A BIDIRECTIONAL, TRANSFER-DRIVEN MACHINE 
TRANSLATION SYSTEM FOR SPOKEN DIALOGUES 
Yasuhiro SOBASHIMA, Osamu FURUSE, Susumu AKAMINE, Jun KAWAI, 
and Hitoshi IIDA 
A TR Interpreting Telecommunications Research Laboratories 
ABSTRACT 
This paper presents a brief overview of the 
bidirectional (Japanese and English) Transfer- 
Driven Machine Translation system, currently 
being developed at ATR. The aim of this 
development is to achieve bidirectional spoken 
dialogue translation using a new translation 
technique, TDMT, in which an example-based 
framework is fully utilized to translate the whole 
sentence. Although the translation coverage is 
presently restricted to conference registration, the 
system meets requirements for spoken dialogue 
translation, such as two-way translation, high 
speed, and high accuracy with robust processing. 
1. INTRODUCTION 
Transfer-Driven Machine Translation\[ll,\[2\],\[9\], 
(TDMT) is a translation technique which utilizes 
empirical transfer knowledge compiled from 
actual translation examples. The main part of 
translation is performed by the transfer module 
which applies the transfer knowledge to each input 
sentence. Other modules, such as lexical 
processing, analysis, and generation, cooperate 
with the transfer module to improve translation 
performance. With this transfer-centered 
translation mechanism together with the example- 
based f'ramework\[3\],\[4\].\[5\], which conducts distance 
calculations between linguistic expressions using 
the semantic hierarchy, TDMT performs efficient 
and robust translation. 
TDMT is especially useful for spoken language 
translation, since spoken language expressions 
tend to deviate from conventional grammars and 
since applications dealing with spoken languages, 
such as automatic telephone interpreta- 
tion\[6\],\[7l,\[S\], need efficient and robust processing to 
handle diverse inputs. A prototype system of 
TDMT which performs bidirectional translation 
(Japanese to English and English to Japanese) 
has been implemented. This bidirectional 
translation system simulates the dialogues 
between two speakers speaking in different 
languages (Japanese and English) using an 
interpreting telephone system. Experimental 
results have shown TDMT to be promising for 
spoken dialogue translation. 
2. TRANSFER-DRIVEN 
ARCHITECTURE 
The bidirectional TDMT system, shown in Figure 
1, translates English into Japanese and Japanese 
into English. Conversion of the translation 
direction is simply done by flipping the mode 
selection switch. Moreover, all of the sharable 
processing modules are used in both translations. 
This bidirectional translation capability, along 
with other features adopted tbr spoken language, 
shows the possibility of two-way dialogue 
translation. 
The transfer module, which is the heart of the 
TDMT system, transfers source language 
expressions into target language expressions using 
bilingual translation knowledge. When another 
language-dependent processing, such as lexical 
processing, analysis, or generation, is necessary to 
obtain a proper target expression, the required 
module is called by the transfer module. In other 
words, all modules in a TDMT system function as a 
part of or hell) the transfer module. This transfer- 
centered architecture simplifies the configuration 
as well as the control of the machine translation 
system. 
English Terminal 
Japanese Terminal 
Fig. 1 Configuration of Bidirectional TDMT 
System 
64 
3. TRANSLATION MECHANISM 
3.1 Examplc-based analysis and transfer 
The TDMT system utilizes an example-based 
framework to translate a sentence. The central 
mechanism of this framework it; the distance 
calculation\[4\],\[5\], which is measured in terms of a 
thesaurus hierarchy. We adopt the calculations of 
Sumita\[51. 
Figure 2 shows an example of the transfer 
knowldege tbr the Japanese pattern "X no Y." (X 
and Y are lexical variables and "no" is an 
adnominal particle. X' represents the English 
translation for X, and the English translations are 
noted in braces after the Japanese words for 
readers' convenience.) 
X no Y --~ 
Y' ofX' ((ronbun {paper}, daimohu {title}) .... ), 
Y' for X' ((beya {room}, yoyaku {reserwltion}) .... ), 
Y' in X' ((Tokyo {Tokyo}, haigi {conference}) .... ), 
X' Y' ((en{yen}, heya {room}) .... ), 
Fig. 2 An Example of Transfer Knowledge 
The first transfer knowledge, "X no Y -, Y' of X' 
(ronbun {paper}, daimoku {title})" represents the 
translation example that a set (ronbun{paper}, 
daimoku{title}) in structure "X no Y' is transferred 
into structure "Y' of X'." Thus, pattern selection is 
conducted using such examples. When the source 
pattern "X no Y" is applied to an input, the transfer 
module compares the actual words tbr X and g with 
the sets of examples, searches for the nearest 
example set with its distance score, and provides 
the most appropriate transferred pattern. 
For example, if the Japanese input is "Kyoto no 
kaigi" and the nearest example set (X, Y) would be 
(Tokyo, kaigi), then the transfer module selects "Y' 
in X'" as the translation pattern and outputs 
"conference in Kyoto." Thus, bilingual transfer 
knowledge consisting of patterns and translation 
examples is used in TDMT to select, the most 
appropriate translation. 
To analyze a whole source sentence and to form its 
target structure, the transfer module applies 
various kinds of bilingual transfer knowledge to 
the source sentencet.ql. Figure 3 is a list of different 
types of bilingual transfer knowledge (patterns 
and words), that are used with examples to analyze 
the Japanese expression "torokuyoushi wa sudeni 
o-mochi deshou ha" and to form its transferred 
result "do you already have the registration tbrm." 
As shown in Figure 4, both the source and target 
structures are formed based on the distance 
calculation. The numbers in brackets represent 
the transfer knowledge pairs in Figure 3. 
X dcshou ka --~ do you X' (1) 
X wa Y - ~ Y' X' (2) 
sudeni X --, already X' (3) 
o-. X -~ X' (4) 
torokuyoushi ~-~ registration form (5) 
mochi -~ have (6) 
Fig. 3 Various Kinds of Transfer Knowledge 
As we have seen, the example-based framework is 
employed in the transfer module of the TDMT 
system, in which bilingual transfer knowledge is 
used for both analyzing and transferring the input 
sentence cooperatively. In other words, in the 
transfer module, both the source and target 
structm'es are formed by applying the bilingual 
transfer knowledge extracted from the example 
database. 
Source Sentence: "torokuyoushi wa sudeni o-mochi deshou ks" \] 
4, 
Source structure Tarzet st,uetu,'c 
×de ho,,l.  \] ---3---- r \] do you X 
(4) ~--~___ (3) ........... (3) -- \[ 
torokuyoushi \] \[ ....... sudcni X \]llii!i!!i!i!iiiiii::il}it ~Y_._X~ I registration fi)rm \] I 
(6) \[ mochi 6 
Target you ah'eady have the registration form" \] 
I 
Sentence: lldo 
Fig. 4 A Transfer Example for "torokuyoushi wa sudeni o-mochi deshou ka" 
65 
3.2 Structural disambiguation 
Multiple source structures may be produced in 
accordance with the application of the bilingual 
transfer knowledge. In such cases, the most 
appropriate structure is chosen by computing the 
total distances for all possible combinations of 
partial translations and by selecting the 
combination with the smallest total distance. The 
structure with the smallest total distance is judged 
to be most consistent with the empirical 
knowledge, and is chosen as the most plausible 
structure. (See \[9\] for details of the distance and 
total distance calculations.) 
For instance, when the pattern "X no Y" is applied 
to the expression "ichi-man en {10,000 yen} no 
heya{room} no yoyaku{reservation}," there are two 
possible structures. 
1) ichi-man en no (heya no~) 
2) ( ichi-man en no heAL ~) no ~ 
The TDMT system calculates the total distance for 
each of the structures 1) and 2) using the bilingual 
transfer knowledge stored in the system. The 
following are the target structures when the 
transfer knowledge in Figure 2 is applied. (Source 
structures 1 and 2 give target structures 1' and 2', 
respectively.) 
1') 10,000 yen (reservation for room) 
2') reservation for (10,000 yen room) 
In this case, (en{yen}, yoyaku{reservation}) in 1 is 
semantically distant from the examples of "X no 
Y," which increases the total distance for struc- 
ture 1. Figure 5 illustrates the two sets of source 
and target structures generated by the transfer 
module of the TDMT system. 
3.3 Sentence generation 
The generation module completes the translation 
of the transferred sentence using target language 
knowledge that is not provided at the transfer 
stage. This module performs the following two 
tasks in cooperation with the transfer module. 
1) Grammatical sentence generation: 
It determines the word order and 
morphological inflections, and generates 
lexically essential words, like articles in 
English, so that the whole sentence is fully 
grammatical. 
2) Natural sentence generation: 
It brushes up the sentence by changing, 
adding, or deleting word(s) so that the whole 
sentence is as natural as a spoken dialogue 
sentence. 
Figure 6 shows an example of Japanese natural 
sentence generation where addition of a polite 
auxiliary adverb and deletion of redundant 
pronouns take place. 
~ "I will send you the form" 
Transfer I "~ 
~:~:~:~:~:~: watasht-waanata-n~ youshi-wo okuru" 
iiiiii!iiiiiiii I {I~,= {to you} {the form} {send} 
Generation Jl clglete ~ ......... ....,.add 
' ~£.,,. ,, ..... fit" 
"youshi-wo o-okuri-shi masu" 
{the form} {send} + politeness 
Fig. 6 An Example of Natural Sentence 
Generation 
en no heya noyoyaku" \] 
! Source Sentence: 
"ichi-man 
i iiii~!~ i~i~!~iii!iii~!i i~ !i !ii':!:E:i! ''''~' ~' ~'"""""""""""""""'""""~' ~"~' ~'~"'~'~":':~i~i ~i!ii~ii!iEi!i~ili!i~!!i~i~ili~iiiii!!~!iiii~i~i~!~i~!iiiiii~!i~ i~ili~ i i~i~i~i ~i~il i~ i lil i~i\[i!ili~i~ilili!ii~iiii!i!iii!iii',i~i ~!ii~iil ~i~ff.i~ii~iiiii~ii!~iiiili~ i" ~':'~' "~': ................. `....`.``.~.......~.~..~...~.~..`....~.`...~.~`~`~.~:~.È~.~.`.~.~!i':i~i~i~i~i~i~i~ii iiiiiiiiiiiiiiiiiii Source structures iiiiiii\[iiiiiiiiiiiiii::~i~i~i:: i~:i::ii :::::7:::::::::::::::: Target structures iiiii:::::::::::: 
::: :: : :::::: : : : :: :::: :::::::::::::::::::: ::: 1) X no Y ......................................................... 1 ) ' ....... 
........................................................................................................ : XY 
iiiilil ,/ " "~ I structure pair .:iil ,-I -~, 
I ichi-man en X no Y (totaldistance = 0.5) :~i 10,000 yen Y for 
~' ' ......................................... iii!::~ ', reservation "~ 
iiiiiii " \[, ~ structure pair 'ii ~ 
iii ichi-manen ~ liiiii:ilil iiili 10000y:n ~ 
-F 
Target a 10,000 yen room" \] Sentence: "reservation for 
Fig. 5 A Disambiguation Example for "X no Yno Z" 
66 
4. IMPLEMENTATION 
4.1 System specification 
The bidirectional TDMT system has been 
developed in Common Lisp and runs either on a 
Sun Workstation or Symbolics Lisp Machine. Ti~e 
dictionaries and the rules are made by extracting 
the entries from the ATR corporaU0l concerning 
conference registration (Table 1). 
Table 1 System Specifications 
Hardware Sun Workstation, 
Symbolics Lisp Machine 
Software Common Lisp 
Translation Japanese to English (J-E), 
English to Japanese (E-J) 
Domain Conference registration 
4.2 System operation 
Figure 7 shows a screen shot of the bidirectional 
TDMT System on a Symbolics Lisp machine. The 
system simulates the communication between an 
applicant (Japanese speaker) and a secretary 
(English speaker). The dialogue history is 
displayed at the bottom of the screen. In the 
screen, Terminal A is the applicant's terminal and 
Terminal B is the secretary's terminal. The 
translated sentences are displayed in reverse video 
(white on black). 
4.3 System performance 
The system has been trained with 825 Japanese 
sentences for J-E translation and 607 English 
sentences fl)r E-J translation. These sentences 
were selected from the ATR corpora and typical 
dialogues about the domain. The system can 
translate sentences with a vocabulary of 1500 
words. In J-E translation, the system on a Lisp 
machine with 10MIPS performance has provided 
an average translation time of 1.9 seconds for 
sentences having an average length of 9.2 words, 
and a success rate of about 70% for open test 
sentences of this domain. 
7-LDM T "" "':'- ......... 
\] >)> analysls 
Clear Window All Carldldate5 OCAL-.lRllltUlORHMlI\[\]rl , 
Model Conversations Morphemes (~1,~\[ IIITI:RJ J.t:" ~-~')" 
Oemo Sentences Structures >> > tf-ar~sfer- 
Free Input ,Show Color lransfer (~,~ ~r~f\[!RJ ,~,~" ~-# ) 
Quit Color Transfer 
TDMT Ll~t~nor ~ EOdhL:'~f~S~'~ll~M ... 
ELEL) H tl.OI)O005 : (~1~,~,~ J,t:') => (I.~I~,L IHTKRJ #,t-) 
{'t) Bt£~i~'-~'h') ~4TOTnL DZStftriCE = O.OOUOO5 
f3 
( TERMIHAL-A )>>> 
( I ERMIIIRL-R >>>> 
1ERMINRI.-II Yes i~ Is) >>>> 
:TERIIIMflL-B May I help you} >:,>> 
>>>> 
(fERMIMRL-fl Ev32.-)~'ff-I~£~ltI:~Z.SL~-C'L ,t 5h,) )>>> 
(IERMIIIRL-B You hu~t u~e a i'egl~t.ration £orm) >>>> 
(TERMI~IIL-8 Do you have one) 
( l ERMItIAL-A L~ 3~#\[-~t" ) 
":GUGb" 
• E- ~J 6 {~,~g~g~ -C ~-)2" 
\[ ,mr, . :1" ) 
\[.m':i;~ Irl:FI t~r;t i1~1 h'~ ~,\]: nt ~ VT t V\[~; \['l/.'~*Fi~ q, i t ,,t m p I 
fran~ let" 
0.13tjtJOUO : ('/H -(-~) :> (I}O .... (J) 
{}*tlOfJtlllO : BI\[: : (3.1~) :> (\[I\[IV not yet) 
TOTAL DlSIl}lICEi - \[¢,O000k}{} 
\[ r31~ ~ rill J'Tl"47ITI ~i\[ l"ql 
~Yes It |.~" 
~May I help yell ~ 
~YmJ mist u~e a re~lstr'atlnn rerM* 
"De yell have el~e N 
III IIII Ir , 
Fig. 7 Screen of the Bidirectional TDMT System 
67 
5. FUTURE WORK 
We have presented an overview of a bidirectional 
TDMT system that simulates communication 
between an English speaker and a Japanese 
speaker. The system performs efficient and robust 
processing by utilizing an example-based 
framework. 
Future work will include: (1) the introduction of 
contextual processing, which can cope with spoken 
dialogue utterances, (2) application of TDMT to 
other language pairs including Japanese and 
Korean, and (3) integration of the TDMT system 
with speech recognition and synthesis processors 
to achieve a fully automatic telephone interpreting 
system. 
\[18\] Waibel, A. and Woszczyna, ~.: "Recent 
Advances in JANUS: A Speech Translation 
System," Proc. of IWST '93 
\[9\] Furuse, O. and Iida, H.: "Constituent 
Boundary Parsing for Example-Based 
Machine Translation," Proc. of COLING-94 
(1994) 
\[10\] Ehara, T., Ogura, K. and Morimoto, T.: "ATR 
Dialogue Database," ICSLP-90, pp.1093- 
1096(1990) 
ACKNOWLEDGEMENTS 
The authors would like to thank Eiichiro Sumita, 
Naoya Nishiyama, Hideo Kurihara, and other 
staff members for their help and collaboration. 
Special thanks are due to Kohei Habara and 
Yasuhiro Yamazaki for their support of this 
research. 
REFERENCES 
\[1\] Furuse, O. and Iida, H.: "Cooperation Between 
Transfer and Analysis in Example-Based 
Framework," Proc. ef COLING-92, pp.645-651 
(1992) 
\[2\] Furuse, 0., Sumita, E. and Iida, H.: "Transfer 
Driven Machine Translation Utilizing 
Empirical Knowledge," Transactions of 
Information Processing Society of Japan, Vol. 
35, No. 3, pp.414-425 (in Japanese) (1994) 
\[3\] Nagao, M.: "A Framework of a Mechanical 
Translation Between Japanese and English by 
Analogy Principle," in Artificial and Human 
Intelligence, Elithorn, A. and Banerji, R. 
(eds.), North-Holland, pp.173-180 (1984) 
\[14\] Sato, S. and Nagao M.: "Toward Memory- 
Based Translation," Proc. of COLING-90 
\[5\] Sumita, E. and Iida, H.: "Example-Based 
Transfer of Japanese Adnominal Particles into 
English," IEICE TRANS. INF. & SYST., Vol. 
E75-D, No. 4, pp.585-594 (1992) 
\[6\] Kikui, G., et al.: "A Spoken Language 
Translation System: ASURA," Proc. of IJCAI 
'93, Vol. 2, pp.1705 (1993) 
\[7\] Morimoto, T., et al.: "A Spoken Language 
Translation System: SL-TRANS2," Proc. of 
COLING-92, pp.1048-1051 
68 
