Generating Non-linear Narrative for Serious Games
with Scenario Templates
Koen Samyn
1
, Ga
´
etan Deglorie
2
, Peter Lambert
2
, Rik Van de Walle
2
and Sofie Van Hoecke
2
1
DAE, University college of West Flanders, Ghent University Association, Botenkoperstraat 2, B-8500 Kortrijk, Belgium
2
Multimedia Lab, ELIS, Ghent University-iMinds, Gaston Crommenlaan 8 Bus 201, B-9050 Gent, Belgium
Keywords:
Serious Game, Domain-specific Modeling Languages.
Abstract:
Complex social interactions between NPC’s and players in a serious game remains a challenge. Standard
game engines are focused on a game world with a fixed set of rules and linear gameplay. The problem is
compounded by the unfamiliarity of many researchers with the basic structure and technicalities of a serious
game. In this paper we propose a method to use linear scenario elements as templates for the generation of
a non-linear narrative. We implement this method by creating an extra layer on top of the existing ATTAC-L
modeling language which is a tool for developing virtual interactive scenarios. Users are thus presented with
a language that offers limited flow control and simplifies the authoring process for the creation of scenario
elements. Our method uses these existing scenario elements together with metadata, and fills in the templated
elements with the contextual information of the current game state. By separating scenario concerns from
non-linear narrative concerns, we hope to make it easier to develop interesting non-linear serious games that
still conform to the requirements of evidence-based serious game research.
1 INTRODUCTION
Non-linear gameplay is a concept that is high on the
wishlist for a lot of game developers. However, an
example of real non-linear narrative is hard to find
in the wide range of commercial or serious games.
This should not come as a surprise because there are
many hurdles to take before linear narrative offers the
necessary consistency in terms of game logic and im-
mersion. Every bifurcation in the story results in a
doubling of the amount of scenarios that are needed
to define the game. Needless to say that a naive ap-
proach towards the implementation of non-linear nar-
rative leads to a huge amount of work, and ultimately
to compromises that reduce the combinatorial explo-
sion.
In this paper, we propose a solution for a non-
linear narrative that is built on top of ATTAC-L
(Broeckhoven and Troyer, 2013). ATTAC-L is a do-
main language for defining virtual scenarios that is
being developed as part of the Friendly Attac project.
The Friendly Attac project studies and develops an
innovative serious game to help youngsters deal with
cyberbullying issues. This is done by allowing young-
sters, through the use of the virtual scenarios, to ex-
perience different roles (bully, victim, or bystander)
Figure 1: Attac-L example.
in cyber bullying incidents during the game, to re-
act to those experiences, and to get adjusted feed-
back based on their individual reactions. This will
increase their empathy, enhance their social skills or
teach/train them relevant coping strategies. ATTAC-
L offers the user (domain experts) a set of graphi-
cal building blocks that can be assembled to create
a simple scenario. An example of a simple ATTAC-L
scenario is presented in figure 1. In this example, a
player sends a tweet to Kate with the message ’Hi’.
In the ATTAC-L language, an action or sentence can
be defined by using a simplified grammatical struc-
ture, in which a subject can declare the execution of a
verb. ATTAC-L also allows the definition of a choice
structure (where the user can pick one of the provided
options), and the definition of synchronous actions,
where all the actions are executed at the same time.
A recent implementation of ATTAC-L for serious
games (Janssens et al., 2014) translates these graphi-
cal building blocks into an XML declarative language
523
Samyn K., Deglorie G., Lambert P., Van de Walle R. and Van Hoecke S..
Generating Non-linear Narrative for Serious Games with Scenario Templates.
DOI: 10.5220/0005363705230530
In Proceedings of the 10th International Conference on Computer Graphics Theory and Applications (GRAPP-2015), pages 523-530
ISBN: 978-989-758-087-1
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
which is interpreted within a serious game engine.
This implementation provides support to test ATTAC-
L scenarios in isolation and also to integrate the sce-
narios in a serious game.
The structure and syntax of the ATTAC-L lan-
guage is simple as to facilitate the creation process of
scenarios for users who are not familiar with game de-
sign. However this simplicity also restricts the range
of possible scenarios and makes it very cumbersome
to introduce non-linear game play. Some notable lim-
itations of ATTAC-L are the lack of state manage-
ment and repetition. To overcome these limitations,
we introduce the concept of a story engine. The
story engine selects a scenario template from a sce-
nario database based on the current state of the player
model and provides the template with contextual in-
formation from the game. The story engine is also
responsible for maintaining the integrity of the sto-
ryline which is defined by additional metadata in the
templates.
Performance and change objectives (Desmet et al.,
submitted) are an additional consideration for the de-
sign of a serious game. In an evidence-based seri-
ous game founded on the intervention mapping pro-
tocol (Bartholomew et al., 2011), performance ob-
jectives are the behaviours we would like to obtain,
whereas change objectives are defined as the determi-
nants that influence these target behaviours. In our
proposal it is possible to link performance objectives
and change objectives to a scenario template. This
linking process makes it possible to track the progress
of the player throughout the duration of the game
and to present the player with the scenarios that are
needed to improve his/her proficiency in the subject
matter.
The remainder of this paper is as follows. In the
next section we discuss the related work. In section 3
we give an overview of the concepts of non-linear sto-
rytelling and game play adaptation. In section 4 we
delve deeper into the technical implementation of our
framework. In section 5 we discuss the adaptability
of the game play that allows us to present the player
a game environment that is tailored to his/her current
skill set. In section 6 we present a testing framework
that validates the selection of scenario templates for a
number of player types. Finally, we present the con-
clusions and future work in section 8.
2 RELATED WORK
In (Prensky, 2005) the author makes the case that dig-
ital game-based learning should maintain the balance
between learning and (fun) gameplay. Although we
present a method for creating interactive and non-
linear stories and not the actual creation of a game,
this balance must be kept in mind when determining
the feature set of the story engine. Another impor-
tant consideration is the payoff versus reward system
which indicates that players are prepared to put in the
work if the reward is sufficient.
In (Greitzer et al., 2007) a cognitive model is pre-
sented that suggests that learners/players benefit from
a layered and non-linear approach to learning. This
approach was applied for the serious game Cyber-
CIEGE by clustering scenarios in layers with greater
scenario complexity in each successive layer. We
present a similar approach but cluster the scenarios
in terms of performance or learning objective. The
CyberCIEGE game does not adopt an adaptive player
model, however we follow some of the recommenda-
tions of the authors for our approach (see section 5
with regards to player driven adaptation of the narra-
tive).
In project Muse (Llobera et al., 2013) a method
is presented that allows content creators to create an
interactive drama (ID) by adopting the concepts of
providence and the Zelig interaction metaphor, which
means that non-player characters can automatically
take on roles that are necessary for the advancement
of the storyline. Our presented aliasing approach (see
section 4.1) is similar but introduces additional con-
ditions on the role taking process.
3 OVERVIEW
An overview of the story engine is presented in fig-
ure 2.
When the game requires the start of a new sce-
nario, a scenario is selected from a database of sce-
nario templates. The selection process takes the cur-
rent state of the player model into account, including
the previous actions and the current location of the
player. The scenario might also require the proximity
or availability of non-player characters (NPC’s). Fur-
thermore the scenario might define the characteristics
(traits and facets, see section 4.1) of these NPC’s.
Once a scenario is selected, the aliases in the sce-
nario template are replaced with the appropriate and
currently available NPC’s and/or items. The scenario
template can require an NPC, with a predefined set
of personality characteristics, and/or a specific state
from the player model. An example scenario template
can define an alias for a character A with a person-
ality that contains the bully trait, and an alias for a
character B that is a victim of bullying. The aliases A
and B are then dynamically replaced with the selected
GRAPP2015-InternationalConferenceonComputerGraphicsTheoryandApplications
524
Figure 2: Story engine overview.
NPC’s by querying the context of the game.
Next the actions in the scenario are played or
executed, and possible options are presented to the
player. The decision of the player will trigger an eval-
uation and leads to an update of the player model.
The player model is used to adapt the probability that
a given scenario template will be selected, given the
performance objective to which it belongs.
4 SCENARIO TEMPLATES
In our implementation, a story is composed of a hi-
erarchical structure of scenario templates. A scenario
template defines via metadata the circumstances, un-
der which a scenario can be selected and executed,
and the requirements for the NPC’s that will take part
in the scenario. Finally, the scenario template in-
cludes the actions of the NPC’s and the possible op-
tions for the player. The following subsections ex-
plain this in further detail.
4.1 Aliasing Approach
A central part of the scenario template is the definition
of an alias. An alias acts as a placeholder for NPC’s
or items. When a scenario template is selected for
execution, the alias is replaced with a concrete NPC
in the game. An example of an alias can be found in
listing 1.
Listing 1: Alias example.
<alias>
<select type="NPC" id="CharacterA">
<trait type="victim"
value="80%" op="higher"/>
<facet type="student" value="true"/>
</select>
</alias>
In listing 1 a game object with type NPC is se-
lected. Within the selector it is possible to add mul-
tiple criteria for this NPC. It is possible to select for
a certain personality trait of the NPC by specifying
the type of the trait (in this case victim), a percent-
age value of the trait (in this case 80%) and an opera-
tor. In the example an NPC with a victim trait higher
then 80% is selected. Additionaly a facet of the NPC
can be selected with the facet element, in this case the
NPC must be a student.
In the remainder of the scenario template, this
NPC can be referenced by its alias CharacterA. This
is shown in listing 2.
Listing 2: Alias useage.
<gamemove subject="player" id="gm1">
<verb type="sendto">
<mediaObject object-class="sms">
<who id="CharacterA"/>
<property key="content"
value="Hi ${CharacterA.name}!"
/>
<ref id="sms1"/>
</mediaObject>
</verb>
</gamemove>
GeneratingNon-linearNarrativeforSeriousGameswithScenarioTemplates
525
In listing 2 a special subject (the player) sends a
message to the CharacterA alias. Within the mes-
sage, an escape sequence can be used to query to
properties of the alias. When the scenario is played
the escape sequence ${CharacterA.name} will be
replaced by the name of the selected NPC.
Finally it is possible to propagate an alias to other
scenarios by defining another query within the select
tag, as shown in listing 3
Listing 3: Propagation example.
<alias>
<select type="NPC" id="CharacterA">
<adopt id="scenario1"/>
</select>
</alias>
The adopt tag instructs the select tag to adopt
the same NPC as the one that was selected for
scenario1. A prerequisite for this tag to work is that
scenario1 must be executed before the current sce-
nario. This can be enforced by the use of requirement
meta-data as explained in the next section.
4.2 Storyline Control
The scenario template system enables the reuse of
story elements in the larger context of the game. The
story engine also provides several possibilities to con-
trol the order of the scenario elements, as defined in
the metadata of the scenario template.
A first important metadata property is the replaya-
bility of a scenario template. An example of a re-
playable type of scenario template is a help scenario
template. In this case the scenario template requires
an NPC with a victim personality type that has lost
an item. The player might opt to help the NPC or to
ignore the NPC. Listing 4 shows an example of a sce-
nario with a replay meta tag.
Listing 4: Help NPC.
<scenario id="help_npc">
<metadata>
<property key="replayable"
value="true"/>
</metadata>
<select type="NPC" id="CharacterA">
<trait type="victim"
value="80%" op="higher"/>
<facet type="student" value="true"/>
</select>
<eventpart type="options">
<option id="1" score="2">
Help ${CharacterA.name}
</option>
<option id="2" score="2">
Ignore ${CharacterA.name}
</option>
</eventpart>
</scenario>
Another possibility is to require the (positive) ex-
ecution of a scenario template A before scenario tem-
plate B can be executed. In the help quest exam-
ple (named help npc) helping the victim is required
to open up the execution of another scenario tem-
plate named explore quest. An additional feature
is that the explore quest scenario can require the
same NPC as the help npc scenario.
Listing 5: Explore quest.
<scenario id="explore_quest">
<metadata>
<property key="replayable"
value="false"/>
<require>
<scenarioref id="help_npc"/>
</require>
</metadata>
<select type="NPC" id="CharacterA">
<adopt id="help_npc"/>
</select>
<!-- npc actions ... -->
</scenario>
In listing 5 the scenarioref tag defines the sce-
nario that is required before this scenario can be
started. Of course it is possible to define multiple sce-
narios as required.
The explore quest scenario in the same list-
ing can be executed unconditionally. However, if
the player did not choose the help of the NPC
in the help npc scenario, the execution of the
explore quest is inconsistent. To remedy this in-
consistency a new metadata tag can be used, as shown
in listing 6, where the explore quest has been updated
with an additional condition.
Listing 6: Explore quest with condition.
<scenario id="explore_quest">
<metadata>
<property key="replayable"
value="false"/>
<require>
<scenarioref id="help_npc"/>
</require>
<conditions>
<faction
idA="CharacterA"
idB="player"
value="friend"/>
</conditions>
</metadata>
<select type="NPC" id="CharacterA">
<adopt id="help_npc"/>
</select>
<!-- npc actions ... -->
</scenario>
GRAPP2015-InternationalConferenceonComputerGraphicsTheoryandApplications
526
With a faction tag it is possible to specify that the
two characters, as defined by the attributes idA and
idB must be friends with each other. The player value
is a special value that references the player and the
CharacterA value is defined by the select tag in the
same scenario. In this case the type of the condition
was set to faction but it is also possible to query spe-
cific attributes of an NPC, or to check the inventory of
a character or player. Some of these possibilities are
shown in listing 7.
Listing 7: Conditions.
<conditions>
<inventory
id="player"
item="key#731"
value="true"/>
<attribute
id="CharacterA"
attribute="health"
value="50%"
op="higher"/>
</conditions>
The trait and facet selectors, defined in sec-
tion 4.1, are valid within the condition tags. The
faction, inventory and attribute tags are also valid
within a select tag.
5 PLAYER-DRIVEN
ADAPTATION
Serious games are used to teach certain behavior or
knowledge to users. To provide an optimal learn-
ing experience, the teacher should adapt to the re-
spective student, e.g. by taking more time to thor-
oughly explain the concepts that the student has trou-
ble with. However, the average video-game does not
adapt gameplay to each individual player. Typically,
one constructs some kind of player-model based on
real-time player input or post-game feedback. The re-
alized model can then be used to tweak certain pa-
rameters of the game (e.g. generating different level
geometry to increase challenge for skilled players). In
an educational context, learning state can be checked
through performance or learning objectives. We use
these performance objectives to track the player’s
learning state in real-time and adapt the occurrence of
learning challenges in order to better suit the problem
areas of player.
In the next subsections we expand on performance
objectives, extracting the player model and probabil-
ity distribution of template selection respectively.
5.1 Performance Objectives
A serious game based on the Intervention Mapping
Protocol needs to define change and performance ob-
jective, and link them to the correct scenario template.
Performance objectives define the desired behavior
for the player in the game world but mostly in the
real world. A game with cyberbullying as topic can
for example define positive bystander behaviour for a
player which means that the player has to disapprove
of bullying behaviour in a visible and public manner.
Change objectives define the underlying determinant
for the behavior. It may be that, in order to disapprove
of bullying in a public way, the player needs to expect
that this will end the bullying without putting himself
at risk. Another change needed in the determinants
could be that the player has to believe the victim is
not to blame for being bullied.
In our implementation a scenario template can ad-
dress one performance objective to induce the desired
behavior together with one (typically) or more change
objectives. This information is stored into the meta-
data of the scenario template. The scenario templates
can then be clustered by performance objective which
enables the evaluation of the player and the player-
driven adaptation of the storyline.
5.2 Player Evaluation
To predict the player model, we assume that a player
has a certain proficiency with each performance ob-
jective. We map this in range of 0 to 1, where 0 states
that the player does the opposite of what the perfor-
mance objective requires, 0.5 that the player acts ran-
domly (i.e. neither in favour nor in opposition of the
performance objective) and 1 that the player fully acts
in accordance with the performance objective.
Our concept is to predict the player model during
gameplay based on the actions performed by the
player. Currently the actions that can be performed
are of a binary nature, acting either for or against the
performance objective. We start by assuming that the
player has a 50% proficiency in all performance ob-
jectives (i.e. starting out without a bias). For each
event, the action is checked to be for or against its
performance objective. If it’s for, we simply increase
the proficiency for that objective by a certain value. If
it’s against, we decrease the proficiency.
This increase/decrease value, named adaptation rate
from here on, is changed according to the current
proficiency level (see Figure 3). For the mapping of
proficiency to adaptation rate, we chose a bell curve
(modelled as 2 sigmoid-functions) because of follow-
ing reasons:
GeneratingNon-linearNarrativeforSeriousGameswithScenarioTemplates
527
Figure 3: Adaptation rate as a function of proficiency.
A high adaptation rate near the center makes the
proficiency go in either direction fast and thus al-
lows the system to quickly revert to 50% when
evaluations alternate (i.e. avoids creating an un-
necessary bias).
A progressively lower adaptation rate near the
ends of the proficiency spectrum creates a barrier
so that truly mastering a performance objective
does not come too easy.
Making the curve flatter near the ends of the profi-
ciency spectrum creates a barrier for return. When
having achieved the far end of the spectrum for
an objective, we don’t want the proficiency to re-
turn in the opposite direction for the slightest op-
posite action. For example having a mastered
an objective, the player’s proficiency should not
drop significantly for a single mistake once and
a while. Similarly having failed an objective, the
proficiency should not go up significantly if the
player happens to make a correct ‘guess’ once in
a while. Only if the player consistently makes
correct choices, the proficiency should change ac-
cordingly.
The peak of the curve is controlled by the learning
rate parameter α. An evaluation using our prediction
model and the current error-rate will be displayed in
the results section.
5.3 Probability Distribution of Template
Selection
The scenario templates are grouped according to per-
formance objective. During gameplay, whenever a
new scenario has to be loaded, a new template is se-
lected at random. This is achieved by first selecting
a performance objective or pool (a set of templates
with the same performance objective) at random and
Figure 4: Proficiency to weight mapping.
from this pool in turn randomly selecting a template.
To adapt the narrative to the player, we change the
probability distribution of the performance objective
pools to better match the problem areas of the player.
As discussed above the predicted model attempts to
estimate the current proficiency of the player. Each
proficiency matches a certain performance objective
and is expressed as a number between 0 and 1. From
these proficiencies we will automatically deduce the
appropriate probability distribution.
In order to inhibit the possibility of creating in-
appropriately large or inappropriately small weight
values when using a simple formula for the weights
(with proficiency as input), we introduce thresholds
to create weight-plateaus (see Figure 4) and avoid
performance objectives with a probability of nearly
0% or nearly 100%. If a proficiency ever enters one
of these plateaus, the performance objective will be
flagged as such (either ’fail’ or ’pass’ can be used to
automatically detect noteworthy events for a supervi-
sor/teacher, especially in the case of ’fail’. Between
the thresholds, the weight value are linearly interpo-
lated, creating a probability distribution that is never
in favour or disfavour of a single performance objec-
tive. Selecting the scenarios using this methodology
will hopefully lead to a more balanced game/learning
experience.
6 RESULTS
To validate our concept we ran several simulations
of our system. The system uses 5 performance ob-
jectives, marked as PO1, PO2, PO3, PO4 and PO5
respectively. Each performance objective has a pool
of 30 scenario templates, making a total of 150 tem-
plates. A single run of the system triggers 100 scenar-
ios, predicting the player model after each event and
adjusting the probabilities accordingly. All result data
is the average of 1000 runs of the system.
A digital player model was created to simulate all
GRAPP2015-InternationalConferenceonComputerGraphicsTheoryandApplications
528
Figure 5: Simulated player proficiency/progression.
Figure 6: Predicted player profiency over time.
interactions. The player is modelled to have profi-
ciency in all 5 performance objectives as has been de-
scribed in subsection 5.2. Table 1 displays the pro-
ficiency of the player for each performance objective
and the matching base learning rate or α-value. In our
current model we assume that each time the player
interacts with a scenario he/she learns from it, repre-
sented as a small increase in proficiency (i.e. learning
rate).
Table 1: Player statistics.
Player PO1 PO2 PO3 PO4 PO5
Starting Profiency 0.15 0.5 0.8 0.9 0.95
Learning Rate 0.01 0.01 0.01 0.01 0.01
The proficiency of the simulated player is tracked
over time, as can be seen in Figure 5. Not surprising,
proficiency in all objectives rises. From the interac-
tion with the scenarios the story engine tries to predict
the player’s proficiencies. For the adaptation rate we
chose an α-value of 0.15. If the α-value is too low,
the prediction will reach the actual proficiency very
slowly; if the α-value is too high, the prediction will
become unstable.
The predicted player model over time is displayed
in Figure 6. The prediction takes about 20 events to
get a good approximate measure of the player’s pro-
ficiencies. The predicted proficiencies are all lower
than the actual proficiencies off the simulated player.
Figure 7: Prediction error over time.
Figure 8: Adaptation of performance objective picking
probability.
Figure 7 displays the absolute difference between the
predicted and actual proficiencies, i.e. error rates. The
average error rate drops to about 15%, this is a rea-
sonable value as a difference of under 20% would not
yield a significant change in the resulting weight of
a performance objective, i.e. no major differences in
the resulting probability distribution.
Finally, the resulting weights of each performance
objective change the overall probability distribution
of template selection (see Figure 8). This new dis-
tribution shows that the story engine creates a per-
sonal game/learning experience without going into
extremes.
7 DISCUSSION
We reduced the technical complexity of writing sce-
narios for serious games by separating the modeling
scenario template language from the serious game im-
plementation. By decoupling the modeling and im-
plementation, a writer can develop a scenario with
only an abstract knowledge of the game world, and
does not need to concern him- or herself with the tech-
nical details of the game engine. A disadvantage of
course is that every character, location and item de-
scribed in the scenario template needs to be provided
in the game world.
The player model introduced in section 5 offers
GeneratingNon-linearNarrativeforSeriousGameswithScenarioTemplates
529
the possibility to test scenario templates in terms of
change and performance objectives. The current ex-
perimental learning rates will, however, need to be re-
fined and validated during user testing.
A potential problem with the current implemen-
tation of the player-driven adaptation lies with antag-
onistic players. If a player selects the wrong option
at every turn of the scenario the proficiency level for
all performance objectives will drop and the scenario
will not advance. It is out of scope of our implemen-
tation to impose a definitive solution to this problem.
Within the current framework the player will have to
work hard to recover from this situation, which might
extend the duration of the game over practical limits.
In most games antagonistic play is avoided by pro-
viding the players with rewards if they make sufficient
progress. This mechanism helps to retain players and
maintain player interest, however for a serious game
an even higher retention rate is required. Most serious
games are played under supervision so there is an op-
portunity to intervene in the case of antagonistic game
play. At any rate we provide a signal that a player is
deliberately failing the game and the decision on how
to act upon this information lies with the (pedagogi-
cal) domain experts.
8 CONCLUSIONS AND FUTURE
WORK
We presented the concept of a story engine to in-
troduce non-linear narrative using scenario templates
and aliasing. The aliasing approach facilitates the de-
sign process by using simpler building blocks (sce-
nario templates) as a way to create a dynamic experi-
ence with an individualized storyline inside the game.
Specific to the topic of serious games and learn-
ing is the creation of a player-driven narrative that
tests the player on his/her skills. This was achieved
through the mapping of performance objectives to
each scenario template using metadata and by creat-
ing a model of a player that is adapted towards.
As for future work, we foresee improvements in
several areas. The learning rate for the simulation is
set to a low and possibly unrealistic value. By com-
paring actual input from a player to our player model
we should be able to derive a better player and game
specific value. Further study is required to create an
algorithm that can deduce an approximate learning
value based on the user input.
Player evaluation is performed in a binary fashion,
to allow for a more diverse way of action we would
include a choice based system. This system would
give the player a set of choices (actions) to perform
based on an occurred event.
Our implementation is used to randomize the oc-
currence of unrelated events. However, to provide a
controlled narrative while still using such a system,
linked templates with smart selection are required. In
such a system the pool of available event templates
would be populated only by those events that can fol-
low the previous one, i.e. an event that should logi-
cally follow from a choice made by the player or a
completely unrelated event.
Our work focuses on the aliasing aspects of char-
acters. However, a story may also require geographic
bifurcations. E.g. the main character could choose a
different path to reach the destination. In this case a
scenario would not only contain aliases for characters
but also for geographic locations (for example hidden
caves, treasures, bridges, . . . ).
ACKNOWLEDGEMENTS
This work was funded by the IWT SBO Friendly AT-
TAC project (http://www.friendlyattac.be/).
REFERENCES
Bartholomew, L. K., Parcel, G. S., Kok, G., Gottlieb, N. H.,
and Fernandez, M. E. (2011). Planning health pro-
motion programs: an intervention mapping approach.
John Wiley & Sons.
Broeckhoven, F. V. and Troyer, O. D. (2013). Attac-l: A
modeling language for educational virtual scenarios in
the context of preventing cyber bullying. In SeGAH,
pages 8. IEEE.
Desmet, A., Van Cleemput, K., Bastiaensens, S., Poels, K.,
Vandebosch, H., Verlogine, M., Vanwolleghem, G.,
Mertens, L., and De Bourdeaudhuij, I. The stepwise
development of a serious game to mobilize bystanders
against cyberbullying among adolescents the friendly
attac project. submitted.
Greitzer, F. L., Kuchar, O. A., and Huston, K. (2007). Cog-
nitive science implications for enhancing training ef-
fectiveness in a serious gaming context. J. Educ. Re-
sour. Comput., 7(3).
Janssens, O., Samyn, K., Van Hoecke, S., and Van de Walle,
R. (2014). Educational virtual game scenario genera-
tion for serious games. In SeGAH, pages 8. IEEE.
Llobera, J., Blom, K. J., and Slater, M. (2013).
Telling stories within immersive virtual environments.
Leonardo, 46(5):471–476.
Prensky, M. (2005). Computer games and learning: Digital
game-based learning. Handbook of computer game
studies, 18:97–122.
GRAPP2015-InternationalConferenceonComputerGraphicsTheoryandApplications
530