'I'RANSTAR ...... 
A COIglERC I AL ENGLI SI\]-,Ctfl NI~E lit SYSTEN 
Dong Zhen Bong 
Language Engineer ing \[,abs 
Ch ina Software Technique Corp, 
TfiANSTAI:t is widely acknowledged as tile 
~flost su¢',c,~ful and l:~)werfut HT syste,i 
i n Ch ina, I t has won favourab I e 
react ions frotll the doltlostral: ions in 
Sin&~l.~re (1986), ttong t(ong (1987) and 
tlanover (1989\]. The syst6~ti has been 
sold to over 50 users l~)th at l~c~,e 
and abroad since its relc~se in 
~pt~be..r 1988. lhe developmnt of 
the s!/sta~, from basic design ~() 
cc~lr~lercial izat iot\], took ~l~ore than 11 
y~rs. 
I. Ctlrre\[\]\[ .... ~(., ~ f i cat ions 
The current speciflca!ions of \[he 
systeal are as fol tows; 
.... BIG ionar ie~ 
.Basic dictionary; 4,5,000, including 
words,phrases, idioms and proverbs etc, 
.T~:hnicat-term dictionaries; 200,000 
t~:',hni(~l tenm, covering a wide range 
(,t' subject fields, sud\] as ecbnOllliC~S, 
co,~put a t i ona I t (~hno 1 ogy, t e 1 eco~un i ca 
t ions, (.,~lr ~lanufacture, acx)tlsl:ics etc. 
--- Hardware 
The ,<.;yste~ll can run on the fol lowing 
machines; IBII AT and its c(~¢~-~tible~, 
VAX series, Wang PC, I BIq S/20, 
Ol~vittJl PC, Itniver~ 68000, etc, with 
COBOL as its prograa~ling lar~guage. 
Tile develoNJent of a new version on 
IJNIX wi t h C language is being under way. 
- Translation Speal 
The translat ion si~zt varies great ly 
with different types of i~achine,s, appro o- 
ximately ranging rr~, i,ooo words/llour 
to 3,000 words/hour. 
.... 'Fans I at ion Accuracy 
General ly, Lhe translaL ion acx:urac, y, 
when \[io pre-exl i t ing and post-edi t ing 
are appl led, is "0t{", as the Si\[lOi"~i<t'.; 
claimed at l|ai\]over Sllov in 1989. "File 
accuracy depends f, uch on the subject 
fields or styh~ of the texts to be 
translated, For example, the translation 
is terribly Ix)or for news rel_~arts. 
--- Faci I i t i~-~ for End-users 
, Inpu{ or output docu,Dent tflanageJ~enf 
tools; to be tlsc~i to help tl,.~rs to do 
pre-~.~litJng (though minilllal), and post-. 
ezliting in two-coltllllll display. 
, Dictionary r~laintenan(.,'e t~x)l: to be, 
used For u~rs to a~llond the diet ionari~ 
in the following rays; <1> to add new 
words or terrm by giving a te~plate, 
rattler than raanual coding for detailed 
dictionary infonmtion; <:2> to ut:~late 
Chinese eauivalents to the English 
enteries, il' they are tltought to be 
unfit to the user's taste. 
2,0very Jew of '\[RANSTAtl 
T\[1ANSTAI\] is a transfer-based, batch- 
opera t J ng un i d i rec t i ona I Eng I J sh-Ch i nose 
l'lf syste~l. Tile configurat ion and Hie 
operat ion process ot" the syste, m are 
outlilled in Fig. 1. 
1 339 
i i ""~ SL G.S. 
- jq _ L _...) 
" -'~,- TLOutput ) 
Fig.1 
ED - English monolingnal dictionary 
BD - English-Chinese bilingual dictionary 
B1 - Rule-type 1 used for analysis 
R2 - Rule-type 2 used for transfer 
SL Input - as "They studied hard." 
SL G.S. - SL gra~atical structure, as "l'hey study hard." 
L.S.S. - Logical semantic structure, as ~ 
Boot, action past 
Agent Manner 
TL G.S. - TL grammatical structure, as Agent < Manner < Action 
TL Output - as "~\]NJJ-~SJo " 
3. Dictionary Configuration and 
Information 
In "rRASSIAB,the basic dictionary or each 
of the technical-term dictionaries 
consists of tvo sub-dictionaries: an 
English monolingual one and an English- 
Chinese bilingual one.Any technical-term 
dictionaries should be used in combina- 
tion with the basic dictionary. When 
the system runs, it should use two 
operating dictionaries. The dictionary 
configuration in TBANSTAR is shown in 
Fig. 2. 
fEnglish monolingual\ 
basic dictionary~ \ 
kE-C bilingual~operating El) 
teclmical-term ~English monolinguaVj~.operatingBD 
dictionary KE-C blli\[Nual ''~ 
Fig.2 
340 2 
\]'he il\]ternal strtlol.ilr'es arid iilforlrlatJoi\] 
{y~,.s for both -the basic dictionary 
and te, uhnical-ter¢ dictionary are just 
• i:he sai~e~ The Jrfforiilat ion giveri to eac:h 
oiltry is al lo£;ated ii\] differerie iriforliia.,- 
l:ion zon(~s: mor'photogicat,, syntact it:, 
,.;emant ic and transfer zenes, Each Entry 
in ED's iiiay have 26 ir\]Pol"l/lat iofi items, 
tr\] tim syslelii's dictionary o:~pilii\]g 
tilafll.la I, t he i rl Potilia t i orl sp~,,c:i P i c~::t tot a 1 s 
abol.lt 630 i t (.;illS, i rioJ tii_1 ing 3<70 i t ellis 
relating lo iliO,,"\[%')togy and syiltax, and 
240 i t tHiS 1. 0 setiiai\] i i os, 
.. % .' ~7' T) 4, Eule I:,doo 
, th,~,;'<,T,tt~ 3f'e S\[Ol"(~(J it} :\ii the. rules of " .... " ;" 
lwo r ~l,:'h"<~ ><.'- ;~,,e Off thelii i,.<, a ,syritax- , . k.'l t(l~.).,'k) t /.\]J I .J 
dr jverl \[ ul .b:l,,,., (,<31)lIB), " ~ .... "~' arid the otlier 
le×icon driven ruteb~se (I.DItB), ll\]e 
forilier deals nith gen(.',ral probleai!s ir\] 
tile \[.trocos,.<:~, whJ le the latter take.':; 
care oi" Sl:~ac'.ific l ingtlistic pheno~mr\]a, 
sucll as idiomatic usage of ~ords, 
verbal plTa~.~s, and aN) igu i ty vllicl\] 
can only be solved by illealls of s~lialit i(', 
analysis, ere, Tile rules in SI)RB 
currerttly total alx)tlt 3,200, and in 
LI)~B, about 2,600, Horeover, 1'}/ANSTA~I'D 
rules cnn t~-'~ diviO:~l ifltO t~'o types 
i!t ler'lllS oC llic:,ir use; iil ihe operat i().r~ 
c;.,.f tile systei~i, Type l is used for 
analysis, ariel I'yB3 2 for trarisfer. Me 
rtitce~ iri S\[)RB are organized in 10 IllaJll 
i\]todules ',:ith 86 Slibifl(xJtile8 included 
(See Tab le ! ), 
Ft.M(I i(;,ii<:~ Si.lblll(~tlle Nllm\[~rs 
i.lliId!o!4Ii-~;(;,r(\] t fea { ff!>,':}rf { 3 
t\]ofilo£.raph d J,%dlliLI J.<Zua { JOI\] 21 
tq ,..m, <. <.,~ "" t...., ,,)r VCT i3a i dot OFIll l IKI { J ()i I 
r" ,:, ~<~+~ %C,).~ lilarl< i l\]g t ks I (Jtlc~,v u. -~ I- ' 
Phrase ~,-'~-~ - '<';r 1 ....... .... !r: ..,, i O~ 7 
Co-ordi£ato r,-,,.~<.t,- ~.t ~c ...... ~,..~., u.,., ~ .g} 4 
' +, ' ',,...~, L~... -, \] --~+,it ar l,., ..... i( reiat iol\] i 9 
\]-.onr,c j t '~. , ", 'o g' i }.I u;,,.,.:, iOli p!IFaSO dl~a,sl\[,la.th.I, ion , 28 
ftar~sfer for <" ~,'-" ~ s.mt ' - "-~ c,, t,4., I t:. edit;, i 7 
Ass-~lb l y o r c..! at!sos j 1 
............................................................................................................ \[ ........................................................... 
Tab I e, 1 
All the-"e< oi ..... '" ..... i.. l:.~ tt%N,.~Imi al-e prOdtlCt iofl- 
f'UlOS~ ~FI1.{OIi i\[1 a l~ohl,,~il!-orJ(-~'nted 
, . cr'/)HT lafl~!..la~t} .... St;Or'Jr, l-,j it.'.. Jfi\[~ a ....... , 
lhc separat ion of' i inSu{si i(.: data ir,'>~ 
hJsor i \[hill J s a(:!i J{:\]v(£:(~ l~at-;!~ i(!t(:.~b i::; 
jdOl'}l j ca I } y r'q ,'v,.~'?,r i .... 
ili\[~eg ghi(.Jl i!~dJcat.<',t; wtiich ,,!;tlh-modcil(~ 
J{ belongs to and ~'l\]el\] it should be 
..... " -".;;:{;'11(I i i , a t ,:.,J P~{' /?\ (_~; \[{i!!?I° '-'i '~i\],:! ~, 
or c;Oitl!\]Iex; <3> ....... j()ri ~,Vlt htu~. . ca<.~ ' ich ...... i r\]\[; i tt(.il{:~ 
a set of ace ions ,:;oncitl(:iod at",t!lal ty by 
subroutines; <4> next-stc:!.~ instrtic:t ion 
,diich ili(Jicatts what is to bO CIOliO 
after tlie oresertt rttte Js ~xeA;tlted, 
5, Parser and Logical Semant its 
T~IANo~Aft m,~ .... a flPSG-like forll4alisl~, 
(,,Jlk.,~ i\[tlort( i:;ttiK:t iofiaJ t(!} l~rtt: tort (JFatlllllaF 
((YltG), The l;ar,:-,t~'l is cha~aotcrist Jc of 
o£ CFR(~, all inK,'rcediate list structure 
wi\]l b~ generato~l~ ~hich cepr~rits 
\[~'0 typOS Of SLFUC\[tlr~:}: SylltaCt iC S\[FitC- 
tt,-e arid logloal se,~antic structure, Me 
nc,~l~:~; i l~ both the s~ructures are cor\]nect- 
e~i in a two-way ,}ode, \]\]itts tl~e iiiotlie\]" 
liod~ ,::an l:}t~' vJsJ \[e(l tltrotlgh I>r datlght -o. 
e-,-' and V J ce versa, The i lll'ot'ltlat i¢)i1 ,l F) 
Jle, ms for each node will increase from 
26 ini~i~.~i once:; tal<e~\] f\[()iJl lhedict Joiial-y 
to 60 It) the IilaXil\]llllit, iflC, II.ldJit~ mat ic 
f~\]t tires arid dyl\]am ic teat uras, The 
logical setmntic structure, si\[~ilar to 
a ca~-structtlre ~rves as l.lie basis (if 
~ransfe.,r'. Logical s(~nan\[ics is rei:~rese, nt- 
~1 by lwo types of setllaflt ic fealure, 
or~eof which is tile se, ltialltic reiatioil 
be\[t#ooft i:he (;()tlc~p{s such as agor}t, 
pat ielll, ~mnner, dural lob, eXc, ii~c 
other is the attributes o.ri tlie CX)fiC~(:}l\]b:;, 
