Stimulating Learning through Educational Games
![I. Explicit Methods
In this part, we present identified methods that have already been used to
model the learner explicitly, along with examples.
Evaluation of questionnaire
Learner modeling can be performed through evaluating the learner's answers
to a questionnaire at the beginning, during or at the end of the educational game. It
can also be done by redirecting learners to an online survey and asking them to
answer questions
For example, Moreno-Ger et al. evaluated the learner’s level of knowledge
as part of learner modelling through a questionnaire and some basic tests to capture
the learner’s initial level of knowledge. The learners were then provided with an
educational game, which also included an in-game exam. The results of that exam
were used to grade the learners. Finally, the final in-game exam was used to assess
the learners’ proficiency. In another example, after completing the educational
game, learners were asked to self-report the mental effort investment level and the
difficulty level associated with the learning task on a 9-point Likert scale [16, 543].
A scale named EGameFlow was used by Fu et al. to measure learners’
enjoyment of educational games. This Likert-type scale contained 42 items, with 1
and 7 representing respectively the lowest and highest degree to which respondents
agree with the items. Similarly, in Pourabdollahian et al., learners were requested
to answer a questionnaire in order to evaluate their level of engagement. The
questionnaire contained 21 questions based on 5-point Likert scale and categorized
according to five classifications (challenge, immersion, interest, purpose and
control) [11, 695]. The data collected from the survey was analyzed by both
descriptive and inferential statistics for evaluating the engagement of learners.
5](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_5.png?v=1)
![Various hardware and software equipment have also been used for modeling
the learners. The behavior of the learners can be recognized through identifying
gestures, body posture and physiological signals. It is done by identifying and
labeling different reactions to game events and then looking for common features
of body expressions.
Rebolledo-Mendez and de Freitas and Jimenez et al. used Electro
Encephalogram readings using Brain Computer Interface (BCI) for psychometric
input to measure attention. BCI offered the possibility of reading electric signals
generated by neural activity in the brain, which could be used to assess the
learners’ attention levels. In another example [17, 268], learner's level of attention
was evaluated using two components. The first component detected user’s gaze
behavior based on input from a web camera. This component used the direction of
eye or head to establish the screen-space coordinates of where the user was
looking. The second component measured attention levels using a
neurophysiological recording device.
Electromyography sensors have also been used to detect several types of
bodily expressions to recognize rapidly changing emotions. Likewise, Amershi et
al. collected data on game events and learners’ biometric expressions to form a
picture of the most common biometric expressions of emotion towards events
within the educational game. Learners’ biometric expressions were recorded using
4 sets of sensors. Each biometric recording was synchronized with logs of the
game events that could stimulate an emotional reaction.
6](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_6.png?v=1)
![II. Implicit and Unobtrusive Methods
Several methods have been used to generate learner models implicitly. In
this section, we will identify and discuss methods that have been used in literature.
Translating learner's actions
In this method, the set of actions made by the learner during the game can be
interpreted and translated into descriptive information useful to model the learner.
In addition, experts or automated systems define an appropriate description for
each set of actions. These descriptions are then used to model the learner. For
example, the learner model in Lessard[15, 97] aimed at creating a representation of
learner’s cognitive traits. To do that, the study established connections between the
behavior of the learner and his/her actions in the game. Specifically, the
manifestations of traits that the learner may exhibit during game-play were
identified. This could provide evidence of the learner’s cognitive abilities.
The learner model’s goals in Conati and Zhao andManske and Conati were
to generate assessment of learner’s knowledge on number factorization.
Knowledge assessment was done by representing the probabilistic relations
between learner’s actions and the corresponding learner’s knowledge.
Learner's motivational state was assessed in Stathacopoulou et al. [18, 203].
For that, the learning environment stored in a log file all the available information
about what a learner was doing, recording each learner action with a time stamp.
Typical examples of learner actions included selection of objects for
experimentation, selection of available tools, mouse moves, mouse drags or clicks
on tools or objects, mouse drags when the learner was trying to draw a vector, as
well as the time when the action was performed. After that, expert teachers defined
actions related to each type of evidence of learner’s motivational state.
7](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_7.png?v=1)
![IV. Interpreting interaction traces
In this method, the game engine monitors and records interactions between
the learner and the game, generating a trace of actions performed by the learner
during the play sessions and tracking a lot of different parameters in the game
(start game, end game, quit game, phase changes, significant variables, time,
performance, tasks completed, etc.). At the end of the game play, these traces can
be delivered to a group of experts or automated systems in order to identify what
information can be extracted from these traces and infer a lot of useful information
for modeling purposes.
For example, learner’s learning style was modeled in Stathacopoulou et al.
by monitoring the learner’s actions over time, where each response, such as a
keystroke, mouse move or drag was timed and recorded [8, 46]. The learning
environment stored in a log file all available information about what a learner was
doing, recording each learner action with a time stamp. Furthermore, this work
used teachers’ expertise in order to select the appropriate measures of learner’s
observable behavior to serve as indicators of the learner’s learning style.
Learner’s learning experiences were assessed in Moreno-Ger et al. by
monitoring and recording interaction between the learner and the educational
game, generating a trace of the actions performed by the learner during the play
sessions, tracking a lot of different parameters in the game and assigning points to
the most relevant actions. These traces were then delivered to the instructors in
order to inform them how the learner interacted with the educational game and
infer a lot of information from these traces for assessment purposes.
The educational game in Hou recorded every complete motion and event
that took place during the game in order to model learner’s behavior. In this work,
a total of 82 unique items were recorded, including various motions (e.g., pressing
a certain button) and events (e.g., completing a certain task) in the game (e.g.,
12](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_12.png?v=1)
![players talking with others, entering a new task, starting a new battle and changing
costumes for the role). These 82 items were divided into 10 behavior categories.
Each category had a name, label, and description [10, 304].
The learner model in Choi et al. generated assessment of learners’
knowledge by keeping track of the learner’s behaviors during the game. The
learner model had access to information such as learner moves and tools accessed
by the learner. The collected data was then interpreted to determine the state of
learner’s knowledge.
In another example Serrano-Laguna et al., high-level assessment report of
the learners was generated by selecting types of traces that must be logged to
facilitate assessment. The system then identified what information could be
extracted from the traces. Finally, the system analyzed learner’s traces and
produced reports for teachers.
Learners’ engagement was identified in Bouvier et al. and Bouvier et al. by
relying on their traces of interactions performed in the learning game. Studies of
Bouvier et al. and Bouvier et al. proposed an approach that consists of three stages
[2, 7]. The first stage aims to determine the high-level engaged-behaviors. The
second stage aims to characterize these engaged-behaviors by identifying the
underlying chains of actions. The last stage aims to detect these chains of actions
among all the actions recorded.
13](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_13.png?v=1)
![IV. Interpreting conversation
In this method, information about the learner is extracted from his/her
communication with a Non Player Character (NPC). The NPC poses questions and
records learner's choices during these conversations and questions answered during
game dialogues.
For example, learner’s learning style and learner’s preferences were
modelled in Moreno-Ger et al. by interpreting conversations between the learner
and NPC during the game play. In particular, the NPC queried the learner on
his/her preferences and asked other questions. Depending on the learner's choices
during these conversations, the game detected his/her learning style and also
detected his/her preferences.
Learner’s emotions were assessed in Muñoz et al. and Muñoz et al. by analyzing
the answers of the learner during the educational game dialogues.
A NPC in Rebolledo-Mendez et al. perpetually monitored the learner and
requested her/him to answer questions. The system collected and processed the
learner responses. After that, it calculated learner's level of motivation [14, 316].
Then it selected one rule from a set of rules taken from theory and reacted with aim
to sustain or enhance the current level of motivation.
14](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_14.png?v=1)
![V. Interpreting learner’s errors
The goal of this method is to retain the information of what the learner has
learnt, and what the learner has learnt incorrectly. In addition, this method records
learners’ misconceptions, errors, number and types of mistakes made by the
learner, learner failure and success, and time taken to complete the game.
For example, the learner model in Virvou et al. generated assessment of the
learner’s level of knowledge by keeping track of right and wrong answers that the
learner had given and the time he/she had spent to give a correct one.In another
example Fareed et al., an evaluation of recall improvement of learner memory was
generated by defining recall accuracy as the ratio of the number of questions
attempted correctly to the total number of questions.
Learner’s failure and success were pursued in Champagnat et al. In the case
of failure, the game had to be tolerant with the learner. This means that the game
allowed the learners to continue and did not stop them from playing or reiterating
the stage, but with penalties (e.g. giving them less points or worse results).The
learner model in Virvou et al. examined the correctness of the learners’ answers in
terms of the learners’ factual knowledge and reasoning they used. In addition, the
learner modeling component performed error diagnosis based on a cognitive theory
of Human Plausible Reasoning Collins and Michalski [19, 427]. In particular,
when a learner was asked a question from a domain, HPR was used to perform
error diagnosis in case of an error and to find out how close the erroneous answer
had been to the correct one.
The learner model in Cheng etal.assessed the learner's degree of mastery by
monitoring his/her performance and comparing it with information stored in a table
called Conditional Probability Table (CPT). The CPT stored the probability of
having a wrong answer even when concepts were seen to be mastered. Taking into
15](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_15.png?v=1)
![account the CPT, the learner model detected evidence of mastering the subject and
also detected evidence of deficiency at a skill or concept level.
An NPC in Rebolledo-Mendez et al. requested the learner to answer
questions. After that, the system processed the learner's responses as either correct
or incorrect and collected contextual information such as number of
correct/incorrect answers and time taken to answer such questions. Based on this
information, the system took decision about the appropriate help.
The learner model in Khenissi et al. detected learner's deficiency in programming
subjects. To do so, the model kept track of the learner’s wrong answers when
he/she arranged a set of unordered instructions in order to form a program [13, 2].
In particular, when the learner moved an instruction in an inappropriate place, the
learner model recorded the incorrect answer and provided the learner with the
appropriate help.
16](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_16.png?v=1)
![VI. Path follow
The goal of this method is to follow the path of the learner during the game
play, record the learner’s trajectory towards a goal and then interpret and infer
information about the learner. In fact, each path has a specific meaning and leads
to specific information about the learner.
For example, Moreno-Ger et al. described the game as state transition
systems in order to assess the learner’s activity inside the game. During the game
play, actions of the learner triggered state transitions and the sequences of actions
led to one or more end states. The game engine kept track of transitions, checked
the states that the game went through and generated reports describing them.A
NPC within the educational game in Rebolledo-Mendez et al. monitored the
learner and identified his/her learning trajectory. The identified learning trajectory
was compared to different predefined learning trajectories. This comparison
assisted the learner model in taking a decision on how to proceed.
In Noguez et al. and Kopeinik et al, a given domain was represented in terms
of its entities, properties, and relationships [7, 109]. The entities and their
relationships provided a general scheme. This scheme facilitated tracking of the
learner’s path during the game play and then inferred useful information about
her/him.
17](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_17.png?v=1)
![VII. Taxonomy of learner modeling methods
In this section, taxonomy of learner modeling methods which use educational
games will be described. This taxonomy was created based on the literature review
discussed in previous sections. Figure 1 presents the organization of the key terms
of our taxonomy from the most general to the most specific.
Learner modeling taxonomy.
From the literature, we observe that learner modeling using educational
games can be achieved explicitly or implicitly. The explicit way can be
accomplished using a questionnaire. Questionnaires can solicit direct and precise
answers, but this would endanger the high level motivation provided by
educational games because such use would require stopping the learner from
playing and requesting her/him to answer questions. Alternatively, the use of
hardware and software equipment provides an explicit way to model the learner
while using educational games [17, 268].This method can provide additional
information for modeling the learner. However, it obviously requires the use of
additional hardware and software equipment. Furthermore, observations collected
using these equipment can be interpreted in different ways and this can endanger
the reliability of the learner models.
Learner modelling can also be achieved implicitly. This is done by recording
actions of the learners while they are using educational games (Action Modelling).
The set of actions made by the learners during the game play can be interpreted
and translated into descriptive information useful for modeling the learners.
Conversation Modeling is an implicit way to extract information about the
learners. It essentially models the communication between the learner and the non-
player character. It records the learner's choices during these conversations and
questions answered during game dialogues. It also records conversations with
fellow players.
18](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_18.png?v=1)
![Moreover, Perturbation Modeling is used to extract information about the
learners in an unobtrusive way. It interprets information of what the learner has
learnt, and what the learner has learnt incorrectly. It also records learners’
misconceptions or errors, and the learner’s failure and success.
Finally, Strategy Modeling is also used to implicitly elicit information about
the learners [5, 16]. It concerns long-term learner behavior during the use of an
educational game as composed of a series of learner actions. Learner strategy can
be monitored by following learner paths inside the game.
19](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_19.png?v=1)
![VIII. Educational games used to model learner
Trade Ruler: Trade Ruler was designed to teach general public about the
importance of trade between countries. The player in this game plays the role of
the “trade ruler” of island. The ruler is responsible for managing the island’s
production on its labor-intensive (jeans) and capital-intensive (cell phones)
products. The ruler has to decide what to trade with its trading partner as some
islands are better for manufacturing labor-intensive products while others might be
advantageous in making capital-concentrated goods. So, the main goal of the ruler
is to maximize the islanders’ welfare.
Hands-on OS game: Hands-on OS game was designed to introduce learners
the common problems associated with the operating system of a computer. The
main goal of this game is to enhance the proficiency of the learner in certain skills
related to the computer’s operating system.
Set Based Concurrent Engineering game: The main goal of this game is to
help practitioners in designing a simplified airplane by following a simplified
approach. The game is divided into two stages. In the first stage, players are
requested to design an airplane for a given list of customer requirements without
following a specified process. In the second stage, players are provided with the
necessary instruments to execute a specified process. Once players complete the
design, a comparison of performances is presented to the players to compare the
two stages in terms of total development cost, time and quality [3, 435].
Learning version of Pacman game (LPG): The LPG aims to motivate
learners to correctly answer the questions of the programming languages. In the
traditional version of the Pacman game, players control the Pacman through a
maze, eating pac-dots. When all dots are eaten, the Pacman is taken to the next
stage. In addition, there are four enemies who roam the maze, trying to catch the
Pacman. If an enemy touches the Pacman, a life-chance is lost. Near the corners of
20](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_20.png?v=1)
![the maze are four dots known as power pellets that provide the Pacman with the
temporary ability to eat the enemies. When all life-chances have been lost, the
game ends. In the Learning version of the Pacman Game, when the Pacman eats a
power star, the learner has to respond to a question (about the programming
language) in order to continue the game having a 'reverse' role (the Pacman can
move freely and eat the enemy for a short period) [13, 3].
Prime Climb: Prime Climb is an educational game designed to help 6th and
7th grade students practice number factorization. Prime Climb consists of series of
mountains. Each mountain is divided into hexes labelled with numbers. Two
players must collaborate to climb these mountains. Each player can only move to a
number that does not share any common factor with the partner’s number. If a
wrong number is chosen, the climber falls and swings from the rope until the
player selects a correct number. During the game, each student has a pedagogical
agent which provides support. This educational game was cited in many works for
different purposes.
Learning version of Second Life: Second Life is an open-ended virtual world
which offers users opportunities to define virtual experiences with other users. In
the learning version, an artificial intelligence NPC poses questions to the learners.
In addition, this NPC uses a predefined set of reactions and has limited
conversation with the learners. The NPC asks questions in a multiple choice
question format .
Talking Island: Talking Island is an educational Massively Multiplayer
Online Role-Playing Game (MMORPG) designed to teach English vocabulary and
conversational skills to elementary school students. It includes elements that are
generally found in an MMORPG (e.g. team-work, battles, pets, and role-playing
situations). The learner can practice the pronunciation through the voice-
recognition module included in this game. IThis module diagnoses the correctness
of the pronunciation and determines whether the learner reaches a new level or
21](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_21.png?v=1)
![passes a quest. In addition, this game allows players to team up, communicate via
voice or texts, and solve quests jointly [18, 210]
Virtual Reality Game for English: It is an educational game that was
designed to teach English orthography and grammatical rules. Learners navigate
through the virtual world and answer questions concerning English spelling. In this
virtual world, there are three types of animated agents: the advisor, the guard (who
acts as a virtual enemy) of a passage and the learner’s companion. The advisor
agent appears in situations where the learner has to read new parts of the theory or
has to repeat parts that s/he appears not to know well. The virtual enemy agent is
responsible for asking questions to the learners. Finally, the virtual companion
appears to make some remarks in a casual way as if a friend was talking to the
learner.
Vectors in Physics and Mathematics: It is a discovery learning environment.
It was designed to help students in constructing the concepts of vectors in physics
and mathematics in secondary schools taking into account the conceptual
difficulties that the students faced. The thematic units of the learning environment
are: Position and Displacement; Motion; Forces and Equilibrium; Forces and
Motion; and, Forces and Momentum. Each of these units contains several
scenarios, which refer to real life situations.
Balance game: The balance game was developed with the purpose of
guiding students to understand the concept of moment. Moment is a quantity
representing the magnitude of a force influencing the rotation of an object. The
object represented in the game is a bar lying on a wheel that acts as the pivot for
the system. Two boxes at the sides of the bar represent two forces acting on the
beam. The learner has to balance and control the movement of the bar by changing
the force acting on one side of the beam. This is made by catching balls of different
weights.
22](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_22.png?v=1)
![ELEKTRA: It is a narrative-driven adventure game. In this game, the learner
has to solve several physics-oriented puzzles. A virtual character (representing the
ghost of Galileo) guides, advises and encourages the learner during the game
(Conlan et al. 2009 ). The ELEKTRA project is accessible online.
Learning Version of Memory Match Game (LMMG) : It is specifically
designed to foster second language acquisition. Memory Match Game is a card
game in which all of the cards are laid face down on a surface. The objective of the
game is to turn over pairs of matching cards with the least possible tries. In the
traditional version of the memory match game, all cards hold only visual
information. However, in the learning version of this game, other types of
information have been added. Specifically, graphics information is kept and
additional sounds, words and mathematical calculations have been added
alongside. The LMMG encourages learners to watch, read and listen to the
contents of each card and then try to match each pair [12, 33].
Conversation modeling
PlayPhysics: It is a role playing game in which the learner plays the role of an
astronaut on a mission. The main goal is to save the mentor (Captain Richard
Foster), who has been trapped on a space station, and reestablish control over the
space station Athena. To achieve the main goals, the learner has to overcome
challenges by applying the principles and the concepts of physics:
23](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_23.png?v=1)
![designed to facilitate the creation of educational games. Games delivered by the e-
Adventure engine are point and click adventure games. During the game, the
learner can learn by interacting with the objects, consultation of in-game books and
conversations with other characters. The < e-Adventure > engine includes
mechanisms that monitors the learner’s activities and then provides adaptation and
assessment.
VR-ENGAGE : It is a virtual world in which the learner navigates through
with the goal of finding the book of wisdom, which is hidden. Several obstacles
challenge the learner while navigating through the virtual world. A dragon guard
(who acts as a virtual enemy that closes doors) poses a question to the learner from
the domain of geography. Depending on the correctness of the learner’s answers,
the dragon allows her/him to continue his/her way through the door, which leads
her/him closer to the “book of wisdom” [9, 573].
Instruction Right Place Game : This game allows learners to benefit from the
drag and drop technology to construct a program (from any programming
language) in an amusing way. This educational game breaks down complex
programming tasks and guides learners through a series of small steps to form a
program interactively. This educational game is accessible online.
Strategy modeling. 80Days : 80Days is an adventure game used to teach geography
for a target audience of 12 to 14 years old and follows European curricula in
geography. The game story is about an alien scout called Feon who kidnaps a Boy
(play character) and travels with him around the world in a spaceship to collect
relevant geographical information. The player assists the alien in exploring the
planet and creating a report about the Earth and its geographical features. In the
course of the game, the player discloses the aliens’ real intention which is to
prepare for the conquest with the earth. The player has to save the planet and the
only way to do it is to draw the right conclusion from the traitorous Earth report.
25](https://docx.uz/documents/39c8a191-32ff-4000-aa97-7e30670b03eb/page_25.png?v=1)
Stimulating Learning through Educational Games