
combineren beide structuren. Toch is er nog een groot aantal spellen waar de
rijkdom van de mechanismen van de speleconomie, ruimte en fysische simulatie niet aansluiten bij de beperkte mechanismen die de spelerprogressie controleren. Door Machinations-diagrammen te combineren met mission- en spacediagrammen, worden twee nieuwe strategieën inzichtelijk waarmee game designers emergence en progression verder kunnen integreren.
Een eerste combinatiestrategie richt zich op slot- en sleutelmechanismen die
eigenlijk zelden tot nooit gebruik maken van feedbackmechanismen. Met Machinationsdiagrammen kunnen designers echter op eenvoudige wijze meer geraffineerde en
emergente spelmechanismes ontwikkelen voor sloten en sleutels. Een tweede
combinatiestrategie ontstaat door de voortgang van spelers als resource te modelleren waardoor het mogelijk wordt deze te betrekken in feedbackmechanismen
op de schaal van het complete level of spel.
De Machinations- en Mission/Space-kaders formaliseren verschillende perspectieven op games. ’Model driven engineering’, een begrip ontleend aan soft-
Nederlandse Samenvatting
265
ware engineering, kan beide perspectieven verenigen onder een formele benadering van game ontwikkeling. Wie games op een goede manier formeel weet te benaderen, kan uiteindelijk bepaalde onderdelen van het ontwikkeltraject automatiseren. Door gebruik te maken van formele grammatica’s en herschrijfsystemen
is het bijvoorbeeld mogelijk om transformaties vast te leggen die game designers
kunnen gebruiken om ruimtes te genereren voor een missie, spelmechanismen
voor een ruimte, of een missie voor spelmechanismen. Door het ontwikkelproces
onder te verdelen in een serie kleine transformaties wordt het mogelijk flexibele software tools te maken die game levels en, tot op zekere hoogte, zelfs hele
games automatisch kunnen (helpen) genereren. Een andere toepassing is het
ontwikkelen van games die zich automatisch aanpassen aan de prestaties van
spelers.
Om het automatiseren van game design dichterbij te brengen is voor dit
proefschrift het prototype Ludoscope verder uitgebouwd. Het stelt designers in
staat grammatica’s en herschrijfsystemen zelf te definiëren. In potentie maakt dit
Ludoscope tot een zeer krachtige, generieke game ontwikkeltool die het mogelijk
maakt om het ontwerpproces op een hoger abstractieniveau te brengen.
De software, ontwikkeld als onderdeel van dit onderzoek, is tegelijkertijd ook
een manier om de theorie uit dit proefschrift uit te proberen en te valideren.
Maar het is niet de enige wijze van valideren die is toegepast. Het overgrote
deel van de ideeën uit dit proefschrift zijn gepresenteerd, uitgeprobeerd en bediscussieerd op lezingen en in workshops. Academici en ervaren game designers hebben zo de mogelijkheid gekregen commentaar te leveren en het werk te
beënvloeden. Daar komt bij dat beide theoretische kaders, maar vooral Machinations, gebruikt zijn in het game development onderwijs aan de Hogeschool
van Amsterdam. Daaruit is gebleken dat de kaders studenten helpen om games
beter te begrijpen en te verbeteren. Vooral doordat ze hen een formeel perspectief bieden op belangrijke aspecten van game ontwikkeling die zonder de
theoretische kaders niet meteen zicht- of tastbaar zijn. De theoretische kaders
stelden de studenten in staat onderdelen van games te simuleren en er mee te
experimenteren.
Het ontwikkelen van games blijft een uitdaging. De theorie en theoretische gereedschappen die in dit proefschrift zijn gepresenteerd kunnen game ontwerpers helpen doordat ze domeinkennis vastleggen, ontwerpers in staat stellen
games in een vroeg stadium te simuleren, en zelfs door gedeeltes van het ontwerpproces te automatiseren. Maar ze kunnen natuurlijk nooit de menselijke
creativiteit en vindingrijkheid vervangen die noodzakelijke ingrediënten blijven
voor het maken van games. Uiteindelijk denk ik dat deze tools en theorie ontwerpers kunnen helpen en een bijdrage kunnen leveren om game design naar een
hoger niveau te tillen.
266
Joris Dormans | Engineering Emergence
Index
400 project, 50
Aarseth, Espen, 2, 135
Adams, Ernest, 4, 6, 7, 44, 48, 60, 71,
80, 81, 97, 110, 114, 187, 209
affordance, 27
agency, 115
Agricola, 142
Alexander, Christopher, 34, 52, 53
America’s Army, 29
Angry Birds, 8, 9, 15
Araújo, Manual, 56
Arinbjarnar, Maria, 114
Arneson, Dave, 130
Ascent, 206
Ashmore, C, 128
Assassins Creed, 12
augmented transition networks, 55
Backgammon, 219
Bahktin, Mikhail M., 116
Bakkes, Sander, 176, 193, 207
Baldur’s Gate, 12
Ball, Philip, 16
Barwood, Hal, 50
Beck, John, 33
Bejeweled, 15
Bewbees, 206
Binmore, Ken, 57
Björk, Staffan, 51–53, 180
Blackjack, 58, 59
Blood Bowl, 250, 251
Bogost, Ian, 25, 27, 29
Boulder Dash, 8, 189
Brom, Cyril, 56
Brown, Alan, 162
Bura, Stéphane, 58, 208
Burke, Edmund, 28
Burke, Timothy, 71
Byrne, Ed, 109, 110, 113
Caesar III, 142–146, 158, 159, 231
Campagne, 206
Campbell, Joseph, 129
Castronova, Edward, 82
Caylus, 77, 78, 142, 146, 251
cellular automata, 16, 186, 210
Chalmers, David J., 19
Checkers, 58
Chess, 12, 13, 16, 34, 59, 84, 88, 107,
147
Chomsky, Noam, 34, 55, 164, 165
chronotope, 116
Church, Doug, 21, 49
Civilization, 15, 21, 142–144, 147, 233,
234
Civilization III, 15, 244, 251
Colossal Cave Adventure, 15
Command & Conquer: Red Alert, 12
commodity flow graphs, 61
complexity, 16
Compton, Kate, 125
Connect Four, 35, 36
control theory, 83, 84
Conway, John, 19
Cook, Daniel, 51, 138
Copier, Marinka, 5
Costikyan, Greg, 16
Counter-Strike, 14
Cousins, Ben, 113
Crawford, Chris, 16, 26, 37, 190
Cubel, José Á Carsı́, 162
data intensity, 37
Descent: Journeys in the Dark, 218,
219
design patterns, 51–53, 91
attrition, 234–238
converter engine, 223–226
dynamic engine, 92, 101, 152, 153,
157, 220–223
dynamic friction, 153, 231–234
engine building, 92, 227–229
escalating complexity, 156, 246–248
escalating complications, 155, 156,
158, 244–246
multiple feedback, 240–242
playing style reinforcement, 248–
251
static engine, 217–220
static friction, 157, 229–231
stopping mechanism, 238–240
trade, 242–244
268
Joris Dormans | Engineering Emergence
Deus Ex, 14, 30, 114, 147–149, 190
Diablo, 30, 31, 37, 186
discrete infinity, 34
DiStefano III, Joseph J., 83
Djaouti, Damien, 180
Donkey Kong, 8
Dormans, Joris, 27, 59, 67, 148, 176,
187, 191, 193, 207
Doull, Andrew, 161
Dungeon Run, 206
Dungeons & Dragons, 130, 186, 187
Elite, 226
emergence, 13–20, 23, 35, 36, 54, 68,
142, 144
Enchanted Hands, 205
Eskelinen, Markku, 2
EverQuest, 14
Eye of the Beholder II, 125
Fable, 190
Fallout 3, 16, 82
Falstein, Noah, 50
Fearsome Floors, 219
feedback, 18, 23, 60, 83, 151, 152, 155,
198
determinability, 88–90
multiple feedback loops, 84
negative, 84
patterns, 91
positive, 84
profiles, 87–91
Flix, 206
Flix 2, 206
formal grammars, 164–166
classification, 165
context-free grammars, 166
context-sensitive grammars, 165
graph grammars, 168
nonterminals, 164
regular grammars, 166
rewrite rules, 165
shape grammars, 170
terminals, 164
unrestricted grammars, 165, 169
Fowler, Martin, 54
Frasca, Gonzalo, 3, 25
Friedman, Ted, 21
Fromm, Jochen, 18
Fron, Janine, 84
Fullerton, Tracy, 4, 48, 83, 109
Gabbler, Kyle, 136
Galloway, Alexander R., 4
game
definition, 5, 22
genre, 10–13
game design documents, 44–46
one-page, 45
game diagrams, 57–59
Game of Goose, 219
Game of Life, 19
game ontology project, the, 50
game studies, 2
game vocabulary, 49–51, 62
gameplay, 5, 23, 35, 68, 198
adaptable, 188
Gamma, Erich, 52
Genette, Gérard, 115
Get H2O, 38, 39, 206
Gips, James, 170
Go, 12, 16, 34, 107, 210
Grünvogel, Stefan M., 53
Grand Theft Auto III, 15
Grand Theft Auto: San Andreas, 14
graph grammars, 168–170
deletion, 169
Guardiola, Emmanuel, 57
Guttenberg, Darren, 62
Half-Life, 12, 21, 130
Half-Life 2, 20, 21, 130, 148
Halo, 12
Heckel, Reiko, 168
hero’s journey, 116
hierarchy of challenges, 114
Holland, John H., 67
Holopainen, Jussi, 51–53, 180
Hooft, Niels ’t, 5
Hopson, John, 97
Huizinga, Johan, 5, 40
human-centered design, 48
Index
Hunicke, Robin, 47
hyperrealism, 30
in medias res, 129
Infinite Boulder Dash, 189, 190, 206
Infinite Mario Bros., 189, 192
innovation, 10, 51
internal economy, 7, 23, 59, 71, 83, 198
Järvinen, Aki, 53
Jakobson, Roman, 28
Jenkins, Henry, 20, 110
Jenkins, Odest Chadwicke, 6, 61, 71
Johnson, Lawrence, 186
Johnson, Steven, 48
Juul, Jesper, 1, 2, 4, 13, 16, 26, 36, 54,
144, 188, 210
Keith, Clinton, 45
Kim, K. L., 27
King, Geoff, 30
Klabbers, Jan, 36
Klevjer, Rune, 25
Klop, Jan Willem, 166, 167
Knytt Stories, 139, 140, 142
Kohler, Chris, 138
Koster, Raph, 3, 57, 61
Kreimeier, Bernd, 45, 49, 52
Kriegsspiel, 32, 37, 97
Lakoff, George, 38
language, 28, 34
Laurel, Brenda, 2
LeBlanc, Marc, 47, 48, 83, 84, 86, 88
level, 23
level design, 109, 199
boss character, 33, 125, 190
bottleneck, 128
branching, 110
hub-and-spoke, 111, 128, 130
kata, 138
kihon, 138
learning curve, 20, 137, 182, 187
mini-boss, 125, 130
pacing, 187
railroading, 21, 110, 130
tasks, 113
269
tutorials, 20
Lewis, Chris, 210
Librande, Stone, 45
Lindley, Craig, 50
Lister, Martin, 37
Little Big Planet, 192
LKE, 206–208
Locke, John, 28
ludology, 2, 50
Ludoscope, 131, 193–196, 205, 211
Machinations
activation modes, 72, 73, 94
activators, 77
artificial players, 96
charts, 94
connection label, 73
converters, 81, 82
drains, 81
end conditions, 94
framework, 68–71
game state, 72, 75
gates, 78–80
inputs, 73
label modifiers, 75
modifiers
intervals, 100
random, 97
node modifiers, 76
outputs, 73
conditional, 78
probable, 78
pools, 73
pull modes, 94
randomness, 97
resource connections, 73
resources, 71
color-coded, 96
sources, 81
state connections, 75–78
static diagrams, 72
time modes, 72, 74
traders, 82
triggers, 77
magic circle, 4, 5, 20
Magic the Gathering, 236, 237
270
Joris Dormans | Engineering Emergence
Malaby, Thomas M., 70
place, 121
Martin, Paul, 131
requirements, 119–121
Mass Murderer Game, 3
strong, 117
Mateas, Michael, 125, 180, 187
weak, 117
set-back, 124
McCloud, Scott, 109
space graphs, 121–125, 131
McGuire, Morgan, 6, 61, 71
tasks, 118
MDA framework, 46–48, 136, 201
unlock relation, 123
mechanics, 6–10, 23, 46, 68
valve, 123
and genre, 12
window, 123
continuous versus discrete, 7–9
controlling progression, 12, 146, 147, Miyamoto, Shigeru, 187
model driven engineering, 162
149, 154–159
model transformation, 162–164, 187, 192
core, 6, 23
monomyth, 129
dialog trees, 148
Monopoly, 71, 73, 75–77, 81–84, 92,
dice, 32
153, 221, 231
double jumping, 139
Murray, Janet, 2, 28, 115
economy building, 142
Myers, David, 28, 36
inventory, 30
jumping, 33
locks and keys, 128, 130, 139, 142, narratology, 2, 50
Natkin, Stéphane, 56, 57
149, 152–154, 174–175, 199
Nelson, Mark J., 180, 187
physics, 7, 209
Nietsche, M., 128
power-ups, 139
Nintendo, 116, 138
social interaction, 12, 210
tactical maneuvering, 12, 210
Pac-Man, 28, 155, 246
types, 12
Pagulayan, Randy J., 48
MineCraft, 186
Panofsky, Erwin, 63
Mission/Space
Peirce, Charles S., 27
entrances, 118
Petri nets, 56, 57
force-ahead, 125
play-centric design, 48–49, 201
framework, 115–117
player modeling, 190
game element, 122
plot, 115
goals, 118
Poole, Steven, 26, 28, 40
hub, 124
Pooley, Rob, 59
inhibition, 117
Portal, 9
lock, 122
Power Grid, 60, 61, 88, 225, 226, 240
lock relation, 123
Prince of Persia, 123
locked short-cut, 124
Prince of Persia: Sands of Time, 130
mission graphs, 117–121, 131
procedural content, 186–188
mission state, 117, 119
mixed-initiative approach, 192, 194
task state, 117–118
process
intensity, 37
tasks, 117
progression,
12–15, 20, 23, 36, 54, 142
mission-space morphology, 129
Propp,
Vladimir,
129
open space, 124
Puerto
Rico,
142,
228
path, 122
pathway, 124
quantifiable outcome, 4, 5
Index
quests, 147
realism, 27
recursion, 55
Rekers, J., 168
representation, 27
rewrite system
confluency, 167
normal form, 167
rewrite systems, 166–168, 187, 192
generating learning curves, 182
locks and keys, 174
Machinations, 180
to generate space, 176
Reyno, Emanuel Montero, 162
Risk, 14, 31, 32, 55, 56, 71, 78, 81, 84–
88, 91, 92, 97, 242
Robot Wants Kitty, 139
Rogers, Scott, 44
Rogue, 186
roguelike games, 186–187
Rollings, Andrew, 4, 6, 7, 44, 48, 60,
71, 80, 81, 97, 110, 114, 187
Rosenblum, Robert, 63
rules, 4, 6, 10
Rumbaugh, Jim, 59
Ryan, Marie-Laure, 2, 40, 110, 191
Saint-Exupéry, Antoine de, 25, 34
Salen, Katie, 4, 14, 48, 57, 83, 84
sandbox games, 15
Saussure, Ferdinand de, 28
Schürr, A., 168
Schaffer, Noah, 48
Schell, Jessie, 110, 176
Seasons, 156–159, 185, 206
Selic, Bran, 59, 162
semiotics, 27
September 12, 3
Settlers of Catan, 62, 76, 82, 91, 92, 97,
222, 223, 229, 244
Shannon, Claude E., 34
shape grammars, 170–172
line-segments, 171
points, 171
quads, 171
271
Sheffield, Brandon, 62
signs, 27
SimCity, 13–16, 100, 159, 228
Simpson, Zack, 71
simulation, 25, 27, 29, 50
iconic, 28, 33
indexical, 30–31, 33, 37, 39
symbolic, 31–33, 37, 39
SimWar, 68, 96, 100–103, 105, 236
Smelik, Ruben, 192
Smith, Gillian, 187, 192
Smith, Harvey, 16, 36, 37
Snakes and Ladders, 219
software engineering, 48, 52, 162
Space Hulk, 237, 238
Space Invaders, 155, 245
Spector, Warren, 30
Spore, 136, 137, 186
SPUG, 137
Star Defender, 119, 120
Star Wars: X-Wing Alliance, 81, 218,
219
StarCraft, 12, 13, 92, 93, 142, 148, 222
StarCraft II, 147, 149
state machines, 53, 54
Stephenson, Neal, 191
Stevens, Perdita, 59
Stiny, George, 170
story, 147, 190
Super Mario Bros., 7, 8, 27, 32, 33, 37,
38, 121, 189
Super Mario Kart, 84
Super Mario Sunshine, 113, 114
Swain, Chris, 48
Sweetser, Penelope, 36, 48
System Shock 2, 115, 164
Tangram, 15
Taylor, M. J., 59
Tetris, 3, 28, 88, 156, 248
The Elder Scrolls IV: Oblivion, 12, 16,
251
The Game of Goose, 154
The Landlord Game, 84
The Legend of Zelda, 12, 40, 125, 174,
188, 253
272
Joris Dormans | Engineering Emergence
The Legend of Zelda: Twilight Princess,
40, 110, 111, 116, 125–127, 133,
138, 139, 141, 142, 148, 151
The Sims, 100
Tic-Tac-Toe, 35, 59
Togelius, Julian, 188, 189
Tolkien, J.R.R., 20
Torchlight, 186
Tower of Goo, 9
Unified Modeling Language, 54, 59, 162
Up the River, 219
Veugen, Connie, 10
Vogler, Christopher, 63, 116, 129
Wade, Mitchell, 33
Ward, J., 26
Wardrip-Fruin, Noah, 27, 147–149, 190
Warhammer Fantasy Roleplay, 154, 155
Weyth, Peta, 48
Wolf, Mark J. P., 4
Wolfram, Stephen, 16–18
Woods, William A., 55
World of Goo, 9
Zagal, José, 51
Zimmerman, Eric, 4, 14, 48, 57, 83, 84
Zork, 15
Bibliography
Aarseth, E. J. (2001). Computer Game Studies, Year One. Game Studies, 1 (1).
URL http://www.gamestudies.org/0101/editorial.html
Aarseth, E. J. (2005). From Hunt the Wumpus to EverQuest: Introduction to
Quest Theory. In Proceedings of the International Conference on Entertainment Computing, Sanda Japan, September 2005 , (pp. 496–506).
Adams, E. (2000). A Letter From A Dungeon. Gamasutra.
URL
http://www.gamasutra.com/view/feature/3424/the_designers_
notebook_a_letter_.php
Adams, E., & Rollings, A. (2007). Fundamentals of Game Design. Upper Saddle
River, NJ: Pearson Education, Inc.
Alexander, C. (1979). The Timeless Way of Building. New York, NY: Oxford
University Press.
Alexander, C., Ishikawa, S., Silverstein, M., Jacobson, M., Fiksdahl-King, I., &
Angel, S. (1977). A Pattern Language: Towns - Buildings - Construction.
New York, NY: Oxford University Press.
Araújo, M., & Roque, L. (2009). Modeling Games with Petri Nets. In Breaking New Ground: Innovation in Games, Play, Practice and Theory: Proceedings of the 2009 Digital Games Research Association Conference, London UK,
September 2009 .
Arinbjarnar, M., & Kudenko, D. (2009). Directed Emergent Drama vs. Pen &
Paper Role-Playing Games. In Proceedings of the AISB’09 Symposium: AI &
Games, Edinburgh UK, April 2009 .
Arvo, J., & Novins, K. (2007). Stratified Sampling of Convex Quadrilaterals.
URL http://www.ics.uci.edu/~arvo/EECS204/papers/quad.pdf
Ashmore, C., & Nietsche, M. (2007). The Quest in a Generated World. In
Situated Play: Proceedings of the 2007 Digital Games Research Association
Conference, Tokyo Japan, September 2007 , (pp. 503–509).
Bahktin, M. M. (1981). The Dialogic Imagination. Austin, TX: University of
Texas Press.
Bakkes, S., & Dormans, J. (2010). Generating Mission and Space for Dynamic
Play Experience. In Proceedings of the GAME-ON Conference, Leicester UK,
November 2010 , (pp. 72–79).
Ball, P. (2004). Critical Mass: How One Thing Leads To Another . New York,
NY: Farrar, Straus and Giroux.
Beck, J. C., & Wade, M. (2004). Got Game: How the Gamer Generation Is
Reshaping Business Forever . Boston, MA: Harvard Business School Press.
274
Joris Dormans | Engineering Emergence
Binmore, K. (2007). Game Theory: A Very Short Introduction. Oxford, UK:
Oxford University Press.
Björk, S., & Holopainen, J. (2005). Patterns in Game Design. Boston, MA:
Charles River Media.
Bogost, I. (2006). Unit Operations: An Approach to Videogame Criticism. Cambridge, MA: The MIT Press.
Bogost, I. (2007). Persuasive Games: The Expressive Power of Videogames.
Cambridge, MA: The MIT Press.
Brathwaite, B. (2010). My games: the mechanic is the message. Private
blogpost.
URL
http://bbrathwaite.wordpress.com/2010/03/14/
my-games-the-mechanic-is-the-message/
Brom, C., & Abonyi, A. (2006). Petri-Nets for Game Plot. In Proceedings of
AISB ’06: Adaptation in Artificial and Biological Systems, Bristol UK, April
2006 , (pp. 6–13).
Brown, A. (2004). An introduction to Model Driven Architecture.
URL
http://www.ibm.com/developerworks/rational/library/3100.
html
Bura, S. (2006). A Game Grammar.
URL http://www.stephanebura.com/diagrams/
Burke, E. (1990). A Philosophical Enquiry into the Origin of our Ideas of the
Sublime and Beautiful . Oxford, UK: Oxford University Press.
Burke, T. (2005). Can A Table Stand On One Leg? Critical and Ludological
Thoughts on Star Wars: Galaxies. Game Studies, 5 (1).
URL http://www.gamestudies.org/0501/burke/
Byrne, E. (2005). Game Level Design. Boston, MA: Charles River Media.
Campbell, J. (1947). The Hero With A Thousand Faces. Princeton, NJ: Princeton Univeristy Press.
Castronova, E. (2005). Synthetic Worlds: The Business and Culture of Online
Games. Chicago, IL: The University of Chicago Press.
Chalmers, D. J. (2006). Strong and Weak Emergence. In P. Clayton, & P. Davies
(Eds.) The Re-emergence of Emergence, (pp. 244–256). Oxford, UK: Oxford
University Press.
Chomsky, N. (1957). Syntactic Structures. The Hague, the Netherlands: Mouton
Publishers.
Chomsky, N. (1959). On Certain Formal Properties of Grammars. Information
and Control , 2 (2), 137–167.
Bibliography
275
Chomsky, N. (1972). Language and Mind, Enlarged Edition. New York, NY:
Harcourt Brace Jovanovich Inc.
Church, D. (1999). Formal Abstract Design Tools. Gamasutra.
URL http://www.gamasutra.com/features/19990716/design_tools_01.
htm
Compton, K., & Mateas, M. (2006). Procedural Level Design for Platform
Games. In Proceedings of AAAI Conference, Boston MA, July 2006 , (pp.
109–111).
Cook, D. (2006). GameInnovation.org. Blog post on LostGarden.
URL http://www.lostgarden.com/2006/04/gameinnovationorg.html
Cook, D. (2007). The Chemistry Of Game Design. Gamasutra.
URL
http://www.gamasutra.com/view/feature/1524/the_chemistry_
of_game_design.php
Copier, M. (2007). Beyond the Magic Circle: A Network Perspective on RolePlay in Online Games. Ph.D. thesis, University Utrecht, The Netherlands.
Costikyan, G. (1994). I Have No Words I Must Design. Interactive Fantasy.
URL http://www.interactivedramas.info/papers/nowordscostikyan.
pdf
Cousins, B. (2004). Elementary game design. Develop, (pp. 51–54).
Crawford, C. (1984). The Art of Computer Game Design. Berkeley, CA:
McGraw Hill/Osborne Media.
URL
http://Vancouver.wsu.edu/fac/Peabody/game-book/coverpage.
html
Crawford, C. (2003a). Chris Crawford on Game Design. Indianapolis, Indiana:
New Riders Publishing.
Crawford, C. (2003b). Interactive Storytelling. In M. J. P. Wolf, & B. Perron (Eds.) The Video Game Theory Reader , (pp. 259–273). New York, NY:
Routledge.
Crawford, C. (2005). Chris Crawford on Interactive Storytelling. Berkeley, CA:
New Riders Publishing.
DeMaria, R., & Wilson, J. L. (2004). High Score! The illustrated history of
electronic games. Emeryville, CA: McGraw-Hill/Osborne.
DiStefano III, J. J., Stubberud, A. R., & Williams, I. J. (1967). Theory and
Problems of Feedback and Control Systems. New York, NY: McGraw-Hill.
Djaouti, D., Alvarez, J., Jessel, J.-P., Methel, G., & Molinier, P. (2008). A
Gameplay Definition through Videogame Classification. International Journal
of Computer Games Technology.
276
Joris Dormans | Engineering Emergence
Dormans, J. (2005). The Art of Jumping.
URL http://www.jorisdormans.nl/article.php?ref=artofjumping
Dormans, J. (2006a). Lost in a Forest: Finding New Paths for Narrative
Gaming. Article posted on Game-Research.org.
URL
http://game-research.com/index.php/articles/
lost-in-a-forest-finding-new-paths-for-narrative-gaming/
Dormans, J. (2006b). On the Role of the Die: A brief ludologic study of penand-paper roleplaying games and their rules. Game Studies, 6 (1).
URL http://gamestudies.org/0601/articles/dormans
Dormans, J. (2006c). Talking to Games: Gameplay as Expressive Performance.
In Proceedings of the International Digital Games Conference, Portalegre Portugal, September 2006 , (pp. 133–143).
Dormans, J. (2008a). Beyond Iconic Simulation. In Proceedings of the IADIS
Gaming Conference, Amsterdam The Netherlands, July 2008 , (pp. 51–58).
Dormans, J. (2008b). Visualizing Game Mechanics and Emergent Gameplay.
In Proceedings of the Meaningful Play Conference, East Lansing MI, October
2008 .
Dormans, J. (2009). Machinations: Elemental Feedback Structures for Game
Design. In Proceedings of the GAMEON-NA Conference, Atlanta GA, August
2009 , (pp. 33–40).
Dormans, J. (2010). Adventures in Level Design: Generating Missions and
Spaces for Action Adventure Games. In Proceedings of the Foundations of
Digital Games Conference Monterey CA, June 2010 .
Dormans, J. (2011a). Beyond Iconic Simulation. Simulation and Gaming, Special
Issue, 42 (5), 610–631.
Dormans, J. (2011b). Integrating Emergence and Progression. In Think Design
Play: Proceedings of the 2011 Digital Games Research Association Conference,
Hilversum the Netherlands, September 2011 .
Dormans, J. (2011c). Level Design as Model Transformation: A Strategy for
Automated Content Generation. In Proceedings of the Foundations of Digital
Games Conference, Bordeaux France, June 2011 .
Dormans, J. (2011d). Simulating Mechanics to Study Emergence in Games.
In Proceedings of AI and Interactive Digital Entertainment Conference, Palo
Alto CA, October 2011 .
Dormans, J., & Bakkes, S. (2011). Generating Missions and Spaces for Adaptable
Play Experiences. IEEE Transactions on Computational Intelligence and AI
in Games - Special Issue on Procedural Content Generation, 3 (3), 216–228.
Bibliography
277
Doull, A. (2008). The Death of the Level Designer.
URL http://pcg.wikidot.com/the-death-of-the-level-designer
Eskelinen, M. (2001). Towards Computer Game Studies. In Proceedings of
SIGGRAPH 2001, Art Gallery, Art and Culture Papers, (pp. 83–87).
URL http://www.siggraph.org/artdesign/gallery/S01/essays/0416.
pdf
Falstein, N. (2002). Better By Design: The 400 Project. Game Developer
Magazine, 9 (3), 26.
Foster, E. M. (1962). Aspects of the Novel . Harmondsworth, UK: Penguin Books
Ltd.
Fowler, M. (2004). UML Distilled: A Brief Guide to the Standard Object Modeling Language. Boston, MA: Addison-Wesley, third ed.
Frasca, G. (1999). Ludology Meets Narratology: Similitude and differences between (video)games and narrative.
URL http://www.ludology.org/articles/ludology.htm
Frasca, G. (2003). Simulation versus Narrative. In M. J. P. Wolf, & B. Perron (Eds.) The Video Game Theory Reader , (pp. 221–235). New York, NY:
Routledge.
Friedman, T. (1999). Civilization and Its Discontents: Simulation, Subjectivity,
and Space. In G. Smith (Ed.) On a Silver Platter: CD-ROMs and the Promises
of a New Technology, (pp. 132–150). New York, NY: New York University
Press.
Fromm, J. (2005). Types and Forms of Emergence.
URL http://arxiv.org/abs/nlin.AO/0506028
Fron, J., Fullerton, T., Morie, J. F., & Pearce, C. (2007). The Hegemony of Play.
In Situated Play: Proceedings of the 2007 Digital Games Research Association
Conference, Tokyo Japan, September 2007 , (pp. 309–318).
Fullerton, T. (2008). Game Design Workshop: A Playcentric Approach to Creating Innovative Games. Burlington, MA: Morgan Kaufman, 2nd ed.
Fullerton, T., Chen, J., Santiago, K., Nelson, E., Diamante, V., Meyers, A.,
Song, G., & DeWeese, J. (2006). That Cloud Game: Dreaming (and Doing) Innovative Game Design. In Sandbox ’06 Proceedings of the 2006 ACM
SIGGRAPH symposium on Videogames, (pp. 51–59).
Gabler, K. (2009). How to make a game in 48 hours. Video keynote for the
Global Game Jam 2009.
URL http://www.youtube.com/watch?v=aW6vgW8wc6c
Galloway, A. R. (2006). Gaming: Essays on Algorithmic Culture. Minneapolis,
MN: University of Minnesota Press.
278
Joris Dormans | Engineering Emergence
Gamma, E., Helm, R., Johnson, R., & Vlissides, J. (1995). Design Patterns:
Elements of Reusable Object-Oriented Software. Boston, MA: Addison Wesley.
Gardner, M. (1970). Mathematical Games: The fantastic combinations of John
Conway’s new solitaire game “life”. Scientific American, (223), 120–123.
URL
http://ddi.cs.uni-potsdam.de/HyFISCH/Produzieren/lis_
projekt/proj_gamelife/ConwayScientificAmerican.htm
Genette, G. (1980). Narrative Discourse. Oxford, UK: Basil Blackwell.
Grünvogel, S. M. (2005). Formal Models and Game Design. Game Studies, 5 (1).
URL http://gamestudies.org/0501/gruenvogel/
Guardiola, E., & Natkin, S. (2005). Game Theory and Video Game, A New
Approach of Game Theory to Analyze and Conceive Game Systems. In Proceeedings of the 7th International Conference on Computer Games, Angoulème
France, November 2005 .
Guttenberg, D. (2006). An Academic Approach to Game Design: Is It Worth
It? Gamasutra.
URL http://www.gamasutra.com/view/news/8908/Student_Feature_An_
Academic_Approach_to_Game_Design_Is_It_Worth_It.php
Heckel, R. (2006). Graph Transformation in a Nutshell. Electronic Notes in
Theoretical Computer Science 148 , (pp. 187–198).
Hodgson, D. S., & Stratton, S. (2006). The Legend of Zelda: Twilight Princess:
PRIMA Official Game Guide. Roseville, CA: Prima Games.
Holland, J. H. (1998). Emergence: From Chaos to Order . Oxford, UK: Oxford
University Press.
Hopson, J. (2001). Behavioral Game Design. Gamasutra.
URL http://www.gamasutra.com/view/feature/3085/behavioral_game_
design.php
Huizinga, J. (1997). Homo Ludens: Proeve Ener Bepaling Van Het Spelelement
Der Cultuur . Amsterdam, The Netherlands: Pandora.
Hunicke, R., LeBlanc, M., & Zubek, R. (2004). MDA: A formal approach to
game design and game research. In Proceedings of the AAAI-04 Workshop on
Challenges, (pp. 1–5).
Jakobson, R. (1960). Closing Statement: Linguistics and Poetics. In T. A.
Sebeok (Ed.) Style in Language, (pp. 350–378). Cambridge, MA: The MIT
Press.
Järvinen, A. (2003). Making and Breaking Games: A Typology of Rules. In
M. Copier, & J. Raessens (Eds.) Level Up Conference Proceedings: Proceedings of the 2003 Digital Games Research Association Conference, Utrecht The
Netherlands, November 2003 , (pp. 68–79).
Bibliography
279
Jenkins, H. (2004). Game Design as Narrative Architecture. In N. WardripFruin, & P. Harrigan (Eds.) First Person: New Media as Story, Performance
and Game, (pp. 118–130). Cambridge, MA: The MIT Press.
Johnson, L., Yannakakis, G. N., & Togelius, J. (2010). Cellular automata for
real-time generation of infinite cave levels. In Proceedings of the Foundations
of Digital Games Conference 2010, Monterey, CA.
Johnson, S. (2005). Everything Bad is Good for You, How Popular Culture is
Making Us Smarter . London, UK: Penguin Boos Ltd.
Juul, J. (2002). The Open and the Closed: Games of Emergence and Games of
Progression. In F. Mäyrä (Ed.) Proceedings of Computer Games and Digital
Cultures Conference, Tampere Finland, June 2002 , (pp. 323–329).
Juul, J. (2003). Just what is it that makes computer games so different, so
appealing? Ivory Tower Column for IGDA.
Juul, J. (2005). Half-Real: Video Games between Real Rules and Fictional
Worlds. Cambridge, MA: The MIT Press.
Juul, J. (2009). A Casual Revolution: Reinventing Video Games and Their
Players. Cambridge, MA: The MIT Press.
Juul, J. (2010). At What Cost Failure? When Games Punish Players. In
Proceedings of the Foundations of Digital Games Conference, Monterey CA,
June 2010 , (pp. 86–91).
Keith, C. (2010). Agile Game Development with Scrum. Upper Saddle River,
NJ: Addison-Wesley.
Kim, K. L. (1996). Caged in Our Own Signs: A Book About Semiotics. Norwood,
NJ: Ablex Publishing Corporation.
King, G. (2000). Spectacular Narratives, Hollywood in the Age of the Blockbuster .
London, UK: I.B. Tauris Publishers.
Klabbers, J. H. G. (2006). The Magic Circle: Principles of Gaming & Simulation. Rotterdam, The Netherlands: Sense Publishers.
Klevjer, R. (2002). In Defense of Cutscenes. In F. Mäyrä (Ed.) Proceedings of
Computer Games and Digital Cultures Conference, Tampere Finland, June
2002 , (pp. 191–202).
Klop, J. W. (1992). Term Rewriting Systems. In S. Abramsky, D. Gabbay, &
M. T (Eds.) Handbook of Logic in Computer Science, Volum2. Background:
Computational Structures, (pp. 1–116). Oxford, UK: Oxford University Press.
Kohler, C. (2005). Power Up: How Japanese Video Games Gave the World an
Extra Life. Indianapolis, IN: Brady Games.
280
Joris Dormans | Engineering Emergence
Koster, R. (2005a). A Grammar of Gameplay: game atoms: can games be diagrammed? Presentation at the Game Developers Conference, San Francisco
CA, March 2005.
URL http://www.raphkoster.com/gaming/atof/grammarofgameplay.pdf
Koster, R. (2005b). A Theory of Fun for Game Design.
Paraglyph Press, Inc.
Scottsdale, AZ:
Kreimeier, B. (2002). The Case For Game Design Patterns. Gamasutra.
URL
http://www.gamasutra.com/view/feature/4261/the_case_for_
game_design_patterns.php
Kreimeier, B. (2003). Game Design Methods: A 2003 Survey. Gamasutra.
URL
http://www.gamasutra.com/view/feature/2892/game_design_
methods_a_2003_survey.php
Lakoff, G. (1987). Women, Fire, and Dangerous Things: What Categories Reveal
about the Mind . Chicago, IL: The University of Chicago Press.
Laurel, B. K. (1986). Towards the Design of a Computer-Based Interactive
Fantasy System. Ph.D. thesis, Ohio State University.
LeBlanc, M. (1999). Formal Design Tools: Feedback Systems and the Dramatic
Structure of Completion. Presentation at the Game Developers Conference,
San Jose CA, March 1999.
URL http://algorithmancy.8kindsoffun.com/cgdc99.ppt
LeBlanc, M. (2004). Mechanics, Dynamics, Aesthetics: A Formal Approach to
Game Design. Lecture at Northwestern University.
URL http://algorithmancy.8kindsoffun.com/MDAnwu.ppt
LeBlanc, M. (2006). Tools for Creating Dramatic Game Dynamics. In K. Salen,
& E. Zimmerman (Eds.) The Game Design Reader: A Rules of Play Anthology, (pp. 438–459). Cambridge, MA: The MIT Press.
Librande, S. (2010). One-Page Designs. Presentation at the Game Developers
Conference, San Francisco CA, March 2010.
URL http://stonetronix.com/gdc-2010/
Lindley, C. (2003). Game Taxonomies: A High Level Framework for Game Analysis and Design. Gamasutra.
URL http://www.gamasutra.com/view/feature/2796/game_taxonomies_
a_high_level_.php
Lister, M., Dovey, J., Giddings, S., Grant, I., & Kelly, K. (2003). New Media:
A Critical Introduction. Abingdon, UK: Routledge.
Locke, J. (1975). An Essay Concerning Human Understanding. Oxford, UK:
Oxford University Press.
Bibliography
281
Malaby, T. M. (2007). Beyond Play: A New Approach to Games. Games and
Culture, (2), 95–113.
Martin, P. (2011). A spatial analysis of the JBA headquarters in Splinter Cell:
Double Agent. In Proceedings of the Foundation of Digital Conference, Bordeaux 2011 , (pp. 123–130).
McCloud, S. (2000). Reinventing Comics: How Imagination and Technology Are
Revolutionizing an Art Form. New York, NY: HarperCollins.
McGuire, M., & Jenkins, O. C. (2009). Creating Games: Mechanics, Content,
and Technology. Wellesley, MA: A.K. Peters, Ltd.
Murray, J. (1997). Hamlet on the Holodeck: The Future of Narrative in Cyberspace. New York, NY: The Free Press.
Myers, D. (1999a). Simulation, Gaming and the Simulative. Simulation &
Gaming: An International Journal , 30 (4), 482–489.
Myers, D. (1999b). Simulations as Play: A Semiotic Analysis. Simulation &
Gaming: An International Journal , 30 (2), 147–162.
Natkin, S., & Vega, L. (2003). Petri Net Modeling for the Analysis of the Ordering of Actions in Computer Games. In Proceedings of GAME-ON Conference,
London UK, November 2003 , (pp. 82–92).
Nelson, M. J., & Mateas, M. (2007). Towards Automated Game Design. In
Proceedings of AI*IA 2007: Artificial Intelligence and Human-Oriented Computing, Rome Italy, September 2007 , (pp. 626–637).
Pagulayan, R. J., Keeker, K., Wixon, D., Romero, R. L., & Fuller, T. (2003).
User-centered Design in Games. In J. Jacko, & A. Sears (Eds.) The Human Computer Interaction Handbook: Fundamentals, Evolving Technologies
and Emerging Applications, (pp. 882–906). Mahwah, NJ: Lawrence Erlbaum
Associates, Inc.
Panofsky, E. (1960). Renaissance and Renascences in Western Art. Stockholm,
Sweden: Almqvist & Wiksell.
Poole, S. (2000). Trigger Happy: The Inner Life of Videogames. London, UK:
Fourth Estate.
Propp, V. (1968). Morphology of the Folktale. Austin, TX: University of Texas
Press.
Rekers, J., & Schürr, A. (1995). A Graph Grammar Approach to Graphical
Parsing. In Proceedings of the 11th International IEEE Symposium on Visual
Languages, Darmstadt Germany, May 1995 , (pp. 195–202).
Reyno, E. M., & Carsı́ Cubel, J. A. (2008). Model-Driven Game Development:
2D Platform Game Prototyping. In Proceedings of the GAME-ON Conference,
Valencia Spain, November 2008 .
282
Joris Dormans | Engineering Emergence
Reyno, E. M., & Carsı́ Cubel, J. A. (2009a). A Platform-Independent Model
for Videogame Gameplay Specifation. In Breaking New Ground: Innovation
in Games, Play, Practice and Theory: Proceedings of the 2009 Digital Games
Research Association Conference, London UK, September 2009 .
Reyno, E. M., & Carsı́ Cubel, J. A. (2009b). Automatic Prototyping In ModelDriven Game Development. ACM Computers in Entertainment, 7 (2).
Rogers, S. (2010). Level Up: The Guide To Great Video Game Design. Chichester, UK: John Wiley & Sons, Ltd.
Rollings, A., & Adams, E. (2003). On Game Design. Berkeley, CA: New Riders.
Rosenblum, R. (1975). Modern Painting And The Northern Romantic Tradition,
Friedrich to Rothko. London, UK: Thames and Hudson Ltd.
Ryan, M.-L. (2001). Narrative as Virtual Reality: Immersion and Interactivity in Literature and Electronic Media. Baltimore, MD: The John Hopkins
University Press.
Saint-Exupéry, A., de (1939). Wind, Sand and Stars. London, UK: Heinemann.
Salen, K., & Zimmerman, E. (2004). Rules of Play: Game Design Fundamentals.
Cambridge, MA: The MIT Press.
Saussure, F., de (1983). Course in General Linguistics. Chicago, IL: Open Court
Publishing Company.
Schaffer, N. (2008). Heuristic Evaluation of Games. In K. Isbister, & N. Schaffer
(Eds.) Game Usability: Advice from the Experts for Advancing the Player
Experience, (pp. 78–89). Burlington, MA: Morgan Kaufman Publishers.
Schell, J. (2008). The Art of Game Design: a book of lenses. Burlington, MA:
Morgan Kaufman.
Selic, B. (2003). The Pragmatics of Model-Driven Development. IEEE Software,
20 (5).
Selic, B., & Rumbaugh, J. (1998). Using UML for Modeling Complex Real-Time
Systems. Retrieved September 8, 2008.
URL http://www.ibm.com/developerworks/rational/library/content/
03July/1000/1155/1155_umlmodeling.pdf
Shannon, C. E. (1950). Programming a Computer for Playing Chess. Philosophical Magazine, 41 (314), 256–275.
Sheffield, B. (2007). Defining Games: Raph Koster’s Game Grammar. Gamasutra.
URL http://www.gamasutra.com/view/feature/1979/defining_games_
raph_kosters_game_.php
Bibliography
283
Simpson, Z. (2000). The In-Game Economics of Ultima Online. Presentation
delivered at the Game Developers Conference, San Jose, CA, March 2000.
URL http://www.mine-control.com/zack/uoecon/uoecon.html
Smelik, R., Turenel, T., Kraker, K. J., de, & Bidarra, R. (2010). Integrating
procedural generation and manual editing of virtual worlds. In Proceedings of
the Foundations of Digital Games Conference Monterey CA, June 2010 .
Smith, G., Treanor, M., Whitehead, J., & Mareas, M. (2009). Rhythm-Based
Level Generation for 2D Platformers. In Proceedings of the International Conference on the Foundations of Digital Games, Orlando FL, April 2009 , (pp.
175–182).
Smith, G., Whitehead, J., & Mateas, M. (2010). Tanagra: A Mixed-Initiative
Level Design Tool. In Proceedings of the Foundations of Digital Games Conference, Monterey CA, June 2010 , (pp. 209–216).
Smith, H. (2001). The Future of Game Design: Moving Beyond Deus Ex and
Other Dated Paradigms.
URL http://www.igda.org/articles/hsmith_future.php
Stephenson, N. (1995). The Diamond Age, or, A Young Lady’s Illustrated
Primer . New York, NY: Bantam Spectra.
Stevens, P., & Booley, R. (1999). Using UML: Software engineering with objects
and components, updated edition. Harlow, UK: Addison Wesley Longman Ltd.
Stiny, G., & Gips, J. (1972). Shape Grammars and the Generative Specification
of Painting and Sculpture. In Proceedings of Information Processing 71 , (pp.
125–135).
Swain, C. (2008). Master Metrics: The Science Behind the Art of Game Design.
In K. Isbister, & N. Schaffer (Eds.) Game Usability: Advice from the Experts
for Advancing the Player Experience, (pp. 119–140). Burlington, MA: Morgan
Kaufman Publishers.
Sweetser, P. (2006). An Emergent Approach to Game Design - Development and
Play. Ph.D. thesis, The Univerisity of Queensland.
Sweetser, P., & Wyeth, P. (2005). GameFlow: a model for evaluating player
enjoyment in games. Computers in Entertainment (CEI), 3 (3).
Taylor, M. J., Gresty, D., & Baskett, M. (2006). Computer Game-Flow Design.
ACM Computers in Entertainment, 4 (1).
Togelius, J., Kastbjerg, E., Schedl, D., & Yannakakis, G. N. (2011). What is
Procedural Content Generation? Mario on the borderline. In Proceedings of
the Foundations of Digital Games Conference, Bordeaux France, June 2011 .
Togelius, J., Preuss, M., & Yannakakis, G. N. (2010). Towards multiobjective procedural map generation. In Proceedings of the Foundations of Digital
Games Conference, Monterey CA, June 2010 .
284
Joris Dormans | Engineering Emergence
Veugen, C. (2011). Computer Games as Narrative Medium. Ph.D. thesis, Vrije
Universiteit Amsterdam.
Vogler, C. (2007). The Writer’s Journey: Mythic Structure for Writers. Studio
City, CA: Michael Wiese Productions, third ed.
Ward, J., Cantor, D., & Carey, B. (2007). The Hard Science of Making
Videogames.
URL
http://www.popsci.com/popsci/technology/
d997f0209dd15110vgnvcm1000004eecbccdrcrd.html
Wardrip-Fruin, N. (2009). Expressive Processing. Cambridge, MA: The MIT
Press.
Wardrip-Fruin, N., Mateas, M., Dow, S., & Sali, S. (2009). Agency Reconcidered. In Breaking New Ground: Innovation in Games, Play, Practice and
Theory: Proceedings of the 2009 Digital Games Research Association Conference, London UK, September 2009 , (pp. 1–9).
Wolf, M. J. P. (2001). The Video Game as a Medium. In M. J. P. Wolf (Ed.)
The Medium of the Video Game, (pp. 13–33). Austin, TX: University of Texas
Press.
Wolfram, S. (2002). A New Kind of Science. Champaign, IL: Wolfram Media
Inc.
Woods, W. A. (1970). Transition Network Grammars for Natural Language
Analysis. Computational Linguistics, 13 (10), 591–606.
Wright, W. (2003). Dynamics for Designers. Presentation at the Game
Developers Conference, San Jose CA, March 2003.
URL
http://www.slideshare.net/geoffhom/
gdc2003-will-wright-presentation
Zagal, J. P., Mateas, M., Fernádez-Vara, C., Hochhalter, B., & Lichti, N. (2005).
Towards an Ontological Language for Game Analysis. In Changing Views Worlds in Play: Proceedings of the 2009 Digital Games Research Association
Conference, Vancouver Canada, June 2005 .
Ludography
Agricola (2007, board game). Lookout Games, Uwe Rosenberg.
Americas Army (2002, PC). U.S. Army.
Angry Birds (2009, iOS, Android, others). Rovio Mobile Ltd.
Ascent (2009, PC). J. Dormans.
Assassins Creed (2007, PS3, Xbox 360). Ubisoft, Inc., Ubisoft Divertissements
Inc.
Baldurs Gate (1998, PC). Interplay Productions, Inc. BioWare Corporations.
Bejeweled (2000, PC). PopCap Games, Inc.
Bewbees (2011, Android, iOS). GewGawGames.
Blood Bowl, third edition (1994, board game). Games Workshop Ltd, J. Johnson.
Boulder Dash (1984, Apple II, Commodore 64, NES, others). First Star Software, Inc.
Caesar III (1998, PC). Sierra On-Line, Inc., Impressions Games.
Campagne (2010, card game). M. van Dodeweerd & J. Dormans.
Caylus (2005, board game). White Goblin Games, William Attia.
Civilization (1991, PC). MicroProse Software, MPS Labs.
Civilization III (2001, PC). Infrogrames Europe SA, Firaxis Games.
Colossal Cave Adventure (1975-76, PDP-10) W. Crowther & D. Woods.
Command & Conquer: Red Alert (1996, PC, PS). Virgin Interactive Entertainment (Europe) Ltd., Westwood Studios, Inc.
Connect Four (1974, board game). Milton Bradley, N. Strongin & H. Wexler.
Counter-Strike (2000, PC). Sierra On-Line, Inc., Valve Corporation.
Descent: Journeys in the Dark (2005, board game). Fantasy Flight, K. Wilson.
Deus Ex (2000, PC, PS2). Eidos Interactive Ltd., Ion Storm Inc.
Diablo (1996, PC, PS). Blizzard Entertainment Inc., Blizzard North.
Donkey Kong (1981, arcade). Nintendo Co., Ltd.
286
Joris Dormans | Engineering Emergence
Dungeon Run (2009, PC). Joris Dormans.
Dungeons & Dragons (1974, pen-and-paper RPG). Tactical Study Rules, Inc.,
G. Gygax & D. Arneson.
Elite (1984, Apple II, Commodore 64, PC, others). Acornsoft Limited.
EverQuest (1999, PC). 989 Studios, Verant Interactive.
Eye of the Beholder II (1991, PC, Amiga). Strategic Simulations, Inc., Westwood
Associates.
Fable (2004, XBox). Microsoft Game Studios, Lionhead Studios Ltd.
Fallout 3 (2008, PC, PS3, Xbox360). Bethesda Softworks LLC, Bethesda Game
Studios.
Fearsome Floors (2003, board game). 2F-Spiele, F. Friese.
Flix (2009, PC). J. Dormans.
Flix 2 (2010, PC). J. Dormans.
Get H2O (2010, board game). 999 Games, Butterfly Works, Mama Bits & J.
Dormans.
Grand Theft Auto III (2001, PS2, PC). Rockstar Games, Inc., DMA Design
Limited.
Grand Theft Auto: San Andreas (2004, PS2, PC, Xbox). Rockstar Games, Inc.,
Rockstar North Ltd.
Half-Life (1998, PC). Sierra On-Line, Inc., Valve L.L.C.
Half-Life 2 (2004, PC, XBox). Sierra Entertainment, Inc., Valve Corporation.
Halo (2001, PC, XBox). Microsoft Corporation, Bungie Studios.
Infinite Boulder Dash (2011, PC). J. Dormans.
Infinite Mario Bros (2006, PC). M. Persson.
Knytt Stories (2007, PC). N. Nygren.
Kriegsspiel (1824, board game). B. von Reisswitz.
Little Big Planet (2008, PS3). Sony Computer Entertainment America, Inc.,
Media Molecule Ltd.
Magic the Gathering (1993, trading card game). Wizards of the Coast, R.
Garfield.
Ludography
287
MineCraft (2010, PC). Mojang AB.
Monopoly (1934, board game). Parker Brothers.
Pac-Man (1981, arcade). Namco Limited.
Portal (2007, PC). Buka Entertainment, Valve Corporation.
Power Grid (2004, board game). 2F-Spiele, F. Friese.
Prince of Persia (1989, DOS, others). Brøderbund Software, Inc.
Prince of Persia: Sands of Time (2003, PC, PS2, Xbox, GameCube). Ubisoft,
Inc., Ubisoft Divertissements Inc.
Puerto Rico (2002, board game). Alea, A. Seyfarth.
Risk (1959, board game). Parker Brothers, A. Lamorisse & M.I. Levin.
Robot Wants Kitty (2010, PC). MaxGames.com, Hamumu Software.
Rogue (1983, PC, others). Artificial Intelligence Design.
Seasons (2010, PC). J. Dormans.
September 12 (2003, PC). Newsgaming.com, G. Frasca.
Settlers of Catan (1995, board game). Franckh-Kosmos Verlag, K. Teuber.
SimCity (1989, Amiga, Commodore 64, PC, others). Infogrames Europe SA,
Maxis Software Inc.
Space Hulk (1989, board game). Games Workshop Ltd.
Space Invaders (1980, arcade). Taito Corporation.
Spore (2008, PC). Electronic Arts, Inc., Maxis Software Inc.
SPUG (2009, PC). Electronic Arts, Inc., Maxis Software Inc.
Star Defender (2002, PC). Awem Studio.
Star Wars: X-Wing Alliance (1999, PC). LucasArts Entertainment Company
L.L.C., Totally Games, Inc.
StarCraft (1998, PC). Blizzard Entertainment Inc.
StarCraft II (2010, PC). Blizzard Entertainment Inc.
Super Mario Bros. (1985, NES). Nintendo Co. Ltd.
Super Mario Kart (1992, SNES). Nintendo Co. Ltd.
288
Joris Dormans | Engineering Emergence
Super Mario Sunshine (2002, GameCube). Nintendo Co. Ltd., Nintendo EAD.
System Shock 2 (1999, PC). Electronic Arts, Inc., Looking Glass Studios, Inc.
Tetris (1986, PC, others). AcademySoft, A. Pazhitnov.
The Elder Scrolls IV: Oblivion (2006, PS3, PC, XBox360). Bethesda Softworks
LLC, Bethesda Game Studios.
The Landlord Game (1904, board game). E. Magie.
The Legend of Zelda (1986, NES). Nintendo Co. Ltd.
The Legend of Zelda: Twilight Princess (2006, GameCube, Wii). Nintendo Co.
Ltd., Nintendo EAD.
The Sims (2000, PC, PS2, Xbox, GameCube). Electronic Arts, Inc., Maxis
Software Inc.
Torchlight (2009, PC, XBox360). Runic Games, Inc.
Up the River (1988, board game). Ravensburger, M. Ludwig.
Warhammer Fantasy Roleplay (1986, pen-and-paper RPG). Games Workshop.
World of Goo (2008, PC, Wii, iOS). 2D Boy.
Zork (1979, PDP-10). T. Anderson, M. Blank, B. Daniels, & D. Lebling.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement