Professional Documents
Culture Documents
New Frontiers in Artificial Intelligence Jurisin SKL Al Biz Tuskuba Lenls Aaa Scidoca Knexi Tokyo November 13 15 2017 Revised Papers Sachiyo Arai
New Frontiers in Artificial Intelligence Jurisin SKL Al Biz Tuskuba Lenls Aaa Scidoca Knexi Tokyo November 13 15 2017 Revised Papers Sachiyo Arai
https://textbookfull.com/product/image-and-graphics-9th-
international-conference-icig-2017-shanghai-china-
september-13-15-2017-revised-selected-papers-part-iii-1st-
edition-yao-zhao/
https://textbookfull.com/product/ambient-intelligence-15th-
european-conference-ami-2019-rome-italy-
november-13-15-2019-proceedings-ioannis-chatzigiannakis/
https://textbookfull.com/product/advances-in-artificial-
intelligence-iberamia-2018-16th-ibero-american-conference-on-ai-
trujillo-peru-november-13-16-2018-proceedings-guillermo-r-simari/
Sachiyo Arai · Kazuhiro Kojima
Koji Mineshima · Daisuke Bekki
Ken Satoh · Yuiko Ohta (Eds.)
LNAI 10838
New Frontiers
in Artificial Intelligence
JSAI-isAI Workshops, JURISIN,
SKL, AI-Biz, LENLS, AAA, SCIDOCA, kNeXI
Tsukuba, Tokyo, November 13–15, 2017
Revised Selected Papers
123
Lecture Notes in Artificial Intelligence 10838
New Frontiers
in Artificial Intelligence
JSAI-isAI Workshops, JURISIN,
SKL, AI-Biz, LENLS, AAA, SCIDOCA, kNeXI
Tsukuba, Tokyo, November 13–15, 2017
Revised Selected Papers
123
Editors
Sachiyo Arai Daisuke Bekki
Chiba University Ochanomizu University
Chiba Tokyo
Japan Japan
Kazuhiro Kojima Ken Satoh
National Institute of Advanced Industrial National Institute of Informatics
Science and Technology Tokyo
Ibaraki Japan
Japan
Yuiko Ohta
Koji Mineshima Fujitsu Laboratories Limited
Ochanomizu University Kanagawa
Tokyo Japan
Japan
This Springer imprint is published by the registered company Springer International Publishing AG
part of Springer Nature
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
JSAI (The Japanese Society for Artificial Intelligence) is a premier academic society
that focuses on artificial intelligence in Japan and was established in 1986. JSAI-isAI
(JSAI International Symposium on Artificial Intelligence) 2017 was the ninth inter-
national symposium on AI supported by the JSAI. JSAI-isAI 2017 was successfully
held during November 13–15 at University of Tsukuba in Tokyo, Japan. In all, 203
people from 15 countries participated.
JSAI-isAI 2017 included seven workshops, where 16 invited talks and 91 papers
were presented. This volume, New Frontiers in Artificial Intelligence: JSAI-isAI 2017
Workshops, is the proceedings of JSAI-isAI 2017. From the seven workshops (JUR-
ISIN 2017, SKL 2017, AI-Biz 2017, LENLS 14, AAA 2017, SCIDOCA 17, and
kNeXI 2017) 30 papers were carefully selected and revised according to the comments
of the workshop Program Committees. The acceptance rate was about 32%. This
resulted in the excellent selection of papers that are representative of some of the topics
of AI research both in Japan and in other parts of the world.
JURISIN 2017 was the 11th international workshop on Juris-informatics.
Juris-informatics is a new research area that studies legal issues from the perspective of
informatics. The purpose of this workshop was to discuss both fundamental and
practical issues among people from various backgrounds such as law, social science,
information and intelligent technology, logic and philosophy, including the conven-
tional “AI and law” area.
SKL 2017 (the 4th International Workshop on Skill Science) aimed to interna-
tionalize research on skill sciences by organizing the meeting. Human skills involve
well-attuned perception and fine motor control, often accompanied by thoughtful
planning. The involvement of body, environment, and tools mediating them makes the
study of skills unique among research on human intelligence.
AI-Biz 2017 (Artificial Intelligence of and for Business) was the second workshop
held to foster the concepts and techniques of business intelligence (BI) in artificial
intelligence. BI should include such cutting-edge techniques as data science,
agent-based modeling, complex adaptive systems, and IoT. The main purpose of this
workshop is to provide a forum for participants to discuss important research questions
and practical challenges in business intelligence, business informatics, data analysis
and agent-based modeling, to exchange the latest results, and to join efforts in solving
common challenges.
LENLS 14 was the 14th event in the series, and it focused on the formal and
theoretical aspects of natural language. LENLS (Logic and Engineering of Natural
Language Semantics) is an annual international workshop recognized internationally in
the formal syntax-semantics-pragmatics community. It has been bringing together for
discussion and interdisciplinary communication researchers working on formal theories
of natural language syntax, semantics and pragmatics, (formal) philosophy, artificial
intelligence, and computational linguistics.
VI Preface
AAA 2017 (the Third International Workshop on Argument for Agreement and
Assurance) focused on argumentation, which has now become an interdisciplinary
research subject receiving much attention from diverse communities including formal
logic, informal logic, and artificial intelligence. It aims at analyzing, evaluating, and
systematizing various aspects of human argument appearing in the media — television,
newspapers, WWW, etc. — and also artificial arguments constructed from structured
knowledge with logical language and inference rules. Their research achievements are
widely applicable to various domains such as safety, political, medical, and legal
domains.
SCIDOCA 17 (the Second International Workshop on Scientific Document Anal-
ysis) focused on the recent proliferation of scientific papers and technical documents
that has become an obstacle to efficient acquisition of new information in various fields.
It is almost impossible for individual researchers to check and read all related docu-
ments. Even retrieving relevant documents is becoming more difficult. This workshop
gathered together all researchers and experts who work on scientific document analysis
from various perspectives, and invited technical paper presentations and system
demonstrations that cover any aspects of scientific document analysis.
HAT-MASH 2016 (Healthy Aging Tech Mashup Service, Data, and People) was
the second international workshop in the series that bridges healthy aging and elderly
care technology, information technology, and service engineering. The main objective
of this workshop was to provide a forum fo participants to discuss important research
questions and practical challenges in healthy aging and elderly care support and to
promote transdisciplinary approaches.
kNeXI 2017 (Knowledge Explication in Industry) was aimed at promoting research
about improving the skill and knowledge of employees in industry. kNeXI focuses on
industry in which the knowledge of employees is important to their daily work. Such
industry includes elderly care, education, health-promotion services, etc.
It is our great pleasure to be able to share some highlights of these fascinating
workshops in this volume. We hope this book will introduce readers to the
state-of-the-art research outcomes of JSAI-isAI 2017, and motivate them to participate
in future JSAI-isAI events.
JURISIN 2017
Workshop Chair
Ken Satoh National Institute of Informatics, Japan
Steering Committee
Takehiko Kasahara Toin Yokohama University, Japan
Makoto Nakamura Nagoya University, Japan
Katsumi Nitta Tokyo Institute of Technology, Japan
Seiichiro Sakurai Meiji Gakuin University, Japan
Ken Satoh National Institute of Informatics, Japan
Satoshi Tojo Advanced Institute of Science and Technology, Japan
Katsuhiko Toyama Nagoya University, Japan
Advisory Committee
Trevor Bench-Capon The University of Liverpool, UK
Tomas Gordon Fraunfoher FOKUS, Germany
Henry Prakken University of Utrecht and Groningen, The Netherlands
John Zeleznikow Victoria University, Australia
Robert Kowalski Imperial College London, UK
Kevin Ashley University of Pittsburgh, USA
Program Committee
Thomas Ågotnes University of Bergen, Norway
Ryuta Arisaka National Institute of Informatics, Japan
Kristijonas Cyras Imperial College, UK
Marina De Vos University of Bath, UK
Phan Minh Dung AIT, Thailand
Randy Goebel University of Alberta, Canada
Guido Governatori NICTA, Australia
Tatsuhiko Inatani Kyoto University, Japan
Tokuyasu Kakuta Chuo University, Japan
Yoshinobu Kano Shizuoka University, Japan
Mi-Young Kim University of Alberta, Canada
Nguyen Le Minh Japan Advanced Institute of Science and Technology,
Japan
Beishui Liao Zhejiang University, China
Hatsuru Morita Tohoku University, Japan
Makoto Nakamura Nagoya University, Japan
VIII Organization
SKL 2017
Workshop Chair
Tsutomu Fujinami Japan Advanced Institute of Science and Technology
Program Committee
Masaki Suwa Keio University, Japan
Ken Hashizume Osaka University, Japan
Mihoko Otake RIKEN, Japan
Yoshifusa Matsuura Yokohama National University, Japan
Yuta Ogai Tokyo Polytechnic University, Japan
Kentaro Kodama Kanagawa University, Japan
AI-Biz 2017
Workshop Chair
Takao Terano Tokyo Institute of Technology, Japan
Workshop Co-chairs
Hiroshi Takahashi Keio University, Japan
Setsuya Kurahashi University of Tsukuba, Japan
Organization IX
Program Committee
Takao Terano Tokyo Institute of Technology, Japan
Hiroshi Takahashi Keio University, Japan
Setsuya Kurahashi University of Tsukuba, Japan
Hiroshi Deguchi Tokyo Institute of Technology, Japan
Reiko Hishiyama Waseda University, Japan
Manabu Ichikawa National Institute of Public Health, Japan
Yoko Ishino Yamaguchi University, Japan
Hajime Kita Kyoto University, Japan
Hajime Mizuyama Aoyama Gakuin University, Japan
Masakazu Takahashi Yamaguchi University, Japan
Shingo Takahashi Waseda University, Japan
Takashi Yamada Yamaguchi University, Japan
LENLS 14
Workshop Chair
Katsuhiko Sano Hokkaido University, Japan
Workshop Co-chairs
Daisuke Bekki Ochanomizu University/JST CREST, Japan
Koji Mineshima Ochanomizu University/JST CREST, Japan
Eric McCready Aoyama Gakuin University, Japan
Program Committee
Katsuhiko Sano Hokkaido University, Japan
Daisuke Bekki Ochanomizu University/JST CREST, Japan
Koji Mineshima Ochanomizu University/JST CREST, Japan
Eric McCready Aoyama Gakuin University, Japan
Alastair Butler Faculty of Humanities, Hirosaki University, Japan
Richard Dietz iCLA, Yamanashi Gakuin University, Japan
Naoya Fujikawa Tokyo Metropolitan University, Japan
Yoshiki Mori University of Tokyo, Japan
Yasuo Nakayama Osaka University, Japan
David Y. Oshima Nagoya University, Japan
Osamu Sawada Mie University, Japan
Wataru Uegaki Leiden University, The Netherlands
Katsuhiko Yabushita Naruto University of Education, Japan
Tomoyuki Yamada Hokkaido University, Japan
Shunsuke Yatabe Kyoto University, Japan
Kei Yoshimoto Tohoku University, Japan
X Organization
AAA 2017
Workshop Chair
Kazuko Takahashi Kwansei Gakuin University, Japan
Workshop Co-chairs
Yoshiki Kinoshita Kanagawa University, Japan
Tim Kelly University of York, UK
Hiroyuki Kido Sun Yat-sen University China
Program Committee
Martin Caminada Cardiff University, UK
Ewen Denney SGT/NASA Ames Research Center, USA
Juergen Dix Clausthal University of Technology, Germany
Phan Minh Dung Asian Institute of Technology, Thailand
Richard Hawkins University of York, UK
C. Michael Holloway NASA Langley Research Center, USA
Antonis Kakas University of Cyprus, Cyprus
Tim Kelly University of York, UK
Hiroyuki Kido Sun Yat-sen University, China
Yoshiki Kinoshita Kanagawa University, Japan
Yutaka Matsuno Nihon University, Japan
Nir Oren The University of Aberdeen, UK
John Rushby SRI International, USA
Chiaki Sakama Wakayama University, Japan
Ken Satoh National Institute of Informatics and Sokendai, Japan
Guillermo Ricardo Simari Universidad del Sur in Bahia Blanca, Argentina
Kenji Taguchi National Institute of Advanced Industrial Science
and Technology, Japan
Kazuko Takahashi Kwansei Gakuin University, Japan
Toshinori Takai Nara Institute of Science and Technology, Japan
Makoto Takeyama Kanagawa University, Japan
Paolo Torroni University of Bologna, Italy
Charles Weinstock Software Engineering Institute, USA
Stefan Woltran Vienna University of Technology, Austria
Shuichiro Yamamoto Nagoya University, Japan
SCIDOCA 2017
Workshop Chairs
Yuji Matsumoto Nara Institute of Science and Technology, Japan
Hiroshi Noji Nara Institute of Science and Technology, Japan
Organization XI
Program Committee
Takeshi Abekawa National Institute of Informatics, Japan
Akiko Aizawa National Institute of Informatics, Japan
Naoya Inoue Tohoku University, Japan
Kentaro Inui Tohoku University, Japan
Yoshinobu Kano Shizuoka University, Japan
Yusuke Miyao National Institute of Informatics, Japan
Junichiro Mori University of Tokyo, Japan
Hidetsugu Nanba Hiroshima City University, Japan
Shoshin Nomura National Institute of Informatics, Japan
Ken Satoh National Institute of Informatics, Japan
Hiroyuki Shindo Nara Institute of Science and Technology, Japan
Yoshimasa Tsuruoka University of Tokyo, Japan
Minh Le Nguyen Japan Advanced Institute of Science and Technology,
Japan
Pontus Stenetorp University College London, UK
kNeXI 2017
Workshop Chair
Satoshi Nishimura National Institute of Advanced Industrial Science
and Technology, Japan
Workshop Co-chairs
Takuichi Nishimura National Institute of Advanced Industrial Science
and Technology, Japan
Ken Fukuda National Institute of Advanced Industrial Science
and Technology, Japan
Program Committee
Satoshi Nishimura National Institute of Advanced Industrial Science
and Technology, Japan
Takuichi Nishimura National Institute of Advanced Industrial Science
and Technology, Japan
Ken Fukuda National Institute of Advanced Industrial Science
and Technology, Japan
Kentaro Watanabe National Institute of Advanced Industrial Science
and Technology, Japan
Yasuyuki Yoshida National Institute of Advanced Industrial Science
and Technology, Japan
Nami Iino National Institute of Advanced Industrial Science
and Technology, Japan
XII Organization
Sponsored By
JURISIN2017
SKL2017
AI-Biz2017
LENLS 14
AAA 2017
SCIDOCA 17
kNeXI2017
Ken Satoh
Masaharu Yoshioka(B)
1 Introduction
operation in 1890 and are revised every year. In contrast, the English version
of the laws is translated from a specific version of Japanese civil laws. Even
though the English version is also updated every year, it has a more uniform
style description than the Japanese version. However, such issues concerning
language have not been thoroughly investigated.
In this paper, to investigate the characteristics of the COLIEE IR task and
the state-of-the-art of proposed IR systems, we analyze all submitted runs for
the COLIEE IR task by using information of the baseline IR system Indri. Indri
is a state-of-the-art IR system based on language modeling. Since Indri employs
phrase matching that is widely used in the submitted system as word n-grams
(sequence of words), the performance of the system is better or almost equivalent
to the baseline system using TF-IDF for English data reported in the overview
papers (baseline for COLIEE 2016 [1] and UA-TFIDF for COLIEE 2017 [2]).
2 COLIEE IR Task
2.1 Task Description
The COLIEE IR task retrieves a static set of relevant civil code articles for
answering the question given as a query. Relevant articles are ones that are
necessary to check the appropriateness of the given question statement. There are
questions that require only one article to check the appropriateness (one relevant
article) and ones that require multiple articles (multiple relevant articles).
The participant systems are requested to return relevant article sets for the
question test data. The performance of the system is evaluated by F-measure
(mean average of precision and recall). In recent years, four competitions have
been conducted. Table 1 gives a summary of the information concerning the
number of questions (number of questions with multiple relevant articles) and
the number of relevant articles in total. (In each competition, a set of questions
from the bar exam of the same year was used: H22 (2010), H24 (2012), H26
(2014), and H28 (2016) for COLIEE 2014, 2015, 2016, and 2017, respectively.1 )
Most of the questions have only one relevant article and less than one-third of
questions have multiple (two or more) relevant articles.
Year Number of questions (with multiple relevant articles) Number of relevant articles
H22 47 (3) 51
H24 79 (12) 102
H26 95 (29) 131
H28 78 (20) 110
1
One COLIEE competition uses one year bar exam data (e.g., H22 for COLIEE 2014)
for IR task and another one year for entailment (e.g. H23(2011) for COLIEE 2014).
Analysis of COLIEE Information Retrieval Task Data 7
Run is a retrieval results (list of candidate relevant articles for each question)
submitted by a participant. Each participating team can submit multiple runs
as candidates for evaluation.2
Table 2 presents information about the runs. Except for COLIEE 2017 [2],
there is no clear description about the language used for generating runs. How-
ever, after checking the papers, we found that most of the submissions used
English data except for Kanolab (H24), HUKB (H26 and H28), and KIS (H28).
In this analysis, since the characteristics of the runs from the same team are
similar, we use the best performance runs for each team for further analysis.
2.3 Indri
legal terms) is used to normalize and segment a query text into a sequence of
normalized Japanese words and no stop-words are used. To construct a retrieval
result, a top-ranked retrieved document for each query is returned as a relevant
article (this approach is the most commonly used among all submitted data).
Table 3 shows the system performance of Indri for the test data. From this
table, it is clear that the difficulty of Japanese and English test data differs year
by year.
First, we compare the rank of relevant articles of the English and Japanese
test data (Table 4). From this table, we can see that the correlations between
the ranks of English and Japanese test data are similar, although there are a
few cases where the results are completely different. H24-27-O6 is a case that
Japanese data is easier than English one (rank 18 for Japanese and rank 495
for English). On the contrary, H26-27-3 is a case that English one is easier
than Japanese one (article 650: rank 9 for English and rank 678 for Japanese;
article 701: rank 300 for English and 98 for Japanese; article 702 rank 561 for
English and 114 for Japanese). Underlines highlight the corresponding parts
among question and article pairs.
– H24-7-O: In cases where (A) and (B) is a married couple, if A engages in juris-
tic act with (C) in the single name of oneself regarding everyday household
matters, the extinctive prescription of C’s credit against B shall be nullified
upon the judicial claim of C’s credit against B.
Answer: 434
(Request for Performance to One Joint and Several Obligor)
Article 434 A request for performance made to one joint and several
obligor shall also be effective with respect to other joint and several
obligor(s).
6
Capital letter such as “A”, “B”, and “C” are used for anonymize original name in
Japanese judicial precedent and also be used in the exam.
10 M. Yoshioka
cupboard and broke it. In this case, if A was free from any negligence, A
may claim compensation for the loss from the owner of the dog.7
Answer: 650,701,702
(Mandatary’s Claims for Reimbursement of Expense)
Article 650 (1) If the mandatary has incurred costs found to be neces-
sary for the administration of the mandated business, the mandatary
may claim reimbursement of those costs from the mandator and any
interest on the same from the day the costs were incurred.
(2) If the mandatary has incurred any obligation found to be neces-
sary for the administration of the mandated business, the mandatary
may demand that the mandator perform the obligation on the man-
datary’s behalf. In such cases, if the obligation has not yet fallen due,
the mandatary may require the mandator to tender reasonable secu-
rity.
(3) If the mandatary suffers any loss due to the administration of
the mandated business without negligence in the mandatary, he/she
may claim compensation for the loss from the mandator.
For the first case, “ ” are translated into “obligor” in the article and
“ ”(obligation) in the question as “credit”. However, Japanese system can
use “ ”(obligation) as a part of “ ”(joint obligor) that are split as
“ ”(joint), “ ”(obligation), “ ”(person; suffix) using Japanese Morpho-
logical analyzers.
7
A part of Japanese questions include description about the explanation of their
question types and this part is not translated into English one. For example, Japanese
question has “
” (There are five descriptions about the legal decision
(a) to (o). Please select a combination of correct description(s) from 1 to 5
below.), but no corresponding description in English one.
Analysis of COLIEE Information Retrieval Task Data 11
For the latter case, the style of writing for article 650 (3) and questions
are very similar (share long phrases) in English and not for Japanese. Another
problem is related to description about the explanation of their question types.
When we use the question that removes this explanation: “
J E
1 2 3 4 5 -10 -30 -50 -100 101- Sum
1 145 10 6 1 1 3 0 0 0 0 166
2 18 13 6 2 0 2 3 0 0 0 44
3 7 6 4 2 0 3 2 0 0 0 24
4 2 3 3 2 4 1 2 0 1 0 18
5 1 0 0 3 3 2 0 0 0 0 9
-10 4 3 3 3 2 5 4 0 0 1 25
-30 5 0 1 0 6 5 16 2 2 3 40
-50 0 1 0 0 0 1 11 3 3 1 20
-100 0 0 0 0 0 1 0 4 7 6 18
101- 0 0 0 0 0 1 0 2 4 23 39
Sum 182 36 23 13 16 24 38 11 17 34 394
Table 4 shows the comparative analysis results between the system based on
English test data and Japanese one. Columns of the table represents rank for the
English test data of the relevant articles and lows represents ones for Japanese
test data. Most of the top one ranked questions for the English are also ranked
as top for the Japanese, but there are several exceptional cases.
Difficult Japanese ones.
Based on the results of Tables 3 and 4, it is therefore necessary to analyze
the runs for English and Japanese data separately.
The numbers of teams that used English data for each run (H22(2010),
H24(2012), H26(2014), and H28(2016)) were 1, 4, 6, and 5, respectively. For
Japanese data, the numbers were 0, 1, 1, 2, respectively. Since the numbers of
Japanese runs are small, we will first focus on the English runs.
12 M. Yoshioka
Table 5 shows the cross table of Indri rank (English) for all relevant articles
and the number of team best performance runs that find those articles. First and
second row of Table 5 represent test data used in the competition and number of
team best performance runs that find those articles (maximum number of teams
are different based on the number of participant teams for each test data 1, 4,
6, 5 for H22, H24, H26, H28) respectively. When there is no team that can be
retrieve relevant articles, it was categorized as 0 (no team). From this table, we
confirmed questions with Indri rank 1 are likely to be retrieved by many teams,
but questions with Indri rank lower than 100 are not retrieved by any teams.
Table 5. Cross table of Indri rank (English) and number of team best performance
runs that find relevant articles
Since most of the runs return only one article per question, the second- and
third-ranked relevant articles are not likely to be retrieved. Therefore, most of the
second- and third-ranked relevant articles are not retrieved by all runs regardless
of Indri rank. To reduce the effect of questions with multiple answers, Table 6
shows the cross table of Indri rank (English) for the top-ranked relevant article
for each question and the number of team best performance runs that find those
articles. From this table, we see that there are several questions whose retrieved
relevant articles differ according to the submitted runs. For example, in the
case of Indri top-ranked questions in H28 data, the number of questions that
are retrieved by all teams (five) increases from 35 to 38, because the following
questions have two different relevant articles that are retrieved by the different
systems.
– H28-11-5 (The statutory lien over movables shall prevail over the pledge of
movables.): Only Indri found 339 and 5 other runs found 334.
Analysis of COLIEE Information Retrieval Task Data 13
Table 6. Cross table of Indri rank (English) of the top-ranked relevant article and the
number of team best performance runs that find at least one relevant article
Table 7 shows the cross table of Indri rank and the results of the best-run of
all submission data using English collection for each year (H22: UA; H24: UA;
H26: iLIS7; and H28: iLIS7 (Year: teamID)). Number of teams for second row of
Table 7 is 0 (the best-run fails to retrieve the article) and 1 (the best-run succeeds
to retrieve the article). From this table, it is clear that the best-run systems are
effective at retrieving articles with higher Indri rank, but it is difficult for those
systems to retrieve articles with lower Indri rank. The characteristics of those
articles are discussed in Sect. 4.
14 M. Yoshioka
Table 7. Cross table of Indri rank (English) and best-run of all submission data using
English collection
Since the number of teams using Japanese test data is just one—except for
H28—and the second-best team in H28 is relatively worse than the best-run of
all submission data and Indri, we use cross tables for the best run (Tables 8 and
9) only for its analysis. The general tendency of the results is almost identical
to the English data.
Table 8. Cross table of Indri rank (Japanese) and best-run of all submission data
using Japanese collection