Robot-Mediated Collaborative Learning

embarrassedlopsidedΤεχνίτη Νοημοσύνη και Ρομποτική

14 Νοε 2013 (πριν από 3 χρόνια και 7 μήνες)

56 εμφανίσεις

Robot
-
Mediated Collaborative Learning


Jun Oshima and Ritsuko Oshima

Shizuoka University

Japan

joshima@inf.shizuoka.ac.jp
,
roshima@inf.shizuoka.ac.jp



Abst
ract:

For examining a possibility of communication robot as agent to improve student
collaborative learning, we implemented agent robots in university students’ collaborative reading
comprehension activity
.

In
jigsaw

group of the collaborative reading comp
rehension, four students
and one robot as a group explained their responsible articles to one another.

We had two questions.
First, do human learners accept the agent robot as their partner? Second, how does the agent robot
influence students’ learning pro
cesses and outcome?

Based on our analysis of learning processes
and outcome, we found that students in this study welcomed the agent robot as their learning
partner and it functioned to facilitate learning as almost equally as did human partners.



Backgr
ound and Research Purpose


Computing technologies are used for facilitating co
-
present collaborative learning. Tangible user interface,
ubiquitous computing and augmented reality are becoming core technologies for supporting co
-
present collaboration
among
students (Dillenbourg, Huang, &
Cherubini

(Eds.), 2009). With these technologies, objects shared in
collaboration become more tangible so that students naturally manipulate them in their collaborative acts. Joint
attention can be easily attained by student
s’ ordinary behaviors of communicating with others. The digital and
natural support for “being together” would have advantage for learners to deeply engage in knowledge creation
practices.

In this study, we take another approach to facilitating co
-
present

collaboration in the classroom by
implementing robots. Although robots are not yet familiar in our ordinary life, advantage of robots (i.e., humanoids)
to assist humans has been discussed for years in the area of human
-
robot interaction (Feil
-
Seifer & Mat
aric, 2005;
Tapus, Mataric, & Scassellati, 2007). Socially assistive robotics (SAR) studies so far have reported successful
implementations for assisting humans through social interaction under a variety of contexts (e.g., care of elderly,
care of individu
als with physical recovery, rehabilitation and training needs, care of individuals with cognitive
disabilities, and so on). Findings by SAR studies suggest that robots would be naturally present in our ordinary life
in the near future, and we consider it p
romising that robots may be used as a technology to support student co
-
present collaboration. The purpose of this study is to explore the effectiveness of robot as agent facilitating student
collaborative learning and propose a direction of research on rob
ot
-
mediated collaborative learning.


Socially Shared Regulation of Learning



Since the early stage of the development of cognitive models of collaboration (
Roshcelle & Teasley, 1995
;
Shirouzu, Miyake, & Masukawa, 2002), the perspective of collaboration ha
s been expanded from how ideas by
different persons are integrated or converged into a new idea toward how a group of persons collaboratively regulate
their group cognition to produce new ideas (
Stahl, 2006;
Hadwin, Järvelä, & Miller, 2011
). In such a new
movement
of theoretical consideration, we refer to the idea of socially shared regulation of learning as a theoretical framework
for designing scaffolds by robots in collaborative learning.


Socially shared regulation of learning (SSRL)
is a regulatory pro
cess model of collaborative learning based
on preceding ideas of self
-
regulated learning (SRL) and co
-
regulated learning (e.g., Hadwin & Oshige, 2011;
Hadwin et al., 2011;
Volet, Vauras, & Salonen,
2009
). In Hadwin et al. (2011), SSRL is defined as “interd
ependent
or collectively shared regulatory processes, beliefs, and knowledge orchestrated in the service of a co
-
constructed or
shared outcome/product” (p. 69). In SSRL, learners are collaboratively involved in planning, monitoring, evaluating
and regulati
ng of socioemotional, cognitive and behavioral aspects of their learning. T
he importance of social
regulatory processes for learning in small group settings

has been preliminarily but empirically supported by several
studies in the last ten years
. For

inst
ance,
it is found that
individual, other, and shared forms of regulation during
group

interactions

can be differentiated (Vauras, Iiskala, Kajamies, Kinnunen, & Lehtinen,
2003
) and these different
types of regulation are used for maintaining group work whe
n confronted with challenge for students (Järvenoja &
Järvelä
, 2009)
. Moreover, studies like Salovaara and Jä
r
velä (2003) and
Vauras et al. (
2003
) manifested that
social
regulation
of learning
is associated with the use of deep
-
level learning strategies an
d transfer
. Rogat and
Linnenbrink
-
Garcia (2011) demonstrated case studies suggesting that (1)
the synergy among the social regulatory

processes of planning, monitoring, and behavioral engagement was
associated with
quality

of group work, and (2)
p
ositive s
ocioemotional interactions an
d collaboration were factors to facilitate
hi
gher quality social regulation.


Although SSRL is still in its infancy, the socially regulatory framework is promising for us to consider
agent
-
based instructions for facilitating co
llaborative learning. One reason is that regulatory processes are focused
on metacognitive aspect of learning rather than learning of content knowledge (Hadwin et al., 2011). Metacognition
is the central target in the knowledge creation practices in collab
orative learning (Paavola et al., 2004). Development
of agents to support students’ SSRL in collaborative learning might provide with further empirical evidence related
to its importance and interplay among different SSRL components for the knowledge creat
ion practice. Another
reason is that the social regulatory framework is reasonable extension from SRL research and there are
many

empirical studies of agent
-
based instructions to facilitate SRL in individual learning (
Azevedo (Ed.),

2007). We can
therefore

develop scaffolds provided by robots on the basis of empirical evidence from studies of facilitating SRL by
agents.


Robot as New Agent for SSRL


Studies of scaffolding for facilitating SRL activities leading to higher conceptual understanding have been
c
onducted for years to make educational implication for designing intelligent tutoring systems with metacognitive
agents in CBLEs (e.g., Azevedo (Ed.), 2007;
Azevedo
,
Cromley, Moos
,
Greene
,

& Winters
, 2011;
Azevedo
,
Crom
ley, & Seibert, 2004
). Scaffoldings f
or SRL activities are either fixed or adaptive. Fixed scaffoldings such as
short lecture of self
-
regulation of learning, a set of sub goals to regulate and monitor student learning and so on are
prepared and provided students in a pre
-
determined way. On th
e other hand, adaptive scaffoldings (content
-
oriented
or process
-
oriented) are provided by a human tutor being present with a learner during learning sessions. A human
tutor evaluates learner’s understanding, suggests her misunderstanding and provides stra
tegies to modify her
misunderstanding if necessary in the content
-
oriented adaptive scaffolding context. In the process
-
oriented adaptive
scaffolding context, a human tutor monitors student’s regulatory process of learning, asks her to assess if her
learni
ng is going well or not, encourages her to use SRL strategies and suggests appropriate ones if necessary.
Comparative studies of fixed and adaptive scaffoldings have concluded that adaptive approach is better in
facilitating student’s SRL activities and co
nceptual understanding (e.g., Azevedo et al., 2004). Studies currently
examine how adaptive scaffoldings should be implemented in CBLEs by developing agent modules (Azevedo &
Strain, 2011).


In

our research project
, based on the preceding research on agent
s in CBLEs, we attempt to create socially
assistive robot as agents to provide learners with adaptive scaffoldings of SSRL during their co
-
present collaborative
learning. We have several reasons to select robots as agents for facilitating SSRL. First, we t
hink that an advantage
of robot agent over human is its participative stance in collaborative learning. Learners at any age usually recognize
instructors or teaching assistants as authoritative persons who know everything in their class. On the contrary, r
obots
may be accepted as assistants or partners by learners because they are ordinarily not seen as intelligent as learners

are.
Learners expect that robots would provide some information that they could take a control.
Therefore
, learners
may keep their i
ntentionality in regulating their collaboration. Second, there may be an advantage of robot over
intelligent agent on PC in co
-
present collaborative learning. Robot’s physical embodiment denotes participative
status as well as physical reality (
Breazeal,

2
002;
Goetz

&

Kiesler,

2002
). With its physical embodiment, robots can
express their engagement in learners’ collaborative learning through verbal and nonverbal channels. Third, by
having multiple robots equipped with the same adaptive scaffolding scripts,
researchers can create a new research
context where they can experimentally control scaffolding across different groups of learners and different years
(Miyake & Okita, 2012). This research infrastructure makes our design
-
based research across multiple yea
rs more
stable and reliable so that we do not have to consider several unavoidable
human
factors that might influence
learners’ performance (e.g., difference in teaching assistants across groups and years).


Research Questions



As a preliminary
step towar
d designing
a learning environment

where agent robots can support students’
SSRL for improving collaborative learning
, t
his study was aimed at exploring
the following research questions
.
The
first question was how agent ro
bots are accepted by learners (uni
versity students in this study)
.

SAR studies
have

suggested
that robots have
productive and assistive communication

with humans

in a variety of contexts (Tapus et
al., 2007)
.

Collaborative learning context, however, has not
yet
been deeply investigated.

W
e implemented agent
robots in an actual collaborative learning where university students were engaged in collaborative construction of
understanding based on articles
,

and
examined

how agent robots could function as partners.
The seco
nd question
was how th
e human
-
robot interaction in a collaborative learning context improves students’ learning processes and
outcome. In a collaborative context in this study, students and agent robots explained their responsible articles to one
another. We examined agent robo
t
’s influence

by comparing students’ note
-
taking strategies, argumentation on
articles and their written discourse
(after their jigsaw group activity)
in
human and robot explanation.


Method



Twelve undergraduate students from engineering and informatics

department took the course

as a
requirement of their teacher certificate program
. They were required to participate in the course as groups in which
they discussed their ideas related to the course content and constructed their shared understanding of the

learning
environment (Bransford, Brown, & Cocking, 1999) and its application to classroom practices. This study was
conducted in the first module of the course (the first one and a half day). In the first module, students were involved
in collaborative
re
ading comprehension

in which they construct
ed

their
collaborative
understanding through
explaining their responsible articles to one another and integrating ideas from different articles.


Coll
aborative Reading Comprehension



Collaborative reading compreh
ension is an activity structure based on Jigsaw method for learners to deeply
engage in their collaborative knowledge building through their understanding of multiple document resources
(Oshima & Oshima, 2011). In this study, students were first placed in
expert

groups. In each
expert

group, three
students collaboratively read and constructed their understanding of a different article that they would explain to
others afterward in
jigsaw

groups. We implemented the reciprocal teaching approach (Brown & Palin
csar, 1989;
Miyake & Shirouzu, 2006) for their collaborative reading such that each student took different roles and was in
charge of different sections of the article. Through their collaboration, each student produced a summary in our
prepared Microsoft
Word template that was used as a handout for the explanation in
jigsaw

groups.


Three
jigsaw

groups were then formed consisting of one student from each expert group and one robot.
Students in the
jigsaw

groups worked to integrate ideas from five differen
t articles explained by the responsible
students and a robot (see Figure 1). Robots were in charge of explaining article #3. After discussing the five articles,
the students reported how ideas from the articles were related to one another and interpreted t
hem with reference to
the basic framework of learning environments in a CSCL system.



Figure 1
. Participatory Structure of the Collaborative Reading Comprehension (left)

and Jigsaw Group Activity with a Robot (right).


Communication Robot as Agent



Ro
bovie
-
W was placed on the desk and remotely operated by
operators,
teaching assistants who had
studied the same content at their laboratory work. Robovie
-
W communicated with learners in the following ways.
First, it spoke a pre
-
set script to explain the ar
ticle (#3) written by the first author. We prepared explanation script
for the article that Robovie
-
W spoke by dividing it into three sections. Each section of script was spoken when an
operator pressed the “play” button on the operating interface. Second,

it spoke sentences that the operators typed on
their laptop computers at the time. Third, Robovie
-
W moved his head and arms to provide nonverbal signs to
students, such as moving its face toward students who were speaking.


After
explaining

each segment o
f the script, Robovie
-
W asked students to monitor their understanding. We
prepared scripts for answering predictable questions such as the meaning of specific terms in the explanation
beforehand. Operators could also respond learners’ questions that were a
nswerable with their knowledge. When
students asked questions that might lead to their deeper conceptual understanding but they could not facilitate by
responding, operators were instructed to recommend learners to ask their instructor for help by pressing

the
help
-
seeking

button.


Data Collection



Three kinds of data were collected for the research purposes. First, students’ discussion in their
jigsaw

groups were video
-
recorded and transcribed. The transcription was used for analyzing
episodes

of how robo
ts were
accepted by students and for examining how students collaboratively constructed their understanding of articles
through their argumentation. Second,
students’

note
-
taking activities on handouts were collected. We identified and
counted different ty
pes of note
-
taking strategies they used.

Third, we also analyzed students’ written discourse in
discussing how ideas from articles were related to the framework of learning environment on a CSCL system

after
their engagement in the collaborative reading co
mprehensi
o
n
. We identified parts of discourse in that students used
their understanding of articles for making inference related to the framework of learning environment.


Results and Discussion

How Robots W
ere Accepted by Students



We here
examine how ag
ent robots were accepted by students in collaborative learning by analyzing
episodes we observed.

First, the implementation of Robovie
-
W made group dynamics more social. Learners
discussed with one another how they like
d

to communicate with the new partner

in their group. The presence of
physical agent naturally activated communication among learners. Second, its physical embodiment had learners’
attention. Its movement to look at a learner explaining was welcomed (e.g., “Oh, he looks at me. So cute.” “He
s
eems to be wondering.”). Third, although robot’s speaking was not smooth and natural, students were quickly
adopted to communicate with robots. They attempted to make eye contact with Robovie
-
W frequently, and naturally
nodded in agreement with it during i
ts explanation of the article. For instance, in a group, learners were engaged in
the following discourse with Robovie
-
W:

Example #1

[After completing a section of explanation]

Robovie
-
W
: This was explanation of this section. Does anybody have a question?

Student
S
: Oh, I wanna ask a question. Let me see…

Student
I
: Yeah, me too. Do you [Student A] have one?

Student
S
: Well, I really try to ask him [Rbovie
-
W], but I guess, I understood what he said.

Student
I
: Your [Robovie
-
W] explanation was really good. W
e got you.


Example #2

In another example,

[After completing a section of explanation]

Robovie
-
W
: This was explanation of this section. Does anybody have a question?

Student A
: Yeah, it was nice. Why don’t you [Robovie
-
W] go on?

These examples suggest tha
t learners accepted robot as their learning partner and did not see authority behind it.
Learners evaluated robot’s explanation as useful resource for them to understand the articles.



In another example, learners attempted to challenge Robovie
-
W by aski
ng a question based on their
understanding.

Example #3

[After completing a section of explanation]

Student C
: Yeah, I have a question. You [Robovie
-
W] told that the authenticity of study materials is
important

for
students to understand why they have to s
tudy or how they are engaged in their meaningful cultural
practices. I agree, but how can you do that in the classroom? How can teachers bring the authenticity for
their students in the classroom?

Student A
: Wow, I wonder if he [Robovie
-
W] can respond it.

Student B
: It may be a bit tough for you [Robovie
-
W].

Robovie
-
W
: Well, let me think… [Operator typed the sentence.]

Robovie
-
W
: This is
a
really good question, but I cannot answer it. Why do not we ask Prof. O for help? [Operator
press the
help
-
seeking

scri
pt button.]

[Loud laugh in the group]

Prof. O
: Do you guys have a question?

Student C
: Yes, we are now discussing how we can bring the authenticity into the classroom practice.

[Discussion continued.]

In this example of discourse, student C brought up his

question that was valuable for his group to further discuss.
The operator first tried to respond by himself but decided to recommend learners to ask the instructor for help. Their
discussion after this episode continued for relatively long period and we c
onsidered that learners’ understanding was
quite deepened. One important role played by robot here was that it detected learner’s question by having them
monitor their understanding and led them to engaging in deeper discussion with seeking help by the ins
tructor.


Whether the Human
-
Robot Interaction Improved Students’ Learning Processes and Outcome



First, we analyzed students’ note
-
taking strategies based on their note
-
taking behaviors on handouts during
explanations. Six types of note
-
taking strategies
(i.e.,
memo,
e
mphasis, figure, underline, area, and
arrow
) were
identified and counted.

Students used their note
-
taking strategies 154 times (mean) in human explanation and 173
times in robot explanation, and there was no significant difference between the
m,

2

= .55,
df

= 1,
p

> .05.



Second,
we counted episodes of argumentation. The episode was defined here as
a
segment of discourse
where students shared the same part of content as target of their discussion and attempted to construc
t their
argument. Stu
dents

engaged in argumentation episodes 6.5 times (mean) in human explanation a
nd 9 times in robot
explanation, there was no significant difference,

2

= .20
,
df

= 1,
p

> .05
.

In each argumentation episode, three
components of argument (i.e., Data, Reasoni
ng and Claim) were identified and counted

(see Figure 1)
.
A 2
(Explainer: Human vs. Robot) X 3 (Component) ANOVA on frequencies
showed
a main effect of Component,
F
(2,40) = 12.72
,
p

< .05
.
A Tukey’s HSD test manifested that students in either explanation c
ontext generated
significantly more claims than the two other.
We further analyzed proportions of students who contributed to three
components of argument in each episode

(see Figure 2)
. A 2 (Explainer) X 3 (Component) ANOVA showed two
main effects,
F
(1,20
) = 8.40,
p

< .05 for Explainer,
F
(2,40) = 26.41,
p

< .05, and the interaction effect,
F
(2,40) =
7.25,
p

< .05.

A Tukey’s HSD test manifested that
significantly more proportion of students contributed to claim in
human explanation than robot explanation.









Third,
w
e analyzed how
students

used their understanding from their collaborative reading comprehension
activity in the nex
t phase of learning. Learners in
jigsaw

groups reported ideas from their discussion as notes in a
CSCL environment. We counted ideas in which learners referred to original articles and made any inference on them

(see Table 1)
. A Chi
-
square analysis
showed
no
significance.

Figure 1
. Mean Frequencies of Three
Components in Argumentation Episodes.

Figure 2
.
Proportions of Students Who
Contributed to Three Components in
Argumentation Episodes.


Table 1:
Frequencies of Learners’ Inference in Discourse Related to Five Articles
.


Article 1

Article 2

Article 3

Article 4

Article 5

14

19

6

14

16



Findings in this study are summarized as follows. First, it was demonstrated that univ
ersity students
accepted agent robots as their partners in collaborative learning. As SAR studies suggest, humans may adapt
themselves into the new situation of human
-
robot interaction in collaborative learning context. Most SAR studies so
far have discuss
ed the single human
-
robot interaction, and our findings could add further evidence of the human
-
robot interaction in collaborative learning.
Second, agent robots could function as learning partners quite equally as
did human learners.
In comparison with hu
man explanation,
robot explanation could elicited equal amount of
meaningful learning processes such as note
-
taking behaviors, argumentation activities and positive participation
(except for making claims), and learning outcome such as their use of underst
anding from learning.

These findings
suggest us that we could further design scaffolding scripts by agent robots to improve students’ learning outcome
through facilitating their SSRL in collaborative learning.


References


Azevedo, R. (Ed.) (2007). Specia
l issue: Understanding the complex nature of self
-
regulatory processes in learning
with computer
-
based learning environments.
Metacognition and Learning
,
2
(2
-
3).


Azevedo, R., Cromley, J. G., Moos, D. C., Greene, J. A., & Winters, F. I. (2011).
Adaptive co
ntent and process
scaffolding: A key to facilitating students’ self
-
regulated learning with hypermedia.
Psychological Test and
Assessment Modeling
,
53
(1), 106
-
140.


Azevedo, R., Cromley, J. G., & Seibert, D. (2004). Does adaptive scaffolding facilitate stu
dents' ability to regulate
their learning with hypermedia.
Contemporary Educational Psychology, 29
, 344
-
370.


Azevedo, R. & Strain, A. C. (2011). Integrating cognitive, metacognitive, and affective regulatory processes with
MetaTutor. In R. A. Calvo & S. K
. D’Mello (Eds.),
New Perspectives on Affect and Learning Technologies

(pp.
141
-
154). New York, NY: Springer Science + Business Media.


Breazeal, C. (2002).
Designing Sociable Robots
. Cambridge, MA: MIT Press.


Bransford, J. D., Brown, A. L., & Cocking, R.

R. (Eds.) (1999).
How people learn: Brain, mind, experience, and
school
. Washington, DC: National Academies Press.


Brown, A. L. & Palincsar, A. S. (1989). Guided, cooperative learning and individual knowledge acquisition. In L. B.
Resnick (Ed.),
Knowing,

learning, and instruction: Essays in honor of Robert Glaser
. Mahwah, NJ: Lawrence
Erlbaum.


Dillenbourg, P., Huang, J., & Cherubini, M. (Eds.) (2009).
Interactive Artifacts and Furniture Supporting
Collaborative Work and Learning
. New York, NY: Springer S
cience + Business Media.


Feil
-
Seifer, D. & Mataric, M. J. (2005). Defining socially assistive robotics.
Proceedings of the 2005 IEEE 9th
Interactional Conference on Rehabilitation Robotics
, 465
-
468.


Goetz, J. & Kiesler, S. (2002). Cooperation with a robo
tic assistant.
Proc. Int. Conf. Computer
-
Human Interaction
(CHI’02)
, 578

579.


Hadwin, A. F., Jäevelä, S., & Miller, M. (2011). Self
-
regulated, co
-
regulated, and socially shared regulation of
learning. In B. J. Zimmerman & D. H. Schunk (Eds.),
Handbook of
Self
-
Regulation of Learning and Performance

(pp. 65
-
84). New York, NY: Routledge.


Hadwin, A. F. & Oshige, M. (2011). Self
-
regulation, coregulation, and socially shared regulation: Exploring
perspectives of social in self
-
regulated learning theory.
Teacher
s College Record
,
113
, 240

264
.


Järvenoja, H., & Järvelä, S. (2009). Emotion control in collaborative learning situations? Do students regulate
emotions evoked from social challenges?
British Journal of Educational Psychology
,
79
, 463

481.


Miyake, N. & O
kita, S. Y. (2012). Robot Facilitation as Dynamic Support for Collaborative Learning. In J. van
Aalst, K. Thompson, M. J. Jacobson, & P. Reinmann (Eds.),
The Future of Learning: Proceedings of the 10
th

International Conference of the Learning Sciences (ICL
S2012)


Volume 2, Short Papers, Symposia, and Abstracts

(pp. 57
-
63). International Society of the Learning Sciences: Sydney, NSW, AUSTRALIA.


Miyake, N. & Shirouzu, H. (2006). A collaborative approach to teach cognitive science to undergraduates: The
lear
ning sciences as a means to study and enhance college student learning.
Psychologia
,
18
(2), 101
-
113.


Oshima, R. & Oshima, J. (2011, April). Knowledge building for pre
-
service teachers through collaborative reading
comprehension.
Paper presented at the Ann
ual Meeting of the American Educational Research Association, New
Orleans, LA.


Rogat, T. K. & Linnenbrink
-
Garcia, L. (2011). Socially shared regulation in collaborative groups: An analysis of the
interplay between quality of social regulation and group pr
ocesses.
Cognition and Instruction
,
29
(4), 375
-
415.


Roschelle, J., & Teasley, S. (1995). The construction of shared knowledge in collaborative problem solving. In C.
O’Malley (Ed.),
Computer
-
supported collaborative learning
(pp. 69

197). Berlin, Germany:
Springer.


Salovaara, H., & Järvelä, S. (2003). Students’ strategic actions in computer
-
supported collaborative learning.
Learning Environments Research
,
6
, 267

285.


Shirouzu, H., Miyake, N., & Masukawa, H. (2002). Cognitively active externalization for s
ituated reflection.
Cognitive Science
,
26
, 469
-
501.


Stahl, G. (2006).
Group cognition: Computer support for building collaborative knowledge
. Cambridge, MA: MIT
Press.


Tapus, A., Matalic, M. J., & Scassellati, B. (2007 March). Socially assistive robotics
: The grand challenges in
helping humans through social interaction.
IEEE Robotics and Automation Magazine
, 35
-
42.


Vauras, M., Iiskala, T., Kajamies, A., Kinnunen, R., & Lehtinen, E. (2003). Shared
-
regulation and motivation of
collaborating peers: A case analysis.
Psychologia
,
46
, 19

37.


Volet, S., Vauras, M., & Salonen, P. (2009). Self
-

and social regulation in learn
ing contexts: An integrative
perspective.
Educational Psychologist
,
44
, 215

226
.


Appendix: Original
(Japanese)
Transcriptions of Episodes of Human
-
Robot Interaction