An Interactive Translation Support Facility 
for Non-Professional Users 
YAMABANA Kiyoshi, MURAKI Kazunori, KAMEI Shin-ichiro, 
SATOH Kenji, DOI Shinichi, TAMURA Shinko 
Information Technology Research Laboratories 
NEC Corporation 
Miyazaki 4-1-1, Miyamae-ku, Kawasaki 216, JAPAN 
{yamabana, k-muraki, kamei, satoh, doi, shinko}~hum.cl.nec.co.jp 
Abstract 
We present an interactive translation 
method to support non-professional users 
to write an original document. The 
method, combining dictionary lookup func- 
tion and user-guided stepwise interactive 
machine translation, allows the user to ob- 
tain clear result with an easy operation. 
We implemented the method as an English 
writing support facility that serves as a 
translation support front-end to an arbi- 
trary application. 
1 Introduction 
With the steady decrease of network communica- 
tion cost and equipment prices, world-wide com- 
puter networks and the number of its users are 
growing very rapidly. However, there is a large 
obstacle against global communication over net- 
works, namely the language barrier, especially for 
non English-speaking people. This is a major reason 
personal EJ (English to Japanese) machine transla- 
tion systems are gaining popularity in Japan. They 
help the user to quickly grasp the content of web 
pages, by providing rough translation. Since speed 
and lexical coverage are most important require- 
ments, conventional automatic machine translation 
systems developed so far are useful for this purpose. 
Contrary to the EJ direction, the major task in JE 
(Japanese to English) direction will be writing short 
original documents, such as e-mail. The most impor- 
tant requirement will be translation quality, because 
the reader is usually different from the MT user. 
To control quality, some kind of human interaction 
will be inevitable. However, interactive support for 
conventional MT systems doesn't seem suitable for 
these users, since they are primarily intended for 
professional translators. Their post-editing function 
often requires working in a special environment that 
requires special training. An interactive, easy-to- 
use translation support facility, targeted for non- 
professional translators, is desirable. 
We may expect that these users have basic knowl- 
edge and ability to read and understand English. 
This expectation is natural and realistic in a coun- 
try like Japan, where all high-school graduates are 
supposed to have completed six year course in En- 
glish. Their reading skill and grammar knowledge is 
usually enough to judge the quality of current MT 
systems, but they may need help from MT systems 
when browsing the Internet. For the JE direction, 
they will not be satisfied with the raw output of con- 
ventional MT systems, but it will be too laborious 
to write down English sentence from scratch. For 
these users, online dictionaries have been used be- 
cause of the reliability of the result. However, in 
spite of abundant information within the dictionary 
such as inflection, verbal case frame, idioms and so 
on, the only electronically available part is spelling 
of translation equivalents (through copy & paste). 
Other information is only presented to be read as in 
the case of a paper dictionary, with all further work 
left to the user. 
In this article, we present an interactive trans- 
lation method and its implementation, which has 
advantages of both a dictionary look-up tool and a 
machine translation system. The system has an in- 
teractive interface similar to Kana-Kanji conversion 
method, and initially serves as a dictionary look- 
up tool. After dictionary lookup, the user can in- 
voke syntactic transformation in terms of grammat- 
ical information in the dictionary. Syntactic trans- 
formation proceeds step by step in a bottom-up 
manner, combining smaller translation components 
into larger ones. This "dictionary-based interactive 
translation" approach allows the user to fully utilize 
syntactic information in the dictionary while main- 
taining clarity of the result. 
In the next section, we give a simple example of 
translation steps and provide a general idea of the 
method. In section 3, we describe the basic model 
and associated operations. Section 4 gives further 
324 
explanation about disambiguation capability of the 
interactive operations. Section 5 discusses exten- 
sions to the basic model to treat linguistic phenom- 
ena such as idiomatic expressions. Section 6 de- 
scribes the current implementation as a front-end to 
an arbitrary application. In section 7, the method is 
compared with former approaches. The final section 
is the conclusion. 
2 An Example 
In this section we show the basic steps of simple 
sentence translation, in order to give a general idea 
about how the method works. 
Consider a situation where the user is writing a 
message in English, using an editor of a mail pro- 
gram. Our system is running as a daemon. While 
the user is typing English characters, the system 
does nothing special and let them through to the 
editor window. The system wakes up when the user 
toggles on Japanese input. The moment the first 
Japanese character is typed in, the main translation 
window is opened, and all subsequent characters are 
typed in to that window instead of the editor win- 
dow. 
Suppose the input sentence is one shown in fig- 
ure 1 (a). As soon as (a) is entered, the dictionary 
look-up function is invoked automatically. A mor- 
phological analyzer recognizes word boundaries in 
the sentence, looks up corresponding word entries in 
the system dictionary, and shows the result in the 
main window (b). 1 Content words are replaced by a 
translation equivalent assumed most plausible, while 
functional words are left unchanged. 
This representation step, in which English words 
(content words) and Japanese words (functional 
words) are mixed, separates steps for word trans- 
lation .and later syntactic transformation, making 
translation steps clearer. Since word order and func- 
tional words carrying grammatical functions are un- 
changed, the user can easily recognize the skeleton of 
the sentence, and clearly grasp the correspondence 
between the original word and its translation equiv- 
alent. This representation is not only to show the re- 
sults of dictionary look-up like (Canisius, 1977), but 
also carries all interactive operations of the method 
with it, and has a double role of showing information 
and being objects of interactive manipulation. 
Translation equivalent alternatives for the cursor 
position word (focus word) are displayed in an alter- 
natives window, appearing nearby that word. Fig- 
ure 2 is a snapshot of the alternatives window for 
ronbun (paper). The second line is highlighted to 
show that it is the current selection. The user can 
1Slanted words show romaji transcription of respec- 
tive Japanese words. They don't appear on a real 
window. 
(a) ~£ t~ ~ tc ~ ~¢ ~b tc 
watashi -wa kare -ni ronbun -o watashi -ta 
I TOP he DAT paper OBJ give PAST 
(b) I ~: he i:- paper ~ give ~v. 
-wa -hi -o -ta 
(c) I gave him a paper 
Figure 1: Translation of a simple sentence 
ronbun 
paper \[noun\] \[typical word 
thesis \[noun\] \[for degree\] 
essay \[noun\] \[general\] 
dissertation \[noun\] \[for degree\] 
Figure 2: Alternatives Window for ronbun 
change the selection simply by a cursor movement or 
mouse click on this window, then the corresponding 
translation equivalent on the main window changes 
synchronously. To see alternatives for another word, 
the user has only to move the cursor to that word on 
the main window. There is also a CD-ROM dictio- 
nary access function, making translation equivalent 
selection easier. In addition, the user can change an 
inflection in a similar manner on an inflection selec- 
tion window, opened by the user's request. 
If the user needs only the result of dictionary 
lookup, s/he can signal the end of translation at this 
point, just after choosing the translation equivalent. 
If a translation is necessary, the user needs to go one 
more step. At the same time as initial translation 
equivalent selection, the system predicts an appro- 
priate area for translation, as shown by an underline 
(b). Just as the translation equivalent selection can 
be freely changed, the area can be changed by drag- 
ging the left or right edge of the underline. After 
the user confirms selections of translation equiva- 
lents and translation area on (b), the user invokes 
translation. The system performs syntactic trans- 
formation using syntactic information in the dictio- 
nary such as the verbal case frame of the main verb 
in the area, shows the result on the main window, 
and replaces the original sentence with the result 
(c). When there is more than one possible transla- 
tion, the different possibilities are shown in an al- 
ternatives window similar to figure 2, allowing the 
user to change the choice. When the user triggers 
translation end, the result is sent to the original ed- 
itor window. The user can continue to work in the 
editor after turning off Japanese input. 
3 Description of the Method 
Most important characteristics of this interactive 
translation method is that the Japanese input is con- 
verted to English in several steps allowing user inter- 
action at each step. In intermediate steps, a mixture 
325 
of target language expression and source language 
expression are shown to give the current status of the 
interactive translation. Translation proceeds from a 
smaller unit as word to a larger unit as sentence, step 
by step in a bottom-up manner. The result can be 
checked and corrected at each step, making it easier 
to obtain a desired result. Interactive operations are 
similar to those of Kana-Kanji conversion, although 
they are further extended to be capable of control- 
ling syntactic transformations. We first describe the 
basic model that determines the scope and timing of 
interaction, then the set of interactive operations. 
3.1 Basic Model 
The basic model of our method is the syntax- 
directed translation scheme with bottom-up at- 
tribute evaluation (see chapter 5 of (Aho et al., 
1986)). In this scheme, an attribute of a syntax tree 
node is calculated from that of the children nodes 
by a semantic rule, associated with the syntax rule 
used to build the node from children. Attributes 
represent partial translation result for the structure 
below the node, and attribute calculation proceeds 
from the lexical nodes to the root node in a bottom- 
up manner. 
Interactivity is integrated with this model by al- 
lowing interactive operation when attribute is cal- 
culated at each node. Before each calculation, the 
system pauses to show an interpretation of the un- 
derlying structure, and allows the user to examine 
and change it if necessary. When interaction is fin- 
ished, the system chooses a next node and pauses 
there. This process repeats until the system reaches 
the root node. Any translation method can be used 
as long as it is compatible with this general scheme. 
Although basic model is as described, it is appar- 
ently too bothersome to give an operation at every 
node. In addition, some nodes only have a formal 
role in the grammar, and are not meaningful to the 
user. For this reason, the nodes at which the system 
automatically pauses for interaction are restricted to 
the node marked as a sentence, and the node that 
dominates a relative clause and its antecedent: in 
short, just restricted to contain one predicate. We 
remark that this restriction is effective only on de- 
fault decision of which node to pause at, and does 
not restrict operations by the user. The system looks 
for a minimal node marked as above, then pauses for 
user operation. At this time, attributes of the focus 
node and lower nodes are still undetermined except 
for lexical nodes. When the user triggers translation, 
undetermined attributes are calculated, then the re- 
sult replaces the tree under the focus node. That 
node serves as a kind of lexical node in subsequent 
translation. 
3.2 Interactive Operations 
The basic interaction model of the method is that 
the system shows current interpretation in the form 
of translation equivalents and translation area, while 
the user responds to it by changing these initial se- 
lections. This set of operations is essentially the 
same as the Kana-Kanji conversion method, and its 
obvious advantage is that everybody who can use 
Kana-Kanji conversion is expected to be well accus- 
tomed to these operations. 
When the system pauses for interaction, it shows 
initial selection of translation equivalents and trans- 
lation area, as in figure 1 (b). Translation equivalent 
selection for all content words, and the designated 
region to be translated next, is shown in a compact 
manner, allowing the user to examine and change 
them before translation. This mixed representation 
level of the source and target language expression 
serves as a playground for all subsequent interac- 
tions. 
After confirming all selections, the user triggers 
translation. Then the original area is replaced with 
resulting English expression. If there are more than 
one possible translation, the system presents them 
in a similar window as alternatives window as in fig- 
ure 2, and the user is allowed to change the system's 
selection by the same interface as translation equiv- 
alent selection. 
Translation equivalent selection enables the user 
to directly manipulate target language expression. 
Selecting before translation is much easier than after 
translation, because the word order and understood 
syntactic structure is that of the user's native lan- 
guage. The meaning of translation area selection is 
also clear. The user should choose the area so that 
it contains necessary and sufficient words to be one 
meaningful expression. Technically, it is bracketing 
by the user. If the user changes the area, the system 
changes the analysis according to the new constraint. 
Further disambiguation capability of this operation 
will be discussed in section 4. 
Other possible interactive operations include edit- 
ing and undoing the translation. The user can freely 
modify the displayed characters at any time, and the 
system responds by invoking an appropriate proce- 
dure, such as morphological analysis. Also, the user 
can go back to any former steps by canceling former 
translations. 
All these operations are optional, except for trans- 
lation triggers to invoke next translation. The 
amount of interaction and timing of translation trig- 
ger is completely up to the user, and s/he can even 
proceed without any modification to the system's 
initial choice. 
Steps of interactive translation can be summarized 
as below. 
326 
1. Type in the sentence. 
2. Repeat until whole sentence is translated. 
(a) Translation equivalent selection 
and translation area are shown. 
(b) Confirm all the selections are 
right. Change them if necessary. 
(c) Trigger translation. 
3. Signal the end of translation. 
3.3 Examples 
Next we turn to more complex examples, and show 
how more than one translation units are combined. 
3.3.1 A Relative Clause 
Figure 3 shows translation steps for a sentence 
with a relative clause. This sentence has a depen- 
dency ambiguity, so we also show how to resolve 
it through interaction. The original sentence (a) 
contains a relative clause with verb kau (buy) with 
an antecedent hen (book). Since Japanese is head- 
final, the sentence-initial case element kare-ga (he- 
SUB J) can be the subject of either kau (buy) or 
yomu (read), causing syntactic ambiguity. 
First, let's suppose kare-ga is assumed to be the 
subject of the relative clause by the system. Then 
the system pauses showing (b), as soon as (a) is in- 
put. In (b), the translation region is assumed to 
be "he-ga buy-ta book". After translation trigger, 
the system pauses showing (c). Please note that 
the underlined part in (b) is replaced by its equiva- 
lent English expression "the book he bought", and 
the whole sentence is underlined now. After another 
translation trigger, (d) is obtained, with missing sub- 
ject filled by some default word. 
Suppose after obtaining (d) the user noticed that 
this interpretation is not what s/he wants, and the 
case element kare-ga should be the subject of the 
verb of the matrix sentence. Then the user triggers 
undo of translation twice, returning to (b). Then 
s/he notice that "he -ga buy -ta book" is treated 
as one phrase, against his/her interpretation. Then 
s/he changes the underlined area to "buy -ta book", 
excluding "he -ga" from the region (e), because this 
is the "correct meaningful phrase" in the user's in- 
terpretation. After translation trigger, (f) follows. 
Note that the subject of the relative clause is supple- 
mented by a default element. Then (g), the desired 
result, follows. 
Generally, if two syntax tree nodes share a child 
leaf node, one is an ancestor of the other. This 
property guarantees that two overlapping transla- 
tion units can always be combined in our stepwise 
bottom-up translation method. 
3.3.2 A Conjunction 
Figure 4 shows translation steps for two sentences 
joined by a subordinate conjunct node (because). 
kate -ga kat -ta hen -o yon -da 
he SUBJ buy PAST book OBJ read PAST 
(b) he 75~ buy fz book ,~ read fg 
-ga -ta -o -da 
(c) the book he bought ~ read 7b: 
-o -da 
• (d) Someone read the book he bought 
(e) he ~ buy ft. book ~ read fc: 
-ga -ta -o -da 
(f) he ;0; the book someone bought ~ read Fd" 
-ga -o -da 
(g) He read the book someone bought 
Figure 3: Relative Clause and Syntactic Ambiguity 
kanojo -ga ki -ta -node watashi-wa 
she SUBJ come PAST because I TOP 
(b) she ~ come f~. ¢)'~I ~$ glad 
-ga -ta -node -wa 
(c) Shecame ©'~I ~ glad 
-node -wa 
(d) She came ¢)'~ I am glad 
-node 
ie) I am glad because she came 
ureshH 
glad 
Figure 4: Treatment of Conjunction 
Component sentences are translated first (c, d), then 
they are combined to produce a complex sentence. 
Here "because" is assumed to be the first alternative 
as translation equivalent for node. 
4 More on Interactive Operations 
The selection of an equivalent translation is more 
than simply choosing among synonyms, as shown in 
(Yamabana et al., 1995). First, part-of-speech of 
translation equivalent may be specified through this 
operation, since translation equivalents with differ- 
ent part-of-speech appear distinctly in the alterna- 
tives window. Second, the translation equivalent for 
functional words can be specified, which can affect 
the syntactic structure of the result. Although func- 
tional words remain unchanged in the intermediate 
representation, some words provide an alternatives 
window when the cursor is placed on them. Third, 
a whole unit with more than one word can be de- 
tected and selected in the same interface as transla- 
tion equivalent for a single word. 
An example for the first and second point is found 
in a translation equivalent set for an auxiliary rareru, 
which is known to be highly ambiguous. Even af- 
ter leaving aside less frequent "spontaneity" usage 
and indirect passivization, there are still at least 
three general interpretations: direct passivization, 
possibility, and honorific. Automatic disambigua- 
tion requires detailed semantic information, espe- 
cially when some case elements are missing or hid- 
den. 
327 
rareru 
be -ed 
Can 
possibly 
it is possible that 
be able to 
\[auxiliary\] 
\[adverb\] 
\[adjective\] 
\[adjective\] 
\[passive\] 
\[honorific\] 
\[capable\] 
\[capable\] 
\[capable\] 
\[capable\] 
Figure 5: Alternatives Window for rareru 
(a) SL I~ ~ ~ ~g ~ ~2 © ~: l~:~z~ 
watashi -wa kare -ga hon -o yomu -no -o tasukeru 
I TOP he SUBJbookOBJ read COMPOBJ help 
(b) I I~ he ;0~ book ~ read © ~ help 
-w& -ga -wo -no -wo 
(c) I ~:~ he reads a book ~ help 
-Wa -0 
(d) I help him to read a book 
Figure 6: Change of Generation Style 
Figure 5 shows the content of the translation 
equivalent alternatives window for rareru. It ap- 
pears when the cursor is placed on that word. If "be 
-ed" is chosen, the auxiliary is interpreted as a pas- 
sive morpheme and treated as such in translation. If 
the second alternative is chosen, it is interpreted as 
honorific. In this case, as the translation equivalent 
is shown as a blank, no morpheme appears in the 
translation. By choosing the third alternative, it is 
translated to an auxiliary "can", showing capability. 
The fourth morpheme translates it to "possibly", an 
adverb. By choosing the fifth alternative, the user 
can specify the result to be a complement of an ad- 
jective "possible". A tree for the structure "it is pos- 
sible that", coded in the dictionary, is synthesized in 
the generation module. 
The third point will be discussed in section 5.2. 
5 Extension from Basic Model 
As explained in section 3.1, the method basically as- 
sumes simple compositionaiity of translation. How- 
ever, this assumption apparently needs modification 
to be applied to broader phenomena. There are 
two major sources of violation. One is inherited 
attributes, corresponding to constraints posed by 
higher nodes to lower ones. Another is idiosyncratic 
violation of compositionality assumption, such as id- 
iomatic expressions. In this section we describe how 
the basic model is extended to treat phenomena that 
violates this assumption. 
5.1 Constraints from Higher Nodes 
One obvious example of this type of violation is in- 
flection. It is not an intrinsic property of a word, but 
a constraint by dominating or governing element. 
For this reason, its calculation is delayed until the 
last phase of generation, when all information are 
gathered at the lexical node. In addition, inflec- 
tion are re-calculated in every translation, even if 
the translation of that word has been already fixed 
by a former translation. 
Another example is constraint posed by a verb 
subcategorization frame to subordinate elements. 
Although syntactic cases can be processed by inflec- 
tion mechanism, constraint of sentence styles, such 
as to-infinitive or gerund, can not be treated in a 
similar manner. Since the sentence is a default paus- 
ing node, subcategorized sentence usually is already 
fixed as a finite form before the constraint is ap- 
plied. To cope with this problem, we provide a 
bookkeeping mechanism that preserves all partial 
syntax trees generated during translation. When 
some grammatical constraint is newly introduced on 
an already translated expression, and if it requires 
structural deformation, the system looks for the reg- 
istered structure and generates it again so that it 
meets the new constraint. 
Figure 6 shows steps to obtain a sentence with an 
embedded clause "I help him to read a book". As 
soon as the original sentence (a) is entered, transla- 
tion equivalent selection and translation region se- 
lection is presented (b). The first region is the com- 
plement sentence "he ga book wo read nd', where 
no is a complement marker. After translation, (c) is 
obtained. Then whole sentence is assumed to be the 
translation region, and (d) is obtained finally. Please 
note the change in the embedded sentence from a fi- 
nite form "he read a book" in (c) to an to-infinitive 
form "him to read a book" in (d), in accordance 
with the grammatical constraint posed by the verb 
"help". 
5.2 Idiomatic Expression 
There are some sets of words that acquire special 
syntactic/semantic behavior when appearing simul- 
taneously. These idiomatic expressions are another 
major source that violates the compositionaiity as- 
sumption of the method. Hereafter, the word "id- 
iomatic expression" is used in a rather broad sense: 
if translation of a combination of words is not pre- 
dictable from their individual behavior, we call it an 
idiomatic expression. 
In one case, cooccurring words determines trans- 
lations of one another, even though their mean- 
ing can be understood compositionally. For exam- 
ple, renraku-wo toru (contact-OBJ take) should be 
translated to "make a contact", not "take a con- 
tact" nor "get a contact". In another case, the whole 
expression can turn into completely another mean- 
ing. For example, ashi-wo arau (foot-OBJ wash) 
328 
can be interpreted as either "wash (one's) foot" or 
"wash one's hands", the latter case losing the orig- 
inal meaning of respective words. Although these 
idiomatic expressions must be recognized and trans- 
lated as one thing, they cannot be registered as one 
word in the dictionary, since their elements can ap- 
pear in a distant position, or they can also have a 
purely compositional interpretation. 
To cope with this problem, we extended the trans- 
lation equivalent selection interface so that transla- 
tion equivalents can be specified as a set for these 
expressions. Translation equivalent for the compo- 
nent words of an idiomatic expression changes syn- 
chronously when one of them is altered. Also, we 
expanded the dictionary and morphological analyzer 
to allow such multi-word translation unit correspon- 
dence. 
We give an example with denwa-wo kakeru, an 
equivalent expression for "make a phone call". This 
is idiomatic because the correspondence between 
kakeru and "make" is peculiar to this interpreta- 
tion. When the expression denwa-wo kakeru is en- 
tered, the morphological analyzer recognizes it as an 
idiomatic expression and retrieves information from 
the idiom dictionary. Figure 7 is a snapshot of al- 
ternatives window for "kakeru", in the idiomatic in- 
terpretation. The second line is highlighted as the 
current selection. The leftmost word "make" shows 
that the current translation equivalent for "kakeru", 
and the third column shows the current translation 
equivalent for the whole expression is "make a phone 
call", an idiomatic interpretation. The alternatives 
window for "denwa" is shown in Figure 8. Here, the 
word "phone call" is highlighted corresponding to 
the interpretation as "make a phone call". When the 
user triggers translation, denwa becomes "a phone 
call", kakeru becomes "make", producing "make a 
phone call" in whole. 
If the user changes the selection to another alter- 
native, say "telephone" at the third line in the al- 
ternatives window kakeru, then the selection in the 
alternatives window denwa also changes to the third 
line synchronously. Translation of denwa as denwa 
shows this word will simply vanish after translation. 
Then the translation of whole expression becomes 
an one word verb phrase "telephone". 
At the first line of both alternatives window, 
the whole original Japanese expression is shown, 
with a slash at the boundaries of words, like 
denwa/wo/kakeru. This alternative allows the user 
to switch from idiomatic interpretation to non- 
idiomatic interpretation. If the user chooses this al- 
ternative, a new alternatives window containing lit- 
eral translation appears as in figure 9. At the same 
time the alternatives window for denwa changes and 
shows literal translations for denwa. The user can 
denwa/wo/kakeru 
make \[verb\] \[make a phone call\] 
telephone \[verb\] \[telephone\] 
call \[verb\] \[call up\] 
Figure 7: Alternatives for kakeru as an Idiom 
denwa/wo/kakeru 
phone call \[countable\] \[make a phone call\] 
denwa \[telephone\] 
denwa \[call up\] 
Figure 8: Alternatives for denwa as an Idiom 
go back to the idiomatic interpretation by choosing 
the alternative denwa+wo+kakeru, at the last line 
of these alternatives windows. 
We remark that this mechanism provides a general 
means to treat translation unit with more than one 
component word. 
6 Implementation 
The method is realized as an English writing support 
software on personal computers. The main function 
is divided into two modules, the interface module 
and the translation module. The interface module 
is in charge of user interaction, morphological analy- 
sis and predicting translation equivalent and region, 
as well as function as a front-end. The translation 
module performs translation of the specified region, 
obeying user specification passed by the interface 
module. The most important requirement for the 
translation module is robustness, in the sense that 
it doesn't drop a word even when specifications are 
contradictory. In that case, the system should serve 
as a simple online dictionary. 
A prominent feature is added in this implementa- 
tion: it works as a language conversion front-end to 
an arbitrary application. The system is placed be- 
tween the keyboard and an application in the data 
flow. It captures Japanese input before they are en- 
tered to an application, converts it into English, and 
then sends the result to the application (figure 10). 
kakeru 
hang \[verb\] 
put \[verb\] 
denwa+wo+kakeru 
Figure 9: Alternatives for kakeru in literal interpre- 
tation 
329 
Any Applications (Mail, Word Processors, etc.) 
t 
Interactive JE Conversion 
t 
Any Kana-Kanji Conversion Program 
t 
Keyboard 
Figure 10: Relation to Other Programs 
This function is realized using a standard hook and 
IME API of the operating system, Microsoft Win- 
dows 95. This feature allows this system used as an 
add-on function of any application, enabling the user 
to work in a familiar document writing environment. 
The system dictionary contains about 100,000 
Japanese entries and 15,000 idiomatic expressions. 
Since there was no source available to build an id- 
iom dictionary of this size, they were collected man- 
ually from scratch following a method described in 
(Tamura et al., 1996). The essence of this method 
is limiting search space utilizing distinguished word 
classes characteristic to idiomatic expressions, re- 
vealed by an intensive analysis of these expressions. 
A CD-ROM online dictionary accessing function 
is also provided to help user's translation equivalent 
selection. 
This software is currently available either as a 
package software or a pre-installed software on per- 
sonal computers. 
7 Discussion 
Interactive method in machine translation have been 
pursued by many people (Kay, 1973; Melby et al., 
1980; Tomita, 1984; Huang, 1990; Somerset al., 
1990a; Boitet et al., 1995). In these approaches, the 
system asks a set of questions to the user to resolve 
ambiguities not solvable by itself. Among problems 
of this approach are, as Melby pointed out, exces- 
sive interaction and necessity for special training for 
interactive operations. 
In our method, interactive operations are initi- 
ated and guided by the user and all interactive op- 
erations are optional, except for a small number of 
translation triggers needed for translating compo- 
nent sentences. The system provides its prediction 
as a default selection, and other possibilities as sec- 
ond or third choices, but the user is free to obey or 
ignore them. If the selection is wrong, the transla- 
tion result becomes wrong, which is a feedback to 
the user. Then the user can undo the translation, 
correct selections, and try again (for example, see 
figure 3). On the other hand, the user has only to 
repeat "next" i~struction to obtain a result of au- 
tomatic translation quality. Frequency and content 
of interaction are determined by the user. In this 
manner, the user and the system are essentially co- 
operative, avoiding the problem of excessive ques- 
tioning by the system. The problem of difficulty in 
learning interactive operations is also avoided since 
our interactions are essentially those of simple Kana- 
Kanji conversion operations. We believe an average 
user can easily learn operations of our system. 
An interactive dependency parser reported in 
(Maruyama et al., 1990), is based on an interface 
like Kana-Kanji conversion, and shares character- 
istics described above 2. However, their method is 
limited to syntactic dependency disambiguation by 
explicitly specifying the words in the dependency re- 
lation, and it is difficult to expand the method to 
handle the types of ambiguity discussed in this pa- 
per. 
A user-driven approach to interactive translation, 
proposed by (Somerset al., 1990b), is based on cut 
and paste operations, where the content of copy 
buffer is translated when it is pasted. This method 
seems to leave too much burden to the user, since 
the user must explicitly specify which portions of the 
text Should be translated, and in what order. Also 
it is not clear how to combine partial translations of 
two overlapping expressions, except for direct edit- 
ing. 
Our stepwise conversion scheme, in which conver- 
sion proceeds from smaller structures to larger ones, 
is a natural conclusion of our try-and-error-based 
conversion approach. As Melby says, a post-editor 
will only improve by a certain increment: if the re- 
sult is completely wrong, s/he will simply abandon 
the whole result. Since it is easier to obtain an ap- 
propriate result for a shorter and simpler structure, 
a result obtained by stepwise conversion tends to be 
of better quality than a result obtained by translat- 
ing the whole structure at one step. In other words, 
our system divides the translation step into smaller 
pieces, and allows post-editing at every step. 
As described before, target users of our method 
are those who have basic knowledge to read and un- 
derstand the target language. According to the tar- 
get language skill of the user, useful support func- 
tion will be different. For example, for a user who 
is competent in English, our system will be useful 
as an online dictionary. While writing in English, 
the user can look up the system dictionary only by 
entering a Japanese word. Then s/he can enjoy easy- 
to-use interactive operations for translation equiva- 
lent selection, inflection selection and CD-ROM die- 
tionary access. When the user find an appropriate 
word, s/he only has to push the return key to enter 
the word into the original application. These users 
2These characteristics are inherited essentially from a 
Kana-Kanji conversion interface. 
330 
will also find it useful to obtain a translation equiv- 
alent expression for an idiomatic expression. These 
idiomatic expressions, either of source language or 
target language, are hard to translate since they do 
not allow literal translation and difficult to find in 
other dictionaries. By combining this idiom dictio- 
nary and translation function, the user can obtain a 
useful skeleton for target language expression. For 
many users, however, the translation function will be 
considered helpful to produce a result of the quality 
level that matches their English reading skill. Suit- 
able usage will be balanced between the user's skill 
and the capability of the system. 
The function as an add-on function to an arbitrary 
software will be an advantage equally for all users, 
enabling them to work in their familiar environment, 
compared to conventional machine-aided translation 
systems that force them to work in an independent 
unfamiliar environment. 
Finally we discuss some remaining problems and 
direction of further work. 
Translation quality needs continuous effort for im- 
provement, in both linguistic coverage and preci- 
sion. Precision of initial prediction of translation 
equivalent and translation area is crucial to the per- 
formance of the system, since they determine the 
quality of default translation. In our experience, the 
users are willing to use interactive operation to im- 
prove translation quality, but never to recover from 
incomprehensible output. 
We also have to mention some ambiguities difficult 
to resolve though basic operations of the method. 
An example is grammatical relation ambiguity be- 
tween a case element and a verb, when the case 
marker is hidden. Generally, the system treats these 
cases by producing all possibilities in the order of pri- 
ority and allowing the user to choose one. However, 
when such ambiguities are multiplied, the number 
of possibilities easily grows large, making selection 
difficult. One possible solution would be to pro- 
vide more disambiguation information, possibly a 
sequence of dialogues, to help the user to make de- 
cision. An important requirement here is that these 
dialogues must not force a response. The user should 
be able to ignore them unless they want to. 
Another further work is expanding the dictionary, 
especially idiomatic expressions. We are also plan- 
ning to add translation examples to the knowledge 
base, so that translation can be performed either 
using grammars or examples in the knowledge base. 
These examples are effective to guarantee correct- 
ness of the result, hence will be useful even for users 
not very familiar in the target language. In this di- 
rection, our system would be expanded as a kind of 
interactive example-based translation support sys- 
tem. 
8 Conclusion 
We presented an interactive machine-aided trans- 
lation method to support writing in a foreign lan- 
guage, which is a combination of dictionary lookup 
and interactive machine translation. The transla- 
tion proceeds as a cooperative process between the 
system and the user, through interactive operations 
similar to Kana-Kanji conversion method. We im- 
plemented the method as a front-end language con- 
version software to an arbitrary application. 

References 

Aho, A. et al. 1986. Compilers. Principles, Techniques, 
and Tools. Addison-Wesley Publishing. 

Boitet, C. et al. 1995. Multilingual Dialogue-Based 
MT for Monolingual Authors: the LIDIA Project and 
a First Mockup. In Machine Translation, vol.9(2), 
pages 99-132. 

Canisius, P. 1977. Automatic Partial Translation in 
a Multilingual Information System. In Overcoming 
the Language Barrier (Third European Congress on 
Information Systems and Networks), vol.1, pages 259-269. 

Huang, X. 1990. A Machine Translation System for 
the Target Language Inexpert. In Proceedings of 
COLING-90, pages 364-367. 

Kay, M. 1973. The MIND System. In Natural Language 
Processing, Algorithmic Press. 

Maruyama, H. et al. 1990. An Interactive Japanese 
Parser for Machine Translation. In Proceedings of 
COLING-90, pages 257-262. 

Melby, A. et al. 1980. ITS: Interactive Translation System. In Proceedings of COLING-80, pages 424-429. 

Muraki, K. et al. 1994. TWP: How to assist English production on Japanese word processor. In Proceedings 
of COLING-94, pages 283-298. 

Somers, H. et al. 1990a. Machine Translation without 
a source text. In Proceedings of COLING-90, pages 
271-276. 

Somers, H. et al. 1990b. A user-driven interactive machine translation system. In Proceedings of 
SICONLP-90, pages 140-143. 

Tamura, S. et al. 1996. Collecting of Verbal Id- 
iomatic Expressions and Development of a Large Dic- 
tionary for Japanese-to-English Machine Translation. 
(In Japanese) In Proceedings o\] 2nd Annual Conven- 
tion of Association for NLP, pages 45-48. 

Tomita, M. 1984. Disambiguating grammatically 
ambiguous sentences by asking. In Proceedings of 
COLING-84, pages 476-480. 

Yamabana, K. et al. 1995. Interactive machine-aided 
translation reconsidered --Interactive disambiguation 
in TWP-- In Proceedings of NLPRS-95, pages 368-373 
