Alves-Oliveira, P., Janarthanam, S., Candeias, A., Deshmukh, A. , Ribeiro,
T., Hastie, H., Paiva, A. and Aylett, R. (2014) Towards Dialogue
Dimensions for a Robotic Tutor in Collaborative Learning Scenarios. In:
23rd IEEE International Symposium on Robot and Human Interactive
Communication (2014 RO-MAN), Edinburgh, UK, 25-29 Aug 2017, pp.
862-867. ISBN 9781479967650 (doi:10.1109/ROMAN.2014.6926361)
This is the author’s final accepted version.
There may be differences between this version and the published version.
You are advised to consult the publisher’s version if you wish to cite from
it.
http://eprints.gla.ac.uk/160838/
Deposited on: 17 April 2018
Enlighten – Research publications by members of the University of Glasgow
http://eprints.gla.ac.uk
Towards Dialogue Dimensions for a Robotic Tutor
in Collaborative Learning Scenarios
Patrı́cia Alves-Oliveiraa , Srinivasan Janarthanamb , Ana Candeiasa , Amol Deshmukhb , Tiago Ribeiroa ,
Helen Hastieb , Ana Paivaa , Ruth Aylettb
Abstract— There has been some studies in applying robots
to education and recent research on socially intelligent robots
show robots as partners that collaborate with people. On
the other hand, serious games and interaction technologies
have also proved to be important pedagogical tools, enhancing
collaboration and interest in the learning process. This paper
relates to the collaborative scenario in EMOTE EU FP7 project
and its main goal is to develop and present the dialogue dimensions for a robotic tutor in a collaborative learning scenario
grounded in human studies. Overall, seven dialogue dimensions
between the teacher and students interaction were identified
from data collected over 10 sessions of a collaborative serious
game. Preliminary results regarding the teachers perspective of
the students interaction suggest that student collaboration led
to learning during the game. Besides, students seem to have
learned a number of concepts as they played the game. We
also present the protocol that was followed for the purposes
of future data collection in human-human and human-robot
interaction in similar scenarios.
I. I NTRODUCTION
Recently, with the explosion of online learning, individualised and intelligent tutoring systems are gaining significant
attention from different stakeholders. There have been some
studies investigating the use of robotic tutors as a new interactive technology for learning [1], [2], [3]. In the EMOTE
EU FP7 project1 , we aim to build robots to be used in
individual and collaborative learning scenarios [4], [5]. The
project aims to design, develop and evaluate a new generation of virtual and robotic embodied tutors with perceptive
capabilities to engage in empathic interactions with learners
in a shared physical space. The work presented in this paper
relates to the collaborative scenario in EMOTE: a gamebased learning environment related to Geography curriculum running on a multi-touch table [4], [5]. This learning
environment is set apart from traditional Intelligent Tutoring
Systems (ITS) in that the robot embodiment provides the
system with the ability to demonstrate intention to interact
and can make use of gaze, mutual eye contact and ostensive
signals (e.g., posture) that have been shown to improve
learning [6]. Besides technologies to enhance learning, it is
important to pay attention to interaction between teachers and
students in classrooms. Such interactions are extremely rich
combining many aspects of human communication through
a INESC-ID & Instituto Superior Técnico, Universidade de Lisboa, Portugal. {patricia.alves.oliveira,tiago.ribeiro,ana.paiva}@inesc-id.pt,
ana.m.candeias@tecnico.ulisboa.pt
b School of Mathematical and Computer Sciences, Heriot-Watt University, Edinburgh. {sc445, a.deshmukh, h.hastie, r.s.aylett}@hw.ac.uk
1 www.emote-project.eu
both verbal and non-verbal cues. Such verbal interaction can
be inherently seen as social, comprising important educational dialogue content [7]. As such, in order to develop an
interactive robotic tutor, it becomes necessary to understand
the multiple dialogue dimensions that teachers and students
share during a learning process[8].
Towards this goal, we designed an experiment to collect
dialogue data from students and teachers playing a collaborative serious game on Sustainable Development, on a large
multi-touch table and a tablet. From this data, we identified
the different dialogue dimensions underlying the teacherstudent interactions. The experiment was conducted in two
schools - one in Edinburgh, UK and another in Lisbon,
Portugal. This paper presents: (1) the experimental protocol
that can be used in future research in similar educational
settings, (2) an analysis of the dialogue dimensions for a
robotic tutor in a collaborative learning scenario from sample
data, and (3) preliminary findings regarding the students
learning gains and teachers comments about collaborative
learning in their interactions with students.
II. R ELATED W ORK
There has been very few studies in applying robots to
education [1]. Recent research on socially intelligent robots
show robots as partners that collaborate with people [9] and
has made the use of robotic platforms in experimental learning more approachable [10]. Also, social supportive behavior
in robotic tutors has been shown to have a positive impact
on student’s learning performance [11]. On the other hand,
serious games - educational games that provide learning
content to players in addition to entertainment - have been
used successfully for learning purposes [12]. Such games
have shown to be very effective in helping students to learn
new concepts [13]. Likewise, recent interaction technologies
(e.g., multi-touch tables) have also proved to be important
pedagogical tools, enhancing collaboration and interest in the
learning process [14], [15].
III. C OLLABORATIVE L EARNING S CENARIO
Our collaborative learning scenario is based on a serious
game called Enercities2 . Enercities teaches sustainable
development through discovery learning. The game presents
opportunities for students to learn concepts such as pollution,
energy shortages, renewable energy, etc. Project Enercities
was co-funded by the European Commission programme
2 www.enercities.eu
Fig. 1.
Screenshot of Enercities game
Intelligent Energy Europe and this project reports that this
serious game made students aware of energy expenditure,
among others [16]. Schools across Europe have already used
Enercities in their curriculum and this game was therefore
chosen as the basis of our collaborative scenario. Within the
EMOTE project, the original online single player game was
adapted to a multiplayer version where participants interact
by using a multi-touch table or a tablet. This was done to
stimulate collaborative learning [17], [4].
Overview of the game The game will be played by three
players and the basic objective is to build a sustainable city
over time (See figure 1). Players can build and improve
structures such as housing, businesses, public amenities, etc
in order to grow the population of the city. The team’s
performance is measured in terms of points they score for
economy, environment and citizen’s happiness. These are
combined to form the overall team score. Players can choose
one of the three available roles: environmentalist, economist
and mayor. The three roles share resources such as oil, power
and money. Some of the structures such as housing and
power plants can be built by all of the players. But there
are some structures that are role specific. For instance, only
the environmentalist can build parks, forests and wildlife
reserves. The economist can build businesses and the mayor
can build public amenities. There are several levels in this
game. The players can advance to the next level when they
achieve the goal population set for each level.
The game finishes when the team completes the final
level of the game. The team loses the game when they run
out of non-renewable resources midway. During the game,
oil decreases over time (yet, its decrease can be slowed
down), money can be earned by building or improving
certain structures, and energy can be generated by building
power plants or improving existing structure. The challenge
is to manage the balance between several interlinked
factors: growing the population, keeping them happy,
keeping the environment green, generating enough power,
building businesses to make money, etc. The game provides
opportunities for students to learn several new concepts and
how they affect each other.
Game Moves During each turn, a player can perform one
of three possible actions: build a structure, upgrade existing
structures, implement a policy or skip the turn. To build
a structure, the player has to select one of the structures
available for his role. He/she can then try placing the
structure at different places in the city to see how it affects
the various indicators and then build it at someplace that
is deemed beneficial. Another option is to perform up to
three upgrades on existing structures to make them more
efficient (i.e., produce less CO2 emissions, consume less
power, etc.). The third option is to implement one of the
five available policies, which improve several aspects and
structures of the city in a single move. The player can also
skip his/her turn. The game allows the players to collaborate
and work as a team to build a healthy and sustainable city. It
also allows players to be competitive in order to maximize
their own individual scores. This presents them with several
opportunities to learn the causal link between the indicators
and the scores.
Robotic Tutor and Enercities Our long term objective is to
build a system where the game will be played by two students
and a robotic tutor (see Figure 2 and attached video) [17].
The objectives of the robotic tutor will be to play the role
of a team member, play the game, contribute positively to
the discussions, and tutor the students on various underlying
educational concepts as the game progresses. In order to
develop the dialogue for the robot tutor, human studies with
a teachers and students were conducted to serve as basis
for the future verbal behavior of the robot tutor. Our study
presents the first step towards this goal by developing the
dialogue dimensions of the collaborative interaction.
Fig. 2.
Robotic tutor playing Enercities with students
IV. O UR APPROACH
Fig. 3.
Experimental layout
A. Participants
The study was conducted in two European cities (Lisbon,
Portugal and Edinburgh, UK). A total of 20 students and 4
Geography teachers participated in our study. Each session
consisted of one teacher and two students. The students
were from early secondary schools, aged between 12 and 15
(M age = 12.9). Table 1 shows the distribution of sessions
for each teacher and the total time duration of the game
interaction.
Teacher
Teacher
Teacher
Teacher
Teacher
A (Edinburgh)
B (Edinburgh)
C (Edinburgh)
D (Lisbon)
No. of
sessions
2
2
1
5
Total interaction
time
40min
40min
20min
100min
TABLE I
T EACHERS AND SESSIONS
B. Experimental setup
The game was presented to the participants on an interactive multi-touch device. In Edinburgh, an 18 inch Windows 8
DELL tablet was used and in Lisbon a 55 inch Multitaction
table was used. In both cases, players could interact with the
game using touch gestures. Three video cameras, one for
each player, were used to capture their physical actions (i.e.
body and facial gestures). Voice recorders and collar microphones were used to capture their speech. The experimental
layout is shown in Figure 3. The teachers, students and their
parents signed a consent form to participate in our study.
Two students and a teacher were paired up as a team to play
the game for each session (see Figure 4).
regarding the interaction of students during the game.
The questionnaire was based on the key-principles of
collaborative learning [18], and contained 3 questions. An
example of a question is the following: “In this collaborative
game students have interacted with each other influencing
their learning process”. The questions rated the interaction
of the students in a 7 point-likert scale ranging from Never
to Always. There was also a section where teachers could
express their qualitatively opinion about the interaction.
Students’ Knowledge Questionnaire A two-part knowledge
questionnaire was designed with the help of a Geography
teacher. The questions were based on the concepts and causeeffect relationships presented in the game. The teacher was
consulted to identify the difficulty levels of the questions,
based on which the question paper was divided into two
parts. Part 1 consisted of 10 easy questions and Part 2
consisted of 10 hard questions. Each question was presented
as a statement to which the student has to answer “yes” if
he/she agrees with the statement and “no” if he/she doesn’t.
In case they are not sure what the answer is, they can
answer “I don’t know”. The main goal of this measure
was to identify learning outcomes related to sustainable
development addressed in the game. This questionnaire was
applied only in the study performed in Scotland and each
student answered this questionnaire before and after playing
the game.
D. Procedure
The present study comprised two phases: the initial phase
where we met the teachers, and the experimental task itself.
C. Questionnaires
We designed two questionnaires, one for students and
another for teachers.
Teacher Questionnaire Teachers’ questionnaire was
developed with the intention of getting their feedback
Phase 1: Meet the teachers
In Phase 1, we met the teachers in order to introduce the
game to them. In this phase, the following issues were
discussed with the teachers:
Game basics: Basics of the game were discussed such the
goals of the team, roles they play, structures that can be
built and upgraded, etc. This was done so that the teachers
got familiarized with the game and could understand the
underlying educational concepts.
Teacher’s role: The teacher’s role in the game was
discussed. Teachers were informed that, although they play
the game as yet another player, they should also be proactive
by presenting ideas, encouraging students to explore the
game, and learn the underlying concepts. They were asked
to balance the fun and learning aspects of the game as they
deemed appropriate. It is to note that teachers always played
the Mayor role in the game.
Demonstration: The game was demonstrated and the
teachers were encouraged to play in order to get familiar
with the interface, the game, and its features.
Phase 2: Experimental task
In the second phase, we conducted the actual experiment
with students and teachers playing together (see figure 4
and video). The experimental task consisted in the following
5-step process:
Step 1. The students were asked to sign a consent form.
Step 2. They were administered a pre-test (in Edinburgh
only).
Step 3. The students and the teachers were given a brief
introduction and a game tutorial for 10 minutes. The game
was demonstrated showing them how it worked and what
each person could do in their turn.
Step 4. For each team, roles were assigned: the teacher
always played the role of the mayor and the two students
were randomly assigned to the other two roles as economist
and environmentalist. Their goal was set to develop a
sustainable city and move up as many levels as possible.
The team played the game for 20 minutes. Their interactions
were video and audio recorded. Experimenters left the room
in order to give privacy to their interaction.
Step 5: The teacher was then given a questionnaire to fill
(in both Lisbon and Edinburgh). Meanwhile, the students
did the post test questions (in Edinburgh only).
V. D IALOGUE C ODING S CHEME
Understanding the dialogue utterances is important to
analyse and understand how the interaction between students and teachers happened in the collaborative learning
scenario. Dialogue utterances from the participants can be
pragmatically grouped together based on different aspects
of the interactions. These are called dialogue dimensions.
For instance, some utterances are about the game, whereas
some other are about the educational concepts underlying
the game. In order to develop virtual agents and robotic
tutors that are capable of performing such conversations with
students in collaborative learning scenarios, it is essential
to identify and understand the different dimensions in such
interactions[19].
Several dialogue coding schemes for analysing studentteacher dialogue interactions have been presented in the past
[20], [21], [22]. These studies considered either individual
learning (student-teacher interaction) or collaborative settings
(students dyads) [23], [24]. To our knowledge, there is no
study in which the dialogues actions and dimensions between
three or more participants in a collaborative learning setting
have been analysed.
In order to identify dimensions in the conversation, two
researchers watched some of the recorded sessions and
identified different dialogue content of the participants that
emerged during the game. The analysis performed by the
two researchers was discussed and combined to establish
dialogue dimensions that occurred during the game. Seven
dimensions were identified. Each of the seven identified dimensions is briefly explained below, followed by transcribed
examples taken from the recorded interaction.
1. Social: Social dialogue dimension includes utterances
such as welcome and goodbye greetings (e.g., “Hello Mr
environmentalist and Ms economist.”).
2. Game Status: This dimension includes dialogues
regarding the availability of resources in the game, the
population, or the scores. It is mainly a set of the utterances
that represents the indicators of the game (e.g., “What is
the environmentalist score?”).
3. Game Rules: Game rules dimension includes dialogue
utterances related to instructions regarding whose turn is,
the interface, and rules of the game (e.g., “The Smile
belongs to the mayor, right?”).
4. Curricular: This dialogue dimension represents the
utterances that links to the curriculum. It also includes the
cost-benefit analysis on proposals and the relation between
the different game indicators and real world concepts (e.g.,
“We shouldn’t build this because we are suppose to create
a sustainable city.”).
5. Strategy: Strategy dimension includes utterances related
to players’ expectations and goals. The strategy includes
players’ utterances about game moves, negotiations (e.g.,
of scores, regarding the city plan), exploring the game and
predicting consequences. The strategy can change along
the game according to each player game status (e.g., “It
is better if you compare other options before you decide,
otherwise my score will be affected.”).
6. Evaluation: Evaluation concerns dialogue related
to players’ auto and/or hetero evaluation of the game
performance. In this type of dialogue dimension players
make judgments about their own performance or about
other players’ performance, for example, regarding scores
evaluation (e.g., “Economist: “My economy sucks.”).
7. Task Unrelated: Task Unrelated dialogue dimension
comprises dialogue that occurs during the game but is not
directly influencing the learning process itself. Requesting
thinking time (stalling), irony, joke and technical dialogue
moves (unresponsive or faulty interface) are also included
in this dimension (e.g., Irony/Joke example: “Put it on the
Fig. 4.
Teacher playing Enercities with 2 students
sea. Just kidding!”).
These dimensions will have to be further analysed and
types of dialogue actions (semantic representation of utterances) need to be identified for each dimension in the near
future.
VI. P RELIMINARY A NALYSIS
Teacher Perspectives: The teachers reported that during
the game the students almost always interacted with each
other influencing their learning process. Also, the students
almost always maintained a balanced dialogue between
team members through the game in teachers’ perspective.
In addition, students always deliberated upon various game
actions, and these deliberations took into account academic
topics, game resources and their scores.
Student’s Learning gains: We analysed the pre- and
post-test knowledge questionnaires from the Edinburgh
dataset to examine how playing the game affected the
students’ learning. The learning gain was calculated from
the pre- and post-test scores based on the following formula
[25]:
Learning Gain (LG) = (Post test score - Pre test score) /
(Max Score - Pre test score) * 100
The above learning gain formula measures the percentage
of concepts that were learned out of the concepts that needed
to be learned after the pretest. For example, a student scored
5 out of 10 in the pre-test signifying that he has to learn 5
more concepts. He then scored 8 in the post-test meaning he
learned 3 out the 5 concepts he had wasn’t aware of at the
time of pre-test. The learning gain is calculated as (8-5)/(105) * 100 = 3 / 5 * 100 = 60%.
Table II shows the mean scores for the pre- and posttest questionnaire from the 10 students in Edinburgh dataset.
Most students found part 1 easy and part 2 a bit more
difficult, which aligns with the earlier assessment of the
teacher. Taking as an example the pre-test questionnaire, they
correctly answered about 8.6 out of 10 in Part 1, and only
answered 4.4 questions (out of 10) in Part 2 correctly. The
learning gains for the two parts were 42.85% and 28.57%,
respectively.
Metric
Mean pre-test score
Mean post-test score
Mean learning gain
Part 1
8.6
9.2
42.85%
Part 2
4.4
6.0
28.57%
TABLE II
L EARNING GAIN
We also analysed the frequency of “I don’t know” answers.
There seemed to be a drastic reduction between the pre- and
post-tests. The number of “I don’t know” answers decreased
by 25 to 40% for both parts of the knowledge questionnaire.
The increase in learning gain and the decrease in uncertainty
can be attributed to knowledge acquired during game-play.
Despite differences in terms of learning gains, results show
that these gains were not significative, p > .05. We attribute
this to the small sample size.
Metric
Mean pre “I don’t know”
Mean post “I don’t know”
Mean relative change
Part 1
1
0.6
-40%
Part 2
3.8
2.8
-24%
TABLE III
R EDUCTION IN UNCERTAINTY
It should also be noted that Part 2 consisted of difficult
questions and to learn all of them the team would have to
play several levels of the game. However, in the Edinburgh
sessions (where the pre- and post-tests were applied), the
students were only able to complete a maximum of 2 levels
(out of 4) during the 20-minute time limit. This could
have impacted their learning of more difficult concepts and
therefore produced less learning gain compared to part 1. We
will analyse this further in future.
VII. C ONCLUSIONS
The present study had three goals: (1) the main goal was
to develop the dialogue dimensions for a robotic tutor in a
collaborative learning scenario; (2) we also aimed to present
preliminary findings regarding the students’ learning gains,
and teachers’ perspective about collaborative learning regarding their interactions; and (3) to describe the experimental
protocol that can be used in future research in human-human
and human-robot interactive educational settings. Seven dialogue dimensions were identified from the teacher-student
interaction data. These dimensions present the preliminary
work done towards building a empathetic social robotic tutorial environment for collaborative learning scenarios within
the EMOTE project. These results serve to inspire future
collaborative learning scenarios using dialogue dimensions
for a robotic tutor inspired in human studies. The identifyed
dialogue dimensions can contribute on the students’ learning
performance complementing the impact of social supportive
behaviour[11], and using games for education[12]. Also, the
preliminary results regarding the teachers’ perspective of
the students’ interaction suggest that students collaborate in
their learning process during the game. Besides, preliminary
results regarding students’ learning outcomes suggest that
students learned the underlying educational concepts during
the game.
Future work: In the future, we will validate the dialogue
coding scheme with the data we have collected from these
studies and identify dialogue actions within each dimension
to represent student and teacher utterances. We also will
annotate important aspects of verbal and nonverbal behavior
of the interaction in this specific learning context. The
annotated dialogue data, along with the outcomes from the
teachers and students will be used to build simulations of the
teacher-student behaviour to learn and test dialogue strategies
for the robotic tutor.
ACKNOWLEDGMENT
This work was partially supported by the European Commission (EC) and was funded by the EU FP7 ICT-317923
project EMOTE and partially supported by national funds
through FCT - Fundação para a Ciência e a Tecnologia,
under the project PEst-OE/EEI/LA0021/2013. The authors
are solely responsible for the content of this publication. It
does not represent the opinion of the EC, and the EC is not
responsible for any use that might be made of data appearing
therein.
R EFERENCES
[1] T. Kanda, T. Hirano, D. Eaton, and H. Ishiguro, “Interactive robots as
social partners and peer tutors for children: A field trial,” HCI, vol.
19(1-2), pp. 61–84, 2004.
[2] E. Hyun, H. Lee, and H. Yeon, “Young children’s perception of irobiq,
the teacher assistive robot, with reference to speech register,” in ICCM,
2012 8th International Conference on, vol. 1. IEEE, 2012, pp. 366–
369.
[3] E. Hyun, H. Yoon, and S. Son, “Relationships between user experiences and children’s perceptions of the education robot,” in Proc. 5th
ACM/IEEE International Conference on HRI.
[4] A. Deshmukh, G. Castellano, A. Kappas, W. Barendregt, F. Nabais,
A. Paiva, T. Ribeiro, I. Leite, and R. Aylett, “Towards empathic artificial tutors,” in Proc. of the 8th ACM/IEEE international conference
HRI, 2013.
[5] S. Serholt, W. Barendregt, T. Ribeiro, G. Castellano, A. Paiva,
A. Kappas, R. Aylett, and F. Nabais, “Emote: Emboided-perceptive
tutors for empathy-based learning in game environment,” in European
Conference GBL, Porto, Portugal, 2013.
[6] G. Csibra and G. Gergely, “Natural pedagogy,” Trends in Cognitive
Sciences, vol. 13, no. 4, pp. 148 – 153, 2009.
[7] R. Tiberius, “The why of teacher/student relationships,” Essays on
teaching excellence, Professional and Organizational Development,
vol. 10(4), 1993.
[8] H. Bunt, “Dimensions in dialogue act annotation,” in Proc. International Conference LRE, 2006.
[9] C. Braezeal, “Role of expressive behaviour for robots that learn from
people,” Philosophical Transactions of the Royal Society B, vol. 364,
pp. 3527–3538, 2009.
[10] I. Leite, G. Castellano, A. Pereira, C. Martinho, and A. Paiva, “Modelling empathic behavior in a robotic game companion for children:
An ethnographic study in real-world setting,” in Proc. ACM/IEEE
International Conference on HRI, 2012.
[11] M. Saerbeck, T. Schut, C. Bartneck, and M. D. Janse, “Expressive
robots in education: Varying the degree of social supportive behavior
of a robotic tutor,” in CHI 2010, USA, 2010.
[12] L. Annetta, “Video Games in Education: Why They Should Be Used
and How They Are Being Used,” Theory Into Practice, vol. 47(3), pp.
229–230, 2008.
[13] A. Brisson, G. Pereira, R. Prada, A. Paiva, S. Louchart, N. Suttie,
T. Lim, R. Lopes, R. Bidarra, F. Bellotti, M. Kravcik, and M. Oliveira,
“Artificial intelligence and personalization opportunities for serious
games,” in Eighth AIIDE Conference, 2012, 2012, pp. 51–57.
[14] S. Higgins, E. Mercier, E. Burd, and A. Hatch, “Multi-touch tables and
the relationship with collaborative classroom pedagogies: A synthetic
review,” International Journal of Computer-Supported Collaborative
Learning, vol. 6(4), p. 515538, 2011.
[15] S. Higgins, E. Mercier, L. Burd, and A. Joyce-Gibbons, “Multitouch tables and collaborative learning,” British Journal of Educational
Technology, vol. 43(6), p. 10411054, 2012.
[16] Knol, E. and de Vries, P, “EnerCities, a serious game to stimulate
sustainability and energy conservation: Preliminary results,” eLearning
Papers, vol. 25, pp. 1887–1542, 2011.
[17] T. Ribeiro, A. Pereira, A. Deshmukh, R. Aylett, and A. Paiva, “I’m the
mayor: A robot tutor in enercities-2,” in 13th International Conference
on AAMAS ’14, In press.
[18] P. Dillenbourg, “What do you mean by collaborative learning?” in
Collaborative-learning: Cognitive and Computational Approaches.
Oxford: Elsevier, 1999, pp. 1–19.
[19] S. Janarthanam, H. Hastie, A. Deshmukh, and R. Aylett, “Towards a
serious game playing empathic robotic tutorial dialogue system,” in
HRI2014, 2014.
[20] N. K. Person and A. C. Graesser, “Fourteen facts about human
tutoring: Food for thoughts for ITS developers,” in Proc. AIED 2003
WK, 2003.
[21] A. C. Graesser, P. Chipman, B. C. Haynes, and A. Olney, “Autotutor:
An intelligent tutoring system with mixed-initiative dialogue,” IEEE
Transactions on Education, vol. 48(4), 2005.
[22] S. Alexander and A. Sarrafzdeh, “Foundation of an affective tutoring system: Learning how human tutors adapt to student emotion,”
International Journal of ISTA, vol. 4-3/4, pp. 355–367, 2008.
[23] C. A. Graesser, K. Wiemer-Hastings, P. Wiemer-Hastings, and
R. Kreuz, “AutoTutor: A simulation of a human tutor,” Journal of
CSR, vol. 1, pp. 35–51, 1995.
[24] C. van Boxtel, J. van der Linden, and G. Kanselaar, “Collaborative
learning tasks and the elaboration of conceptual knowledge,” Learning
and Instruction, vol. 10, pp. 311–330, 2000.
[25] A. C. Graesser, G. Jackson, E. Mathews, H. Mitchell, A. Olney,
M. Ventura, P. Chipman, D. Franceschetti, X. Hu, M. Louwerse,
N. Person, and the Tutoring Research Group, “Why/AutoTutor: A test
of learning gains from a physics tutor with natural language dialog,”
in Proc. 25th Annual Conference of the CogSci. Boston, MA, 2003.