QRedit: An Integrated Editor System to Support Online Volunteer Translators

1. Introduction
We are currently developing an English-to-Japanese
computer-aided translation (CAT) system with the aim
of aiding online volunteer translators, who are involved in
translating online electronic documents in their free time. In
the following, we will first describe the characteristics of target
users, and discuss the basic concept of the system within which
the choice of functions and system specifications has been
made. Then we will describe the prototype CAT system which
we are now experimentally providing to a limited number of
online translators.
2. Background and basic concepts
Many CAT systems have been developed to date [see
Kay 1997 for background information and Gow 2003
for a comparison of several CAT systems]. While some existing
CAT systems have proved useful for some translators and
translation companies, volunteer translators who work online
do not use these systems [Kageura, et. al. 2006; see also Fulford
and Granell Zafra 2004 in relation to the situation of freelance
translators in Britain], for a variety of reasons: they are too
expensive for personal use (except for Omega-T, which is free
[Omega-T 2007]); they have more functions than are necessary;
such translators are not under pressure from managers to keep
a translation log in order to control quality, etc. This shows that
some sort of CAT-lite system is required by online translators,
which shares some basic functionalities with existing
commercial CAT systems but with a different emphasis and
overall design principle [Boitet, et. al. 2005].
We have therefore been engaged in developing a CAT system
for online volunteer translators, based on the principle of
maximally aiding the translators' work flow by removing
existing obstacles, rather than providing extra functions which
translators have never used. Thus we are taking a bottom-up
approach in developing system functions, reflecting the concrete
requirements of translators. Another essential principle we have
adopted is that it should be translators who make decisions, not
the system.
After consulting some 15 volunteer translators working online,
we identified the following issues as being particularly
important:
1. Most English-to-Japanese volunteer translators do not have
native level command in the source language (English). As
a result, unlike such CAT systems as TransType
[Macklovitch 2006] which assume that the target users are
bilingual, well-trained and professional translators, and
therefore try to reduce the time used for inputting the target
language text, the main point to be tackled in our
environment is making the reference lookup process as easy
as possible.
2. When selecting translation expressions, translators examine
various possibilities. When necessary, they look up more
than one dictionary and check other information resources
such as print encyclopaedias or Wikipedia, etc. The CAT
system should facilitate the translation process by reducing
the effort involved in looking up multiple resources. As it
is not the mission of the system to decide on behalf of
translators, good output for the system is considered to be
a range of candidates and information that translators can
take into account in making decisions, and not simply the
same expression as translators chose.
3. Some online translators do not use online dictionaries,
because doing so breaks the rhythm of translation and
hinders the construction of the target text. Partly because
of this, some prefer using print or standalone electronic
dictionaries. A few use dictionaries built into the text editor,
but none of them are fully satisfied with the dictionary
look-up environment.
4. There is a pressing need for improved idiom and phrasal
look-up. The importance of this function derives from two
factors: (a) many translators, even experienced ones, have
relatively less knowledge of idioms than of words, and (b)
some idioms may not be identified as such by translators,
because they make sense without an idiomatic interpretation.
This leads to translation mistakes.
5. As our system does not aim to choose and restrict
information on behalf of translators by using "sophisticated"
NLP techniques, the amount of information it provides will
naturally tend to increase. As a result, the user-interface
becomes an important issue. 3. The prototype system QRedit
Based on these concepts and requirements, we are currently
developing a prototype system that supports online
translators. The system runs on the TOMCAT server and users
access the system through a Web browser. The overall image
of QRedit is shown in the figure. The browser screen is divided
into two areas: (i) the source text area, and (ii) the target text
area. The users can choose between horizontal and
perpendicular division of the areas, i.e. source text area on the
bottom and target language area on the top, or source text area
on the left and target language area on the right. The two areas
are linked with a synchronised scroll function.
3.1 Functions in the source text area
After a translator inputs the URL of a Web page (in which case
the system analyses the tags and extracts the textual area
automatically) or copies and pastes text into the source text
area, the system activates the dictionary look-up functions.
When the user clicks on a particular word in the source text
area with the mouse, the system shows the translation candidates
in a pop-up window.
Dictionary look-up functions
The system displays translation candidates from the dictionaries
incorporated into the system [Sanseido 2004; Eijiro 2006]. The
system does not only incorporates simple word look-up
functions but also incorporates flexible idiom look-up functions.
The idiom look-up functions can match such idiom occurrences
as "He said that with his big fat tongue in his big fat cheek"
with the dictionary entry "with one's tongue in one's cheek."
This function has not been realised in any English-Japanese
MT systems we have checked, and while some CAT systems
realise similar functions through approximate matching, they
do not specifically target the look-up of idioms with their
variations. The system alerts users to idioms by marking them
with an underline.
Displaying translation candidates
The system can display information in two ways. One is a small
pop-up window displayed within the source text area, which
includes only target word candidates. The other is a large pop-up
window that displays all the information given under the
headings in the dictionaries. The latter is particularly useful
when translators wish to examine related information in detail
before deciding on a translation, while the former is convenient
for more straightforward dictionary look-up.
3.2 The connection between the source text area
and the target text area
Translators can paste a selected expression from the list of
candidates in the pop-up area to the target text area by clicking
the mouse. The system also provides automatic transformation
of numerical expressions to Japanese conventions and the
pasting of HTML tags. This mouse operation does not affect
the keyboard operation, and whatever the user does with the
mouse, the keyboard cursor always stays on the target text
editing area, so that the user can input translated text
continuously.
3.3 Target text area
The target text area consists of the open source Web editor
FCKeditor, which offers WYSIWYG textual decoration and
editing, as well as saving and loading functions. The text can
be saved in HTML as well as in basic textual format. The target
area is split into paragraph spaces which correspond to the
source text.
4. Conclusions
We are currently running a prototype system of QRedit
that incorporates these functions, and accumulating
feedback from a few online volunteer translators. The feedback
that we have obtained so far can be categorised into two types:
1. The need to refine existing functions and improve basic
usability, e.g., to improve the accuracy of the extraction of
the textual area when the user specifies an URL for the
source text area.
2. The need to incorporate higher-level functions, e.g. to allow
the user to specify the register of the text that (s)he is
translating, in order to have the system block irrelevant
dictionary information from being displayed.
3. In addition, we are developing a module that automatically
compiles bilingual technical terminologies from the Web,
a module that detects existing translation pairs and recycles
the translation information, and a system that detects
candidates for transliterated expressions of proper names
from the Web, all of which will be integrated into the
QRedit environment.
Bibliography
Boitet, Christian, Youcef Bey, and Kyo Kageura. "Main
Research Issues in Building Web Services for Mutualized, Non-commercial Translation ." Proceedings of the SNLP: 6th
Symposium on Natural Language Processing . 2005.
Eijiro. Accessed 2006-10-31. <http://www.eijiro.j
p>
FCKeditor. Accessed 2006-10-31. <http://www.fcked
itor.net>
Fulford, Heather, and Joaquin Granell Zafra. "The Uptake of
Online Tools and Web-based Language Resources by Freelance
Translators: Implications for Translator Training, Professional
Development, and Research ." Proceedings of the Second
International Workshop on Language Resources for Translation
Work, Research and Training . 2004. 37-44.
Kageura, K. et. al. "Improving the Usability of Language
Reference Tools for Translators ." Proceedings of the 10th of
Annual Meeting of the Japan Association for Natural Language
Processing . 2006. 707-710.
Kay, Martin. "The Proper Place of Men and Machines in
Language Translation." Machine Translation 12.1-2 (1997):
3-23.
Macklovitch, Elliott. TransType2: The Last Word. LREC, 2006.
167-172.
Omega-T. 2007. <http://www.omegat.org/omegat/
omegat_en/omegat.html>
Sanseido. Grand Concise English Japanese Dictionary. Tokyo:
Sanseido, 2006.