Supporting Multi-Language Communication in Children`s

Master Thesis
Supporting Multi-Language
Communication
in Children’s Workshop
Supervisor
Professor Toru ISHIDA
Department of Social Informatics
Graduate School of Informatics
Kyoto University
Shohei HIDA
February 12, 2016
i
Supporting Multi-Language Communication
in Children’s Workshop
Shohei HIDA
Abstract
Recently, the importance of education for international understanding is increasing. In an international field, the lack of international understanding causes
negative feelings such as misunderstanding and prejudice. In this situation, intercultural collaborations for children’s international understanding education
are held by some schools and international Non-Profit Organizations(NPOs).
In intercultural collaborations for international understanding education, participants who speak different languages learn from from cooperative works such
as problem-solving discussions.
Previous researches supported face-to-face multi-language communication
with machine translation, but these researches focused on support only for
adults.
However, unlike support for adults, in a children’s multi-language
communication, children’s specific problems are likely to occur and a multilanguage and intercultural situation which is unfamiliar to children increases
such problems. Therefore, detecting children’s typical problems in a multilanguage environment and developing a support system to solve these problems are needed. Hence, this research analyzes communication in a children’s
multi-language workshop and detects problems in a children’s multi-language
environment. Moreover, this research improves the multi-language discussion
support system and applies it to the real field in order to evaluate the system.
In this research, we deal with following two issues.
Analysis of children’s multi-language communication based on a real
field
This research points out children’s specific problems in multi-language communication environment from the analysis of communication in actual children’s
multi-language workshops.
Improvement of the multi-language discussion support system for
children
This research proposes and implements a multi-language discussion support
ii
system in order to solve problems and support children’s international understanding education.
We found that problems are caused by translation’s accuracy but also by
the obstruction of discussion by children’s selfish behaviors. For example, some
children disturbed discussion by suddenly posting meaningless text and consecutive pictograms. In addition, excessive self-assertion and extremely inactive
participants were also pointed out. These behaviors are considered as a factor that increases the difference of utterances opportunity among a team as
well as individual personality or typing skill. A dominant participant has the
possibility of imposing his/her opinion and making the atmosphere discourages
contribution from inactive participants. Hence these behaviors are not undesirable for a children’s international understanding education field. That is why
we regard this behavior as problems.
In this research, we share analysis and subjective problems of the field in
order to detect problems to solve by the system. Then, we propose a multilanguage discussion support system for children to solve these problems. We
implement user a interface that promotes correction of children’s wrong behaviors by themselves and elicits various ideas and interactions. We introduce the
system to multi-language workshops in order to evaluate the system.
The main contributions of this research are described as follows:
Finding problems in children’s multi-language communication
This research clarifies problems in children’s multi-language communication
by analyzing actual multi-language workshops. In a children’s multi-language
communication environment, sudden selfish behaviors such posting meaningless
texts and consecutive pictograms disturb discussion.
Proposal and implementation of multi-language discussion supporting system for children
This research implements a multi-language discussion support system which
introduces Control of Error in the Montessori method as design background
in order to promote equal discussion and to elicit various opinions. Improvements caused children’s behaviors such as voluntary control of their utterances
frequency and taking care about others, and makes discussion more active.
iii
児童のワークショップにおける多言語コミュニケーション支援
肥田 昌平
内容梗概
近年，国際理解を目的とした教育の重要性が高まっている．国際的なフィール
ドでは異なる文化や言語に対する理解の欠如が誤解や偏見といったわだかまり
を生む原因となる．そうした中で特に児童に対する国際理解教育のための異文
化コラボレーションが学校や国際的な活動をおこなう非営利活動法人等によっ
て開催されている．国際理解教育のための異文化コラボレーションでは言語混
合で問題解決型の議論や共同タスクの遂行を通じて学ぶ．
対面での多言語コミュニケーションを機械翻訳を用いて支援する研究がある．
しかしこれまでの研究では支援対象として大人が想定されてきた．一方で大人
に対する支援とは異なり児童の多言語コミュニケーションでは児童固有の問題
が生じ，異文化・多言語といった慣れない環境がこれを増長する可能性がある．
したがって児童特有の問題の特定とそれらを解消する支援システムの開発が必
要である．
そのため本研究では児童のための多言語ワークショップを対象として，そこで
発生するコミュニケーションの分析をおこなう．分析で特定した児童の多言語
コミュニケーション環境における課題を解決するための多言語議論支援システ
ムの開発によって多言語コミュニケーション環境の改善をおこない，再度フィー
ルドに対して適用することで評価と考察をおこなう．
本研究で取り組む課題は以下の 2 点である．
実フィールドに基づく児童の多言語コミュニケーションの分析
児童の多言語ワークショップという実際のフィールドから取得したデータの
分析によって多言語コミュニケーション環境で発生する児童特有の問題点を指
摘する．
児童のための多言語議論支援システムの改善
分析によって特定された問題から，それらを解消し，児童の国際理解教育の
ための多言語コミュニケーション環境において適切な支援をおこなうための多
言語議論支援システムの提案と実装をおこなう．
ワークショップでおこなわれた多言語コミュニケーションの分析によって機
械翻訳の精度に起因する問題に加え，児童の利己的行動による議論の阻害が発
iv
見された．例えば突発的に無意味な文字列を投稿したり，悪戯に連続投稿を重
ねたりすることで全体の議論を阻害する行動が見られた．また過度な自己主張，
あるいは極度に消極的な参加者の存在も指摘された．これらは性格やタイピン
グ能力等様々な要因によって発生する発言機会の偏りを増長する一因と考えら
れる．支配的な参加者の存在は意見の押し付けや内向的な参加者が更に発言し
辛い状況を作りだす可能性があるため，特に国際理解のための児童教育現場で
は好ましくない．したがってこれらを問題であると考えた．
本研究では分析によって発見した児童の多言語コミュニケーションの課題と
フィールドの主観的な課題意識のすり合わせからシステムで解決すべき課題を
指摘する．またそれらを解決する児童のための多言語議論支援システムの提案
をおこなう．多言語議論支援システムでは児童が自身の誤りを自発的に修正す
るような行動を促進したり，多様な意見の創出やインタラクションの増加を促
すユーザインタフェースを実装する．実装した多言語議論支援システムを再度
多言語ワークショップに導入し，得られたコミュニケーションおよび参与観察
の記録を分析することで評価と考察をおこなう．
本研究での主な貢献は次の 2 点である．
児童の多言語コミュニケーションにおける課題の特定
児童の多言語ワークショップの現場で実際におこなわれた多言語コミュニケー
ションの分析から問題点を明らかにした．児童の多言語コミュニケーション環
境において無意味な文字列や絵文字の連続投稿が見られる等，突発的な利己的
行動が議論を阻害していることが分かった．
児童のための多言語議論支援システムの提案と実装
モンテッソーリ教育法における Control of Error を基本設計に据え，議論参
加者の平等な発言機会や多様な意見の創出を促進する，児童のための多言語議
論支援システムの提案と実装をおこなった．参加者の投稿状況の可視化等の改
善により，参加児童が自発的に発言量を調整したり，消極的な参加者に対する
配慮を見せる等の他者に対する協調的な行動が見られ，議論が活性化した．
Supporting Multi-Language Communication
in Children’s Workshop
Contents
Chapter 1
Introduction
1
Chapter 2
Related works and field
4
2.1
Related works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
2.2
Target area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
2.3
Field: Kyoto Intercultural Summer School for Youths . . . . . . .
6
Chapter 3
The first experiment
10
3.1
Outline of the first experiment . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2
The multi-language discussion support system in the first experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3
3.4
Result and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3.1
Chain of utterances . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3.2
Difference of post opportunity . . . . . . . . . . . . . . . . . . 15
3.3.3
Field interview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Problems in children’s multi-language communication . . . . . . . 19
Chapter 4
Design of the multi-language discussion support system
22
4.1
Design background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Chapter 5
The second experiment
27
5.1
Outline of the second experiment . . . . . . . . . . . . . . . . . . . . . . 27
5.2
Result and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.3
5.2.1
Effect of visualizing ratio of posts . . . . . . . . . . . . . . . . 28
5.2.2
Chain of utterances . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2.3
Thread analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Chapter 6
Conclusion
47
Acknowledgments
50
References
51
Appendix
A.1
A.2
A-1
Hearing sheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A-1
A.1.1
Hearing sheet for team leaders . . . . . . . . . . . . . . . .
A-1
A.1.2
Hearing sheet for technical staffs . . . . . . . . . . . . . .
A-7
The ethnography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-11
Chapter 1
Introduction
Recently, importance of education for international understanding is increasing. In the international fields where gathering various backgrounds people,
full understanding of other cultures and languages is needed, and lack of international understanding causes negative feelings like misunderstanding and
prejudice. Importance of education for international understanding especially
for children is increasing so that this is described in course of study of periods for
integrated study for elementary established by Ministry of Education, Culture,
Sports, Science and Technology in Japan1) . There are some activities providing intercultural collaboration fields to children for international understanding
education by some schools and international Non-Profit Organizations(NPOs).
In such fields, cooperative communication like problem-solving discussions and
cooperative tasks are more important than competitive communication like a
debate and a speech contest. Participants in such fields learn from such type of
communication.
Because the participants of an intercultural collaboration have different
mother languages, they have to communicate each other somehow. In many
cases, all participants use English as a common language to solve this problem[1].
However, communication in English is difficult for most of non-native speakers
[2]. Particularly the case of children who have lower language skill than adults,
this difficulty becomes higher. Therefore, if they use English as a common
language, in some case, some participants’ influence decline and they cannot
contribute discussion. To solve this problem, some researcher studied supporting method by interpreter or machine translation, so we have to decide supporting method depending on communication style in target fields. Because
in face-to-face communication like problem-solving discussions and cooperative
tasks, which are usually used in international education fields for children, participants interact each other, machine translation support is effective if certain
extent accuracy is ensured. This is because in such type of communication,
rapid translation is more important than translation accuracy[3].
1)
http://www.mext.go.jp/
1
Some researchers have been studying and developing multi-language communication support systems in order to help communication in intercultural
collaboration fields. For example, Aiken et al. developed the decision-making
support system[4], and Yamashita et al. developed the chat system integrated
machine translation[5]. Besides, communication analysis focuses on different
cultures have been studied, for example, Nguyen et al. analyzed text-based communication between Chinese and American with the instant messaging tool[6].
However, these previous researches focus on supporting and analyzing environments for adults. Environments for children are expected to have different
problems from adults.
Thus, this research analyzes multi-language communication on support system in the children’s multi-language workshop. In this research, we analyze
contents of posts sequence between two utterances and frequency of each user’s
utterances on multi-language communication support system to find problems.
Moreover, we share analyzed problems and subjective problems by field organizer to detect problems in children’s multi-language environment. Furthermore, we propose and implement the multi-language discussion support system
for children to solve detected problems.
This research deals with following two issues.
1. Analysis of children’s multi-language communication based on actual field
Systems that are assumed to be used in actual field like an intercultural
collaboration are affected by constraint of a field and unexpected events.
Therefore, this study makes much account of field research and applies the
system to the field and collects data from there. Moreover, we analyze
data collected from field research and detect children-specific problems of
communication in children’s multi-language environments.
2. Improvement of the multi-language communication support system for children
This research proposes and implements the practical multi-language discussion support system for children’s international understanding education
based on the field analysis.
This thesis is organized as follows: In chapter 2, we introduce related works
2
about the discussion support system which promotes harmony discussion and
multi-language meeting support system with machine translation, and show
difference from this research. Moreover, we also introduce the target field of this
research. In chapter 3, we explain outline and analysis of the first experiment
to design the multi-language discussion support system and show problems
in children’s multi-language communication. Chapter 4 introduces design and
implementation of the multi-language discussion support system based on the
first experiment analysis. In chapter 5, we describe analysis and consideration
about result of the second experiment introduced the improved multi-language
discussion support system. Chapter 6 concludes this thesis.
3
Chapter 2
Related works and field
In this chapter, we introduce the related works of discussion support system by visualizing social interaction and multi-language group support system. In addition, we explain the field, Kyoto Intercultural Summer School for
Youths(KISSY).
2.1
Related works
The research by Kim et al. developed discussion support system by visualizing social interaction between participants[7]. They support group meeting to
detect group dynamics in the meeting like each participant’s speaking time or
body movement and provide feedback to all member in the meeting. They proposed Meeting Mediator, which captures body movement, proximity to other,
and speech characteristics such as speaking speed and tone of voice by the Sociometric badge, electronic sensing device, and provides real-time feedback by
visualizing this data on the mobile phone of each user.
Figure 1 is visualization on mobile phone by Meeting Mediator. Meeting
Mediator represents four participants as colored squares in corner of screen. The
color of central circle gradually becomes green depending on interaction level
and the location of the circle change dependent on balance in participation. And
they further promote balanced speech by displaying each participant’s speaking
Figure 1: Visualization on the mobile phone[7]
4
time through the thickness of the line connecting the central circle with square
representing each member.
They make an experiment to 36 groups composed of 144 participants(71
male, 73 female, mean age 27.7). They compared 2 types of group meetings,
18 groups used Meeting Mediator and other groups did not. They analyzed
difference of behavior between 2 types of groups. In the groups using Meeting Mediator, system captures speech characteristics and body movement, and
feedback this data in real-time.
As a result of experiment, they found reduction the amount of overlapping
speaking time and increase social interaction. This means the quality of discussion increased. In addition, they found non-dominant participants behaved
more like the dominant person by feedback of social interaction. In the other
words, Meeting Mediator reduced difference between non-dominant person and
dominant person.
Aiken et al. developed group support system called Polyglot that provides
automatic translation[3]. Polyglot is face-to-face, multilingual group support
system with 41 languages based on Google Translate. This system is simple
multilingual chat system that can translate and display text in each user’s
native or preferred language.
They conducted experiment that translate 40 languages to English and evaluate this accuracy. 40 undergraduate business students measured they could
understand translation result or not. They reported 82.86% of the text could
be understood.
In addition, they let 83 undergraduate business students in six groups exchange comments in four languages(French, German, Italian and Spanish) with
Polyglot. After discussion, students reported being able to understand as well
90% of comments and almost all of students regarded support system with machine translation was useful. They reported machine translation is effective to
support communication between English and some European languages.
5
2.2
Target area
We explained related works about the support system visualizing participants’
social interaction to improve quality and equality of discussion and the group
support system integrating machine translation. However, these previous works
are supposed to be used by adults. In children’s multi-language communication,
there is a possibility of occurrence of other children’s specific problems, and
adapting support method for adults is difficult.
Therefore, in this research, we detect the problems in children’s multilanguage communication, and propose and implement the support system to
solve this problem. Actually, this research focuses on 8 to 14 year-old children.
This age is called “late school periods” and in this age, children mature ability
of considering other people and sociability to live in communities[8]. Therefore,
this research focuses on children in late school periods and develops the system to support cooperative multi-language communication for them. Besides,
this research focuses on face-to-face discussion by small team consist of 5 to
7 children with machine translation. In a fields for international understanding, equal discussion which all members have equal opportunity to express their
opinions is more important than discussion which some dominant participants
have strong influence and only their opinions decide direction of team. Moreover, this research supposes that adult staffs support children for their smooth
communication.
2.3
Field: Kyoto Intercultural Summer School for Youths
This research address Kyoto Intercultural Summer School for Youths(KISSY)
held by NPO Pangaea, intercultural collaboration event for children, as experiment field. KISSY is the boarding intercultural event for about 1 week. Children
who are from Japan, Korea, Cambodia, Kenya and United States make teams
and discuss each other in order to learn from communication beyond language
barrier. KISSY was held two times in the past and NPO Pangaea will hold once
a year in the future. Table 1 shows participants detail who joined KISSY 2014
and KISSY 2015. Children’s names are anonymized as their mother language
and serial numbers and team leader’s names are also represented as character
6
string ”TeamLeader” and serial numbers.
Table 1: Participants detail in KISSY 2014 and KISSY 2015
Red Team
Blue Team
Yellow Team
Green Team
KISSY 2014
KISSY 2015
Japanese-1,2,3,4
Japanese-14,15,16
Korean-1
Korean-8,9,10
Khmer-1
English-2
TeamLeader-1
TeamLeader-5
Japanese-5,6,7
Japanese-17,18,19
Korean-2,3
Korean-11,12,13
English-1
Khmer-3
TeamLeader-2
TeamLeader-6,7
Japanese-8,9,10
Japanese-20,21,22
Korean-4,5
Korean-14,15,16
Khmer-2
Khmer-4
TeamLeader-3
TeamLeader-8
Japanese-11,12,13
Japanese-23,24,25
Korean-6,7
Korean-17,18,19
TeamLeader-4
English-3
TeamLeader-9
KISSY has the activity called ”workshop”. In the workshop, about 7 children and 1 adult staff make teams. They are assigned creative task along
theme1) provided by workshop’s organizer with PicoCricket2) and some materials such as papers and expanded polystyrenes which prepared by workshop’s
organizer beforehand. Participants use the KISSY system to communicate with
other language children in the workshop. The KISSY system has the multilanguage discussion support system with machine translation, the idea voting
system for brainstorming, the making multi-language worksheet function to
1)
2)
2014’s theme was “bond” and 2015’s was “We Love”
http://picocricket.com/
7
brief outcome of team activity, all members’ profile pages and the learning
greeting system. In the workshop, children basically communicate through the
multi-language discussion support system. The multi-language discussion support system uses language-grid[9] to translate texts. Figure 2 is overview of the
KISSY system.
Translation
request
Request
HTTP
Access
Participants using
web browser
KISSY
System
SOAP
Access
Translation
result
Response
Language Grid
Figure 2: Overview of the KISSY system
Figure 3 shows a actual discussion scene using the KISSY system and final
artwork created by children. Participants present their artwork after workshop
activity.
(b) Artwork created by children
(a) Discussion scene
Figure 3: Scenes of the workshop(These photos are provided by NPO Pangaea)
In this research, we set the KISSY’s workshop activity to target field and
we collect and analyze data from KISSY 2014 as the first experiment. We
detect problems of children’s multi-language communication by analyzing data
8
collected from the first experiment and improve the multi-language discussion
support system. Furthermore, we regard KISSY 2015 as the second experiment
and introduce the improved multi-language discussion support system. And we
analyze the data collected in the second experiment to evaluate the effect of the
improved multi-language discussion support system.
9
Chapter 3
The first experiment
This research conducts the first experiment to design the multi-language discussion support system based on actual data and phenomena in the field. This
chapter shows the outline, result and analysis of the first experiment.
3.1
Outline of the first experiment
NPO Pangaea held the KISSY 2014 in July 31, 2014 to August 5, 2014 at Shirankaikan, Kyoto University in Japan. 23 participants: 13 Japanese, 7 Korean, 1
Kenyan and 2 Cambodian children joined KISSY 2014, and about 15 adult
staffs came from around the world to support children. This research regards
KISSY’s workshop activity as the first experiment to design the multi-language
discussion support system for children. In the first experiment, participants
were gave creativity tasks with team.
In the workshop activity, 23 children are divided into 4 teams: Red Team,
Blue Team, Yellow Team and Green Team. 4 teams have equal languages
distribution as much as possible and children performed all activity with their
team during workshop. However, because a Kenyan participant could not join
this event unexpectedly, Green Team had only 5 members, that was less one
person than other teams. In addition, each team has one adult staff called team
leader to support children’s communication and to facilitate team discussion. In
the first experiment, Blue Team’s team leader was English and Korean bilingual
but other teams’ team leader was English and Japanese bilinguals. Table 2
shows language configuration in each team in the first experiment.
Table 2: Language configuration in each team in the first experiment
English
Japanese
Korean
Khmer
Total
Red Team
0
4
1
1
6
Blue Team
1
3
2
0
6
Yellow Team
0
3
2
1
6
Green Team
0
3
2
0
5
10
3.2
The multi-language discussion support system in the
first experiment
Participants of the first experiment use Web system called the KISSY system
to communicate with other language speakers. They use the multi-language
discussion support system(Figure 4) when they discuss each other for their
creativity task.
As Figure 4 shows, the multi-language discussion support system that were
used in the first experiment has the area displaying texts posted by all team
members in the right side of screen. The multi-language discussion support
system has a simple user interface that participants post their utterances on
this area, and when a new text is posted, this text is displayed at the highest
position in this area. To post new message, a user inputs text at text box on
the left side of screen or chooses a pictogram and submits it. Posted texts will
be translated by The Language Grid[9] and displayed in language selected by
each user. User can see his/her own posts by red color.
Figure 4: The multi-language discussion system used in the first experiment
(This screenshot is provided by NPO Pangaea)
11
3.3
Result and analysis
In the first experiment, 1583 posts were sent in total except team leader’s posts.
Each team posted the following number of texts: Red Team 651, Blue Team
117, Yellow Team 410 and Green Team 405.
We analyzed data collected from the first experiment. After experiment,
we extracted and analyzed data posted on the multi-language discussion support system from the KISSY system’s database to detect problems in children’s multi-language communication. Moreover, we interviewed to organizer
of KISSY, NPO Pangaea and got hearing sheets to adult staffs joined the experiment. We considered these data comprehensively and briefed to ethnography
that usually used in cultural anthropology. Furthermore, we interviewed to
field’s organizer again. In field research, problems recognized by field have high
priority to solve, so we found problems to solve by integrating description in
the ethnography and subjective awareness of the issue by field’s organizer.
3.3.1
Chain of utterances
We focus on chain between two utterances in order to find failure pattern of conversation in the first experiment. In the conversation analysis, adjacency pairs
are regarded as important concept[10]. Adjacency pairs divide two utterances
into “first pairs part” and “second pairs part”. Specific types of utterances(for
example, Questions or Offer) occur in other specific types of utterances(for example, Answers or Acceptance/Refusal). For example, in the first experiment,
✓
following adjacency pairs were posted1)2) .
TeamLeader-4: Do you represent How do culture?[Wh-Question]
Japanese-13:
✏
Holding Hands With that, you feel the bite on
earth[Answer(Idea)]
Japanese-11:
Dance[Answer(Idea)]
TeamLeader-4: What dance?[Wh-Question]
Japanese-11:
✒
1)
2)
You were Japan, Toka Bon Odori[Answer(Idea)]
We define Wh-Question as question which expects to be responded specific opinion
These texts were posted in Japanese.
12
✑
Turn-taking is also important concept in conversation. In conversation, only
one person is allowed being current speaker and sequence of conversation are established by reiteration of turn-taking[11]. There are two patterns to distribute
turn: one is that current speaker selects next speaker, another is that someone are selected by himself as next speaker. Contents and contexts of current
utterance or opinion of current speaker decide this distribution way. At above
case, TeamLeader-4 did not select next speaker explicitly but Japanese-13 was
selected by himself as next speaker. After that, Japanese-11 was also assigned
by himself and posted answer to TeamLeader-4’s question. Thus, utterance
sometimes links to several sequence pairs but speaker is always determined one
person. If some people speak simultaneously, one person is selected as next
speaker immediately.
The conversation on the multi-language discussion support system is also
seem to be consist of these iteration of turn-taking and making adjacency pairs.
Thus, we examined if these conversational function are completed or not. In this
research, we investigate that how many failure patterns of sequence of utterances, for example the case that there are no answer after question, are there.
Furthermore, we considered failure patterns of communication by qualitative
analysis.
Conversation on the multi-language discussion support system for creativity
task often begin from “Yes/No-Question”, “Wh-Question” or “Offer”. These
types of utterances cause response such as “Answer”, “Opinion(Idea)” or “Acceptance/Refusal” and conversation consists of iteration of these pair. Thus,
we labeled utterances type: “Yes/No-Question”, “Answer”, “Wh-Question”,
“Opinion(Idea)”, “Offer”, “Acceptance”, “Refusal” plus other utterances type:
“Repair”, “Pictogram” and “Other” to all texts posted in the first experiment.
Failure is the case that “Yes/No-Question”, “Wh-Question” and “Offer” get
no response, so we investigated patterns of these cases. Besides, “Repair” is
labeled to not only correcting one’s own posts but also procedure solving all
problems on conversation such as rephrasing and asking again[12]. Because we
found some users often divide their sentences, we regard this type of posts as
“Repair”.
13
As a result, there are total 287 posts labeled “Yes/No-Question”, “WhQuestion” and “Offer”: Red Team has 130 posts, Blue Team has 20 posts,
Yellow Team has 60 posts and Green Team has 77 posts. Moreover there are
total 113 posts(39.4%) that get no responses: Red Team has 61 posts(46.9%)，
Blue Team has 2 posts (1.0%)，Yellow Team has 21 posts(35.0%)，Green Team
has 29 posts(37.7%)
To detect characteristic failure patterns this research did qualitative analysis
of message contents and ethnography research, we will explain later. that finds
behaviors and phenomena that do not appear in statistical analysis. As a result,
we found characteristic pattern as follows:
1. Input errors and bad machine translation result
Some failure patterns are caused by bad translation result. Some users
often made mistype because they are not good at inputting with keyboard
and input colloquial text. In such case, machine translation result will be
not able to understand. To make matters worse, important word like an
user’s name are sometimes translated as bad result.
2. Unclear destination of posts
The system used in the first experiment does not have function to indicate
destination. Some users input destination user’s name directly in sentence
but most posts do not have such information.
3. unrelated posts from discussion topic
Some posts that are not related with a discussion topic, especially posted
by team leader, are sometimes not able to get any responses. In that case,
other posts related with topic can get some responses.
4. Turning out from screen by other posts
The multi-language discussion support system’s user interface has limited
area to show text. Hence texts will be turned out from screen immediately,
if several texts or pictograms posted in a short time. That is considered
why some posts did not notice by anyone.
Failure pattern 1 especially disturbs interaction between other language participants. In addition, failure pattern 1 was not improved in spite of the field’s
organizer asked children to take care for input text to fit machine translation.
14
Few children concerned that other members understood their texts. Besides,
on the multi-language discussion support system, children hardly ask someone
again and correct text posted by themselves, which is often seen at ordinary
conversation. That is why if machine translation output bad results, almost all
of this text is ignored.
Failure pattern 2 was recognized by field staff as problem. For example, at
second day, after workshop, TeamLeader-1 posted following text to their team
✓
members.
• 会話が誰に対してのものなのかわからないので、名前を入れるか、皆と
✏
入れるかしてください。(Since I do not know who your post refers to,
please write the name or “Dear all members”.)
• 特に、誰かの意見に賛成したり反対したりするときは、名前を入れてく
ださい。(In particular, when you agree or disaree with someone, please
✒
put the name.)
✑
Failure pattern 4 is seemed to be caused due to system user interface, because
the system has limited area to display texts. Even though someone posts a text
related with discussion, if some users post simultaneously or particular user
posts many times, the text is turned out from text display area. Many children
were not good at inputting keyboard and they could not take care to other
children’s opinions. That is seemed to be a part of factor.
In addition, we also observed the cause of failure pattern 4 that some children suddenly posted many pictograms and meaningless texts. We found these
behaviors disturb team discussion by occupying text display area. Besides, as
Figure 5 shows, when one child begin to post many meaningless text, sometimes
other children are affected and also start to post such type of text. And the
case of posting many pictograms is same.
3.3.2
Difference of post opportunity
Participants of this research’s field have different age, gender, mother language,
language skill and so on. This difference is considered to affect to each partici-
15
Figure 5: Meaningless text(This screenshot is provided by NPO Pangaea)
pant’s contribution to team discussion. In general, dominant participants have
negative effect on team discussion such as discouraging contribution of other
members to discussion and imposing their opinion[13]. It is depend on kind of
discussion that dominant participants are good or not but in the international
understanding education for children, discussion that all members have same
opportunity of express their opinions regardless of any individual characteristic
is better than discussion that dominant participants have strong effect to team
discussion. Therefore, this research focuses on difference of post opportunity
between participants in same team.
We compare the number of posts between same team on the multi-language
discussion support system in the first experiment. For example, Figure 6 shows
the number of posts by each participant in Red Team.
Japanese-4 and Korean-1 posted large number of text through three days.
These Participants posted about 60% of Red Team’s total posts; on the other
hand, Japanese-1 posted a few through three days. Particularly, the second day
of the first experiment was important day that almost all of this day is used to
discuss about their artwork’s concept, each child’s role and so on. Japanese-2
who posted only 5 text at the first day increased his posts to 53, but Japanese-1
16
and Japanese-3 were still inactive and posted very few: Japanese-1 posted 9 at
the first day and 16 at the second day, Japanese-3 posted 11 at the first day
and posted 15 at the second day.
In terms of contents of messages, Japanese-4 and Korean-1 were very aggressive about discussion. Both of them posted a lot of opinions about creativity
Firstday
Secondday
Thirdday
Thenumberofposts
160
140
120
100
80
60
40
20
0
Japanese-1Japanese-2Japanese-3Japanese-4 Khmer-1
Korean-1
Username
Figure 6: The number of posts by each participant on each day in Red Team
Firstday
Secondday
Thirdday
Thenumberofposts
160
140
120
100
80
60
40
20
0
Japanese-11
Japanese-12
Japanese-13
Korean-6
Korean-7
Username
Figure 7: The number of posts by each participant on each day in Green Team
17
task and responded to other children’s opinions. Particularly, Korean-1 posted
“First, let’s define the role (original language is Korean)”, “Presenters who want
to? ? Please raise your hand.(original language is Korean)” and so on. She
often decided direction of team discussion and it is like team leader. On the
other hand, Japanese-1 hardly posted her opinions and her posts were very few
through three days.
This tendency was not peculiar for Red Team and seen in other teams. As
Figure 7 shows, in Green Team, Japanese-12 and Korean-7’s posts were not
increased through three days.
Furthermore, we found some users did not post at all in the specific periods.
For example, in Red Team, Japanese-2 never posted since about 1:50pm for one
hour in the second day. In Green Team, Korean-7 post nothing in spite of other
members discussed about their creativity task in the particular period. In such
case, it is better that team leaders or other members notice about such quiet
person and encourage to post his/her opinions or if he/she get some accident,
support to solve this. However, we could not found such type of behavior.
3.3.3
Field interview
Because this research considers field’s opinion is important and improves the
multi-language discussion support system with field organizer, we interviewed
NPO Pangaea, KISSY’s organizer, after the first experiment and are provided
hearing sheet to adult staffs joined KISSY 20141) . We got following main subjective problems and opinions about the multi-language discussion support system
used in the first experiment.
• Ability of keyboard typing affects the number of utterances.
• Mistranslation often disturbed team conversation, especially, when sentence indicating agree or disagree were mistranslated, participants con-
fused. Moreover, when some participants use strange phrase, translation
result tended to be bad accuracy. To make matters worse, the theme of
KISSY 2014, “bond”, which means connection, was translated as different
1)
Hearing was carried out by NPO Pangaea
18
meanings.
• They wanted function of editing translation results or retranslation.
• If translation response is delay, some children type text randomly or clicked
somewhere.
• They concern the difference between children who post many times and
children who are inactive and shy
• It is better that the multi-language discussion support system have function
of indicating particular post mentions to other specific post.
• When children are creating artworks, they often talk with same mother
language participants.
3.4
Problems in children’s multi-language communication
We found children’s behaviors that suddenly post a lot of texts and pictograms
and post Meaningless texts disturbed team discussion in children’s multi-language
communication environment. Children in late school periods have big individual difference of developmental stage and some children do not have ability of
collective behavior. Moreover, some children have excessive interest in information system like a computer and the multi-language discuss support system.
These children might do selfish behavior that differs from team purpose like
mischief suddenly, and if these behaviors become excessively, it disturbs communication in a group. We think these behaviors will be increased in unfamiliar
environment for children such as intercultural and multi-language environment
due to strain and anxiety. Furthermore, we point out these individual selfish
behaviors might affect other children. If effect of these behaviors to group is
low at first, it propagates other children and will be higher. In such case, there
is possibility that whole group will be confused. We consider children’s selfish
behaviors that were observed in the first experiment caused by these children’s
specific characteristics.
In addition, we confirmed that machine translation problem affects interaction especially between other language participants. Moreover, in children’s
multi-language communication, even though participants use machine transla19
tion, they hardly input desirable texts for machine translation and machine
translation accuracy became low level. Children often use informal expression
and do typing errors. Furthermore, if children face difficulty of understanding
other’s posts because of translation’s bad accuracy, they hardly listen again.
Then, we enumerate problems to solve on the multi-language discussion support system to help children’s multi-language communication. We described
the ethnography to share the problems of the system with the field organizer.
Ethnography is method describing and modeling phenomena occur in field and
its output[14]. In the situation that developed artifacts like information systems
are introduced into fields and affect human activity, ethnography deals with
relationship between artifacts and human or some humans through artifacts.
Ethnography is described how actions and behaviors of people are configured
influencing complexly to artifacts and other people in the fields. Actually, description of ethnography consists of field notes which is accumulation of memo
and document recording experience by field works[15], however in this research,
we described the ethnography from data from the KISSY system’s database,
interview for the field organizer and staffs.
In addition, we associated problems or considerations to phenomena that
are described in the ethnography based on analysis or some data sources. For
example, we associated consideration of factor of problem that “Korean-6’s utterances were difficult to judge who are mentioned, and moreover, translation
accuracy was bad. That is considered as reason why she could not get any responses.(original correspondence table is written in Japanese)” to a description
of the ethnography that ”but this time, even if all members logged in the chat
system, Japanese talked with only Japanese and Korean-6 and Korean-7 could
not get any responses.(original ethnography is written in Japanese)”
We interviewed the field’s organizer, NPO Pangaea, again to compare the
description of ethnography and subjective problems of the field. We extracted
problems to solve by giving priority to problem in ethnography and field’s subjective opinion. Table 3 show extracted problems and corresponding the description of ethnography or the correspondence table.
20
Table 3: Problems and description of the ethnography or the correspondence
table
Problems
Description of the ethnography or the correspondence table
Problem 1: Team discus-
because Korean-6 and Japanese-13 posted
sion is disturbed by mean-
about 20 pictograms repeatedly, conversation
ingless texts and repeated
scrolled and turned out from screen.
pictograms.
Problem 2:
Team leader
cannot aware of trouble
Korean-7 did not post anything more than 30
minutes of later workshop.
that occurred in team because of busyness of facilitation
Problem 3: Children who
problem that Korean children had difficulty of
are not good at inputting
inputting with Japanese keyboard was occurred
with keyboard are unable to
keep up with discussion.
Problem 4: User cannot in-
because post’s destination was unclear, commu-
dicate post’s destination.
nication often failed, thus TeamLeader-1 dictated children to indicate destination.
Problem 5: User lost topic
Korean-6 posted opinions to no longer finished
of discussion because of
topic and because no one responded to her,
rapid conversation.
she indicated discontent with following message
“No answer No
(original language is Ko-
rean)”
Problem 6: Texts are in-
because translation accuracy was bad, there
comprehensible because of
were no response to Korean-6
bad accuracy of machine
translation.
21
Chapter 4
Design of the multi-language discussion support system
We improve the multi-language discussion support system to solve the problems detected in Chapter 3. This chapter shows design background and actual
implementation of the improved multi-language discussion support system.
4.1
Design background
We introduce Control of Error, which is the important concept for The Montessori methods, in order to improve the multi-language discussion support system.
The Montessori methods rely on a “prepared environment” that helps children to develop their own skills. Because children have ability of self-growth,
adult staffs should be not teacher but facilitator who provides a free area
where promotes children’s sense experiences and the foundation for children’s
learning.[16][17]. This is established by Maria Montessori for educational method
for an infant or a kindergartner in early 20th century but now it is also applied
for after late school periods children[18]. Control of Error is important concept in The Montessori method [19]. This concept means that providing chance
to correct children’s errors and mistakes by themselves is more effective for
children’s growth than correcting by teachers.
The multi-language discussion support system developed by this research
aims to promote equal discussion regardless of individual characteristics. As we
described before, in an international understanding education field for children,
cooperative discussion is needed. In this case, all members’ contribution elicits various opinions hence providing equal utterances opportunity is important
regardless of individual characteristics such as language, language skill, personality, maturity of developmental stage, cultural background, age, gender and so
on. Therefore, design of the multi-language discussion support system in this
research regards factor that disturbs equal discussion as Error on Control of
Error. In detail, Error is selfish behaviors such as sudden and excessive consecutive posts or meaningless text posts and behaviors such as excessive utterances
that makes dominant person in team and deviation of utterances opportunity
22
among team.
From analysis of the first experiment, we clarified that in a group of children who are immature developmental stage, strain and anxiety make children
extremely inactive and cause behaviors like a mischief disturbing purpose of
community. Though, these behaviors cause a little obstacle for a group at first,
these behaviors may affect other children and cause confusion of whole group.
Even though facilitator supports a group, if he/she cannot find and correct these
behaviors immediately, big confusion might be occurred. Therefore, we expect
children’s Control of Error to spontaneously correct their behaviors disturbing
discussion. Eliminating factors disturbing discussion is effective not only for
educational purpose but also for increase in quality of discussion.
Hence, this research introduces Control of Error as design background and
improves problems described in Chapter 3.
4.2
Implementation
Based on design background explained at section 4.1, we improve the multilanguage discussion support system to solve problems extracted from analysis
of the first experiment. Figure 8 shows the screenshot of the improved multilanguage discussion support system.
As the first improvement, we visualize utterances status of team members
except team leader. In general, providing feedback of group status helps participants to modify their behaviors and increase participants’ satisfaction and
group performance[20]. Thus, in this case, the multi-language discussion support system aims to correct individual user’s behaviors and increase quality
of team discussion to real-time display ratio of posts of each participant. We
define “ratio of posts” as follow: ratio of posts is percentage of the number
of each user’s posts in team’s total posts excepted team leader’s over the last
three hours. We define value of ratio of posts as five levels according to table 4,
and implement the user interface that displays this value as a bar graph in real
time. From now, we call this bar graph as “ratio of posts bar”. Ratio of posts
bar’s color and length will change depend on value of ratio of posts. If ratio of
posts is relatively high in team, bar’s color will change to orange and red from
23
Figure 8: The improved multi-language discussion support system(This screenshot is provided by NPO Pangaea)
default color, blue. This function let participants notice utterances status of
each member. We expect all user to keep their bar’s color within blue.
Table 4: Value of ratio of posts and bar’s color
Value of ratio of posts
Ratio of posts bar’s color
Level 1
Less than 4%
Red
Level 2
4% or more and less than 7%
Orange
Level 3
7% or more and less than 27%
Blue
Level 4
27% or more and less than 42%
Orange
Level 5
42% or more
Red
Visualization of members’ utterances status by ratio of posts bar expected to
improve problem 1 in Table 3. If some participants suddenly posted meaningless
posts repeatedly, their ratio of posts will increase and ratio of posts bar become
24
longer and red(or orange). In such situation, participants are expected to avoid
more post and to behave to return color to blue due to Control of Error.
Furthermore, visualization of ratio of posts is also expected to affect to
problem 2 and problem 3. Because team leader can understand ratio of posts
of all members at once, they can aware of quiet children. That leads other
members to notice about quiet person, and promoting to post makes discussion
more equal. Team leaders are also able to deal with accident like computer
troubles immediately.
Second improvement is modifying post display area. In the first experiment,
the multi-language chat system has simple vertical timeline but we modified the
user interface to giving all users their own display area. In Figure 8, balloon by
the side of box showing user’s face photo displays only each user’s own posts.
This modification prevents from occupation of the post display area by specific
user and reduces effect of problem 1. Moreover, user can see all members’ posts
at once. It is prevent discussion from be dominated by particular users, and
team leader becomes being able to aware of silent children’s opinions. We expect
participants’ interest in other’s opinions will be increased by this improvement
because they have more chance to see various opinions.
Third improvement is implementation of mention function and reply function. When user post text with clicking specific user’s face photo, this post
indicates referring clicked user(we call this function “mention function”). And
when user post text with clicking specific text, this post indicates replying
clicked text(we call this function “reply function”. Text with mention function
is displayed with the mentioned user’s face photo at end of the text, and text
with reply function have replied message at end of text.
We introduce user interface that emphasize the topic of discussion against
problem 5. In the first experiment, the multi-language discussion support system had limited area to display text and this area shared with all members, so
posted texts were turned out immediately. That is reason why users often lost
current topic. This is problem that leads to breaking down utterances pairs, so
we implement the user interface having area to show current topic to emphasize
it. In the first experiment, most of texts, which were regarded as topic, are
25
posted by team leaders, thus only team leaders can post topic message on the
improved system.
Additionally, because solving problem 6 by improving user interface is difficult, field’s organizer asks participants for inputting suitable texts for machine translation. Moreover, we registered some words such as proper noun to
dictionary and used Language Grid’s dictionary combine service to get good
translation result as much as possible.
26
Chapter 5
The second experiment
To evaluate how effect the system improvement to children’s multi-language
communication, we conducted the second field experiment. This chapter shows
the result, analysis and consideration of the second experiment.
5.1
Outline of the second experiment
Same as the first experiment, we regard KISSY 2015’s workshop activity by
NPO Pangaea as the second experiment. KISSY 2015 was held in July 31,
2015 to August 7, 2015 at Kyoto University Cock Tower Centennial Hall and
Shiran-Kaikan in Japan. 28 participants: 12 Japanese, 12 Korean, 1 Kenyan,
2 Cambodian and 1 American took in part of KISSY 2015 and about 15 adult
staffs supported children. Participants were given team creativity task. KISSY
2015’s workshop had almost the same timetable as KISSY 2014’s timetable, thus
Participants of the second experiment had almost same time to use discussion
as the first experiment.
In the workshop activity of KISSY 2015, same as the first experiment, 28
participants were divided into 4 teams: Red Team, Blue Team, Yellow Team and
Green Team, and 4 team have equal languages distribution as much as possible.
Each team had 1 adult staff as team leader to support children and to facilitate
team activity. In the second experiment, all team leaders were Japanese and
English bilinguals and their mother language were Japanese. However, only
Blue Team was assigned 2 team leaders and TeamLeader-6 were in charge of
first day and second day and TeamLeader-7 were third day. Table 5 shows
language configuration in each team in the second experiment.
In the second experiment, participants used the Web system that improved
KISSY system has used in KISSY 2014. Particularly, to discussion with team
members, they used the multi-language discussion support system introduced
at Chapter 4.
27
Table 5: Language configuration in each team in the second experiment
5.2
English
Japanese
Korean
Khmer
Total
Red Team
1
3
3
0
7
Blue Team
0
3
3
1
7
Yellow Team
0
3
3
1
7
Green Team
1
3
3
0
7
Result and analysis
In the second experiment, 3871 texts are posted in total1) . Each team posted the
following number of posts: Red Team 585, Blue Team 737, Yellow Team 1356
and Green Team 1193. The improved multi-language discussion support system
aims to realize equal discussion and has the user interface prevent children from
posting too much but actually, the second experiment had more posts than the
first experiment. In the first experiment, 1583 text were posted. Because the
second experiment has more participants than the first experiment, we compare
the number of post of per participant. Result is that the first experiment
is 68.3 and the second experiment is 138.3. This result means improvement
of multi-language discussion support system does not disturb participation to
team discussion.
5.2.1
Effect of visualizing ratio of posts
We analyze effect of visualizing participants’ ratio of posts. At first day of the
second experiment, the multi-language discussion system hid the ratio of posts
bar and showed at second day.
Analysis about transition between the first day and second day
First of all, we compare the first day, which hid ratio of posts bar, and the
second day, which showed bar. The most time of third day was assigned to
creativity task and some user hardly used the multi-language discussion support
system, so we ignore the third day on this analysis. We found characteristic
tendency showing Figure 9 and Figure 10. As Figure 9 shows, Korean-9 posted
1)
Excepting team leader’s posts
28
Firstday
Secondday
Thenumberofposts
300
250
200
150
100
50
0
Korean-8
English-2
Korean-9
Korean-10 Japanese-14Japanese-15Japanese-16
Username
Figure 9: The number of posts among Red Team in each day
Firstday
Secondday
350
Thenumberofposts
300
250
200
150
100
50
0
Korean-17
English-3 Japanese-23 Korean-18 Korean-19 Japanese-24Japanese-25
Username
Figure 10: The number of posts among Green Team in each day
extremely many times at first day but at second day, he reduced his number of
posts close to other members. Same tendency could be seen for Korean-18 in
Figure 10.
First, we focus on Korean-9 in Red Team. Transitions of ratio of posts in
Red Team at first day and second day are shown by Figure 11(The first day is
Figure 11a and the second day is Figure 11b.).
29
Korean-8
English-2
Korean-9
Japanese-14
Japanese-15
Japanese-16
Korean-10
100
90
80
70
60
50
40
30
20
0
1
14
27
40
53
66
79
92
105
118
131
144
157
170
183
196
209
222
235
248
261
274
287
300
313
326
339
352
365
378
391
10
(a) First day
Korean-8
English-2
Korean-9
Japanese-14
Japanese-15
Japanese-16
Korean-10
100
90
80
70
60
50
40
30
20
0
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
101
106
111
116
121
126
131
136
141
146
10
(b) Second day
Figure 11: Transition of ratio of posts in Red Team
As Figure 11a shows, Koran-9 kept posting with high frequency from the
beginning of workshop and he kept his ratio of posts in 60% to 70% whole
day. Korean-9’s ratio of posts was level 5 all time of first day. Tendency of
his message’s contents was that he posted a lot of opinions to team leader’s
30
question. This attitude seems to be active for team discussion. However, as
result that more than half of his posts were pictogram suggests. he suddenly
posted a lot of pictograms, meaningless texts like “Carry kyakyakya kyakyakya
kyakyakya kyakyakya(original language is Korean)” and same word such as
“Rice(original language is Korean)” and “Hello(original language is Korean)”.
These posts were not related with topic of discussion and team activity. In
the other hand, other members’ ratio of posts were relatively low, especially,
English-2 and Korean-10 kept their ratio of posts value around 2%.
Figure 11b shows transition of ratio of posts in Red Team at the second
day, which showed ratio of posts bar. Same as first day, Korean-9 uttered very
aggressively from the beginning of second day but his ratio of posts gradually
became low and reduced to nearly other user’s ratio of posts. In terms of
contents of messages, sudden posts of many pictograms had disappeared and
meaningless posts had reduced. Moreover, English-2, Korean-8 and Korean10 were low ratio of posts at the beginning but from around noon, their ratio
of posts were improved and kept their ratio of posts’ level 3 in the other word
blue. Furthermore, we got estimation from Red Team’s team leader that he and
their team members encouraged quiet Kenyan participant(English-2) to post his
opinion. Actually, English-2’s ratio of posts gradually increased in second day.
Figure 12 shows Green Team’s transition of ratio of posts.
Tendency of first day is gradual increasing of Korean-18. Same as Korean-9,
Korean-18 behaved posting same text repeatedly, and end of the discussion,
suddenly posted many pictograms and his ratio of posts increased. No one was
very inactive but Japanese-23 and Japanese-24 posted a little, so their ratio of
posts were around 5% in a whole discussion.
On the other hand, all members kept their ratio of posts level 3(cf. Table
4) in second day. Korean-18 who had high ratio of posts in first day gradually
reduced his ratio of posts and he kept his ratio 10% to 20%. Only Japanese-25
had relatively low ratio of posts at the beginning but after noon he kept blue
color of his ratio of posts bar.
Participants who posted relatively large number of messages at first day had
tendency that they suddenly posted meaningless texts and pictograms repeat31
Korean-17
English-3
Japanese-23
Korean-19
Japanese-24
Japanese-25
Korean-18
100
90
80
70
60
50
40
30
20
0
1
24
47
70
93
116
139
162
185
208
231
254
277
300
323
346
369
392
415
438
461
484
507
530
553
576
599
622
645
668
10
(a) First day
Korean-17
English-3
Japanese-23
Korean-19
Japanese-24
Japanese-25
Korean-18
100
90
80
70
60
50
40
30
20
0
1
19
37
55
73
91
109
127
145
163
181
199
217
235
253
271
289
307
325
343
361
379
397
415
433
451
469
487
505
523
541
10
(b) Second day
Figure 12: Transition of ratio of posts in Green Team
edly. This tendency is same as Problem 1 explained in Chapter 3. At second
day, after displaying ratio of posts bar, these participants reduced such meaningless posts and their ratio of posts became close to other members. However,
ratio of posts bar did not eliminate children’s selfish behaviors such as post-
32
ing meaningless texts completely, thus we found some children became selfish.
Nevertheless, even if in such case, these users’ ratio of posts kept with in level
3.
Difference between the largest contributor and the smallest contributor
Next, we compare the second experiment with the first experiment. This research defines participant who posted the largest number of posts as the largest
contributor and participant who posted the smallest number of posts as the
smallest contributor. We examined difference between the largest contributor
and the smallest contributor. We focus on the second day of experiment because the second day was assigned as main time to discuss about their creativity
task on both of experiments. Table 6a shows the largest contributor and the
smallest contributor’s the number of posts and ratio of posts in the first experiment. Table 6b shows the second experiment’s value. Moreover, Figure
13 shows difference of ratio of posts between the largest contributor and the
smallest contributor.
In the second experiment, each team has 7 children but in the first experiment, each team has 5 or 6 children. This time, we regard 7 children as criterion
and correct ratio of posts value in the first experiment. In the case of 6 children team, correction value is multiplied six seventh. Similarly in the case of 5
children team, correction value is multiplied five seventh.
In the team consist of 7 members, if all members ratio of posts are equal,
their ratio of posts are 14.29%. Thus, when both of the largest contributor and
the smallest contributor’s ratio of posts are close to 14.29% and difference of
both ratio become smaller, discussion will be more equal. In the first experiment, mean of the largest contributor’s corrected ratio of posts is 30.78%(SD:
8.88) and in the second experiment is 26.15(SD: 4.93). The largest contributor’s
ratio on the second experiment is lower than the first experiment. Moreover, in
terms of the smallest contributor, the first experiment’s mean is 4.11(SD: 2.93)
and the second experiment’s mean is 6.98(SD: 0.31).
As a result, the mean difference of ratio of posts between the largest contributor and the smallest contributor decrease 7.50%.
33
Table 6: The number of posts and ratio of posts at second day of the first
experiment and the second experiment
(a) The number of posts and ratio of posts at second day of the first experiment
Red Team
Blue Team
Yellow Team
Green Team
The total number of
posts
252
8
223
131
The number of posts of
the largest contributor
84
4
48
61
Ratio of posts of the
largest contributor
33.3%
50.0%
21.5%
46.6%
Ratio of posts of the
largest contributor (correct value)
28.6%
42.9%
18.4%
33.3%
The number of posts of
the smallest contributor
15
0
21
6
Ratio of posts of the
smallest contributor
6.0%
0.0%
9.4%
4.6%
Ratio of posts of the
smallest contributor (correct value)
5.1%
0.0%
8.1%
3.3%
(b) The number of posts and ratio of posts at second day of the second experiment
Red Team
Blue Team
Yellow Team
Green Team
The total number of
posts
117
375
442
476
The number of posts of
the largest contributor
32
118
122
86
Ratio of posts of the
largest contributor
27.4%
31.5%
27.6%
18.1%
The number of posts of
the smallest contributor
8
25
33
33
Ratio of posts of the
smallest contributor
6.8%
6.7%
7.5%
6.9%
Report about qualitative analysis
We report result of qualitative analysis about how visualization of ratio of
34
45
Ra5oofposts(%)
40
35
30
25
20
15
10
5
0
2014 2014 2014 2014
Red Blue Yellow Green
2015 2015 2015 2015
Red Blue Yellow Green
Figure 13: Difference of ratio of posts between the largest contributor and the
smallest contributor
posts affect to children’s utterances frequency to support our assertion that visualization and feedback improve discussion quality. Then, we report children
or team leader’s behavior refers to ratio of posts bar from participant observation, movie data took by video camera and message contents sent on the
multi-language discussion support system. In the second experiment, we were
allowed to take a video only to Yellow Team.
At 11.20am, second day of the second experiment, after displaying ratio of
posts bar, we observed Japanese-20 asked team leader “赤とかオレンジってな
に？ (What does red or orange mean?)”. Moreover, we could find Japanese-21
said “下がった．(My bar’s length has became smaller)”. Furthermore, 11.37am
on this team, Japanese-20 said “これ以上やると赤になる．(If I post more, my
bar’s color becomes red)” and “青になった．よかった．(I have felt relieved because my bar’s color has became blue.)” and these utterances obviously indicate
that children are interested in length and color of their ratio of posts bar.
In addition, some team leaders who belong to team that we could not take
videos reported that children indicated their interest in keeping their bar’s color
blue.
We also found posts that referred to ratio of posts bar in conversation on
35
the multi-language discussion support system. In the Yellow Team, at second
day, Khmer-4 who is from Cambodia was very inactive and hardly posted her
opinions. Because her ratio of posts bar was red color, Korean-15 sent text
“Let such a person red gauge conversation! !(original language is Korean)” and
“I’ll say reducing others comments(original language is Korean)”. TeamLeader8, who is Yellow Team’s team leader, asked what these posts mean by word
of mouth to Korean-15 and he answered that he promotes participant whose
bar’s color is red(this time, he meant Khmer-4) to post more opinions and he
restrains himself from sending text for a while to hear other’s opinions. These
utterances indicate that children have interest in not only themselves but also
other participants(moreover, in this case, other language participants), and this
attitude seems to increase equality and quality of discussion.
Ratio of posts bar was also referred by team leader. As we explained above,
Korean-15’s posts, which are “Let such a person red gauge conversation! !
(original language is Korean)” and “I’ll say reducing others comments(original
language is Korean)”, were asked these means by team leader, and after asking,
team leader told this posts’ intention to other Japanese members. Additionally, we found the scene from the video that at 11.40am of the second day,
TeamLeader-8 said “Khmer-4 を待った方がいいのかな．(I wonder if I should
wait opinion from Khmer-4)”.
In the other teams, which we could not take a video, some team leaders
reported that some of their team members were interested in keeping their ratio
of posts’ bar to blue color. Besides, we introduce the characteristic example
indicating contribution of ratio of posts bar. TeamLeader-5, Red Team’s team
leader, reported due to implementation of displaying ratio of posts, he was able
to elicit good opinion from inactive Kenyan child(English-2). This report seems
to indicate the occurrence that although, English-2 was very inactive at the
beginning of the workshop, other team members and team leader took care
about him owing to ratio of posts color and encouraged posting. As a result,
they could elicit idea based on Kenyan traditional background1) and this idea
1)
This idea is “fly whisk” because in Kenya, fly whisk is regarded as symbol of peace.
36
was actually adopted to their final artwork.
These qualitative analysis reports indicate visualization of ratio of posts promotes children’s Control of Error and children reduce their own post frequency
voluntarily. However, in particular team, team leader reports that some children
were interested in increasing their bar’s length. This report means visualization
sometimes affects adversely against our intention.
5.2.2
Chain of utterances
Same as the first experiment, we examine how many posts were ignored although these posts required response and analyze weather multi-language discussion support system improved some failure patterns or not. We attach
labels, “Yes/No-Question”, “Answer”, “Wh-Question”, “Opinion(Idea)”, “Offer”, “Acceptance”, “Refusal”, “Repair”, “Pictogram” and “Others” to all posts
sent in the second experiment same as the first experiment. We regard the case
that “Yes/No-Question”, “Wh-Question” and “Offer” get no responses as failure.
As a result, there are total 476 posts labeled “Yes/No-Question”, “WhQuestion” and “Offer”. Detail is follows: Red Team has 75 posts, Blue Team
has 114 posts, Yellow Team has 149 posts and Green Team has 136 posts. And
we found 162 posts(33.2%) get no responses. Detail is follows: Red Team has 24
posts(32.0%), Blue Team has 41 posts(36.0%), Yellow Team has 41 posts(27.5%)
and Green Team has 52 posts(37.7%).
In the first experiment, ratio of failure pattern is 39.4% and in the second experiment, it is 33.2%. Thus multi-language discussion support system improves
only 6%. Still some translation result was not good and children inputted text
with informal grammar and colloquial expression. That is why multi-language
discussion support system could not reduce break down of adjacency pairs drastically. Particularly, we observed many mistypes. Children often use informal
expression, so machine translation had difficulty to translate these texts with
high accuracy. For example, English speakers often use “i” instead of “I” to
means the first person. In this case, a machine translation could not output
available text. For example, the text “i think pigeon is the best in japan but
37
in kenya is both a fly whisk and a white flag” by English-2 translated “i 考察
ハトは日本で最もよいけれども、ケニアで飛行泡立て器および白い旗です” in
Japanese.
Besides, we point out the possibility that topic emphasizing user interface
sometimes causes break down of adjacency pairs. This user interface contributes
children to understand current discussion topic but sometimes this might en✓
✏
courages children to ignore some posts not related discussion topic like greeting.
TeamLeader-5: I’ll put the one which has formed now and the one
which isn’t made in order.[Topic]
⋮
Korean-9:
How do I play 0 won keonsaep is made incidentally,
as an amusement park?[not related with topic]
⋮
English-2:
a flywhisk and a pegion but not completed[related
with topic]
Japanese-14:
When the sun goes around, it’s unready that Sound
rings.[related with topic]
✒
⋮
In this case, Korean-9’s posts are not related with current topic. As a result,
✑
this Korean-9’s post got no responses but some posts sent after this post referred
to current topic. This pattern seems to emphasized topic prevent other posts
to get responses.
Moreover, children could not use reply function intuitively, because of multilanguage discussion support system’s user interface. This caused difference of
frequency of using reply function. The number of posts with reply function by
each team’s were follows: Red Team was 14 posts, Blue Team was 188, Yellow
Team was 245 and Green Team was 2. That is why many posts did not indicate
their destination. In addition, users must watch all members’ displaying posts
area to read all members’ opinions. In some case, it increases difficulty of seeing
all members’ opinions and leads to break down of adjacency pairs.
38
5.2.3
Thread analysis
We regard set of posts related the same topics as thread and analyze thread
structure. Thus, we attach link between related posts on the multi-language
discussion support system and make groups consist of these related posts. In
this research, these groups of related posts are defined as thread. We compare
thread features made on the first experiment and the second experiment to
analyze effect of multi-language discussion support system’s improvement to
multi-language interaction between children.
At fist, we construct the threads from post log data in the first experiment
and the second experiment. All of posts in the first experiment and part of
the second experiment are not attached destination information. Furthermore,
posts with mention function have information that who is referred by this post
but do not have which text is referred. Thus, we connect two posts based on
rule describing below.
At first, we set following rules for posts that have no reference information,
in other word, posts that do not send with reply or mention function. These
types of link is showed by black dotted line.
1. In the case that message contents indicate destination, we attach link between two posts according to following rules. Besides if there are some
candidates of destination post, “second pair part” of adjacency pairs has
higher priority. In addition, newer post has higher priority.
(a) If particular post has a same or related word with a precedent post,
former post is regarded referring to latter post. If post has some words
related with several precedent posts, former post refers to several posts.
(b) If particular post indicates user’s name in its message, this post is
regarded referring to suitable pair that is posted by indicated user. If
post indicates several users’ names, this post is considered referring
some posts.
2. In the case that the message is considered as “second pairs part” but destination is not clear, Following rules are adapted.
(a) If second pairs part contributor is involved to some threads and these
threads have “first pairs part”, and it has no response, second pairs
39
part is regarded linking to this first pairs part.
(b) The other case, post links to the newest first pairs part in whole conversation.
3. Pictograms do not refer to any posts.
4. If a user posts text to complement his precedent post, former post is consider to link latter post.
Next, we set following rules for posts with mention function, which indicate
only referred user’s name. These types of link is showed by red dotted line.
1. If particular post with mention function and its destination user’s post have
same or related word, former post links to latter post.
2. If “second pairs part” indicates specific user as target and this user posted
suitable post as “first pairs part”, former post is regarded referring latter
post.
3. Pictograms with mention function link to the newest post that is sent by
indicated user.
4. The other case, post does not refer to any other posts.
Finally, because posts with reply function have complete destination information, we use this original information and this type of link is showed by black
solid line.
Based on rule above, we construct thread like Figure 14.
TeamLeader-7 What was Concept of this work?
Korean-12
Earth and it is one of us,
Japanese-19
Japanese-17
Friend
Korean-11
Maybe.... It was a man with'd harmony with nature?
It's same as Japanese-19! I'm a friend.
Korean-12 One with a friend? ?
Korean-11
Not even do that
Figure 14: A part of thread
Second, we analyze threads based on the rules, and we compare the first ex40
periment and the second experiment quantitatively. We report the total number
of posts constructed in the first and second experiments and also compare the
number of posts composing one thread. This result shows whether interactions
were increased or not and how characteristic of each thread changed. Table 7
shows this result.
Table 7: Comparison between the first and second experiments
The number
The first ex-
The number of people
The number of posts
composing one thread
composing one thread
of threads
Mean
SD
Mean
SD
171
3.2
1.4
4.7
3.7
330
3.5
1.9
6.0
6.5
periment
The second
experiment
In the first experiment, total 171 threads(Red Team 66, Blue Team 19, Yellow Team 41, Green Team 45) are constructed. On the other hand, in the second
experiment, total 330 threads(Red Team 51, Blue Team 76, Yellow Team 121,
Green Team 82) are constructed. Compared with the first experiment, second
experiment makes about twice threads and this result indicates interactions on
the multi-language discussion support system were increased.
However, between the first experiment and the second experiment, the mean
number of participants composing one thread increase slightly from 3.2 to 3.5.
The mean number of posts also increased a little, 1.3, from 4.7 to 6.0, but as Figure 15 shows, the frequency distribution of the number of posts composing one
thread indicates large threads consist of 10 and over posts highly increased(the
first experiment: 13, the second experiment: 55). Because thread consist of only
two posts increased exceedingly, mean value of the number of posts composing
one thread(the first experiment: 55, the second experiment: 121).
This result indicates improvement of the multi-language discussion support
system made team discussion more active but from only quantitative results, we
41
TheFirstExperiment
TheSecondExperiment
Thenumberoftreads
140
120
100
80
60
40
20
0
2
3
4
5
6
7
8
9
10and
over
Thenumberofpostscomposingonetread
Figure 15: Frequency distribution of the number of posts composing one thread
cannot understand how type of posts composing large threads. Thus, we analyze
contents of messages and threads to examine actual tendency and characteristics
of large threads.
A lot of ideas and opinions make large threads
First, we found ideas and opinions to one Wh-Question post increased in
large threads. Following example is part of conversation posted in Green Team
in the second experiment. In this example, TeamLeader-9 asks for specific ideas
about theme “We Love” and children posted total 10 or more ideas. We found
many such type of threads in the second experiment.
We can also found posting a lot of ideas to one Wh-Question in the first
experiment, but in the first experiment, the average number of ideas posts to
one Wh-Question is 3.4 and in the second experiment, this number increase to
7.3. Moreover, the Wh-Question that has the largest number of ideas got 15
ideas in the first experiment but in the second experiment, the Wh-Question
got the largest number of ideas had 31 ideas.
42
✓
TeamLeader-9: What are the thing this team likes and a thing?
✏
(original language is Japanese)[Wh-Question]
English-3:
i like ice cream!
TeamLeader-9: The family, a friend, animal and our team were
given yesterday, weren’t they?(original language is Japanese)
Koren-18:
Ice cream.(original language is Korean)
Koren-18:
It is my favorite year(original language is
Korean)
Korean-17:
I think that perhaps the family would not be
valuable(original language is Korean)
Koren-18:
Sun(original language is Korean)
Japanese-24:
I’m a friend.(original language is Japanese)
Koren-18:
I love it(original language is Korean)
Korean-17:
I think that loved the sun and the moon(original
language is Korean)
Japanese-25:
✒
A friend is important.(original language is Japa✑
nese)
We consider increase in ideas and opinions is one of the reason for increase
in threads consist of 10 and more posts. Question emphasized as discussion
topic sent by team leaders got especially a lot of posts. The average number of
posts composing thread begin from team leader’s topic post is 14.3.
Reaction to other’s opinions make large threads
We found increase in reaction to other’s specific ideas and opinions and most
of these reference posts were expressing agreement. In the first experiment, the
number of reaction posts simply express only agreement or disagreement is 62.
On the other hand, in the second experiment, these types of posts is 144(in this
case, we except reactions by only a pictogram).
Moreover, other than reactions expressing agreement or disagreement, we
observed detailed discussion about someone’s opinions. For example, posting
question or improvement plan for other opinions makes discussion more detail
43
and threads larger. Following example is part of conversation in Blue Team.
In this case, after Korean-12 posted his idea of creating a globe, some users
✓
✏
pointed out problem of Korean-12’s idea and they discussed about this opinion.
Korean-12:
Let’s make the earth a Styrofoam ball!(original
language is Korean)
Japanese-18:
If a globe goes around, doesn’t a person fall?
(original language is Japanese)
TeamLeader-6: The idea of Korean-12 is wonderful. But when the
earth goes around, animal and man may fall down.
(original language is Japanese)
Korean-12:
Would not you firmly in the bond or glue gun? ?
(original language is Japanese)
Korean-11:
But doneunde quickly...(original language is
Korean)
Korean-12:
Freezing fall Anjo(original language is Korean)
⋮
✒
5.3
✑
Discussion
We implemented the user interface that visualizes children’s ratio of posts and
feedbacks in real time in order to cause children’s Control of Error. We consider this user interface reduced difference of the number of posts among team
between the first experiment and the second experiment, although sample data
is a few and situation of two experiments such as participants and discussion
theme is different. Moreover, between the first day and second day of the second experiment, we observed visualization changed children’s behavior. In the
second experiment, we hid the ratio of posts bar in first day and showed it in
second day. Between these two days, characteristic changes appeared in two
children’s behaviors. In detail, after beginning visualization of ratio of posts,
children who sent too much posts and kept their ratio of posts high level in
first day gradually reduced their post frequency and their ratio of posts became
close to other children. Moreover, we reported change of children’s behaviors
44
from qualitative analysis. Children who are mentioned above by us changed
their tendency of message contents. At first day, they often posted meaningless
text and too many pictograms but at second day, these types of posts were
restrained. In terms of other children, we found children who posted relatively
many times said that he reduces posts and focuses on reading other’s opinions. Furthermore, we also found behaviors encouraging inactive children to
post opinions. These qualitative reports support hypothesis that visualization
and feedback of ratio of posts reduce deviation of discussion and promotes equal
discussion. These result that we described above strongly supports visualization of ratio of posts draws out children’s Control of Error and children become
more cooperative. However, this user interface could not eliminate discussion
gap between team members completely and we still found sometimes meaningless texts and large amount of pictograms were posted, especially end of a day.
We point out the possibility that visualization of ratio of posts was not able to
prevent children from disturbing discussion when children lost interest in team
discussion and creativity task. This consideration indicates that we have to
consider about how to prevent children from losing interest in team discussion
whole day. Moreover, we got report that some part of children competed their
bar length. This behavior seems that children’s interest in changing their bar
length and color affect contrary to design intention.
Besides, improvement of the multi-language discussion support system affects children’s interaction and increases large threads. In the first experiment,
problem that when children are not used to a computer and a lot of texts posted
in a short time, children sometimes lost discussion topic is pointed out. Thus
we implemented a user interface to emphasize topic in order to prevent children
from losing important information such as current topic. As a result we found
more ideas posted to question and particularly, thread begin from emphasized
topic post consist of many posts. This result indicates that emphasized topic
contributes to elicit more ideas.
Furthermore, we consider implementation of individual posts display area
contributes to cause more interaction. In the second experiment, more reaction
especially agreement to other members’ opinions were posted and we can also
45
found children posted question or improvement plan to other’s opinions. Participants could see all members posts at once regardless of amount of utterances
for individual posts display area, hence we consider even inactive children could
get other’s response if their opinions is good. Moreover, this implementation
supports children to know various opinions and it increased children’s interest
in other opinions.
46
Chapter 6
Conclusion
This research points out problems of children’s multi-language communication
from analysis to the children’s multi-language workshop. Based on analysis, this
research designs and Improves the multi-language discussion support system
for children. We introduce the improved system to children’s multi-language
workshop again to evaluate this system. The main contributions of this research
are described as follow:
1. Detecting of problems in children’s multi-language communication
This research conducted field experiment in the children’s multi-language
workshop to detect problems of children’s multi-language communication
environment. The experiment let 8 to 14 year-old children communicate
each other with multi-language discussion support system and collected
conversation data to execute creativity task.
We analyzed communication data and found behaviors that in children’s
multi-language communication environment, children sometimes posted
meaningless texts and too much pictograms. We consider this behaviors
are caused by children’s individual maturity of developmental stage and
regard this type of behaviors as sudden selfish behavior. Selfish behaviors disturb team discussion, in detail, some children’s sudden consecutive
posts occupied post display area of the discussion support system and in
this time, children had a difficulty of keeping up with discussion. Moreover,
we point out a possibility that children’s selfish behaviors produce atmosphere that inactive participants are hard to involve to discussion. That
is why selfish behaviors are regarded as problem in international education
field where is considered equal discussion is important. Moreover, we found
these selfish behaviors affect other children and might cause confusion of
team discussion.
Besides, we also point out that children often use informal expression and
this decreases machine translation’s accuracy. We examined failure pattern
of adjacency pairs, 39.4% of pairs were broken. Children often omit subject
of their sentence and use colloquial expression. This is not suitable to
47
machine translation and is considered as a part of factor of breaking down
of adjacency pairs as well as occupation of posts display area.
We enumerated problems described above and multi-language discussion
support system’s problems and shared these problems to give priority of
problems to solve. Some problems that are revealed by analysis are also
recognized by field, thus we discussed with field’s organizer to detect high
priority problems to solve.
2. Design and implementation of the multi-language discussion support system for children
Based on analysis, this research designs and implements the multi-language
discussion support system for children. We introduced Control of Error
that is important concept in Montessori methods as basic concept. This
concept aims to let children to correct their irrelevant behaviors by themselves. Moreover, improvement of system has intention that increases interaction and reducing deviation of utterances frequency among team.
The improved multi-language discussion support system visualizes children’s ratio of posts among team and feedback in real time. This improvement reduced 7.50% difference of ratio of posts between the largest
contributor and the smallest contributor from the first experiment. Furthermore, we also found tendency that children who posted too much at the
beginning of workshop gradually reduced their post to average value. In addition, we observed posts promote other participants’ opinions or restrain
their own posts like “Let such a person red gauge conversation! ! (original language is Korean)” and “I’ll say reducing others comments(original
language is Korean)”. These altruistic behaviors seem to be caused by
improved system and contribute to increase quality of discussion.
Additionally, user interface of individual posts display area seems to enable
participants to reach all members’ opinions. As a result, ideas and opinions to one question and reaction to other’s posts increased. This result
promotes eliciting various ideas and detailed discussion.
Due to design and implementation of multi-language discussion support system for children, this research contributes result described above. However,
48
this system does not restrain problems such as children’s selfish behaviors completely. The system still has problems for example, some children were interested in increasing their ratio of posts bar’s length and this behavior is contrary
to intention of the system. Moreover, we still found some patterns that cause
breaking down of adjacency pairs, especially bad input text often decreased
machine translation’s accuracy, hence we have to consider the system that promotes children to input text optimized machine translation. In the field based
system design, repetitive examination and improvement are important, thus
future works are repetitive analysis to children’s multi-language workshop and
modeling of the design.
49
Acknowledgments
First of all, I would like to express my sincere gratitude to my supervisor,
Professor Toru Ishida at Kyoto University, for his valuable advice, and the opportunity to conduct this research. I would like to express my appreciations to
the advisers, Associate Professor Michinori Hatayama and Associate Professor
Hiromichi Mitamura at Kyoto University for helpful advice. I would like to express my acknowledgement to NPO Pangaea including President Yumiko Mori
and Vice-President Toshiyuki Takasaki for technical advice and coordinating
experiment in KISSY. Moreover, I would like to express my gratitude to Researcher Masayuki Otani, Researcher Takao Nakaguchi and Researcher Katsuya
Takanashi for valuable advice and helpful discussion. Finally, I would like to
thank all members in Ishida and Matsubara laboratory.
50
References
[1] Nickerson, C.: English as a lingua franca in international business contexts,
English for Specific Purposes, Vol. 24, No. 4, pp. 367 – 380 (2005).
[2] Takano, Y. and Noda, A.: A temporary decline of thinking ability during
foreign language processing, Journal of Cross-Cultural Psychology, Vol. 24,
No. 4, pp. 445–462 (1993).
[3] Aiken, M., Park, M. and Lindblom, T.: Integrating machine translation
with group support systems, International journal of business and management, Vol. 5, No. 5, p. p25 (2010).
[4] Aiken, M. W., Martin, J. S., Paolillo, J. G. and Shirani, A. I.: A group decision support system for multilingual groups, Information & management,
Vol. 26, No. 3, pp. 155–161 (1994).
[5] Yamashita, N. and Ishida, T.: Effects of machine translation on collaborative work, Proceedings of the 2006 20th anniversary conference on Computer supported cooperative work , ACM, pp. 515–524 (2006).
[6] Nguyen, D. T. and Fussell, S. R.: Effect of message content on communication processes in intercultural and same-culture instant messaging conversations, Proceedings of the 2013 conference on Computer supported cooperative work , ACM, pp. 19–32 (2013).
[7] Kim, T., Chang, A., Holland, L. and Pentland, A. S.: Meeting mediator:
enhancing group collaborationusing sociometric feedback, Proceedings of
the 2008 ACM conference on Computer supported cooperative work , ACM,
pp. 457–466 (2008).
[8] Gardner, H.: Developmental Psychology, Addison Wesley Publishing Company, Boston (1998).
[9] Ishida, T.: The Language Grid , Springer (2011).
[10] Schegloff, E. A. and Sacks, H.: Opening up closings, Semiotica, Vol. 8,
No. 4, pp. 289–327 (1973).
[11] Sacks, H., Scheglpff, E. A. and Jefferson, G.: A Simplest Systematics for the
Organization of Turn-Taking for Conversation, Language, Vol. 50, No. 4,
pp. 96–735 (1974).
51
[12] Schegloff, E. A., Jefferson, G. and Sacks, H.: The preference for selfcorrection in the organization of repair in conversation, Language, pp.
361–382 (1977).
[13] Desanctis, G. and Gallupe, R. B.: A foundation for the study of group
decision support systems, Management science, Vol. 33, No. 5, pp. 589–
609 (1987).
[14] 京都大学フィールド情報学研究会: フィールド情報学入門, 共立出版 (2009).
[15] Emerson, R., Fretz, R. and Shaw, L.: Writing Ethnographic Fieldnotes,
University of Chicago Press (2011).
[16] Montessori, M.: The Montessori Method , Standard Pubns Inc (2007).
[17] Burnett, A.: Montessori Education Today and Yesterday, The Elementary
School Journal , Vol. 63, No. 2, pp. 71–77 (1962).
[18] Lillard, P. P.: Montessori today: A comprehensive approach to education
from birth to adulthood , Random House LLC (1996).
[19] Montessori, M.: The absorbent mind. Adyar , India: The Theosophical Publishing House (1949).
[20] Diehl, E. and Sterman, J. D.: Effects of feedback complexity on dynamic
decision making, Organizational Behavior and Human Decision Processes,
Vol. 62, No. 2, pp. 198–215 (1995).
52