A Support System for Revising Titles
to Stimulate the Lay Reader’s Interest in Technical Achievements
Yasuko Senda†‡, Yasusi Sinohara†, and Manabu Okumura§
†System Engineering Research Laboratory,
Central Research Institute of Electric Power Industry, Tokyo Japan
‡Department of Computational Intelligence and Systems Science,
Tokyo Institute of Technology, Tokyo Japan
§Precision and Intelligence Laboratory, Tokyo Institute of Technology, Tokyo Japan
Abstract
When we write a report or an explanation
on a newly-developed technology for readers
including laypersons, it is very important to
compose a title that can stimulate their in-
terest in the technology. However, it is dif-
ficult for inexperienced authors to come up
with an appealing title.
In this research, we developed a support sys-
tem for revising titles. We call it “title revi-
sion wizard”. The wizard provides a guid-
ance on revising draft title to compose a title
meeting three key points, and support tools
for coming up with and elaborating on com-
prehensible or appealing phrases.
In order to test the eﬀect of our title revision
wizard, we conducted a questionnaire sur-
vey on the eﬀect of the titles with or with-
out using the wizard on the interest of lay
readers. The survey showed that the wiz-
ard is eﬀective and helpful for the authors
who cannot compose appealing titles for lay
readers by themselves.
1 Introduction
When we read a document, we usually read its
title first, and then we read the body text only
if the title catches our interest. Therefore, when
we write a report or an explanation on a newly-
developed technology intended for readers in-
cluding laypersons, it is very important to com-
pose a title that will stimulate their interest in
the technology. However, technical specialists
are not necessarily good at composing appeal-
ing titles, because it isn’t clear what sort of titles
will stimulate the interest of lay readers in the
technology.
In the field of NLP and linguistics, there are
few researches which help the specialists com-
pose appealing titles for lay readers. Several
researches have been reported on title genera-
tion (Jin and Hauptmann, 2000) (Berger and
Mittal, 2000) and readability of texts (Minel et
al., 1997) (Hartley and Sydes, 1997) (Inui et al.,
2003). However, the researches on title gener-
ation focus on generating a very compact sum-
mary of the document rather than composing
an appealing title. The previous researches on
readability mainly see it as comprehensibility
rather than interestingness.
In this regard, our previous study (Senda and
Sinohara, 2002) clarified what sort of content
and wording in titles are eﬀective in stimulat-
ing lay readers’ interest in the technology by an
analysis of a parallel corpus of Japanese techni-
cal paper titles and Japanese newspapers head-
lines. The study categorized the eﬀective con-
tent and wording of the titles into the following
three key points.
Key Point 1 (for Wording) Instead of tech-
nical terms, use synonymous plain terms
even where the plain term is not synony-
mous with the technical term in a precise
sense.
Key Point 2 (for Content) Describe what
the technology is for, rather than what
the technology does.
Key Point 3 (for Content) Describe the ad-
vantages of the technology, rather than
the method of realizing the technology.
Our next goal is to enable inexperienced au-
thors to compose a title according to these key
points. To this end, we developed a support
system for revising titles. We call it “title revi-
sion wizard”. The wizard provides a guidance
on revising draft title to compose a title meet-
ing the key points, and a few support tools for
coming up with and elaborating on appealing
phrases. In this paper, we report on the title
revision wizard, and a questionnaire survey on
the eﬀect of titles composed with and without
using the wizard on the interest of lay readers.
2 Method for Revising Titles
Intended for Lay Readers
It is diﬃcult for inexperienced authors to change
their “specialist-centered mind-set” and come
 5 V G R     + P R W V  U Q O G  R J T C U G U  K P V Q  V J G  V K V N G  V G O R N C V G
 5 V G R     4 G X K U G  V J G  R J T C U G U  K P  1 $  2  U N Q V U
 5 V G R     4 G X K U G  V J G  R J T C U G U  K P  1 2  2  U N Q V U
 5 V G R     5 G N G E V  V J G  V K V N G  H T Q O  R Q U U K D N G  V K V N G U  I G P G T C V G F  K P  V J G  C D Q X G  U V G R U
 / G V J Q F  V Q   5 J Q T V G P   V J G  4 C F K Q C E V K X G  * C N H  N K H G                  D [   / G V C N N K E  ( W G N  ( $ 4
 = 1 D N K I C V Q T [  2 C T V  
 1 $  2  ?  = 1 R V K Q P C N  2 C T V  
 1 2  2  ?
 / G V J Q F  V Q   5 J Q T V G P   V J G  & W T C V K Q P  Q H  4 C F K C V K Q P
 / G V J Q F  V Q   5 J Q T V G P   V J G  5 V Q T C I G  2 G T K Q F  Q H  4 C F K Q C E V K X G  9 C U V G    D [   $ W T P Q W V
 6 G O R N C V G  Q H  1 $  2  H Q T  - G [  2 Q K P V  
 % J C P I G  V G E J P K E C N  V G T O U  V Q  U [ P Q P [ O Q W U  R N C K P  V G T O U
 / G V J Q F  V Q   5 J Q T V G P   V J G  5 V Q T C I G  2 G T K Q F  Q H  4 C F K Q C E V K X G  9 C U V G  V J G  5 V Q T C I G  2 G T K Q F  Q H  4 C F K Q C E V K X G  9 C U V G / G V J Q F  V Q   5 J Q T V G P 
 ' Z R N C K P  Y J C V  V J G  V G E J P Q N Q I [  K U  H Q T   T C V J G T  V J C P  Y J C V  V J G  V G E J P Q N Q I [  F Q G U
   D [   $ W T P Q W V  V J G  5 V Q T C I G  2 G T K Q F  Q H  4 C F K Q C E V K X G  9 C U V G  V J G  5 V Q T C I G  2 G T K Q F  Q H  4 C F K Q C E V K X G  9 C U V G  / G V J Q F  V Q   5 J Q T V G P 
 6 G O R N C V G  Q H   1 $  2    1 2  2   H Q T  - G [  2 Q K P V  
 % J C P I G  V G E J P K E C N  V G T O U  V Q  U [ P Q P [ O Q W U  R N C K P  V G T O U
 ' Z R N C K P  V J G  C F X C P V C I G U  Q H  V J G  V G E J P Q N Q I [   T C V J G T  V J C P  V J G  O G V J Q F  Q H  T G C N K \ K P I  V J G  V G E J P Q N Q I [
 / G V J Q F  V Q   5 J Q T V G P   V J G  5 V Q T C I G  2 G T K Q F  Q H  4 C F K Q C E V K X G  9 C U V G    D [         
 6 K V N G  
 6 K V N G  
 6 K V N G  
 6 K V N G  
 6 G O R N C V G  Q H  1 $  2  H Q T  - G [  2 Q K P V  
 6 G O R N C V G  Q H   1 $  2    1 2  2   H Q T  - G [  2 Q K P V  
 
 & T C H V  8 G T U K Q P 
 9 J K V G  I T Q W P F  K U  U N Q V  H Q T  K P R W V V K P I  C P F  T G X K U K P I
  2 C T C N N G N  % Q T R W U  Q H
 6 K V N G U  C P F  * G C F N K P G U 
 6 G E J P K E C N  6 G T O U
 % J G E M G T % J G E M G T
 / G P W  1 R V K Q P U  Q H
 V J G  2 J T C U G U  7 U G F  K P
 0 G Y U R C R G T  * G C F N K P G 0 G Y U R C R G T  * G C F N K P G
 / G P W  1 R V K Q P U  Q H
 V J G  2 J T C U G U  7 U G F K P
 5 W R R Q T V  6 Q Q N U  2 T Q X K F G F
 K P  V J G  9 K \ C T F
Figure 1: Steps in the Procedure for Revising draft title Using the Wizard
up with appealing titles for lay readers even
when they know the three key points.
In our title revision wizard, therefore, the au-
thors first input draft title into the title tem-
plate, and then compose candidates of title by
revising the phrases of the draft title according
to the wizard’s guidance with the help of the
support tools provided.
The steps in the wizard’s procedure are il-
lustrated in Figure 1
1
. In the following, we
explain the steps with reference to Figure 1.
2.1 Inputting Some Phrases into the
Title Template
In step 1, the user inputs draft title into the
title template displayed in the wizard window.
The title template consists of an “OBligatory
Part” (OB-P) and an “OPtional Part” (OP-P).
The OB-P phrase describes what the technology
does, and OP-P phrase describes the method
used to implement the technology or the advan-
tages of the technology.
2.2 Revising the Obligatory Phrases
In step 2, the wizard presents only OB-P to the
user, and hides OP-P. The user revises only the
OB-P phrase according to the two key points.
First, the user changes technical terms to the
synonymous plain terms according to key point
1
The sample titles presented in Figure 1 are trans-
lated into English from original Japanese titles.
1. In the template 2-1 in step 2, the technical
term “the Radioactive Half-life” is changed to
the plainer term “the Duration of Radiation”
from this viewpoint.
Secondly, the user changes the phrase in the
OB-P slots to describe the purpose of the tech-
nology rather than what the technology does,
according to key point 2. In the template 2-2
in step 2, the phrase “Shorten the Radioactive
Half-life” is changed to the phrase “Shorten the
Storage Period of Radioactive Waste” from this
view point.
These revised OB-P phrases without optional
phrase are recorded as candidates of “simple”
title for future selection in step 4. At the end
of step 2, for future revision in step 3, the user
selects one title that he/she deems the better
one from these candidates of title.
2.3 Revising the Optional Phrases
In step 3, the wizard presents new title combin-
ing OB-P phrase (selected at the end of step 2)
and the OP-P phrase (inputted in step 1) as a
draft title. The user revises only the OP-P draft
phraseaccordingtothetwokeypoints.
First, the user changes technical terms to syn-
onymous plain terms according to key point 1.
In the template 3-1 in step 3, the technical term
“Metallic Fuel FBR” in the slot is changed to
the plainer term “Burnout” from this viewpoint.
Secondly, the user changes the phrase in the
slot to describe the advantages of the technology
rather than the method of realizing the technol-
ogy according to key point 3. In the template
3-2 in step 3, the OP-P phrase is changed to the
phrase “by 1/10000” from this viewpoint.
The title combining the title selected in step
2 and each phrase revised at step 3 are recorded
as candidates of title for future selection in step
4. Before next step, the user can return to step
2 to select another OB-P phrase and revise OP-
P phrase attached to the OB-P phrase in step
3 again if he/she likes.
2.4 Select the Title from Title
Candidates
In step 4, the user selects one title from the
candidates of title composed in the above steps.
2.5 An Example of the Wizard Window
Figure 2 is a screenshot of the Wizard. This
screenshot shows the window in step 2 described
in section 2.2. The Japanese text in upper pane
of the window is the explanation of key point
1. The template for the OB-P Phrase for key
point 1 is displayed at the bottom of the win-
dow. The buttons in the center of the pane are
for accessing the support tools for coming up
with and elaborating the input phrase. Detail
of the support tools will be given in the next
section.
Title revision wizard is implemented in PHP
(Hypertext Preprocessor). Users can access the
wizard using a web browser such as Internet Ex-
plorer.
3 Support Tools provided in the
Wizard
In this section, we explain the three support
tools provided in the title revision wizard.
3.1 Database of Paper Titles and
Newspapers Headlines for Related
Technologies
It is diﬃcult for inexperienced authors to come
up with a phrase meeting the three key points.
We therefore prepared a parallel corpus of titles
and headlines for related technologies in order
to provide clues for coming up with an appropri-
ate phrase. The database consists of about 420
titles and 440 headlines. They were categorized
into 150 groups on technology that has been de-
veloped in a research institute, and covers sci-
ence and technology in general. The database
is (b) in Figure 1.
Figure 3 shows an example of the database
window. The upper pane in the window shows
the phrases in titles and the headlines describ-
ing related technologies. The lower part of the
window presents search boxes for menu-based
retrieval and keyword-based retrieval. The pull-
down menu options are organized by technical
fields. Users can search the clues for revising the
title of the draft version from these search boxes
as well as by scrolling through the window.
The role of this parallel corpus is basi-
cally the same as the one of “Example-Based
Translation Aid (EBTA)” (Sato, 1992) (Furu-
gori and Takeda, 1993) (Kumano and Tanaka,
1998). EBTA researches have shown that par-
allel translation examples (pairs of source text
and its translation equivalent) are very helpful
for translators to translate the similar text be-
cause parallel translation examples give them
useful clues for translation. From the viewpoint
that paper titles and newspapers headlines for
related technologies are also regarded as paral-
lel translation examples (pairs of text for spe-
cialists and its translation equivalent for lay
readers) describing newly-developed technolo-
gies, the our database is expected to be helpful
for the user of the wizard.
3.2 Technical Terms Checker
The author should use comprehensible term for
lay readers in order to compose the titles meet-
ing the key point 1. In order to avoid incompre-
hensible terms, it is important to identify diﬃ-
culty level of a term that authors want to use in
his/her title. We, therefore, prepared “techni-
cal term checker” estimating the diﬃculty level
of a term.
It has been reported that human recognition
level of a term correlates with its frequency of
appearance (Homes and Solomon, 1951). From
that standpoint, it is considered that frequency
of a term on the academic website represents
the recognition level of a term for the special-
ists of the field, and that frequency of the term
on the general website represents the one for
laypersons of the field. On the basis of this con-
cept, our “technical term checker” estimates the
diﬃculty level of a technical term on the basis
of the frequencies on the academic and general
website, and then inform the users about the re-
sult on three level (“plain”, “may be diﬃcult”,
“diﬃcult”).
 5 W R R Q T V  6 Q Q N U  H Q T  K P R W V V K P I
 6 K V N G  6 G O R N C V G
Figure 2: Screenshot of Title Revision Wizard
,(9*/ �	6?,: �-69 �2,@>69+ �9,;90,=(3 �(5+ �4,5< �9,;90,=(3 �69.(50A,+  �)@ �;,*/50*(3 �-0,3+: �
,*/50*(3 �0,3+ /, �	 � �/9(:,: �05 �(7,9 �0;3,: /, �	 � �/9(:,: �05 �,>:7(7,9 �,(+305,:
Figure 3: Parallel Corpus of Titles and Headlines for Related Technology
3.3 Menu Options of the Phrases Used
in Newspapers Headlines
In order to meet key point 3, the authors have to
consider how to represent the advantages of the
technologies from various viewpoints, and come
up with the phrase that can concisely represent
the advantages. However, it is considered that
those who have a detailed knowledge of a techni-
cal field are unaccustomed to represent the ad-
vantages of the technologies in a title because
previous research showed that most of technical
paper titles in Japanese does not included such
a phrase (Senda and Sinohara, 2002).
We therefore prepared the pull-down menu
options of the phrases representing technical ad-
vantages used in newspapers headlines. The
menu options presents about 70 phrases catego-
rized by the way to describe the advantages of
technology. For example, in our menu options,
the following phrases are listed by the following
group: high density, high concentration, high
accuracy, . .. , low price, low cost, low pollution,
.. ., long-lived, long distance, . .. , short time,
short duration, . .., ; etc
2
.
Figure 4 shows the menu options installed at
the title template for key point 3. This menu
options can help the users come up with or elab-
orate on the phrases by oﬀering various expres-
sions and viewpoints.
4 Eﬀect of Title Revision Wizard
In order to test the eﬀect of the title revision
wizard, we conducted an experiment which had
17 technical researchers revise their titles with
and without using our wizard. We then con-
ducted a questionnaire survey on the eﬀect of
the titles on the interest of lay readers.
4.1 Outline of the Experiment
We conducted the experiment according to the
following procedures.
Experiment 1 (Ex 1)
1. We showed each subject his technical pa-
per title (published after 2000) on his
own developed technology. We asked him
to imagine himself writing an explanation
of the same technology for lay readers.
2. We asked each subject to compose three
candidates of title for the explanation by
his own eﬀort (that is, without our wiz-
ard) within the 20-minute time limit.
Experiment 2 (Ex 2)
1. After Ex 1, we input original technical
paper title into first title template of our
title revision wizard (for draft title), and
presented each subject with the wizard.
2. We asked each subject to compose candi-
dates of titles for the same document us-
ing the wizard and select thee titles from
the candidates within the 20-minute time
limit.
The research fields of subjects were physics,
electrical engineering, material science and me-
teorology. There was no intermission between
Ex 1 and 2.
2
These sample phrases are translated into English
from original Japanese phrases.
4.2 Outline of the Questionnaire
Survey
In the questionnaire, each respondent was
shown seven titles per one technology that each
subject developed. They consisted of an origi-
nal technical paper title, three titles composed
at Ex 1, and three titles composed at Ex 2. Each
respondent was asked to select three-most inter-
esting titles from these 7 titles per one technol-
ogy.
Respondents to the questionnaire were 108
persons and general monitors of an Internet re-
search firm. We asked them to answer our ques-
tionnaire on a Web page which was prepared for
this survey.
This means, of course, that the results of the
questionnaire only contain responses from peo-
ple who have some skill at using the Internet,
rather than the public at large, because all re-
spondents were able to use E-mail and access
the web page. However, at least, the result
helps to write document on a Web page which
explains new technology for lay readers.
Moreover, the result is also an exercise in
reaching the general lay readers over the Inter-
net, and the use of the Internet is expected to
be increasingly important in reaching the gen-
eral public.
4.3 Analysis of Experiment and
Questionnaire Survey
Figure 5 shows the (average) share of the vote
which the respondents select three-most inter-
esting titles from each author’s seven titles (by
method of composing). Figure 5 indicates that:
 The average share of the vote of the titles
revised using the wizard are distributed
within the highest range.
 The average share of the vote of the titles
revised without using the wizard are dis-
tributed within the second highest range,
 The share of the vote of original techni-
cal paper titles are distributed within the
third highest range,
We checked if there is a significant diﬀerence
between the (average) share of the vote of orig-
inal technical paper titles and the titles revised
without using the wizard by t-test. As the re-
sult, we confirmed that there is a significant dif-
ference between these titles (the significant level
is 1%). We also checked if there is a significant
diﬀerence between the share of the vote of the
titles revised with and without using the wizard
<33 �+6>5 �4,5< �6- �;/, �7/9(:,: �+,:*90)05. �;/, �(+=(5;(., �6- �;/, �; ,*/5636.@
Figure 4: Pull-down Menu Options Organizing the Phrases Used in Newspaper Head-
line
:�:�:�:�:�:�:w:�:�:�:�:�:�:�:w:�:�:�:�:�:�:�:w:�:�:�:�:�:w:�:�:�:�:�:�
:�:�:�:�:�:w:�:�:w:�:�:�:w:�:�:�:�:w:�:�:w:�:�:�:w:�:�:�:�:�::�:�:w:�:�:�:�:�:w:�:�:�:�:w:�:�:�:�:�:�:w:�:�:�:�:�:�:�:�:w
:�:�:�:�:�:�:w:�:�:w:�:�:�:�:�:�:�
:�:�:| :�:�:| :�:�:| :�:�:| :�:�:| :�:�:|
:�:�:�:�:�:�:w:�:�:�:�:�:�:�:w:�:�:�:�:�:w:�:�:�:�:�:�
:�:�:�:�:�:�:�:�:w:�:�:�:�:�:�:�:�:�:w:�:�:�:�:�:w:�:�:�:�:�:�
:�
:�
:�
:�
:�
:�
:�
:�
:�
:�
:�:�:|C>:�:�:|
::�:�:�:�:�:�:�:w:�:�:�:�:�:w:�:�:w:�:�:�:w:�:�:�:�:�
:w::�:�:�:�:�:w:�:�:w:�:�:�:w:�:�:�:�:�
Figure 5: Histogram of Share of the Vote
which the Respondents Select Three-
most Interesting Titles from Each Au-
thor’s 7 Titles (by Method of Composing)
by t-test. As the result, We also confirmed that
there is a significant diﬀerence between these
titles (the significant level is 5%).
These results show that the titles composed
using the wizard could stimulate the lay readers’
interest the most among each subject’s seven
titles. In other words, the authors can stably
compose the more appealing titles for lay read-
ers with using the wizard than without using
the wizard.
If an author could compose eﬀective titles by
himself, he might not take an advantage from
the wizard. We, therefore, focused only on
seven subjects who could not compose the ti-
tles stimulating a majority of the respondents’
interest without the wizard, and analyze the ef-
:�:�:�:�:�:�:�
:�:|
:�:�:|
:�:�:|
:�:�:|
:�:�:|
:�:�:|
:�:�:|
:�:�:|
:�:�:|
:�:�:�:�:�:�:�:w:�:�:�:�:�:w:�:�:�:w:�:�:�:�:�:�
:�:�:�:�:�:w:�:�:�:w:�:�:�:�:�:�
:�:�:�:�:�:�:�:w:�:�:�:�:�:w:�:�:w:�:�:�:w:�:�:�:�:w
:�:�:�:�:�:�:�:w:�:�:�:�:�:�
Figure 6: Average Share of the Vote of
the Titles Composed with and without
Using Wizard
fect of their titles on the respondents’ interest
in more detail.
We, then, compared the average share of the
vote for the titles which each of the seven au-
thor revised with and without using the wizard.
figure 6 shows that all the subjects can stably
compose the more appealing titles with using
the wizard than without using the wizard. We
checked if there is a significant diﬀerence be-
tween the share of the vote of the titles revised
with and without using the wizard by t-test. As
the result, We also confirmed that there is a sig-
nificant diﬀerence between these titles (the sig-
nificant level is 1%).
As the result of the above analysis, we con-
firmed that the title revision wizard can help
the users who cannot compose eﬀective titles
by their own eﬀorts.
5 Conclusion and Future Work
We emphasized that, in order to stimulate the
lay reader’s interest in newly-developed technol-
ogy, it is very important to compose an appeal-
ing title, however it is diﬃcult for inexperienced
authors to come up with an appealing title.
In this research, we developed a title revi-
sion wizard that provides a guidance on revising
draft title and a few support tools for coming
up with and elaborating on appealing phrases.
Moreover, we verified the eﬀect of the title re-
vision wizard. As a result, the titles composed
using the wizard can stimulate the lay reader’s
interest more than the titles composed without
using the wizard. In particular, our title revi-
sion wizard can help the users who cannot com-
pose eﬀective titles by their own eﬀort.
In future work, we will analyze the diﬀerence
of the expression of the titles composed with
and without using the wizard, and investigate
what sort expression is eﬀective to lay readers.
Acknowledgements
The authors would like to express our grati-
tude to Ms. Tomoko Tsuchiya, Ms. Motoko
Kosugi, and Ms. Tomoko Mitamura of Central
Research Institute of Electric Power Industry
for their valuable advices for our questionnaire
survey, and Mr. Masahito Tanaka of Denryoku
Computing Center, Ltd. for his valuable sup-
ports to implement our title revision wizard. In
addition, the authors would like to express our
gratitude to anonymous reviewers for their sug-
gestions to improve our paper.

References

Adam L. Berger and Vibhu O. Mittal. 2000.
Ocelot: A system for summarizing web pages.
In Proc. of the 23rd Annual International
ACM-SIGIR Conference on Research and
Development in Information Retrieval, pages
144–151, Athens, Greece.

Teiji Furugori and Akiko Takeda. 1993. An
example-based system of writing english sen-
tences for japanese english users. Literary
and Linguistics Computing, 8(2):85–90.

James Hartley and Matthew Sydes. 1997. Are
structured abstracts easier to read than tradi-
tional ones? Journal of Research in Reading,
20(2):122–136.

Davis H. Homes and Richard L. Solomon. 1951.
Visual duration threshold as a function of
word-probability. Journal of Experimental
Psychology, 41:401–410.

Kentaro Inui, Atsushi Fujita, Tetsuro Taka-
hashi, Ryu Iida, and Tomoyo Iwakura.
2003. Text simplification for reading assis-
tance: A project note. In Proc. of The Sec-
ond International Workshop on Paraphras-
ing: Paraphrase Acquisition and Applications
(IWP2003), pages 9–16, Sapporo, Japan.

Rong Jin and Alex G. Hauptmann. 2000. Ti-
tle generation for spoken broadcast news us-
ing a training corpus. In Proc. of the 6th In-
ternational Conference on Spoken Language
Processing (ICSLP), pages 680–683, Beijing,
China.

Tadashi Kumano and Hideki Tanaka. 1998.
Translation examples browser: Japanese to
english translation aid for news articles. In
Proc. of NLP+IA 98/TAL+AI 98, pages 96–
102, Moncton, Canada.

Jean-Luc Minel, Sylvaine Nugier, and Gerald
Piat. 1997. How to appreciate the quality
of automatic text summarization? examples
of fan and mluce potocols and their results
on seraphin. intelligent scalable text summa-
rization. In Proc. of 35th Annual Meeting of
the ACL Workshop Intelligent Scalable Text
Summarization, pages 25–30, Madrid, Spain.

Satoshi Sato. 1992. Ctm: An example-based
translation aid system. In Proc. of COLING
1992, pages 23–28, Nantes, France.

Yasuko Senda and Yasusi Sinohara. 2002.
Analysis of titles and readers for title gen-
eration centered on the readers. In Proc. of
COLING 2002, pages 421–424, Taipei, Tai-
wan.
