alignment of player and non-player character assertiveness
TRANSCRIPT
Alignment of Player and Non-Player Character
Assertiveness Levels
Antonio Penalva Carneiro Pacheco
Thesis to obtain the Master of Science Degree in
Information Systems and Computer Engineering
Supervisor: Prof. Carlos Antonio Roque Martinho
Examination Committee
Chairperson: Prof. Daniel Jorge Viegas GoncalvesSupervisor: Prof. Carlos Antonio Roque Martinho
Member of the Committee: Prof. Maria Beatriz Carmo
October 2018
Acknowledgments
I would like to thank my parents for their support over all these years. I would like to thank my
mother for her love, guidance and encouragement when I needed it the most. I would like to thank my
father for being a role model in being dedicated and striving to achieve more. Without you, this project
and the rest of my academic career would not have been possible. I would also like to thank my siblings
and my grandmother for being so patient, and helping me value what is actually important.
I would like to give my regards to my friends and colleagues, my high school friends, Migs, the
Dark Things About team, the Kaikan boys, and everyone from Laboratorio de Jogos, your support and
fellowship made this journey much more joyful.
I would also like to thank Prof. Ryunosuke Kikuchi, for his interminable drive to help others, and
whose help was essential for my studies in Japan. You are an inspiration as a person and an academic.
I would like to acknowledge the support and guidance given by Prof. Tatsuya Nomura, from Ryukoku
University. I am very grateful that you accepted me as your pupil, and leading me towards studying The
Media Equation changed dramatically the direction of this work.
This Thesis would not have been possible without my dissertation supervisor Prof. Carlos Martinho.
Thank you for your insight, availability and incredible dedication. Your immense support and great
feedback was greatly appreciated.
Last but not least, I would like to give special thanks to Fanny Chabeau, for her continuous support,
companionship and comprehension. Thank you for always making my days brighter. Merci pour tout.
To each and every one of you – Thank you.
Abstract
Video game development is one of the fastest growing businesses, and appealing to a large audience is
one of the key factors to a successful game. Recent research in the field of video game design proposes
adapting games to the player’s profile, which has the potential to broaden a game’s target audience.
Companion Non-Player Characters (NPCs) are an element of games that, when present, have a large
potential influence on the player’s experience.
This work, takes the Media Equation finding that the law of similarity-attraction applies to relation-
ships between media and people, and applies it to video game NPCs. We propose adapting companion
NPCs to the player’s profile, more specifically, aligning the NPC’s assertiveness level to the player’s own
assertiveness level. We also present a methodology for the Media Equation’s findings to be tested in
the context of video games, which can be used for future research endeavors. To prove this hypothesis,
we developed a testbed game, a 2D puzzle platformer with a companion NPC. We also developed two
versions of the NPC’s behavior, one for each end of the assertiveness scale.
We conducted a 2x2, between-subjects experiment (n=48), in which Assertive and Non-Assertive
subjects were randomly matched with one of the NPCs. Subjects recognized the NPC’s personality type,
giving a significantly higher assertiveness score to the NPC endowed with assertive characteristics. Non-
Assertive players reported significantly higher Tension scores when interacting with the Assertive NPC
than when interacting with the Non-Assertive NPC. However, based on assertiveness level alignment,
there was neither a significant difference in the enjoyment of the experience nor in the player’s affinity
for the NPC.
Keywords
Video Games; Game Adaptation; Non-Player Character; Media Equation; Assertiveness
iii
Resumo
O desenvolvimento de videojogos e um dos setores com crescimento mais rapido, e um dos fatores prin-
cipais para um jogo ser bem sucedido e apelar a um vasto publico alvo. Alguma investigacao recente,
no campo de design de videojogos, propoe adaptar jogos ao perfil do jogador, o que tem o potencial de
alargar o publico alvo de um jogo. Um elemento de jogos que, quando presente, tem grande potencial
para influenciar a experiencia do jogador sao Non-Player Characters (NPCs) companheiros.
Este projeto baseia-se numa das descobertas da Media Equation, que a lei da semelhanca-atracao se
aplica a relacoes entre pessoas e computadores, e aplica a descoberta a NPCs de videojogos. Propomos
adaptar NPCs companheiros ao perfil do jogador, especificamente, alinhar o nıvel de assertividade do
NPC com o do jogador. Para provar esta hipotese, desenvolvemos um jogo puzzle platformer 2D com um
companheiro NPC. Desenvolvemos tambem duas versoes de comportamento para o NPC, uma para cada
extremo da escala de assertividade. Apresentamos tambem uma metodologia para investigacao posterior,
que consiste em aplicar resultados da Media Equation a videojogos.
Conduzimos um estudo 2x2, entre-sujeitos (n=48), em que participantes Assertivos e Nao-Assertivos
foram aleatoriamente emparelhados com um dos NPCs. Os participantes reconheceram o tipo de per-
sonalidade do NPC, classificando o NPC dotado de caracterısticas assertivas como significativamente
mais assertivo. Os jogadores Nao-Assertivos relataram nıveis de Tensao significativamente superiores
quando emparelhados com o NPC Assertivo. Contudo, a analise dos resultados baseados no alinhamento
dos nıveis de assertividade nao revelou nenhuma diferenca significativa nem no nıvel de satisfacao da
experiencia do jogador nem no nıvel de afinidade do jogador perante ao NPC.
Palavras Chave
Video Jogos; Adaptacao de Jogos; Non-Player Character; Media Equation; Assertividade
v
Contents
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Dissertation Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Related Work 7
2.1 Player Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Gameplay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.2 Physiology and Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.3 Player Profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Models in Psychology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1 The Five Factor Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1.A Revised NEO Personality Inventory . . . . . . . . . . . . . . . . . . . . . 11
2.2.2 Myers-Briggs Type Indicator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Player Models in Game Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1 BrainHex Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 Quantic Foundry’s Gamer Motivation Profile . . . . . . . . . . . . . . . . . . . . . 13
2.4 Game Adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.1 Difficulty balancing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.2 Playtesting Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.3 Monetization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.4 Personalized Content Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 The Media Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5.1 “Can computer personalities be human personalities?” . . . . . . . . . . . . . . . . 16
2.5.2 Contemporary Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 Non-Player Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6.1 Alyx Vance (Half-Life 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
vii
2.6.2 Elizabeth (Bioshock Infinite) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.7 Game Experience Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.7.1 Networked Minds Measure of Social Presence . . . . . . . . . . . . . . . . . . . . . 21
2.7.2 Game Experience Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3 Implementation 25
3.1 Design and Initial Prototyping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Game Engine and Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.1 NPC Script . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3 Interaction Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.1 Normal Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.2 Final Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 NPC Assertiveness Expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4 Studies 37
4.1 Preliminary Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1.1 Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 Main Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.1 Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.3 Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2.4 Manipulation Check . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2.6 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5 Conclusions 47
5.1 Summary of Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.2 Work Accomplished . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
A Appendix A - Questionnaires 60
A.1 Preliminary Study Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
A.2 Main Study Questionnaire - Assertive NPC . . . . . . . . . . . . . . . . . . . . . . . . . . 62
viii
A.3 Main Study Questionnaire - Non-Assertive NPC . . . . . . . . . . . . . . . . . . . . . . . 64
B Game Levels 67
C Dialogue Scripts 71
ix
x
List of Figures
2.1 AI companions in successful video games. . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1 First room. Showing the cave setting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 Cave Escape’s logo. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Game’s mechanical components. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4 Box and trampoline puzzle solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 NPC and Player character, left and right respectively. . . . . . . . . . . . . . . . . . . . . 31
3.6 Dialogue box with player options. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.7 Single line Dialogue box. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.8 End scene player options. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.9 Game ending progress. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.10 NPC logic for door choosing process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1 Sample’s gaming habits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Mean perception of NPC assertiveness score as a function of player and NPC assertiveness
level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
B.1 First level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
B.2 Second level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
B.3 Third level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
B.4 Fourth level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
B.5 Fifth level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
B.6 Sixth level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
C.1 Preliminary study’s dialogue instances. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
C.2 Main study’s dialogue instances, part 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
C.3 main study’s dialogue instances, part 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
xi
xii
List of Tables
2.1 Revised NEO Personality Inventory (NEO PI-R) facets of the Five Factor Model (FFM)
personality traits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.1 Participant distribution in experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2 Treatment Groups used in Kruskal-Wallis H test. . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 Mann-Whitney U test results for each Game Experience Questionnaire (GEQ) component,
separated by “aligned” observations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
xiii
xiv
Acronyms
AB5C Abridged Big Five-Dimensional Circumplex
FFM Five Factor Model
GEQ Game Experience Questionnaire
GMP Gamer Motivation Profile
IPIP International Personality Item Pool
IST Instituto Superior Tecnico
MBTI Myers-Briggs Type Indicator
MOJO Montra de Jogos
NEO-FFI NEO Five Factor Inventory
NEO PI-R Revised NEO Personality Inventory
NPC Non-Player Character
RPG Role Playing Game
xv
xvi
1Introduction
Contents
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Dissertation Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1
2
1.1 Motivation
What if a single game could appeal to all types of players?
This might sound like a utopian idea, and in fact, it is. However, there are some steps that can be
taken to move in that direction, and an important one is adapting the game to the player.
In the decade of 1980, fan-made enhancement kits were the first manifestation of the community’s
demand for games to better suit them, and, ever since, there has been an interest and subsequent business
in changing and customizing games to different players’ desires. Enhancement kits ranged from simply
speeding-up the game or making it look slightly different to complete unofficial expansion packs, such
as Crazy Otto, an expansion pack for Namco’s iconic Pac Man(1980)1, (which eventually became Ms.
Pac-Man(1981)). Nowadays we have even larger examples of success cases of fan-made modification
endeavors, such as Defense of the Ancients (DotA)(2003) or DayZ: Battle Royale(2013), a mod for
DayZ (in-development)(which also started as a mod, for Arma 2 (2009)), which inspired a whole new
genre of games that has recently become one of the most popular genres, Battle Royale.
In the game development industry, one of the responses to this demand was the popularization of mod
support. More recently, with platforms like Valve’s Steam Workshop further facilitating the creation and
distribution process. Another approach was to start dynamically adapting the game to the player, which
puts the challenge of personalizing the game in the game developers’ hands. Some games adapt their
difficulty level, for example, Resident Evil 4 (2005) adapts the AI’s behavior to the player’s performance
[Future Press, 2005]. Other games have recently started adapting their monetization tactics to each
player to maximize profits. This type of adaptation is used primarily in free-to-play games, such as Clash
of Clans(2012) or League of Legends(2009). These tactics are described in more detail in section 2.4.
Although traditionally only seen as a simple user-selected difficulty adjustment, adaptation has been
shown to increase player enjoyment [Aponte et al., 2011]. However, there are other and more promising
ways to adapt the game to the player. These approaches suggest considering the player’s performance,
tactics, strategies and profile as listed by [Bakkes et al., 2012], and [Karpinskyj et al., 2014], and changing
the game to better fit the experience the designers intend the player to have.
In this work we are particularly interested in games with Non-Player Character (NPC) companions,
and how to adapt the companions’ interactions with the player. The interaction between human and
non-human entities has been the subject of a lot of research, ranging from the psychology field to more
computer science related fields, such as robotics.
In the psychology field, the Media Equation studies suggest that, to a certain degree, people attribute
to computers, personalities similar to those of real people, even when interacting solely through text
[Reeves and Nass, 1996]. One of the discoveries of these studies is that the law of similarity-attraction,
which says that individuals prefer to interact with others who are similar in personality to themselves,
1The games referenced in this document are listed in the Game Bibliography at the end of the document.
3
also applies to human-computer relationships, [Nass et al., 1995] (this experiment is presented in detail
in Subsection 2.5.1). The video game equivalent to the computers in this experiment are NPCs, however,
NPCs can interact with the player through more varied and impactful ways, and in different contexts.
Given these findings, and our intentions of adapting the game to the player, testing the same premise
with NPCs in video games seems like the most logical step.
In game design though, the interaction between the player and NPC is a fairly recent research subject,
but a promising one. Nowadays, the interaction between player and NPCs usually does not take into
account the player’s profile. When the player interacts with an NPC for the majority of the game,
this oversight is a missed opportunity for engaging the player in a less frustrating or more satisfying
way. An important dimension of interaction that has been the subject of research is social presence,
that is, “the degree to which a user feels access to the intelligence, intentions, and sensory impressions
of another” [Biocca, 1997]. Social presence is also used as a quality measurement for the interaction
between human and non-human entities.
1.2 Problem
We want to adapt games to the player’s profile effectively.
Adapting a game to the player’s personality model or motivation profile is a complex endeavor but
one with great potential [Yannakakis et al., 2013]. However, adapting an entire game is, in most cases,
a close to impossible feat. So, the logical question to be asked is: Which aspect of the game should be
adapted to the player? In games with companion NPCs, adapting the NPCs seems to be the logical
step, since they interact with the player throughout a large part of the game, and have great potential
to influence the player’s decisions. So the research question becomes:
How should we adapt companion NPCs to the player’s profile?
1.3 Hypothesis
We propose that, in the context of games with NPC companions, adapting the NPC’s assertiveness to
align to the player’s to be effective in making the player’s experience more positive.
We propose that, in games in which players spend a large portion of the game interacting with a
companion NPC, the developers focus on understanding the player’s personality and adapt the NPC’s
personality accordingly. Furthermore, based on the findings of [Nass et al., 1995], we propose that the
adaptation of the NPC will have a more positive effect when aligning the NPC’s personality traits with
those of the player.
4
From the Media Equation study mentioned previously, we predict that the player will be able to easily
identify the NPC’s assertiveness level and that when NPCs’ assertiveness aligns with the player’s, the
player will not only have a more positive overall experience, but will also like the NPC more. In other
words, we predict that the law of similarity-attraction can be applied to Player-NPC relationships.
We are also interested in observing which particular components of the experience are improved by
adapting the NPC to the player.
The formal hypotheses are the following:
• H1. The players will perceive the NPC that exhibits assertive characteristics as assertive and the
NPC that exhibits non-assertive characteristics as non-assertive. (Manipulation Check)
• H2. When the NPC’s assertiveness level is aligned with the player’s, the player perceives their
game experience as more positive than when they’re not aligned.
• H3. When the NPC’s assertiveness level is aligned with the player’s, the player prefers interacting
with the NPC more than when they’re not aligned.
1.4 Dissertation Outline
In Chapter 1 we introduce the motivation behind our work, the problem raised by it, and the formal
hypotheses based on the research done. The chapter ends with the document’s delimitation.
In Chapter 2, We will first consider player modeling and the various ways it is implemented, focusing
mainly on player profiling. Then, we will survey personality models in psychology and game design,
discussing some of their strengths and shortcomings. After that, we will describe the different uses of
game adaptation, focusing on Personalized Content Generation. In the following section, we will focus
on the Media Equation, its implications for our work, taking a closer look at one of its studies on the
relationship between people and media, and summarizing some of the recent research being conducted on
the subject. Next, we will analyze NPCs and some scenarios in successful video games with companion
NPCs that could potentially have had their players’ experiences improved if adaptation was used. We
will also review game experience evaluation in the gaming industry and previous studies that try to
measure the player’s experience. Finally, we will summarize the research done and discuss the knowledge
we gained and how it pertains to the decisions made in the design, implementation and evaluation of our
solution.
In the following chapter, Chapter 3, we will describe the implementation process. We will start by
defining our approach, followed by the design decisions on the testbed game’s concept and setting. In
the same section we will describe the game engine that was developed and the mechanics implemented,
including the logic behind the NPC’s behavior. Then, we will describe, step by step, the interaction sce-
5
narios that were created. In the following section we describe the manipulation of the NPC’s assertiveness
expression, followed by a summary of the chapter.
In Chapter 4, the two studies conducted are described. First, the preliminary study, which was used
as an initial test of our methodology for assertiveness expression. Then, the main study, in which we
tested our hypotheses with the fully developed game and NPC behavior.
Finally, in Chapter 5 we discuss the implications of our results, and some routes in which this work
could be continued and improved upon.
6
2Related Work
Contents
2.1 Player Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Models in Psychology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Player Models in Game Design . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Game Adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 The Media Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.6 Non-Player Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.7 Game Experience Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
7
8
In this chapter, we will take a look at each different aspect of our study, explain it and seek further
knowledge on the subject. Since our study is on adaptation to players, we will start by explaining player
modeling and its most promising methods: gameplay derived data, physiological input and movement
detection, and player profiling. Our study also has a basis in previous work done in psychology, so we will
be taking a look at some research done in the field on personality profiling, describing the prominent Five
Factor Model (FFM) and the Myers-Briggs Type Indicator (MBTI). Then, we will discuss some popular
player models and other promising ones, taking a more careful look at International Hobo’s BrainHex,
and Quantic Foundry’s Gamer Motivation Profile (GMP). After describing player modeling, we will look
at what can be done with the information gathered from the player as we take a look at game adaptation,
its most common uses, such as difficulty balancing and playtesting analysis, and other promising ones,
such as personalized content generation.
We will then change the focus towards the relationship between media and people, focusing mainly on
Media Equation related studies, to provide some context in which to insert this work. We will continue by
analyzing the work on companion NPCs, taking a deeper look at works that study personality expression
and behavior adaptation. For knowledge on how to gather results in our experiment, we will explore
the different methods and techniques to measure game experience and its different aspects. We will
then describe the Game Experience Questionnaire (GEQ) and the Networked Minds Measure of Social
Presence, and understand how to measure social presence and player’s experience in video games. Lastly,
we will discuss what we learned from the research on these topics, and what information we’re going to
use to inform our proposed solution.
2.1 Player Modeling
Video game adaptation has many uses and ways of being applied, however, it always relies on having
some sort of player model to adapt the game to. This model could simply be gathered from the player’s
conscious choice of difficulty in the game’s menu and stay static throughout the game’s length, or it
could be gathered through imperceptible data collection about the player’s physiology or actions inside
the game, and be refined over the course of the gameplay [Bakkes et al., 2012].
2.1.1 Gameplay
This method of player profiling tries to analyze patterns in the player’s actions and infer the current
experience being provided by the game. The metrics used for gameplay-based player modeling can
emerge from the player’s interaction with any element of the game, from NPCs to the level itself. Other
data such as time taken to complete a task or which weapon was chosen can also be considered. This
data can then be used to infer the player’s experience.
9
The main problem of this approach is that the player’s experience is never directly observed, only
inferred through game metrics. This can become an issue in certain situations where one behavior could
have multiple interpretations. For example, if a player is not interacting too much with the game, it
could mean that they are deep in thought, amazed by the experience, or simply just bored and not
paying attention [Yannakakis et al., 2013].
2.1.2 Physiology and Movement
There are multiple ways players express emotional states, physiology being one of them. This might
reflect on changes in the player’s facial expressions, speech, heart rate or posture. There are multiple
research efforts exploring the relation between different gameplay scenarios and physiological responses.
In [Yannakakis et al., 2010] and [Tognetti et al., 2010], heart rate, respiration and skin conductance are
some of the physiological types of data collected.
Besides physiological data, another physical aspect that can be analyzed is the player’s movement.
Data such as facial expressions, eye-tracking and body posture can be mapped to basic emotions and
cognitive processes [Pantic and Caridakis, 2011], [Asteriadis et al., 2008].
Although more reliable than gameplay, this method has two large flaws: the first one is requiring
extra, often large and expensive equipment to make the measurements with; and the second is often
being seen as invasive by players.
2.1.3 Player Profiling
A very promising aspect of player modeling is player profiling, which uses psychologically and sociolog-
ically verified player profiles to provide a player model that models internal traits of the player, such as
their personality and motivations. This type of player modeling has the advantage of being supported
by a large body of psychological knowledge, and has been developed and refined over a long period of
time. In this field, there are different aspects through which to model the player. Player preferences,
motivations and personality are the main aspects used to profile the player.
“Player profiling is of a different calibre, though comprised of predominantly ongoing re-
search. By incorporating psychologically-verified knowledge in player models, as well as knowl-
edge on player experience and satisfaction, player profiling may potentially have a substantial
(and more directly noticeable) impact on the experience that users have with a gaming sys-
tem.” [Bakkes et al., 2012]
Works like [Bartle, 1996], which divides online role-playing game players in “achievers”, “explorers”,
“socialisers” and “killers”, are based on the player’s preferences. Using player’s motivations, Quantic
10
Foundry’s GMP which provides a relative percentile score for each motivation (see section 2.3.2), com-
paring the player’s motivations to those of everyone else that took the survey.
There has also been successful research in using personality models directly taken from the psycho-
logical field, and applying them to game adaptation. [van Lankveld et al., 2011] found a statistically
significant correlation between in-game behavior data and the Five Factor Model. In [Yee et al., 2011]
a statistically significant correlation was also found between player’s chat logs in Second Life(2003) and
the FFM.
We will take a closer look at both psychology models, and game specific player models in Sections 2.2
and 2.3, respectively.
2.2 Models in Psychology
In this section, we will describe two personality models from psychology, which were the inspiration and
served as a basis for a lot of the work done on personality in video games.
2.2.1 The Five Factor Model
The Five Factor Model, or more commonly referred to as “Big Five” is a theory that divides a per-
son’s personality in five main components. These components were discovered and refined by multiple
independent empirical studies over a period of 50 years.
The model’s components are, Extraversion, how outgoing the person is and how likely they are to seek
someone else’s company; Agreeableness, how empathic and compassionate the person is towards other
people; Conscientiousness, how focused, determined and self-disciplined the person is; Neuroticism, how
emotionally stable a person is, or how likely they are to feel emotions such as fear and anxiety; and
Openness to Experience, the person’s tendency to think outside of the box and experience new things.
Its use in interpersonal relationships has been the focus of studies such as [McCrae and Costa, 1989b],
in which, using joint-factor examination with Interpersonal Circumplex theories, the two dimensions of
the FFM that were found to be of greater influence were Extraversion and Agreeableness.
2.2.1.A Revised NEO Personality Inventory
The Revised NEO Personality Inventory (NEO PI-R) is a personality inventory that measures the FFM’s
personality traits, however, in addition to this, also reports on six facets for each of the FFM traits [Costa
and McCrae, 1992], these facets can be seen in Table 2.1.
This inventory is comprised of a 10-item scale for each of the facets seen in Table 2.1, which is a
contributing factor to its completion time of 45 to 60 minutes. A shorter version has been developed by
11
Neuroticism Extraversion Openness Agreeableness ConscienciousnessAnxiety Warmth Fantasy Trust CompetenceAngry Hostility Gregariousness Aesthetics Straightforwardness OrderDepression Assertiveness Feelings Altruism DutifulnessSelf-Consciousness Activity Actions Compliance Achievement StrivingImpulsiveness Excitement Seeking Ideas Modesty Self-DisciplineVulnerability Positive Emotions Values Tender-Mindedness Deliberation
Table 2.1: NEO PI-R facets of the FFM personality traits.
the same authors, Costa & McCrea, the NEO Five Factor Inventory (NEO-FFI), which in turn has only
60 items in total, and a completion time of 10 to 15 minutes.
2.2.2 Myers-Briggs Type Indicator
Developed by Katharine C. Briggs and her daughter Isabel B. Myers, MBTI is an introspective ques-
tionnaire that indicates people’s psychological preferences. It is based on Carl Jung’s conceptual theory,
which proposed four psychological functions by which humans experience the world, sensation, intuition,
feeling and thinking1. Carl Jung also suggests that one of these four dimensions is dominant for a person
most of the time( [JUNG, 1971]). The MBTI defines four dichotomies, they are as follows, Extraversion-
Introversion, Sensing-Intuition, Thinking-Feeling and Judging-Perceiving. This theory has received a lot
of criticism, the main arguments that are given against it are its lack of empirical evidence, validation
and its poor reliability, [McCrae and Costa, 1989a], [Pittenger, 2005].
2.3 Player Models in Game Design
In this section, we will look at psychological models that have been widely used in game design.
A popular model used in game related research is Cloninger’s Temperament and Character Inventory,
which profiles people on three main dimensions of temperament. It is used in games research for its
easiness of application to gaming motivations and actions. There are also models developed specifically
for game design, such as Demographic Game Design 1(DGD1) and Demographic Game Design 2(DGD2),
which served as basis for the previously mentioned models in this section. These models divide players
into four categories, similarly to Bartle’s taxonomy of player types, which we briefly described previously.
DGD1 was an attempt to adapt the Myers-Briggs Type Indicator to gaming, however, as stated by one of
its creators2, an accurate survey for this model was never successfully developed. DGD2 was International
Hobo’s attempt to further develop a player type theory, eventually shifting towards their trait theory:
BrainHex.
1https://www.myersbriggs.org/my-mbti-personality-type/mbti-basics/home.htm?bhcp=1 - last accessed 30/09/20182http://onlyagame.typepad.com/only_a_game/2005/08/towards_dgd2.html - last accessed 29/12/2017.
12
2.3.1 BrainHex Model
The BrainHex Model is a player satisfaction model created by International Hobo Ltd, the gaming
consulting firm behind the DGD1 model. BrainHex was derived from extensive study of neurobiological
research studies and was also directly influenced by the consulting firm’s previous surveys, such as the
DGD1 survey (which led to the creation of the DGD1 model) and the DGD2 survey( [Nacke et al., 2013]).
This player model uses seven key elements in the human nervous systems, which correspond to seven
different player types.
Each of these player types defines the player’s motivations and behavior. The Seeker likes finding
wonderful things, either strange or familiar. They often demonstrate curiosity, have a highly sustained
interest, and seek to stimulate their senses. The Survivor likes being put under menacing and scary
threats to be able to escape, they also like high-stakes and adrenaline rushes. They tend to push the
limits in order to be in dangerous situations and then feel safe again. The Daredevil likes thrills and
risk taking. They enjoy facing tough challenges while staying in control of the situation, like high-speed
chases or navigating through dizzying platforms. The Mastermind has fun solving complex puzzles and
coming up with strategies to overcome the problems they are faced with. These players enjoy puzzle-
solving, efficient decision-making and complex strategy games. The Conqueror likes a challenge, they
like to struggle before they win. They enjoy defeating seemingly impossibly difficult foes, overcoming the
struggles and beating other players. The Socialiser likes to connect with other people. Talking, helping
or just hanging around people they trust. They tend to trust other people and get angry when that trust
is broken. The Achiever likes to achieve explicit goals, motivated by long-term achievements. They strive
to fully complete a game, they are very focused on reaching the next goal and are often very obsessed.
2.3.2 Quantic Foundry’s Gamer Motivation Profile
The GMP is an attempt to create a game-specific motivational structure. It is based on 12 major
motivations, which are then grouped in pairs based on factor analysis of how they cluster together.
These motivation groups are Immersion, Creativity, Action, Social, Mastery and Achievement( [Yee,
2016]).
The Immersion cluster is composed of Fantasy, which is the desire to become someone else, some-
where else, and Story, the importance of an elaborate storyline and interesting characters. The Cre-
ativity cluster is composed of Design, which is the appeal of expression and deep customization, and
Discovery, the desire to explore, tinker, and experiment with the game world. The Action cluster is
composed of Destruction, which can be simply described as the enjoyment of chaos, guns, and explosives,
Excitement, the enjoyment of games that are fast-paced, intense, and provide an adrenaline rush. The
Social cluster is composed of Competition, which is the enjoyment of competition with other players
(duels or matches), and Community, the enjoyment of interacting and collaborating with other play-
13
ers. The Mastery cluster is composed of Challenge, which is the preference for games of skill and
enjoyment of overcoming difficult challenges, and Strategy, the enjoyment of games that require careful
decision-making and strategic thinking. The Achievement cluster is composed of Completion, which is
the desire to complete every mission, get every collectible, and discover hidden things, and Power, the
importance of becoming powerful within the context of the game world.
Another aspect that validates the GMP is its correlation with the FFM. These findings were reported
on their development blog3. A clear correlation between the Action-Social cluster and Extraversion, and
the Immersion-Creativity cluster and Openness was found. They also found a weaker, but still relevant
correlation between the Mastery-Achievement cluster and Conscientiousness.
The main problem of this model is its black box approach. Since it was developed by a private
company, Quantic Foundry, the access to its item’s distribution is protected. Moreover, the results given
to the questionnaire are in percentiles that rank the person compared to the rest of the population.
From their official website4 : “A percentile of 80% means you scored higher than 80% of gamers.”. This
means that the score any person gets when entering the same answers can vary over time. It also means
that, since the questionnaire is of voluntary participation, driven mostly out of curiosity, the population
sample might be heavily biased towards the type of gamer that would take that sort of survey, hence not
representative of the whole population.
2.4 Game Adaptation
In this section, we will describe the various applications of game adaptation, namely, difficulty balancing,
playtesting analysis, monetization, and personalized content generation, focusing mainly on the latter.
2.4.1 Difficulty balancing
It is clear to see from works which explore M. Csikszentmihalyi’s theory of flow [Csikszentmihalyi, 1990]
in games, that the challenge and difficulty has to be in the right spot throughout the game to maximize
player engagement in the task. This is the most popular and prominent form that game adaptation has
taken in the last few years [Aponte et al., 2011].
Games actively track how the player is doing and balance the difficulty of the tasks accordingly. One
way of easily achieving this is to increase the health and damage output of enemies, which games in
the The Elder Scrolls series do. However, games like Resident Evil 4 go a step further and increase
how aggressive enemies are towards the player if they are progressing through the levels too easily and
ammunition for the weapons they use the most is made scarcer [Future Press, 2005].
3https://quanticfoundry.com/2016/01/05/personality-correlates/ - last accessed on 19/12/2017.4https://apps.quanticfoundry.com/profiles/gamerprofile/rrcfXBpubyYDyT2Gs8SZQ6/ - last accessed on 21/09/2018.
14
2.4.2 Playtesting Analysis
When testing if the experience that is being provided is the intended one, it is often difficult to pinpoint
which aspects of the game or which game mechanics are the problem. Considering the player’s model
may help analyzing playtesting sessions and identifying what pleases a particular player or player type.
2.4.3 Monetization
Recently, free-to-play games have become increasingly successful on, not only mobile platforms (e.g. Clash
Royale(2016), Clash of Clans), but also PC (e.g. League of Legends, Fortnite(2017)). In this business
model, playing the game is free, so in order to be profitable, they often sell additional content and other
in-game services to the player through microtransactions. To maximize the potential for the players to
spend money on the game, player modeling can be used to help improve the understanding of why players
pay, and help identify the players that are more likely to pay. This information on player habits and
intentions might then be used to adapt the market’s content, prices or promotions to increase profits.
2.4.4 Personalized Content Generation
Personalized Content Generation is the procedural generation of content based on a player model. Pro-
cedural Content Generation is the automatic, or semi-automatic creation of content through algorithms.
It first appeared in videogames in the decade of 1980, with games like Rogue(1980) and Elite(1984),
but has become even more relevant in the last few years. This increase in popularity comes from the
time and cost reduction of development and design of large amounts of game content, and how it helps
to personalize the experience according to the player model during playtime [Yannakakis and Togelius,
2011].
Personalized Content Generation is a fairly recent research topic, with works like [Shaker et al., 2010]
and [Jennings-Teats et al., 2010], which demonstrate that platformer levels can be generated online based
on a player model, and works like [Riedl et al., 2011] in which a narrative is adapted using a player model
to maximize satisfaction. In [Hastings et al., 2009], adapting guns in a space shooter to a player model
is demonstrated to let players find new and appealing content based on their past preferences. In [Dias,
2010] the player’s profile informs the game to adapt its Heads-Up Display, difficulty and control scheme.
Player models can also serve as a basis for adapting NPC behavior. Even though agent believability
and player modeling are obviously linked, research on this implementation in games is not vast. However,
the few efforts that have been made have shown positive results. By imitating the player’s behavior,
NPCs believability has been proven to increase in [Hastings et al., 2009] and [Munoz et al., 2013]. Bots
for Unreal Tournament(1999) using a similar approach have even managed to pass the Turing test [Karpov
et al., 2012].
15
2.5 The Media Equation
The Media Equation is a theory that states that people tend to treat computers and media as people
or real places. The studies that were conducted towards the Media Equation book, used the following
method: take previous findings from the psychology and sociology fields on interpersonal relationships;
change one of the “humans” in the experiment’s statement to “computer”, e.g. “people like people that
flatter them” becomes “people like computers that flatter them”; replicate the same methodology and
replace one of the members of the relationship with a computer.
2.5.1 “Can computer personalities be human personalities?”
One study of the Media Equation in particular, [Nass et al., 1995], tests the law of similarity-attraction
between computers and people. As mentioned previously in the document, this law states that individuals
prefer to interact with others who are similar in personality to themselves. As mentioned previously in
Subsection 2.2.1, the two most relevant personality traits from the FFM in interpersonal relationships
are Extraversion and Agreeableness. Nass and his team chose to focus on Extraversion in their study,
which ranges from dominance in the positive end, to submissiveness in the opposite end.
The study was a 2x2 between-subjects experiment (n=48). The Bem-Sex Role Inventory was used
to choose subjects that fit into a dominant category and a submissive category. The dominant and
submissive subjects were randomly assigned a computer that exhibited submissive or dominant behavior.
The experiment consisted of the “Desert Survival Problem”, in which the player was assisted by the
computer to order a list of objects, in order of usefulness, in the hypothetical case of being stranded in a
desert. In order to express the computer’s dominant/submissive behavior, the experimenters manipulated
the following:
• The phrasing of the text displayed by the computer - the dominant computer used strong language,
assertions and commands, whereas the submissive computer used weaker language, suggestions and
questions;
• The confidence level expressed by the computer - the computer’s opinion was accompanied with
a 10-point scale of confidence. The dominant computer presented an average confidence level of
8.0 with a standard-deviation of 0.8, conversely, the submissive computer displayed an average
confidence score of 3.0 with a standard-deviation score of 0.8;
• The order of interaction - the dominant computer would always interact and give their opinion first,
in contrast, the submissive computer always presented their opinion after the subject had already
presented theirs;
16
• The name given to the computer - the dominant computer was given the name of Max, and the sub-
missive computer was named Linus, which were both confirmed by a pretest to suggest dominance
and submissiveness, respectively.
This study found that to convey personality we do not need very complex agents, realistic visuals,
or deep logic and artificial intelligence. Posing that “(...) even the most superficial manipulations are
sufficient to exhibit personality with powerful effects.”. Moreover, subjects preferred the computer that
was similar to them in personality, and were more satisfied with the whole experience.
There are a few things to consider when taking this study into account, namely the change in the
perception of computers from the year the study was conducted in, 1995, to nowadays. As comput-
ing devices become increasingly present in today’s society, and the general public become increasingly
computer-savvy. The first instinct towards this, is to assume that people nowadays are more skeptic
towards accepting computers as one of their own, however, according to [Johnson and Gardner, 2005],
this change might have the opposite effect. We will discuss this research in further detail in the next
section.
Another concern is the effectiveness of choosing Extraversion as the measure to align to the player’s
personality. A previous study [Paunonen and Ashton, 2001] suggests that using the personality traits’
facets, instead of the whole personality trait, leads to better results in predicting behavior. With this in
mind, in this work, we will focus on a facet of Extraversion, rather than the whole trait. Assertiveness is
the facet we will be focusing on.
2.5.2 Contemporary Research
Nowadays, Media Equation research includes efforts in robotics, such as [Konok et al., 2018], which studies
which qualities dogs have, that make people be fond of them and how to implement the same behavior
in social robots. In artificial agents, for example, in [Bosse et al., 2018], the idea of bad “consequential”
agent is presented, this is, an agent that is able to physically threaten human beings, which found a
non-conclusive relation between consequential agent and believability. Other research efforts focus even
on the influence of social media reviews on online purchase intention of movie tickets, in [Fu et al., 2018],
which found a highly significant positive effect of the law of similarity attraction, previously discussed in
this section.
In game design, some research has been conducted. In [Johnson and Gardner, 2005], the research in
“team formation between humans and computers” found a bias against computer teammates. However
they also found that highly-experienced users tend to accept a computer more easily as a teammate and
treat them more negatively than less-experienced users do, due to what the researchers call “The Black
Sheep Effect”. In the same work, further applications of the Media Equation findings on game design
are proposed. Among these, the application of the findings of the study presented above, in Subsection
17
2.5.1, to games.
2.6 Non-Player Characters
We will now take a look at the research that has been done in demonstrating companion NPC personality
and adapting its behavior to the player. We will then give two examples of scenarios in successful games
with NPC companions which we believe could have had their player experience improved by applying
NPC adaptation.
NPCs have been present in games since the early days of tabletop games, in games such as Dungeons
& Dragons(1974). It is defined in Oxford Dictionary as “A character in a role-playing or video game that
is not controlled by a player of the game” and it is a common tool used by game developers. They can
be used as coaches, opponents or companions to the player [Bakkes et al., 2012].
As an opponent, the NPC’s role is to try to match the player’s skill and provide a suitable challenge,
since it has been shown in [Scott, 2002], that if the player finds the opponents too weak, they lose interest
in the game, and if they find it too difficult, it also has been shown in [Livingstone and Charles, 2004]
and [van Lankveld et al., 2010], that the player is prone to getting frustrated and quit playing the game.
This is where player modeling comes in, helping the developers predict and monitor the player’s skill level
and adapt to it dynamically. This has traditionally been a complex AI problem and has been applied to
a wide range of game types, from Role-Playing Games to Real-Time Strategy games.
As a coach, the NPC is used to redirect the player’s attention and focus, or encourage a certain type
of behavior. When coupled with player profiling, this type of NPC can be very effective in games which
have a training purpose and personalized coaching is often a requirement.
When NPCs act as companions to the player, they are used to help, motivate or even guide the player.
It is often the case though, that the player becomes frustrated with the NPC’s behavior for an action that
goes against the player’s intentions. For example, if the player is trying to act stealthily and the NPC
rushes in to try to eliminate some threat and cause mayhem, the player’s experience might be negatively
affected. That is where player modeling can help. With player modeling in mind, the companions have
the role of behaving according to the player’s expectations made easier. By understanding the player’s
motivations or preferred behavior, the task of deciding how to act becomes quite simple.
Research on companion NPC demonstrated personality and adaptation to the player has been growing
over the past few years, with works like [Martins, 2017], [Chowanda et al., 2016] and [Filipe, 2015] showing
some success in demonstrating personality traits through NPC behavior, and works like [Doirado and
Martinho, 2010], which successfully adapts Fallout 3 ’s(2008) companion dog, Dogmeat, to better predict
player intentions and behave accordingly.
18
We will now provide two scenarios in commercially successful games with NPC companions that we
believe could have been improved if adapted to the player’s model.
2.6.1 Alyx Vance (Half-Life 2)
Half-Life 2 (2004) is a critically acclaimed game released in 2004 by the gaming industry giant, Valve.
Alyx Vance, shown in figure 2.1(a), is the NPC companion that the player interacts with, for a large
part of the game, especially during the game’s expansion, Half-Life 2: Episode Two(2007), in which Alyx
accompanies the player for the majority of its length. She is sensitive, cheerful and has a sense of humor,
evidenced by some jokes throughout the game’s course. The relationship established between Alyx and
the player is one of mutual help, as she saves the protagonist on several occasions, and vice-versa.
The scene chosen from this game is in Half-Life 2: Episode Two. The scene is one in which Alyx is
providing sniper support from a vantage point, while the player progresses through the level. Alyx proves
herself to be a great sniper, helping the player by killing as many enemies as she can have visual contact
on, for some players, this might be the ideal behavior for the NPC. However, imagine a scenario in which
a very competitive player finds themselves in this scene. The player might become irritated with Alyx’s
actions, for “stealing” the joy of killing those enemies from him. In this scenario, considering the player’s
gaming motivations could potentially prevent frustration and irritation towards not only the NPC, but
the game itself.
2.6.2 Elizabeth (Bioshock Infinite)
In Irrational Games’ 2013 hit game Bioshock Infinite(2013), you have Elizabeth (seen in figure 2.1(b))
as your companion for the majority of the game. She has a very well defined personality: very jovial,
compassionate, trusting and loves to daydream. This personality is very clearly conveyed throughout
the whole time the player interacts with her, through subtle voice queues, body language and behavior
towards the player.
There is a scene in the game in particular that demonstrates where adaptation could have impact on
player enjoyment. In this scene, Elizabeth kills a character and runs away in shock of what she has done.
In making her run away, the designers are clearly guiding the player to chase after her and this might be
a suitable expected behavior for some players. Now imagine a player that enjoys collecting everything
they can (which the game strongly encourages) being placed in this situation. They might feel inclined to
postpone chasing after Elizabeth in order to search for anything they might otherwise overlook. And in
fact, there are some items the player would miss if they immediately chased after her. So in this scenario,
the player is forced into a situation in which they have to choose between being faithful to their gaming
motivations and being faithful to how they feel at the time.
19
(a) Alyx Vance (b) Elizabeth
Figure 2.1: AI companions in successful video games.
2.7 Game Experience Evaluation
Evaluating the user experience in games has been done since the beginning of game development. It has
however evolved from just playing the game and trying to figure out why it was not fun, to scientifically
validated methods and rigorous analysis. This science driven gameplay and level testing received a huge
boost in the development of Microsoft Studios’ Halo 3 (2006)5. There are numerous methods being used
in the industry nowadays. These methods are used to understand the different contributing aspects of
the gaming experience. The user experience in games can be evaluated through different concepts, such
as immersion, fun, engagement, flow and playability [Takatalo et al., 2010].
The evaluation method of the user experience depends on the development stage of the game. We
will, however, focus on the methods applied during the implementation and testing phases. Some of the
methods that have been proven to be useful during this stage are [Bernhaupt, 2010]:
• Play testing is a play session with players, used to understand what is not working, or needs
improving in the game. It can be designed to focus on a specific aspect of the game, and the
information collected varies from in-game data, biometric measurements or any of the following
items.
• Video coding is useful to detect patterns in large amounts of video data. It can also be used to
detect out-of-game physical behavior and verbal expressiveness.
• Questionnaires are very versatile, and usually scientifically validated. They can be used to mea-
sure a multitude of dimensions of the player’s perception of the experience.
5https://www.wired.com/2007/08/ff-halo-2/ - last accessed 29/12/2017.
20
We will now study in greater detail some questionnaires that not only measure game experience, but
also the measure some dimensions of the interaction with other social beings. In particular, we will
describe the GEQ and Networked Minds Measure of Social Presence. Before we look at the question-
naires more in-depth, we need to understand the concept of social presence and its uses in measuring
interactions between humans and technology.
“Social presence is the degree to which users of a medium feel that mediated others are
spatially co-present, psychologically accessible, and behaviorally interactive.” [Biocca and
Chad Harms, 2003]
Social presence has been used to measure users’ perception of different technological artifacts in multiple
studies. In [Lee and Nass, 2005], the perceived social presence created by machine-generated voices is
measured, while in works such as [Leite, 2013] and [Schermerhorn et al., 2008] it is used as a measure of
how humans perceive social robots. In the same fashion, we will use social presence as a measurement of
the player’s perception of the NPC.
With this concept in mind, and our goal of measuring the quality of the player’s experience while
playing a game, we will now describe two questionnaires that measure these elements.
2.7.1 Networked Minds Measure of Social Presence
Networked Minds Measure of Social Presence is a questionnaire, that, although not directly designed for
games, is used to assess the user’s sense of social presence when interacting with others in a non-physical,
mediated way. It has been widely as the basis for evaluating interactions with synthetic characters, such
as social robots, in [Leite, 2013], as well as multiple other scientific research efforts.
It measures six dimensions of social presence:
• Co-presence is the degree to which the user feels as if they were in the same physical space as the
other entity.
• Perceived attentional engagement is the degree to which the users feel they paid attention to
the other entity, and the other entity paid attention to them.
• Perceived emotional contagion is the degree to which the user perceives that emotional states
transfer from the user to the other entity and vice versa.
• Perceived comprehension is the degree of mutual comprehension and understanding between
the user and the other entity.
• Perceived behavioral interdependence is the degree that the user perceives their behavior to
be dependent on the other entity’s actions and vice versa.
21
• Symmetry of social presence is the degree to which the user perceives that the other entity
shares the same state of social presence.
2.7.2 Game Experience Questionnaire
The Game Experience Questionnaire was designed to measure various aspects of the game experience,
dividing it into four different modules, each using a 5-point Likert scale. The modules are: the core
module, the in-game module, the post-game module, and the social presence module. This questionnaire
is widely used in video game based research.
The core module concerns the experience during the gameplay, and reliably distinguishes between
seven different dimensions of the player experience, these dimensions are: Sensory and Imaginative Im-
mersion, Tension/Annoyance, Competence, Flow, Negative Affect, Positive Affect and Challenge.
The in-game module, which is a shortened version of the core module, and also measures the same
seven components.
The post-game module, which concerns how the player felt after they have stopped playing, consists
of the following four components: Positive Experience, Negative Experience, Tiredness and Returning to
Reality.
The social presence module, which is used to assess psychological and behavioral involvement
of the player with other social entities, consists of three components: Empathy, Negative Feelings and
Behavioral Involvement.
2.8 Discussion
We started this proposal by ascertaining the importance of our work, we came to find that there was a
significant oversight towards NPC adaptation by the game development industry, and that it has great
potential to influence and improve the player experience.
We then started this section with the goal of understanding the different aspects that our work would
rely on. We first discussed player modeling in games, and its different methods of gathering information
about the player, and concluding that player profiling is one of the methods with the most potential to
directly affect the player’s experience. We then looked at the work done by the psychological field on
modeling personality and other psychological traits of people, and the subsequent work made in player
profiles, such as BrainHex and the GMP.
We then looked at what the information gathered from the methods described in the previous Section
could be used for, as we go into game adaptation and its main applications, focusing on personalized
content generation and found work related to NPC behavior adaptation.
After, we switched focus onto research on the interaction between people and media. The Media
22
Equation, being a cornerstone research effort in the field, was also dedicated a Section, in which we paid
closer attention to a study that suggested that the law of similarity attraction also applies to people’s
relationships with computers. We found that focusing on a facet of Extraversion seems to help with
predicting behavior, so we chose to focus on Assertiveness. And then looked at the Media Equation-
based contemporary work being done, including efforts in applying its findings to game design.
We then tried to gain a better understanding of NPCs, and we found the three roles they could
play, one of them being the companion, which has a lot of potential for improvement. Following that
logical path, we tried to look at commercially successful games that had companion NPCs and see where
adapting the companion could potentially have a positive effect on the player’s experience, finding two
scenes in Half-Life 2 and Bioshock Infinite which have the potential of being frustrating or annoying for
the player, which could be improved by using adaptation to the player’s profile.
Since our work relies on measuring the quality of the player’s experience, we took a look at how game
experience is evaluated with scientific accuracy, and came to find a multitude of methods for measuring
the player’s experience, from physical reactions, in-game data, biometric data or even perception of the
experience. We also described social presence, its use in human and non-human interaction and why
it is relevant for our work, concluding that it is a useful measurement of the player’s perception of the
NPC. Finally, we looked at two questionnaires, the Networked Minds Measure of Social Presence, which is
specifically designed for mediated interactions, and the GEQ, which is widely used to measure a multitude
of dimensions of the player experience, including a social presence module.
23
24
3Implementation
Contents
3.1 Design and Initial Prototyping . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Game Engine and Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Interaction Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 NPC Assertiveness Expression . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
25
26
Recalling our hypotheses from Subsection 1.3, we wanted to test whether the alignment of the player’s
and the NPC’s assertiveness levels increases player enjoyment. In order to test this, we needed a game
with an NPC companion. One option would have been to choose a pre-existing game that has an NPC
companion and good mod-support. However, upon extensive research, no such game was found under
ideal circumstances. Therefore, instead of using a pre-existing game, we chose to develop a 2D puzzle
platformer from scratch. This also allowed us to have as much control over the scenario design and NPC
behavior as possible.
In this chapter, we are going to describe each of these segments of the game. We start by describing the
game’s concept and the setting it takes place in. Then, in Section 3.2 we will describe the game engine and
mechanics, which include the Player controls, the NPC’s movement and behavior, the puzzle mechanics,
the level design and the dialogue system. In Section 3.3 we will describe the design and implementation
of all the interactions with the NPC. Then, in Section 3.4, we detail how the manipulation of the
NPC’s behavior to express assertiveness was done. Finally, in Section 3.5 we will summarize this chapter,
recalling the main points of the implementation phase and providing some insight on it.
3.1 Design and Initial Prototyping
The game created is called Cave Escape1. We wanted to create a setting for the game that made the player
feel like they were in the same situation, or at least similar, to the NPC. However, one characteristic we
wanted to avoid was creating the notion of the characters being part of a team, therefore avoiding a main
effect, as the one suggested in [Johnson and Gardner, 2005]. To do this, the setting had to be somewhat
neutral in team formation queues, conversely we wanted to avoid settings such as sports based settings
and military based settings. The “trapped in a cave together” seemed like a good compromise between
having the player and NPC cooperate naturally and not invoking team based reactions. The name given
to the game, Cave Escape, is meant to reinforce the premise that the player and NPC are stuck in the
cave together.
Another requirement for the concept was creating a context in which the player and NPC could
interact repeatedly. The two doors were our answer to that requirement. Standing side by side, and
only letting one person through before closing, the doors prompted a brief discussion between the two
characters, on which door each of them would go through.
An element that was left out of the scope of this research was the NPC’s gameplay behavior, this is,
how the NPC performed the actions the player was required to perform to progress in the game.
The development process of the game was iterative, starting with the design and paper prototypes
of the game, incrementally implementing mechanics and finally designing and implementing of the in-
teraction scenarios. In the initial implementation phase, two prototypes were developed and tested. We
1https://youtu.be/lEkBwDwNa2g - Video of the game with assertive NPC.
27
Figure 3.1: First room. Showing the cave setting.
used the Unity engine to develop the game, because of the existing familiarity with the tool, easy and
fast C# script development, 2D game development support and the vast community and subsequent vast
source of knowledge about the engine. When the first player controls prototype was tested, only the
keyboard controls were implemented. The informal testing for this prototype was done with four people,
and the results revealed a demand for gamepad controls, as well as overall more intuitive controls. The
second prototype implemented included only two levels seen in Figures B.1 and B.2 (all levels of the game
created for this work can be seen in Appendix B), and two mechanics. This prototype was tested with
five people, and these results, gathered informally, were the first indication of the game’s appeal. This
feedback also helped with the level design and tutorial process.
3.2 Game Engine and Mechanics
Figure 3.2: Cave Escape’s logo.
The purpose of the game mechanics in this experiment is to give a backdrop for the interactions
between the NPC and the player. With this in mind, we wanted the puzzles to remain simple enough
not to overshadow the interactions, but still engaging enough for the player experience to be positive.
The game is based on a couple of mechanics from classic puzzle platformers such as Portal(2007),
however, the player’s movement is similar to that of Braid(2008). They are able to move to either side,
28
and jump. They can also pick up and drop boxes, and enter doors.
Figure 3.3: Game’s mechanical components.
The main goal of each level is to open the locked doors, seen in Figure 3.3(3), to progress to the next
level. To unlock the doors, the player has to activate the triggers, seen in Figure 3.3(2), spread around
the level by placing the boxes, seen in Figure 3.3(1), on them. Above each door, there are indicators
representing the amount of triggers associated with it, and their state (Figure 3.3(4)). This was done to
help the player keep track of what they have left to do. Once all the triggers in the room are activated,
the door opens to let the player through.
There are three other mechanics that are introduced in the game, trampolines (Figure 3.3(6)), check-
points (Figure 3.3(5)) and spikes (Figure 3.3(7)). The trampolines boost the player and boxes in the
direction they are facing, allowing the player and boxes to reach otherwise unreachable places. Although
the player is able to use the trampoline while holding a box, they get weighed down by it, and are
propelled with less intensity than otherwise. The boxes on the other hand get a larger impulse than the
player. This fact is the solution to the first puzzle that joins trampolines and boxes, since the player has
to place the box, on its own, on the trampoline, and then get up on their own (seen in Figure 3.4).
The checkpoints and the spikes work together. The checkpoints save the game as the player passes
by them, and the spikes are the only thing in the game that ”kills” the player and sends them back to
the last checkpoint. However, these checkpoints are also used if the player decides to press the “Restart
29
Figure 3.4: Box and trampoline puzzle solution.
the level” button.
3.2.1 NPC Script
We developed a simple reactive script for the NPC’s movement and behavior, responding to triggers
placed in the level. The way the NPC starts talking, jumps, changes the direction their facing and enters
doors, is by setting these triggers off.
Besides those triggers, the NPC also has a movement script similar to the one developed for the player.
It allows the NPC to jump, walk towards a goal and change direction. During the door choosing process,
the NPC’s behavior is dependent on its assertiveness level. This behavior is described in more detail in
Section 3.4.
We also developed a movement algorithm to be used only in the final scene. With this script the
NPC’s walking speed is based on the player’s position: slower if the player is trailing them, and faster if
the player is ahead. This script was developed to make the ending as ambiguous as possible when both
characters run for the door. For more details on this scenario, see Subsection 3.3.2.
The visuals for the NPC were made to be similar to the player character’s and neutral in conveying
personality and team identity. Both the player and NPC can be seen in Figure 3.5.
3.3 Interaction Scenarios
We developed a simple dialogue system through which the NPC and the player could interact.
30
Figure 3.5: NPC and Player character, left and right respectively.
The dialogue input is parsed from a .txt file in order to allow changing between different dialogue
versions as easily as possible. This feature was crucial for the first study we conducted, which we describe
in detail in Section 4.1, since there were four different dialogue scripts to use: one for each combination of
Assertive/Non-assertive x Friendly/Unfriendly (all of the dialogue scripts used in this work can be seen
in Appendix C).
In the dialogue interactions, when the player is prompted to give an answer, they have four options to
choose from (seen in Figure 3.6). The four options were the combination of agreeing or disagreeing and
non-assertive or assertive phrasing, to provide the player a way to express their own assertiveness level.
Figure 3.6: Dialogue box with player options.
3.3.1 Normal Interactions
In the full game, the player and the NPC interact in five different rooms throughout the game. In three
of them, the player and NPC have to choose between themselves who goes through which door. These
rooms have the same dialogue structure:
1. Introduction and small comment by the NPC (Figure 3.7);
31
2. Simple reply by the player;
3. NPC’s comment on the door choosing process;
4. Statement of intention by the NPC;
5. Player’s decision;
6. NPC’s response to the decision.
Figure 3.7: Single line Dialogue box.
In the third interaction room however, although it starts the same way as the other three, when it’s
time to discuss the doors, the NPC suggests a game of Rock, Paper, Scissors. This was done to add
some variety to the door choosing process, and avoid repetitiveness. This interaction is, as all dialogue
in the game, adapted to the NPC’s assertiveness level. The assertive NPC gives the idea of Rock, Paper,
Scissors as a command rather than a suggestion or offer, as the non-assertive NPC does.
3.3.2 Final Interaction
The final interaction adds a twist to the formula. Instead of choosing between two doors, the player
and NPC are given only one door for both of them. Making the situation one of competition instead of
cooperation. In this scenario, the structure varies in the second half of the interaction. It follows the
following structure:
1. Small comment by the NPC on the previous level;
2. Simple reply by the player;
3. NPC notices the singular door;
4. NPC comments that only one person can escape;
32
5. NPC states that they want to be the one to go through the door;
6. Player’s decision to try to run for the door or stay behind.
Figure 3.8: End scene player options.
The player is free to stay behind or run for the door, regardless of what option they choose (seen in
Figure 3.8). This is, even if they choose the “Go ahead, I would rather stay.” option, the game still allows
them to run for the door, and in the case of choosing the “I’m not going to stay here either!” option,
they are still allowed to stay behind and sacrifice themselves.
Following the player’s decision, the NPC walks towards the door, regardless of the player’s choice.
The NPC walks slowly if the player decides not to run for the door and stay behind, and, before entering
the door, turns back and thanks the player for their “sacrifice”. As the NPC enters the door, the game
fades to black and ends. If, instead the player decides to run for it, the NPC is programmed to follow
the player as closely as possible - walking slowly if the player is trailing them, and faster if the player is
ahead, and as both of them get close to the door, the game starts slowing down incrementally, and fades
to black just before they get to the door, giving the game an ambiguous ending. This was done to avoid
having the idea of defeat or victory influence the player’s experience.
3.4 NPC Assertiveness Expression
In order to express assertiveness, we manipulated the NPC’s text according to the following parameters:
1. The phrasing of the text used by the NPC - In accordance with the theory on assertiveness expression
studied in Section 2.5, the assertive NPC uses assertions and statements rather than questions and
33
(a) Running for the door. (b) Fading to black. (c) ”The end” screen.
Figure 3.9: Game ending progress.
suggestions. In contrast, the non-assertive NPC uses questions, suggestions and seems uncertain of
they are saying;
2. The name of the NPC - The name given to the assertive NPC was the same as the corresponding
assertive computer in [Nass et al., 1995], “Max”. In the same vein, the non-assertive NPC was
given “Linus” as their name;
3. The NPC’s response to the player’s answer - In the first two interactions, the NPC responds to the
player’s decision, these replies depend on the NPC’s assertiveness level, and the player’s decision,
this is, whether the player goes with or against the NPC’s intents. The assertive NPC expresses
their dissatisfaction if the player chooses the door the NPC declared as their intended door. If
the player goes against the NPC’s choice in the first two interactions, the assertive NPC revolts
and imposes their choice on the player, as seen in the flowchart in Figure 3.10. Conversely, the
non-assertive NPC makes an effort to not seem bothered by the player’s choice when it opposes
theirs, and accepts whatever the player decides;
4. The order of declaration of intent - The assertive NPC states its preference for a door right away,
before the player has any chance to express themselves. On the other hand, the non-assertive NPC
is ambiguous when presenting their intentions, and instead asks the player for their choice, and if
they would like to choose.
We used three of the four parameters used in the Media Equation study mentioned previously, leaving
out only the “confidence level”. This was left out due to the differences between the interactions. In
[Nass et al., 1995], the computer gives their opinion on a subject matter. However, in our interaction
scenario, the NPC merely expresses a choice, based on instinct and blind preference, which means that
the “confidence level” is incongruous with our scenario.
34
Figure 3.10: NPC logic for door choosing process.
35
3.5 Summary
We started this chapter by deciding to implement a game with an NPC companion with which we could
test our hypotheses. Then, in Section 3.1, we described our reasoning for the choice of setting and context
for the game, we also discussed the design, initial prototyping and their corresponding testing phases. In
Section 3.2 we enumerated and explained all the mechanics created for Cave Escape and the logic behind
each level. We also described the NPC’s movement script. Next, in Section 3.3 we described all the
interaction scenarios between the NPC and the player, including the default interactions’ structure and
the variations on it (Rock, Paper, Scissors, and final interaction - single door). Finally, we enumerated
the NPC’s elements that were manipulated to express assertiveness.
36
4Studies
Contents
4.1 Preliminary Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Main Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
37
38
In this chapter we will describe the two studies conducted towards this work, and discuss the results
gathered. First, in Section 4.1 we will present the preliminary study conducted to test the degree to
which the dialogue conveys assertive and non-assertive personalities, and the subsequent changes that
were made based on the results. Then, in Section 4.2, we will describe the main study that was conducted
with 48 people to test our hypotheses, present the results obtained, describe the statistical analysis used
and reflect on the emerging implications.
4.1 Preliminary Study
One of the concerns we had when developing the dialogue system, was the relationship between assertive-
ness expression and other personality traits, mainly the friendliness trait.
To test if there was a correlation between these personality traits, we held an open test during the 2018
Montra de Jogos (MOJO), a yearly game demonstration fair held at Instituto Superior Tecnico (IST)’s
Taguspark campus. We also used this experiment to test the player controls’ quality, the first two levels,
and the overall quality of the game with a larger sample than the previous informal tests referred in
Chapter 3.
In this study, we had 20 participants, out of which 6 were female, 14 were male, and the ages varied
between 15 and 25 (mean=21.65, s=2.92). Each person only played the game once with a specific NPC
personality. Given the purpose of this study, we wrote four different dialogue scripts(see Figure C.1):
• Assertive-Friendly;
• Non-Assertive-Friendly;
• Assertive-Unfriendly;
• Non-Assertive-Unfriendly;
The text was manipulated to express assertiveness according to the theory presented in Section 3.4, and
friendliness through “reverse-engineering” of the friendliness items in the questionnaire we will describe
next.
4.1.1 Questionnaire
To measure the NPC’s assertiveness and friendliness levels in this experiment we used the Abridged
Big Five-Dimensional Circumplex (AB5C)’s assertiveness and friendliness scales (see Appendix A.1).
The AB5C was chosen because it relates assertiveness and friendliness in the context of interpersonal
relationships [Hofstee et al., 1992]. These scale were composed of 10 items and the assertiveness scale
39
has a 0.75 cronbach alpha, while the friendliness scale has a 0.85 one. A 5-point Likert-scale was used to
rate each of the items of these scales, ranging from “Very inaccurate” to “Very accurate”.
Since the items are written for self-assessment purposes, and we intend to measure the NPC’s per-
sonality traits, we converted the 20 items into the third-person using the International Personality Item
Pool (IPIP)’s guide found on their official website1.
The questionnaire was administered in accordance to the methods described in the official IPIP
website. Each of the 5 levels of the Likert-scale were assigned a score from 1 to 5, based on whether they
were +keyed, or -keyed. The +keyed items are scored in ascending order, while the -keyed items are,
in contrast, scored in descending order. For example, the fourth level of the Likert-scale, “Moderately
accurate”, when scoring a +keyed item, is assigned a score of 4, and scoring a -keyed item, it is assigned
a score of 2.
4.1.2 Results
The two scales used gave us an assertiveness score and a friendliness score, which was the mean of the
corresponding items. Given our small treatment groups (n=5), we used a Kruskal-Wallis test, and found
that there was a statistically significant difference in the assertiveness scores between the different groups,
X 2(3)=8.215, p=0.042, with a mean rank assertiveness score of 17.00 for Assertive-Friendly, 8.30 for the
Assertive-Unfriendly, 7.80 for the Non-Assertive-Friendly and 8.90 for the Non-Assertive-Unfriendly.
Given this result, we used Mann-Whitney U tests on pairs of gathered scores. Comparing the
assertiveness scores between the Assertive-Friendly and Non-Assertive-Friendly NPCs, with means of
4.08 and 3.34, respectively, the Assertive-Friendly NPC was ranked significantly higher than the non-
assertive (Mann-Whitney U =1.5, p=0.02 two-tailed). However, when comparing assertiveness scores of
the Assertive-Unfriendly and Non-Assertive-Unfriendly NPCs, there was no significant difference (Mann-
Whitney U =12.5, p>0.05).
Another Mann-Whitney test indicated that the assertiveness score of the Assertive-Friendly NPC is
significantly higher than that of the Assertive-Unfriendly NPC(Mann-Whitney U =1.5, p=0.02 two-tailed)
4.1.3 Analysis
The version of the game used in this fair only had two levels and two rooms. This meant that the
interaction period with the NPC was merely 4 dialogue sequences.
The friendly NPCs had assertiveness scores more closely aligned with the ones that they were intended
to convey. This is, the Assertive-Friendly NPC had a higher assertiveness score than the Assertive-
Unfriendly NPC, and the Non-Assertive-Friendly one had a lower assertiveness score than the Non-
Assertive-Unfriendly NPC. From which we concluded that higher levels of friendliness convey different
1https://ipip.ori.org/Third-Person-Items.htm - last accessed 25/09/2018
40
levels of assertiveness better. We believe this correlation happens because higher levels of friendliness
translate into more opportunities for interaction. Given the low amount of interactions in this version of
the game, this might have had a bigger effect than usual on the perception of the NPC’s personality.
Given these results, we decided to set the level of friendliness for our main study on the positive end
of the scale, and maintain a similar level for both the assertive and non-assertive personalities.
4.2 Main Study
The two dialogue scripts used in the main study’s game can be seen in Figures C.2 and C.3. In the rest of
this Section, we will describe the study conducted to test if the player’s assertiveness level being aligned
with the NPC’s increases the player’s enjoyment of the game and affinity for the NPC. We will start with
describing the questionnaires used. Then, we will detail the procedure used in the experiment. Finally,
we will present and analyze the results obtained.
4.2.1 Questionnaire
The questionnaire used was composed of three sections, player identification and assertiveness self-
assessment, game experience and NPC social presence evaluation, and NPC assertiveness assessment
(see Appendixes A.2 and A.3).
In the first section, for the assertiveness self-assessment, we used the NEO PI-R’s 10-item assertiveness
scale, which has a cronbach alpha of 0.84 (reported by its authors), introduced by “How accurately do
you believe that each of these sentences applies to yourself.”, and evaluated by a 5-point Likert scale.
The second section of the questionnaire is composed by the GEQ’s In-game Module and Social Pres-
ence Module. The Social Presence Module was adapted to include the NPC’s name, instead of the word
“other” in the phrasing of the items. The items in this section are introduced with the following sentence:
“Please indicate how you felt while playing the game for each of the items:”, and are scored with a 5-point
Likert scale from “not at all” to “extremely”.
In the last section, we measure the NPC’s assertiveness with the same 10-item assertiveness scale that
was used in the first section of the questionnaire, however, to the same effect of the personality items
used in the preliminary study, the items were converted to third-person phrasing.
4.2.2 Procedure
The participant was allowed to be a part of the experiment either remotely or in person. This was done
for two reasons, to allow those that did not have the means to play on their own hardware to participate,
and because participants were more likely to participate if they could do so at their own time, or at home.
We will now describe the procedure taken with each participant.
41
1. Introduction - The participant, upon opening the online questionnaire, was explained that they
would be part of an experiment to test a game, that the experience was of voluntary participation
and took around 25 to 30 minutes;
2. Initial Questionnaire - The participant was prompted to fill the first section of the questionnaire,
the identification and assertiveness self-assessment;
3. Starting the game and getting to know the controls - The participant was told of the various options
of playing with a game controller or keyboard and mouse. Then, upon opening the game and
pressing “Play”, was explained the controls of the game;
4. Escaping the Cave - The participant played the game from start to finish;
5. Final Questionnaire - When the participant was finished playing the game, they filled the two last
sections of the questionnaire, described previously in Subsection 4.2.1.
6. Saying goodbye - Finally, the participant was thanked for participating.
4.2.3 Sample
The participation in this experiment was voluntary. The participants were randomly approached, in
person and via social media, and asked to participate in our experiment.
We had a sample of 48 people, aged between 21 and 29 (Mdn=23.75), distribution of gender was 7
females and 41 males. Regarding game experience, most of the sample is an avid gamer, choosing the
“I reserve time in my schedule to play video games.”, as seen in Figure 4.1(a). In Figure 4.1(b) we can
observe the experience the participants had with the puzzle-platformer genre. The discrepancy in game
and genre experience was not caused by the methods used to disclose the experiment, given its publication
through generally game neutral means. We believe this difference comes from the increased likeliness of
more game-savvy people to participate in such an experiment voluntarily.
In our sample, the median for assertiveness score (calculated in the same manner as in Subsection
4.1.1), was 3.35, with a standard deviation of 0.60. This median is the cutoff point used to divide our
sample into assertive and non-assertive players.
Given our sample size of 48 participants we can observe the four treatment groups and how many
observations per treatment group we had in table 4.1.
Assertive NPC Non-Assertive NPCAssertive Player 8 16
Non-Assertive Player 16 8
Table 4.1: Participant distribution in experiment.
42
(a) Distribution of participants’ experience withvideo games.
(b) Distribution of participants’ experience withthe 2D puzzle platformer video game genre.
Figure 4.1: Sample’s gaming habits.
4.2.4 Manipulation Check
Normalization analysis of the NPC’s assertiveness scores, using a Shapiro-Wilk test revealed them to be
approximately normal. To analyze the player’s perception of the NPC’s assertiveness, we used a two-way
ANOVA model, considering the following independent variables:
• Player’s assertiveness level (non-assertive, assertive);
• NPC’s demonstrated assertiveness level (non-assertive, assertive).
Consistent with Hypothesis H1, the assertive NPC was perceived as significantly more assertive than
the non-assertive NPC, F (1,44)=33.467, p<0.001 (see Figure 4.2). There was no main effect for the
player’s assertiveness level, F (1,44)=1.987, p=0.166, and no significant interaction effect was found,
F (1,44)=0.482, p=0.491.
4.2.5 Results
We will now describe the analytical process of the results from GEQ’s in-game module and social presence
module. We started by trying to apply ANOVA on Ranks to the data, however, a normal distribution
could not be achieved. Therefore, we used a Kruskal-Wallis H test, separating the data by each treatment
group seen in Table 4.2.
This analysis revealed that there was a statistically significant difference in Tension scores between
the different groups, X 2(3)=13.513, p=0.004, with a mean rank Tension score of 30.44 for group A, 19.91
for group B, 31.63 for group C and 13.50 for group D.
Applying a Mann-Whitney U test with Bonferroni correction for each pair of groups, revealed that
Non-Assertive players report significantly (p<0.0125) higher Tension scores when interacting with the
43
Figure 4.2: Mean perception of NPC assertiveness score as a function of player and NPC assertiveness level.
Assertive NPC Non-Assertive NPCAssertive Player Group A Group B
Non-Assertive Player Group C Group D
Table 4.2: Treatment Groups used in Kruskal-Wallis H test.
Assertive NPC (Mean rank=15.56) than with the Non-Assertive NPC (Mean rank=6.38), U =15.00,
p=0.002.
Hypothesis H2 predicted that the players would perceive their game experience more positively when
their assertiveness levels were aligned with the NPC’s own assertiveness level. For this Hypothesis, we
have to take into account the seven components in the GEQ’s In-game module.
Hypothesis H3 predicted that the players would prefer to interact with the NPC, when their as-
sertiveness levels were aligned with the NPC’s own assertiveness level. For this Hypothesis, we are going
to analyze each of the three components of the social presence module of GEQ.
For both Hypothesis H2 and H3, and since the normality assumption for ANOVA wasn’t met, we used
a Mann-Whitney U test, separating the observations into “aligned” (groups A and D) and “not-aligned”
(groups B and C) to test the effect of assertiveness alignment.
There was no significant difference between the aligned and not-aligned groups, as seen in Table 4.3.
4.2.6 Analysis
A potential reason for the significantly higher Tension score registered when Non-Assertive players inter-
acted with the Assertive NPC, is the NPC’s imposition of their choice on the player. The player, being
44
Mean rankAligned
Mean rankNot Aligned
Mann-Whitney U p
Competence 28.47 22.52 192.5 .152Sensory And Imaginative Immersion 24.06 24.72 249.0 .877Flow 26.19 23.66 229.0 .551Tension 21.97 25.77 215.5 .352Challenge 26.09 23.70 230.5 .572Negative Affect 19,88 26.81 182.0 .086Positive Affect 29.03 22.23 183.5 .104Empathy 27.03 23.23 215.5 .374Negative Feelings 21.34 26.08 205.5 .266Behavioral Involvement 26.94 23.28 217.0 .393
Table 4.3: Mann-Whitney U test results for each GEQ component, separated by “aligned” observations.
Non-Assertive, would feel frustrated if they wanted the same door, but would not express that feeling
to avoid conflict. For context, the items that score this component are “I felt frustrated” and “I felt
irritable”.
An aspect that is worth considering when reasoning about the results for Hypotheses H2 and H3 is
the difference in contexts of interaction between the Media Equation study [Nass et al., 1995], and our
scenario. The scenarios of interaction implemented in this work introduces certain elements that might
have an effect on the player’s relationship with the NPC. Namely, a potential conflict, when both parties
want the same door, which introduces a competitive component to the interaction, in contrast to the
interaction in [Nass et al., 1995] which is merely cooperative. Another aspect that was not present in
the Media Equation study’s interaction was the potential consequences of the conversation. From the
player’s point of view, the choice of door could lead to different levels, which meant that this decision
could lead to an easier level, and that the NPC wanting one of the doors might mean they know which
level is easier.
4.3 Summary
In this Chapter, we described the two formal studies conducted to validate our implementation, and
test our hypotheses. In the first study, presented in Section 4.1, we tested the effect, different levels of
friendliness have on the player’s perception of the NPC’s assertiveness level. We found that a higher level
of friendliness amplifies assertiveness expression, given the significantly higher level of assertiveness regis-
tered for the Assertive-Friendly NPC over the Non-Assertive-Friendly one, and there being no difference
in assertiveness levels between the Assertive-Unfriendly and Non-Assertive-Unfriendly NPCs.
In Section 4.2, we presented the main study in this work, in which we put our hypotheses to the
test. Hypothesis H1 proposed that the participants would perceive the Assertive NPC as more assertive
than the Non-Assertive NPC. In accordance with this hypothesis, the Assertive NPC was rated as
45
significantly higher than the Non-Assertive NPC. Hypothesis H2 predicted that the alignment of the
assertiveness levels of the player and NPC would result in the player rating their experience as significantly
more enjoyable. Hypothesis H3 predicted an increase in the player’s reported social presence when the
assertiveness levels of the NPC matched theirs. Contrary to both Hypotheses H2 and H3, there was no
significant difference in the player’s enjoyment and reported social presence with relation to the alignment
of assertiveness levels. However, a significant increase was found in the player’s reported Tension score
when a Non-Assertive player was paired with the Assertive NPC over the Non-Assertive NPC.
46
5Conclusions
Contents
5.1 Summary of Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.2 Work Accomplished . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
47
48
In this chapter we will attempt to summarize our work, including the testbed game developed, the
main experiment and the implications of the results obtained. Then, we will review our main contributions
and accomplishments in Section 5.2. Finally, in Section 5.3, we will present some possibilities on how to
improve the work that we started, and other research routes that might be interesting to pursue.
5.1 Summary of Dissertation
We began this work with the intent of testing if the law of similarity attraction would apply to player-NPC
relationships. With that goal in mind, we started by reviewing the state of the art in player modeling
and game adaptation, while also studying research on the relationships between people and media. We
also reviewed recent work done in the field of NPCs and the different techniques used to measure game
experience.
We developed a testbed game called Cave Escape. This game is a 2D puzzle-platformer with a
companion NPC, which provides a context for repeating interpersonal interactions, in the form of the
discussion and subsequent decision of which door each character will go through. The companion NPC
expresses their personality merely through the manipulation text interactions. Given the encapsulation of
the dialogue tree in a .txt file, writing and implementing new behaviors is easily accessible. Accompanying
the development of the testbed game, informal playtesting sessions were conducted to ensure its player
controls and level design quality.
Then, we conducted a user experiment in MOJO (n=20) and confirmed that a higher level of friendli-
ness allowed the NPC’s assertiveness to be perceived more accurately. With this in mind, we implemented
two friendly behaviors for the companion NPC, an Assertive behavior and a Non-Assertive one.
We conducted a 2x2, between-subjects experiment (n=48) to test the hypotheses posed earlier in
the document. We found that the players perceived the NPC endowed with assertive characteristics as
significantly more assertive than the NPC that exhibited non-assertive characteristics. We also found
no difference in the levels of enjoyment of the experience and social presence, when the player’s and
NPC’s assertiveness levels were aligned. However, we found significantly higher Tension scores when
Non-Assertive players were matched with the Assertive NPC.
5.2 Work Accomplished
In this section we will describe the highlights of the work and achievements accomplished in the devel-
opment of this project:
• Research Methodology - We proposed a research methodology of taking Media Equation findings
and applying them to video games. We explain this methodology in further detail in Section 5.3.
49
• Testbed Game - We developed a 2D puzzle platformer with a companion NPC. This game
serves as a testbed game for NPC based experiments, as their dialogue can be easily changed. The
game also features two interchangeable behaviors for the NPC that have been validated to exhibit
Assertive and Non-Assertive behaviors.
• Development Process - The development process of the testbed game was iterative, with informal
and formal testing phases to guide it in the right direction. The quality of the game experience
was controlled, and informally confirmed as good enough, but not too good, in order not to have a
substantial effect on the player’s experience and block the remaining dimensions of the experiment.
• Findings - We found that the NPCs that had higher levels of friendliness, had its assertiveness levels
perceived more accurately by the player. We also found that to avoid provoking higher Tension in
Non-Assertive players, developers can endow their companion NPCs with Non-Assertive personality
traits. Lastly, when aligning the NPC’s and player’s assertiveness levels, we found no significant
difference in the player’s enjoyment of the experience and reported social presence, compared to
when the levels were not aligned.
5.3 Future Work
From the research methodology described in the previous section, and work mentioned in Subsection
2.5.2, we believe that taking other Media Equation implications and applying them to video games might
lead to worthy research endeavors. A possible experiment design could be extrapolated from the Media
Equation’s experiment design itself. Taking findings from the Media Equation replacing the computer
with an NPC, and inserting the interaction into a video game context.
One of the findings is that gain/loss theory applies to people’s interactions with media, this is, people
will be more attracted to computers that initially dislike them and end up liking them, than ones that
consistently like them from the start [MOON and NASS, 1996]. This could easily be applied to video
games, for example, in an Role Playing Game (RPG), developing the two behaviors described previously
for an NPC, and the same effect of gain/loss theory is observed.
Another interesting finding is that people feel like they owe a favor to a computer, after said computer
has done them a favor [Nass and Moon, 2000]. This effect can be replicated in a video game context.
For example, have, for half of the sample, an NPC help the player with a problem, such as a fight or
puzzle and afterward, have the same NPC ask the player for another favor, such as a search for as many
materials as possible(it should be a favor in which the player could choose to which degree they want to
help). For the other half of the sample, reverse the order of the favor exchange, and observe if there is a
significant difference in the amount of work performed by the player.
50
Another route to follow is focusing on other facets of Extraversion and Agreeableness, given that in
interpersonal relationships, they seem to be the most important personality traits.
In this work, we left out of the scope some of the elements NPCs possess, which are at game designers’
disposal, and instead manipulated only the NPC’s dialogue. Some other characteristics that could be
manipulated to express personality traits are:
• The visuals of the NPC, including physical appearance, expressive animations, both facial and
bodily;
• Voice-overs, intonation, higher and lower pitched voices;
• Gameplay behavior, for example, acting more sure of a guess in a puzzle, asking the player which
role they would like to take in a cooperative puzzle, running into a mob of enemies, etc.;
As mentioned previously in Section 4.2.6, the interaction context in this work might have had an effect
on the player’s experience, therefore, different types of interactions should be tested. For example, fully
cooperative efforts, or direct competition environment, or love interest, or even a nemesis relationship, in
which the NPC plays the player character’s archenemy.
51
52
Bibliography
[Aponte et al., 2011] Aponte, M.-V., Levieux, G., and Natkin, S. (2011). Difficulty in Videogames: An
Experimental Validation of a Formal Definition. In Proceedings of the 8th International Conference
on Advances in Computer Entertainment Technology, ACE ’11, pages 49:1–49:8, New York, NY, USA.
ACM.
[Asteriadis et al., 2008] Asteriadis, S., Karpouzis, K., and Kollias, S. (2008). A Neuro-fuzzy Approach
to User Attention Recognition. In K\rurkova, V., Neruda, R., and Koutnık, J., editors, Artificial
Neural Networks - ICANN 2008: 18th International Conference, Prague, Czech Republic, September
3-6, 2008, Proceedings, Part I, pages 927–936. Springer Berlin Heidelberg, Berlin, Heidelberg.
[Bakkes et al., 2012] Bakkes, S. C. J., Spronck, P. H. M., and van Lankveld, G. (2012). Player behavioural
modelling for video games. Entertainment Computing, 3(3):71–79.
[Bartle, 1996] Bartle, R. A. (1996). Hearts, Clubs, Diamonds, Spades. Technical report, MUSE Ltd,
Colchester, Essex.
[Bernhaupt, 2010] Bernhaupt, R. (2010). User Experience Evaluation in Entertainment. In Bernhaupt,
R., editor, Evaluating User Experience in Games: Concepts and Methods, pages 3–7. Springer London,
London.
[Biocca, 1997] Biocca, F. (1997). The Cyborg’s Dilemma: Progressive Embodiment in Virtual Environ-
ments [1]. Journal of Computer-Mediated Communication, 3(2):0.
[Biocca and Chad Harms, 2003] Biocca, F. and Chad Harms, P. (2003). Guide to the Networked Minds
Social Presence Inventory v. 1.2: Measures of co-presence, social presence, subjective symmetry, and
intersubjective symmetry.
[Bosse et al., 2018] Bosse, T., Hartmann, T., Blankendaal, R. A. M., Dokter, N., Otte, M., and Goed-
schalk, L. (2018). Virtually Bad: A Study on Virtual Agents That Physically Threaten Human Beings.
In Proceedings of the 17th International Conference on Autonomous Agents and MultiAgent Systems,
AAMAS ’18, pages 1258–1266, Richland, SC. International Foundation for Autonomous Agents and
Multiagent Systems.
53
[Chowanda et al., 2016] Chowanda, A., Flintham, M., Blanchfield, P., and Valstar, M. (2016). Playing
with social and emotional game companions. Lecture Notes in Computer Science (including subseries
Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 10011 LNAI:85–95.
[Costa and McCrae, 1992] Costa, P. T. and McCrae, R. R. (1992). Revised NEO personality inventory
(NEO PI-R) and NEP five-factor inventory (NEO-FFI) : professional manual. Odessa, Fla. (P.O. Box
998, Odessa 33556) : Psychological Assessment Resources.
[Csikszentmihalyi, 1990] Csikszentmihalyi, M. (1990). Flow: The Psychology of Optimal Experience.
Perennial Modern Classics. Harper & Row.
[Dias, 2010] Dias, R. (2010). inFlow Adapting Gameplay to Player’s Personality. M.Sc. thesis, Instituto
Superior Tecnico, Universidade de Lisboa.
[Doirado and Martinho, 2010] Doirado, E. and Martinho, C. (2010). I mean it!: detecting user inten-
tions to create believable behaviour for virtual agents in games. Proceedings of the 9th International
Conference on Autonomous Agents and Multiagent Systems: volume 1-Volume 1, (Aamas):83–90.
[Filipe, 2015] Filipe, A. (2015). CoBelievable – The Effect of Cooperation in Believability. M.Sc. thesis,
Instituto Superior Tecnico, Universidade de Lisboa.
[Fu et al., 2018] Fu, S., Yan, Q., and Feng, G. C. (2018). Who Will Attract You? Similarity Effect
Among Users on Online Purchase Intention of Movie Tickets in the Social Shopping Context. Int. J.
Inf. Manag., 40(C):88–102.
[Future Press, 2005] Future Press (2005). Resident Evil 4: The Official Strategy Guide. Future Press.
[Hastings et al., 2009] Hastings, E. J., Guha, R. K., and Stanley, K. O. (2009). Automatic content
generation in the galactic arms race video game. IEEE Transactions on Computational Intelligence
and AI in Games, 1(4):245–263.
[Hofstee et al., 1992] Hofstee, W. K., de Raad, B., and Goldberg, L. R. (1992). Integration of the Big
Five and Circumplex Approaches to Trait Structure. Journal of Personality and Social Psychology,
63(1):146–163.
[Jennings-Teats et al., 2010] Jennings-Teats, M., Smith, G., and Wardrip-Fruin, N. (2010). Polymorph:
A model for dynamic level generation. Proceedings of the 6th AAAI Conference on Artificial Intelligence
and Interactive Digital Entertainment, AIIDE 2010, pages 138–143.
[Johnson and Gardner, 2005] Johnson, D. and Gardner, J. (2005). Effects of team-based computer in-
teraction: The media equation and game design considerations. Lecture Notes in Computer Science
54
(including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 3711
LNCS:468–479.
[JUNG, 1971] JUNG, C. G. (1971). Collected Works of C.G. Jung, Volume 6: Psychological Types.
Princeton University Press.
[Karpinskyj et al., 2014] Karpinskyj, S., Zambetta, F., and Cavedon, L. (2014). Video game personali-
sation techniques: A comprehensive survey. Entertainment Computing, 5(4):211–218.
[Karpov et al., 2012] Karpov, I. V., Schrum, J., and Miikkulainen, R. (2012). Believable bot navigation
via playback of human traces. Believable Bots: Can Computers Play Like People?, 9783642323:151–
170.
[Konok et al., 2018] Konok, V., Korcsok, B., Miklsi, d., and Gcsi, M. (2018). Should We Love Robots?
The Most Liked Qualities of Companion Dogs and How They Can Be Implemented in Social Robots.
Comput. Hum. Behav., 80(C):132–142.
[Lee and Nass, 2005] Lee, K.-M. and Nass, C. (2005). Social-Psychological Origins of Feelings of Presence:
Creating Social Presence With Machine-Generated Voices. Media Psychology, 7(1):31–45.
[Leite, 2013] Leite, I. (2013). Long-term interactions with empathic social robots. PhD thesis, Instituto
Superior Tecnico, Universidade de Lisboa.
[Livingstone and Charles, 2004] Livingstone, D. and Charles, D. (2004). Intelligent Interfaces for Digital
Games. AAAI-04 Workshop on Challenges in Game AI, pages 1–5.
[Martins, 2017] Martins, R. (2017). “PONTiFF’ PersONaliTy Framework For Companion Characters.
M.Sc. thesis, Instituto Superior Tecnico, Universidade de Lisboa.
[McCrae and Costa, 1989a] McCrae, R. R. and Costa, P. T. (1989a). Reinterpreting the Myers-Briggs
Type Indicator From the Perspective of the Five-Factor Model of Personality. Journal of Personality,
57(1):17–40.
[McCrae and Costa, 1989b] McCrae, R. R. and Costa, P. T. (1989b). The structure of interpersonal
traits: Wiggins’s circumplex and the five-factor model. Journal of Personality and Social Psychology,
56(4):586–595.
[MOON and NASS, 1996] MOON, Y. and NASS, C. (1996). How “Real” Are Computer Personalities?:
Psychological Responses to Personality Types in Human-Computer Interaction. Communication Re-
search, 23(6):651–674.
[Munoz et al., 2013] Munoz, J., Gutierrez, G., and Sanchis, A. (2013). Towards Imitation of Human
Driving Style in Car Racing Games. In Believable Bots, pages 289–313. Springer, Berlin, Heidelberg.
55
[Nacke et al., 2013] Nacke, L., Bateman, C., and Mandryk, R. (2013). BrainHex: A neurobiological
gamer typology survey. Entertainment Computing, 5.
[Nass and Moon, 2000] Nass, C. and Moon, Y. (2000). Machines and Mildlessness: Social Responses to
Computers. Journal of Social Issues, 56(1):86–103.
[Nass et al., 1995] Nass, C., Moon, Y., Fogg, B. J., Reeves, B., and Dryer, D. (1995). Can computer
personalities be human personalities? International Journal of Human-Computer Studies, 43(2):223–
239.
[Pantic and Caridakis, 2011] Pantic, M. and Caridakis, G. (2011). Image and Video Processing for Af-
fective Applications. In Cowie, R., Pelachaud, C., and Petta, P., editors, Emotion-Oriented Systems:
The Humaine Handbook, pages 101–114. Springer Berlin Heidelberg, Berlin, Heidelberg.
[Paunonen and Ashton, 2001] Paunonen, S. V. and Ashton, M. C. (2001). Big Five factors and facets
and the prediction of behavior.
[Pittenger, 2005] Pittenger, D. (2005). Measuring the MBTI and coming up short. Consulting Psychology
Journal: Practice and Research, 57(3):210 – 221.
[Reeves and Nass, 1996] Reeves, B. and Nass, C. (1996). The Media Equation: How People Treat Com-
puters, Television, and New Media Like Real People and Places. Cambridge University Press, New
York, NY, USA.
[Riedl et al., 2011] Riedl, M., Thue, D., and Bulitko, V. (2011). Game AI as storytelling. Artificial
Intelligence for Computer Games, pages 125–150.
[Schermerhorn et al., 2008] Schermerhorn, P., Scheutz, M., and Crowell, C. R. (2008). Robot Social
Presence and Gender: Do Females View Robots Differently than Males? ACM/IEEE International
Conference on Human-Robot Interaction, page 263–270.
[Scott, 2002] Scott, B. (2002). The Illusion of Intelligence. Ai Game Programming Wisdom, 1:16–20.
[Shaker et al., 2010] Shaker, N., Yannakakis, G. N., and Togelius, J. (2010). Towards Automatic Person-
alized Content Generation for Platform Games. Proceedings of the 6th AAAI Conference on Artificial
Intelligence and Interactive Digital Entertainment, AIIDE 2010, (Hudlicka 2008):63–68.
[Takatalo et al., 2010] Takatalo, J., Hakkinen, J., Kaistinen, J., and Nyman, G. (2010). Presence, In-
volvement, and Flow in Digital Games. In Bernhaupt, R., editor, Evaluating User Experience in
Games: Concepts and Methods, pages 23–46. Springer London, London.
56
[Tognetti et al., 2010] Tognetti, S., Garbarino, M., Bonanno, A. T., and Matteucci, M. (2010). Enjoy-
ment Recognition From Physiological Data in a Car Racing Game Categories and Subject Descriptors.
Computational Intelligence and Games (CIG), pages 321– 328.
[van Lankveld et al., 2010] van Lankveld, G., Spronck, P., van den Herik, H. J., and Rauterberg, M.
(2010). Incongruity-Based Adaptive Game Balancing. In van den Herik, H. J. and Spronck, P.,
editors, Advances in Computer Games: 12th International Conference, ACG 2009, Pamplona Spain,
May 11-13, 2009. Revised Papers, pages 208–220. Springer Berlin Heidelberg, Berlin, Heidelberg.
[van Lankveld et al., 2011] van Lankveld, G., Spronck, P., van Den Herik, J., and Arntz, A. (2011).
Games as personality profiling tools. 2011 IEEE Conference on Computational Intelligence and Games,
CIG 2011, pages 197–202.
[Yannakakis et al., 2010] Yannakakis, G. N., Martınez, H. P., and Jhala, A. (2010). Towards affective
camera control in games. User Modeling and User-Adapted Interaction, 20(4):313–340.
[Yannakakis et al., 2013] Yannakakis, G. N., Spronck, P., Loiacono, D., and Andre, E. (2013). Player
Modeling. In M. Lucas, S., Mateas, M., Preuss, M., Spronck, P., and Julian, T., editors, Artificial and
Computational Intelligence in Games, volume 6, pages 45–59. Schloss Dagstuhl–Leibniz-Zentrum fuer
Informatik, Dagstuhl, Germany.
[Yannakakis and Togelius, 2011] Yannakakis, G. N. and Togelius, J. (2011). Experience-Driven Procedu-
ral Content Generation. IEEE Transactions on Affective Computing, 2(3):147–161.
[Yee, 2016] Yee, N. (2016). The Gamer Motivation Profile: What We Learned From 250,000 Gamers. In
Proceedings of the 2016 Annual Symposium on Computer-Human Interaction in Play, CHI PLAY ’16,
page 2, New York, NY, USA. ACM.
[Yee et al., 2011] Yee, N., Harris, H., Jabon, M., and Bailenson, J. N. (2011). The Expression of Person-
ality in Virtual Worlds. Social Psychological and Personality Science, 2(1):5–12.
57
58
Game Bibliography
Pac-Man (1980). Game developed by Namco, published by Namco and Midway Manufacturing
Ms. Pac-Man (1981). Game developed by Midway Games, published by Midway Manufacturing and
Namco
Defense of the Ancients (2003). Mod for Warcraft III, developed by Eul, Feak, S., and IceFrog.
DayZ: Battle Royale (2013). Mod for DayZ, developed by ”PlayerUnknown” Greene, B..
DayZ (2013). Mod for Arma 2, originally released in 2013 as a mod for Arma 2. Standalone currently
in early access, with full release planned for 2018, developed by Bohemia Interactive, published by
Bohemia Interactive.
Arma 2 (2009). Game developed by Bohemia Interactive, published by Bohemia Interactive.
Resident Evil (2005). Game developed by Capcom Production, published by Capcom.
Clash of Clans (2012). Game developed by Supercell, published by Supercell.
League of Legends (2009). Game developed by Riot Games, published by Riot Games.
The Elder Scroll Series. Game series developed by Bethesda Softworks and Bethesda Game Studios,
published by Bethesda Softworks.
Clash Royale (2016). Game developed by Supercell, published by Supercell.
Fortnite (2017). Game developed by Epic Games and People Can Fly, published by Epic Games.
Rogue (1980). Game developed by A.I. Design, published by Epyx.
Elite (1984). Game developed by Braben, D. and Bell, I., published by Acornsoft, Firebird and
Imagineer.
Dungeons & Dragons (1974). Board game designed by Gygax, G. and Arneson, D., published by TSR
and Wizards of the Coast.
Fallout 3 (2008). Game developed by Bethesda Game Studios, published by Bethesda Softworks.
Half-Life 2 (2004). Game developed by Valve Corporation, published by Valve Corporation.
Half-Life 2: Episode Two (2007). Game expansion for Half-Life 2, developed by Valve Corporation,
published by Valve Corporation.
Bioshock Infinite (2013). Game developed by Irrational Games, published by 2K Games.
Halo 3 (2006). Game developed by Bungie, published by Microsoft Game Studios.
Portal (2007). Game developed by Valve Corporation, published by Valve Corporation, Electronic
Arts, Microsoft Studios.
Braid (2008). Game developed by Jonathan Blow, Number None, Inc., Hothead Games.
59
AAppendix A - Questionnaires
60
A.1 Preliminary Study Questionnaire
A.2 Main Study Questionnaire - Assertive NPC
A.3 Main Study Questionnaire - Non-Assertive NPC
66
BGame Levels
Figure B.1: First level.
67
Figure B.2: Second level.
Figure B.3: Third level.
68
Figure B.4: Fourth level.
Figure B.5: Fifth level.
69
Figure B.6: Sixth level.
70
CDialogue Scripts
71
Figure C.1: Preliminary study’s dialogue instances.
72
Figure C.2: Main study’s dialogue instances, part 1.
73
Figure C.3: main study’s dialogue instances, part 2.
74