D~IGN ISSUE OF ECTST 
IIuang Jianshuo 
Language Training Center 
South China University of Technology 
GuamQzhou 510641, P.R.C. 
ABSTRACT 
This note presents an overview of the 
English-Chinese Translalion System for 
Tourists (~I~q) currently under de- 
velopment at South China University of 
Technolo~/. A brief description of 
the bilingual dictionazv is given, 
followed by descriptions of grammar 
rules representation and the n~in 
processes involved in translation. 
S~TEM CONFiGUR&TIOR 
ECTST consists of a translation pro- 
gram, a bili,4Tual diction~w and a 
rule-data base. The program body is 
composed of a sequence of modules, 
performing translation in four phases: 
SL sentence 
Initiation ~ 
Tt, sentence 
The initiation pha~e is concerned with 
input of the SL sentence, diotionsry 
look-up and morphological processing. 
The analysis phase involves syntactic 
and semantic parsing, which are accom- 
plished through linguistic models and 
case frame. The transfer phase com- 
prises rules for converting the parsed 
SL sentence into the TL sentence. ~e 
generation phase, applying the con- 
tents obtained from the previous pro- 
cess, generates the TL sentence. 
ECTST is implemented in PASCAL. The 
software is separated from the lin- 
guistic data, so that any language can 
be accepted if the grammar rules and 
vocabulary are given. Likewise, the 
system program are free to o:hange 
their components with no undue res- 
trictions that may be imposed by the 
linguistic components. 
DIdTIONAP¥ ~,ONT~PFS 
The dictionary is bilingual: it con- 
tains morphological, syntactic and se- 
mantic information needed for the 
analysis and generation of a sentence. 
This includes information about the 
category of words, their semantic fea~ 
tures and case frames. Entries in %he 
lexicon are of eight types, and their 
semantic features are based on graded 
concepts. The case frame provides a 
means to find out the logical rela- 
tionship between components in the 
sentence. 
At present, the dictionary contains a 
limited set of lexical entries, which 
are grouped into models with the ini- 
tial letter as the index for access° 
GRAMMAR RUL~ 
In ECTST, rules are represented in the 
form of meta-model. It may contain 
one or more data items as shown below: 
A \[~\](a) B FY\](b) C \[~3(c) 
376 
~,'here #~B,C Pepresent in:Formation on 
caLe.~oz'y of the word, while x,ytz aald 
a, t h, c are parameters ind J. oatin~ multi- 
p\].e pieces of informat ion amd %z:ee 
stl.uoture ,~eparate\].y. In many oases 
semantic f'eatures or inf'ormatzon on 
~qram:matioa\] ,'~eT4J.er~ number or ea,,~e ca.n 
i,e useO as para;'~etez~,:~. 
,generally, hmxlreds or even thousamds 
of grm'm;ar rules are necessa~?/ for 
lan~1~.a.~e analysis \[~!d tramsforma, tion. 
I% is impr'aotical %0 check whether all 
these rules oa/% be applied to all in- 
termediate processings. To make %he 
pro.certain:?/ effi<~ien%,ru\] es ~z\]-e, qroupe@ 
into packe1:,s~ each degding wigh speci- 
fic oorrespondir~£, iirl£<uistic phenomena 
rai\].ging from ~.~ phrai',e, a clause to 
a ~:, entence. 
The work of a~a/ysis :is perfo>med !<; 
the parse r.ioc\]ule, \]:% constJtl~tes a 
unified syntactic al.,c, semantic anaJ.y- 
si;~ of "the iz'pui sen%£.t:ce, ~:d its 
Ot'df.\]:\\]~ \[erVCr: ~;;~ 82\] : '""~ * " ~ " ~' .:L.i.~,, ,,0 <,r&,i.. ~ t:\] ~. 
Th,:> r';~r'.'i>': >regr~..~: .J~%eeprels 1:lye ir- 
a-r: ,-e>q:e>.ce g-:~<:tCC2"~5.E,~ ~0 't\]e SVU.J.\], ~\],-- 
b.\].,' ii:':"..\]rtio ~.~.O<!e\]g. q'}_c.ge oo!~e J...'l 
s e ;.ers,\] v&pl £/::i_e£' ~ ~R: o,: &s ~ne% a- 
,I. ~; J.,LF: ~'\[C.. 
rT'h,- rnp:-:e :40d,t\]O; OOY.d-:i~4.;:h 0~' 5 T<ur})er 
o1:' ro:~9ines ~ each 1:'espo.:t~:J\]::le for r~ 
sT,eoi<~l wozJ' . O,':,e z'o!:i.ne < Or ex.~'~T~\].e 
!::oY.fOr'r~F{ :..-iopphol o~(:i, o&l ,:~L.s.\].$~s i's r.)f 
W':)\]:'!:~\[: ~ ,'-~-<.0 ;, ::.es. , ~Of 
~a:~%.io r.ng:i.~7")" oT ::><t"~::'se,:'~ :'ti:{! nN- 
other, f'o:r ap2J.ioatior~ of o~:-;e fra.o.e£, 
and so on /~n,~! ~,s; s &re ,---,,-~i e, ~ -',- -~ 
:\['roTr~ JJ~\]e SLI\]?SeG~ s truoture %Lrou.d: %he 
de~c r-;irtlotur.e. A dee-~ Ptrt~etu.re ma~.; 
comprise roles of siz e:'~entia\], types: 
SU\]!j eat, (.;:8J eat ~ \]~Vol,~ement, \[:: ITna%i on, 
ST/i.~e and CAU,~alion.o They make up a 
basic fra,me wit\]~ the ~%'EDic~tor as 
shown ~n F:j.~;o 1. 
In ap£1icstion, a. role rosy come into 
it. ~, subc-:~%eqories v,d",er:,ever neoess~;, 
Slq'ua%ion, for e:<amoie~ may fa_l\], into 
-~r,~ , . - ri, ~ TIi'Je, LO,.a,~xor~,~ \]).,:,<pee c,r .gONdition~ 
:(NVo\].vement into I{vfiipient~ PARt:Lot- 
pant or GOAl ~6md CAUsation ln%o \]9.A~i;:~ 
T U\].q.-,ose or. R EF.uIt ano so One 
',::~ U \] ', \] R, £, 0 i',,\] 
Fig° 1 
in analysis, the deep structure is 
obtained via preferei'~tial weightincc 
oacala.ted from ~rammatical data m~d 
frame labels. 'T}~e fo!lowi~c sent~noe~ 
for exa;nF,\].e would resul% in %'.co dif~ 
feren% intern~l s!ruotures even ±hough 
their slEPf&oe struct1~res ~re sJ.mi\]ar. 
I. We found John ve~, well. 
2. We know John ve.~ ~ wello 
in the case frame, tLeir deep struc- 
tures would be: 
When an intern~l strimture results, i% 
serves as the basin for transfer. 
TRM~SFF~ AND GIi~i~I~ATiON 
In the transfer sta~e, transformation 
is accomplished in two steps" firs% 
the internal structure of the Fi, sen- 
teuoe is generated into a tree, with 
nodes that indicate their proper me- 
msntic and gr~r.atica\] order in th~ TL 
sentence; then lexicon rules arc in- 
voked to transfer the SL entrT- ~ on the 
basis of its context, in%o %he TL 
entry. The advantage of such a pro- 
cedure are obvious. It ~kes the 
r~u\].es more flexib!e9 especially those 
oopin¢ S with sentences with similar in- 
t ernal s tructl/res. Moreover p ~ es 
can be added, modified or changed as 
needed with no resultant effect on one 
another. 
When a target tree is built up, it is 
then scanned by way of left recursion 
377 
and with words appended to its termi- 
nal nodes a~ required° Finally~ a 
oh~r~cter string is obtained; its out- 
put is ~ sentence in TL. 
CONCLUSION 
Efforts made over the past decade h~ve 
achieved co~miderable progress in 
machine translation. First syntactic 
parsing was pursued, then semantic and 
context analysis was advocated. ECTST 
has profited from both theories. 
Natural language is essentially 
discrete information system. It often 
al\]ows multiple syntactic interpret&- 
tions. To mini,~ize the possibility of 
multiple interpret~.tion, we introduce 
a mioimum amount of semantic informa- 
tion and .adopt the case frame. This 
provides flexible f~cility in .<vnt~c- 
tic analysis and helps to distinguish 
st~otura/ mmbiguity, in this ap- 
proach, the translation accuracy is up 
to 9C,~)~ or more, which can be raised if 
correspondzng i~'orn~ztion and/or rules 
are modified ar/l supplemented. 

References

I. Pan Yu a~nd Yuan Y. Sung (87). 
"Automatic Ehglish~Chinese Transl.- 
tot," Procesding of 1987 Interna- 
tional Conference on Chinese in- 
formation Processing, oP. 52-61 

2. Jonotha~ Sloctun Siemens Corpora- 
tion and Wit_field S. ~ennett (82). 
"The L}IC Machine Tr~iation S~3~s- 
tern: An A~plication o~ \[~tate-of-the 
Art Text an<-!. ~,~eA~omal l,az:g1~age Pro- 
cessing Teokniques to the Transl::~- 
Lion of Technio~l ~,'~un;~a&s~" Work-. 
i,.~" }:'aper L~{3-$2-I July 19,:.92 

3. :'~ei-Du ! i ~ud G-ax2z q. Erickson. 
" ~,\[oduJ.ar Diotions/.~ Desi.~n For 
Chznes e-To-~\]r~lish !'~aohine Transla- 
tion." GARJA~ Resesmoh Inc. 990 
Highland Drive, Solaria Be~ch, C~li- 
........ , o 249-252 florr.za y,-~ 5-,:-47~, ::'.'P. 

4.  Huang, Jianshuo 87. "Language
}"eat',,zre, ~ and Semantic Inform&tion 
in /"\])," Proceedings of" ICCIP' 37t 
pp. 110-116

~-' ~,ta~oto .!{~,AO (~;92)~ "A \[:urvey of 
Natural Language Processing and Ma-
chine Tz,~lsla,tion in Jaymn~" Compu-. 
f ~ y~ ter ,~\]cienoe :! Techz,~ology~ PO. <>4-(,, 
T. Kil a,.~w},~( e{J ~ ) OHMSHA, I,TD. m 
Nort h-Hell, ~u~d Pub. Co. 

6. !.!ako%o ~\[A(}AO ('~"~,.>o) . "'~,,1,\].I~. ent S+~tu~'. 
and ~,n& l,%:,'t:zre Trends ~,~, ~4aohJ.~e 
Trans\].a,tion~" Future Genera%ions 
Computer \[:7<stem 2 (1986) PP.7{-,}%2o 
Elsevier Science ~ublishers B.V. 
(1,\]ortb Hol!\[~d) 
