English-Chinese Machine Translation System IMT/EC 
Chert Zhsoxlong and Gee Qingshi 
Institute of Computing Technology 
Chinese Academy of Science 
BelJlng, PRC. 
Abstract 
IM'I/EC is an English-Chinese machine translation 
system ~,hich integrates some outstanding features of 
the case grammar end semantic grammar lnto a uniform 
frame, LISeS various kuowledgo In the disamblguation, 
and tries to modify the object language by itself. In 
this poF,er,we first introduce IMT/EC's design motive- 
tiorl and overall architecture, then describe the 
deslgn philosophy of its translation mechanisms and 
thelr procesging algorithms. 
J,The design nlotivation 
\]'he design of the IMT/EC system are motivated to 
develop new approaches to the Engllsh-Chinese machine 
translation, such as, to provide the system with 
powerful analysis meohanisnls end MT knowledge base 
menagonlerit system , as well as some exceptional pro- 
cessing and learning meohanlsms, that is, to make the 
system ba intelligent. In addition, it also tries to 
inregret9 as many advantages of conventional machine 
translation systems into a single system as possible, 
such as, to provide the system with powerful mechon-. 
Isms for the processing of various ambiguities and 
contextu,~l relations. The design of the IMT's trans- 
lation mechanisms are based on the following consl- 
derQti on ~;, 
(1) St-analysis 
In the development of machine translation system, 
in order to disambiguato the source language, we 
have to {molyze the input deeply to get the internal 
meonlng representation of the source language. 
However, the deeper we aaalyze the input, the more we 
lose the clues about how to express the translation, 
also, th(it it results in extremely poor or no trans- 
lations ef sentences for which complete analyses can 
not be (h~rlved\[Slocum 85\]. To find a suitable analys- 
is depth so as to get both clues about how to express 
the trorl~;lation of the input and to disombiguate the 
input conlpletely Is almost impossible. In the IMT/EC, 
we try t(, design a simple grammar analysis mechanism 
-- SC-gr'(~mmar auulysls mechanism to inherit both the 
outstanding features of case grammar analysis and se- 
mantic grammar analysis so as to produce a high qua- 
lity translation, 
(2) Multl-language translations oriented 
In present technical conditions, it is impossible 
to design a general internal meaning representation 
for all natural languages. Thus, the knowledge based 
multl-language oriented machine translation system is 
difficult to be marketed in the near future. A feasl- 
ble way ~or designing multl--longuage oriented machine 
translatJorl systems might be to separate the proces- 
sing mechanisms from the language specific rules |as 
King et oI. ~5\], thet ie, t O apply the same process- 
ing meotlonlsm with different language specific rules 
for d~ffErent natural language pair translations. In 
the 1NI'/EC, we develop a general rule representation 
form for the representation of various knowledges 
used in the translotlon. Knowledges for different 
language palr translations are stored in the differ- 
ant packages of the knowledge base IMT-KB. The know- 
ledge base are organized In multi-package end multi- 
level way so as to store rules for the translation of 
different language pairs and different phases of the 
processing. Thus, the system can be easily extended 
for' multi-language trarlslatlon purposes. 
(3) Diversity processing 
As the dlsemblguation rules are rather words spe- 
cific, ±% is difficult to manage them in the same 
way. To deal wlth this problem, we store these rules 
in their respected word entries end classify them es 
several categories in the IMT/EC, Each category cor- 
responds tO 0 general subroutine epplleatlon mechan- 
ism, which apply the word specific rules and subrout- 
ines in the processing of translation. The subrout- 
ines are stored in a natural language specific sub- 
routine package. Some word specific subroutines are 
directly stored in the respected word entry. 
(~) Powerful exceptional processing 
Since the natural language phenomena ore so abund- 
ant that any existed machine translation system can 
not process all the phenomena, it is essential to 
provide an exceptional processing mechanism in the 
system to deal wlth exceptional phenomena. As IMT/EC 
incorporates some learning mechanisms, thus, it is 
more powerful in dealing with the exceptions than 
others. 
(5) Automatic modification of the translation 
Generally speaking, machine translation system can 
only produce rigid translatlons, it is a desire that 
MT systems be able to modify the output by itself so 
as to produce more fluent translations. IMT/EC tries 
to apply same common sense knowledge and linguistic 
knowledge of object language to disamblguate the 
input end modify the translations, thus, to improve 
the translatlon quality. 
In the following paragraph, we focus on the trans- 
lation procedure of the system end the algorithms 
related to it ignoring the knowledge base organiza- 
tion and management mechanisms. 
2. The overall architecture of the system 
The architecture of the IMT/EC system is as 
follow, 
\] knowledge base management 
Engllsh Input System IMT-KB \[ 
~ f Knowledge~ Morphological Analysis l . I 
I ~ Applicatzon J & 
Dictionary Retrlving~ / p._I_T_T..~1~ 
t 
I ~;n:l;;";'s ' L \~/ ___ B°se.~_ 7 
\[ --~~ /~ / Augmentation-~ _ / Dlsambigu~on r / I\k & Modifiootio~ f*~ 
| & Transfer L / I \~ 
' .... ',, I 
I - Acquisitioo ) 
I the Tren?lotlonJ / 
t 
/ 
/ 
Fig. The architecture of the IMT/EC 
117 
As the rule bose and dlctlonarv in a machine 
trenslatlon system is so vast that it ls impossible 
for human beings to find the confllotion end implica- 
tion among the rules. To modify a rule in the knowle- 
dge base often results In many side effects on other 
rules. Thus, it Is necessary to provide a self re- 
organization and refinement mechanisms in the knowle- 
dge bose. 
In the IMT/EC, we design a special knowledge base 
management system IMT-KB to manage ell the knowledge 
used In varlous processing phases of the transla- 
tion. In addition, IMT/EC also provides o knowledge 
bose augmentation and knowledge acquisition environ- 
ment for the system to augment system performance by 
itself and for the users to improve the knowledge 
base. 
The col1 relations connected by dotted llnes in 
the figure above ore executed only when the user sets 
the learning mechanisms in working status. These 
mechanisms can acquire new knowledge in the dynamic 
interactive, static interactive,or disconnected ways. 
They ore primarily used to resolve the exaeDtlenel 
phenomena in the translation. 
Dynamic Interactive Learning (DIL): Whenever the 
system encounters c sentence out of its processing 
range, it produces various possible translations for 
each segment of the sentence and interacts with human 
beings when necessary to select on appropriate trans- 
lation of the segment and combine them to get 
o correct translation of the sentence. At the some 
time, it also creates'some new rules to reflect the 
selections. That is, it learns some new knowledge. 
Static Interactive Learning (SIL): Whenever the 
system encounters a sentence out of its processing 
range,it records down the sentence and its appearance 
context in e file. After the text has been translated, 
it begins to analyze the sentence in detail to get 
various possible translations for each segment of the 
sentence and interacts with human beings when necess- 
ary to get appropriate translations of the segments 
and combines them to get a correct translation of the 
sentence. At the same time, it also creates some new 
rules to reflect the selections, thus, to learn new 
knowledge. 
Disconnected Learning (DL): Whenever the system 
encounters o sentence out of its processing range, it 
analyzes the sentence in detail to get all the possi- 
ble translations, and then evaluates these transla- 
tions according to the preference rules stored in.the 
IMT-KB to select on appropriate translation and 
modify the related rules used in the analysis to 
reflect the selections. It skips over sentences which 
the translation con not be determined by the prefe- 
rence rules Instead of interacting with human beings. 
5. The translation procedure 
IMT/EO's tronslatlon procedure is divided into 
several phases, i.e.,morphology onalysls and diction- 
cry retrlvlng, SC-grammor anolysls,dlsamblguotlon and 
transfer, modification of the tronslatlon etc. 
The communlcotions between tronslotlon mechanisms 
and the knowledge bose ore performed by the knowledge 
base management system IMT-KB, these operations 
includes getting a se~ of related rules and returning 
some Information for the modification as well aS 
augmentotlon of the MT knowledge bose. 
5.1. Morphology analysis and dictionary retriving 
In the IMT/EC, words in most common uses con be 
retrived by either their base forms or their surface 
forms, whlle most of the other words can only be re- 
trieved by their base forms. The tasks of the morpho- 
logy analysis ore to process the prefix, suffix, and 
compound words. Since these processlngs ore complete- 
iv natural language specific, in order for the proce- 
ssing mechanisms to be language independent, we deve- 
lop a language independent morphology analysis me- 
chanism to opply the language specific morphology 
rules In the morphology analysis, 
The morphology analysis rule form is 
<surface pattern> -> <conditions> I <result> 
118 
Here, 
<surface pattern> is the surface form of the word 
to be analyzed, 
<conditions> is the oppllcotlon condltlons of. the 
rule, 
<result> Is the definltlon of the word base form 
analyzed. 
For example, 
(1) (" s)-> (verb -) I (def("), SV) 
(2) (" s)-> (noun *) I (clef(*), PN) 
(3)(-1 - "2)-> (word -1)(word "2)1 
((def(morpholog v w1), 
def(morphologv "2)), CaM) 
Here, *, -1 and *2 are variables Indlcating that 
it con be bounded to any sub-character string of the 
word to be analyzed, def(X) is the definition of X in 
the IMT-KB, SV, PN, cam are surface 'features of the 
word. 
Rule (I) Indlcotes that when the last character of 
the surface form of a word is 's' and the remained 
character string * in the word is o verb, then its 
surface feature is the slngulor verb form (SV) of the 
verb *. Thus, It returns the value of 
(def(*), SV) 
as result. 
Rule (2) indicates that when the lost character of 
the surface form of a word is 's' and the remained 
character string * in the word is o noun, then its 
surface feature is the plural noun form (PN) of the 
noun *. Thus, it returns the value of 
(def(*), PN) 
as result. 
Rule (5) indicates that when the character string 
of o word comprises a character '-', the left pert ~1 
and the right part w2 of '-' ore both words, then it 
lso compound word of ~I and w2.Thus, it applies mor- 
phology rules to analyze the word ~1 and *2, and re- 
turns the value of 
((f(morphology -1),f(morphology w2)),COM) 
as result, 
Suppose that, 
SX indicates that X is o variable, 
#X returns the character llst of X, 
&X returns the lost character of X, 
>X returns the first part of rule X or the 
first element of a list, 
<X returns the remained port of X which 
(>X" <X o X), 
f(X,V) returns the first different item palr of 
X ond Y, 
lookup(X) looks up the dictionary and returns the 
deflnltlon of the word X, 
search(X) returns the morphology rules which leclu-. 
des character X, 
check(X) tests whether two elements of the item 
pair X is uniflable or nag, 
null(x) tests whether llst X IS empty, 
apply(g,x) returns the result of g(X), 
t(X) tests whether result X needs further 
onolysls and performs recurslve analysis 
when necessary. 
The algorithm for morphology analysis and diction-- 
erv retrievlnq is as follow. 
INITIALIZE 
$X <- #word; 
SP <- search(& SX); 
$P <- $PU search(> $X); 
$result <- ( ); 
for $rule m SP do 
{MATCH 
SPAT <= > Srule; 
$COND <- >"< $rule; 
$RES <- <=< $rule; 
Loop 
$patr <-- f(SPAT, SX); 
if (null($polr)) goto TEST; 
if (not(check($pair))) break; 
SPAT <- $PAT~ Spelt; 
iPAIR .I. <-. iPAIR l.U ($polr}; 
gate Loop ; 
rEST 
for $CONDI e$COND, do 
($PROP <- lookup(>e < ($CONDL)); 
if (not(apply(> $CONDI, SPROP)) brook; 
} 
|:~ESUI.T 
iPROP <-- lookup (> ($RES~'$PA);R L)); 
iresult <-- iresult U {($RES~ iPA:\[R L, 
iPR(IP ) } ; 
) 
if (pull($res~it)) return word 
else returrl t($resu\].t); 
El~ll), 
3.2. St--Grammar Anulys:Ls 
The St-grammar enolysis mechanism of IMT/EC 
U|)plle~ th( SC--rolos stored ill the iMT-KB to dlsumbl- 
gLlate the ~.trueturol embigultles of the input senten-- 
cos end predm;es the structural description for them. 
"ihe grammar |lOS some outstending features of the case 
grommet ond semuntic gr'ommor. The rule form ls cs 
follow, 
<S-STRUCTURE> ~> <S-ENVIRONMENT>I 
< R=SI'RUCTURE >, 
<I~- ENVIRONMENi> 
<TRANSFER>. 
Itore, 
<S-.SIIRUCrUI~E> mid <S-ENVIRONMENT> are rule condi- 
tions which defines the current strtlctLIr(\]l form end 
contextual foattlres of the input, <R-.STRUCTURE> und 
<R-ENVIRONNiENT> ore result strueturul form ~nd 
corltextuel features of the input, <TRANSFER> ere the 
trensformotlo~Is related to the rule. 
lho structural forms , <S-SI'RUCTURE> end 
<R-STRUCTURE> ore represented os strings of syntogmos 
arid words.\]he contextuel envlronments,<S-ENVIRON~ENT> 
and <R-ENViRONMENT> ore represented os vectors, of 
which each element corresponds to on inter-sententtal 
reletion or e specla\] eeoc, their values ere used to 
resolve the ellipsi.s,dnephore, tense and espocts etc. 
it is the principal contextual processing mochenisms 
in the IMI/EC. 
Since the contextual vector is used only os a 
supplo*~lent to the pure semantic grammar ona\]vsls, 
espeele\].\].y in the processing of contextual relations, 
it is riot necessary to analyze the Irlput to the 
extent the|. one con get ell the semantic relations of 
the input. Thus, the vector processing formalism is 
completely acceptable. 
Two example rules ore as follow, 
NP VP -> A I S, change(B1,×), INP IVP. 
in NP -> A1 I PP, chonge(B2,X), zel INP nuei. 
St-grammar onelysis mechanisms receive the results 
of morphology analysis or previous SO-reduction, send 
the messages to the IMT-K8 to get releted rules, end 
apply these rules to ,reduce. the input until o non- 
terminal symbol S is reduced, thus, to produce the 
structural description o1' the input. 
The SO-grommet analysis algorithm of the system 
is:. 
(I) \]in the entries of' the \[MT-KB dictionary, we 
stored not only the word meanings and their disembi-- 
guotlon conditions, but oleo SO-phrase end sementlc 
rules specJflo to the entry word. When onulyzlng e 
sentence, the system first retrieves the SC-phrose 
rules specific te the words appeared irl the sentence, 
end ~pplim{ these rules to find a list of possible 
phrases of the sentence from the context of the words 
in the senl;eneo. 
The phr~se list returned is os follow, 
X, (i, , J, ) 
X~(±~,J:) 
x~(i.,j~). 
Here, XI, X~., ...,Xm ore phrase syutogma Identif$ers, 
i}, ~) ..... J.~ arid J~, Jz ..... J~ ere ending posi- 
trons of the phrases in the input sentence. 
(2) Find a list of expectation pathos from the 
phrase \[List as follow,. 
"'~') X~;(J,~l k,! X~l(m,4t m) X, (i,J,) 
vl~), ×~/(l,Jm i X~)(Jm÷l,k ) ~ U"~+I,n#) 
P(w ~ ) P(w~+ L) .,, p(w~z. # 
(Here, P(w) is the word w itself or its property, t 
is the current onalysls position which initial veluo 
is ~,I is the expects|lee length defined by the user) 
end order them by means of the phrase ending post-- 
tlons n=,n~ ..... n~ from lerger to smeller. These 
pothes are used as heuristics in the ano\].ysis ef the 
sentence. We try one new patti ot one becktrecking. 
(3) Send the ano\].yzed component 
M = V,(...) V~(...) ... V~(...) 
cndIiihe current expectetion path to theIIMT---KB to 
retrleve the'SC-ruIes'which heed pc|terns contain 
sub-string of {~ 
,(...) ... v~(...) x~(..:) ... xz(...) 
~1 Path = or 
(. ) v~( ) P (w~ ) ... P (w~.~) 
and organize these rules in a list according to their 
preferences from higher" Lo lower. Then, it tokes one 
rule from the rule llst at one buckfirecklng and go to 
(~) to appl V the rule to reduce the input. 
If no rule in the llst con be successfully applied 
to reduce the input, the system gets the next expec- 
tation path from (2)end repeet (3). If all the expec- 
tation pethes have been tried end no successful rule 
has been applied, it returns 1;o the last analysis 
position to re-analyze the input. If the current erla-- 
lysis position Is the beginning of the sentence, the 
system coils the exception processing ,lechenism to 
deal with this un--analyzable sentence. 
(l~) Match the rule head pattern with the current 
form of the input sentence. If there is a sub-pattern 
of the current sentence pettern that can match the 
rule heed, then go to (5) else get the next rule from 
(3) end tries to re-.match them. 
(5) First, odd some newly formed phrases into the 
phrase list in order For the backtracking of the one 
lysls, then coll the cese enelys±s mechanism to check 
the eurreet analysis results und the current form of 
tile sentence 'to Fill in the rose freme A, B in the 
rule end the context vector. The case anelysis algo- 
rithm is described In the following paragraph. 
(6) Check A end the context vector to see whether 
their values are unlfleble. If they are unifieble, 
then go to (71, else get the next rule from (3) and 
returns to (4). 
(7) Store the backtracking informetion into the 
temporary stock, substitute the reducing part of the 
current sentence form with the reduced form, change 
the current analysis position to the last word oF the 
newly reduced syntagma,cnd change the related element 
values o£ the context vector aecording to the element 
values of B. 
If the current position is not the end of o sen- 
tence, then go to (2), 
If the current position is the end of a sentence 
end the current form of the sentence is net S, ~hon 
go to (2), 
If the current position is the end of e sentence 
end the current form of the sentence is S, then go 
to (8), 
(8)Call the semantic processing mechanism to check 
the result of the nnslysls to see whether ~t violates 
the English collocetlon rules. If the result violetes 
the collocation rules, the system recovers to the 
status before the last reduction and gets 'the rlext 
rule from (5) to r'e-onoiyze the input. Otherwise,there 
wlll be two cases, 
a, If the user only needs the most adequate 
• trensletlon, the system proceeds to analyze the next 
sentence. 
b. If the user needs ell possible trcnslations, 
the system records down ~he current result end rose- 
vers to the stetus before the Zest reduction end gets 
119 
the next rule from (5) to re-analyze the input in 
order to get other onolysls results. 
AS we have mentioned before, the case analysis in 
the SC-analysls is only a complement to the semantic 
analysis. It is mainly used to des1 with the context 
relation and a§pect, tense, modal etc. Thus, the 
system only needs to analyze those cases which can be 
used in those purposes. It ls much simpler than the 
case analysis in the case grammar analysis. 
The case analysis in the SC-enalysls ls performed 
by the following algorithm, 
(1) Get the case expressions defined in the 
el'emerita of" vector A and B. The form of the element 
expressions of A and g ls 
s~\[i\]:E 
Here, S~\[I\] indicates the element case Cdentlfler(S#) 
of the case frame A or g is corresponded to the case 
identifier sill of the system case frame, l.e.,system 
context vector. E is the expression used to get the 
value of the respected case. 
(2) Retrieve the definition of the case identifi- 
ers from the system case frame and organize these 
case identifier into a 1let according to their prefe- 
rences from higher to lower. The form is, 
(S\[il\].subject:EI,S\[12\].obJect:E2 ..... S\[lm\].Em .... ) 
(5) Evaluate the value of the elements in the case 
identifier llst, and flll them Into the respected 
position in the case frame A end B. There are many 
cases in the evoluatlon. 
a. E ls a constant, returns E, 
b. E ls empty, evaluate the case value according 
the definition of the case identifier, 
c. If the case identifier lsa syntagma idetlfl- 
er, then finds the vclue of the identifier 
from the analyzed input according to the 
heuristics provided by the expression E, 
d. If the ease identifier is o sementlc identi- 
fier, then call the semantic mechdnlsm to get 
the value which can be filled into the case 
identifier from the input according to the 
heuristics provided by the expression E, 
e. For other case identifiers,call thelr respec- 
ted Subroutines to get the value of the case. 
These subroutines are defined by the rule 
designer. 
The case analysis in the SC-analysis con solve the 
elllpsls, anaphora, and other contextual problems. 
5.5. Semantic dlsambiguatlon and transformation 
The SO-rules define not only the relations for the 
syntagma reduction, but also contextual vector value 
changes with respect to the reduction of o sentence, 
and the rules related transformations. 
The transformation operation defined in the SC~rule 
is in the follewlng forms, 
IX IX ... IX 
Here, IX , IX .... , IX are translations of the 
syntagmas X , X , ..., X in the rule head. Their 
positions indicate the positions of the translations 
of the syntogmas. There will also be some indicators 
In the string which are used to indicate positions of 
the translations for inserting tense, voice, modal 
modiflers.These indlcotors are used as the heuristics 
of the semantic processing. 
The transformetlon in the IMT/EC ls relatively 
slmple. It travels over the whole anolysls tree from 
top to down, left to right, transfer every node when 
the node is £raveled.J'he result of the transformation 
is the Chinese utterance of the sentence. 
Rules with same head patterns may have different 
case frames A and B,in this case, they may correspond 
to different transformation operations. These rules 
ore defined as two different rules by the rule deslg- 
ne'r. Whfle in the IMT-KB, the system stores them as 
one rule wlth many candidate right patterns. Whenever 
the head pattern is successfully matched, the system 
sequentially checks these candidates until one of 
them is satisfied and records down the current suc- 
cessful position so that backtracking mechanism can 
120 
get the other candidates when necessary. 
The tasks of the semantic processing in the IMT/EC 
ore to check the results of the analysis to see whe- 
ther they satisfy the syntax or semantic collocation 
rules defined in the IMT-KB, to produce the suitable 
modifiers for expressing the tense, volce, aspects 
and so on In the Chinese. In some cases,lt also apply 
the well formed world knowledge deflned in the IMT-KB 
to eliminate some 11legal expressions and extend the 
meanings of some ambiguity words. 
Slnce the SC-analysls is based on the semantic 
grammar analysis, most of the syntax and semantic 
ambiguities are solved in the reduction operations. 
Even though the case analysls in SC-analysls ls aimed 
mainly to resolve the contextual problems, they can 
also solve some ambiguities among o sentence. That is, 
the semantic processing in the IMT/EC ls orlented to 
speclflc ambiguities and lnter-sententlal case value 
evaluations. Though the processtngs are different in 
different phases of the trons1otlon, they can be 
categorized as, 
(1) determining the value of o specific semantic 
identifier, such as tlme adverbial, place edverblal, 
anophoro etc. 
When o specific semantic identifier is concerned, 
the semontlc, processlng mechanisms first finds the 
key word which con match the semantic identifier from 
the sentence,such as word wlth tlme,plaoe properties, 
then get the phrase which comprises the key word in 
the sentence, and return the phrase as the value of 
the identifier. 
Only simple anaphoro phenomena are considered in 
the IMT/EC. They are processed in two different ways. 
One is to compare the synonyms to flnd the. anaphorn 
words, the other Is to flnd the suitable anaphoro 
content through the position relations, such ca, in 
some specific context the word 'which' can refer to 
the noun phrases immedlotel y before it. 
(2) checking the collocation of syntogmas. 
There ore three possible categories of collocation 
In the analysls results, 
<1> X W -> (W => CI) 
<2> W Y -> (W => C2) 
<5> X W Y -> (w => C 3) 
Here, X, Y may be strings of words or syntagmos, W 
Is a specific word. The above expressions means that, 
<1> W appears after string X and functions as 
speech CI, 
<2> W appears before string Y and functions as 
speech C~, 
<5> W appears between strlng X and Y and func- 
tions as speech C~. 
The related word definition in the IMT-KB diction- 
ary is as follow, 
W := C, (E, => MI) 
(E~ :> M~) 
c: (E~ => M~) 
Cm (E~ => M~) 
Here, C is the speech category, E is the context 
structure of word W, M Is the meaning of word W. 
The semantic processing mechanism retrieves the 
collocatlon rules specific to words of the sentence 
from the IMT-KB, and applies these rules to check the 
analysis result to see whether there is any Violation 
between the analysis result and collocation rules. If 
there ls, returns fell. 
(5) cheoklng the distant contextual relatlons. 
There are also three possible categories of 
distant contextual relations appeared in o sentence, 
X ... W \[m\] -> (W => C~) 
W ... Y \[n\] -> (W => C~) 
X ... W ... Y \[m,n\] -> (W => C~) 
Here, X, V, W, C have the same meanings as in the 
(2). n, m are optional, they defines the relative 
position between the word W and XIY. When n, m = 0, 
they ore the cases described in (2), When n, m is not 
defined, they indicates any position before/after the 
word ~V. These distant contextual relation rules are 
defined la the It~-I'--Kg In the same way as in (2). 
If m and/or u are present, the semantic processing 
mochenisdl finds m/rl word before/after tile word W in 
the sentence, and tries to reduce that word and its 
adjacent words inca X or' Y. If they can be reduced, 
end tile word W functioas as the same category as 
defined in the rule, trlen Successes, e18o eliminates 
tlm analysis, 
if .iaud n are not ~efined, then try to find the 
word before/after tile word W which con be reduced 
into X or Y together with its adjacent words. If there 
ere no such element in tile sentence, then returns 
f(lil. 
(ll)cre,3ting Chinese modifiers to express the tense. 
voice, modal arid so on Grid insert these modifiers in 
tile t r(3Jisloticn according to tile position mark 
Ip'pearoci iu the rule. 
the 9recessing procedure is as follow, 
a. GeL ClIO niorks of the tense, voice, modal etc. 
b. Coll. trio' correspelld/ng sabrautines defined by the 
rule designer to del.orlidtle on appropriate modifi-- 
or 1-'or tile mclrk. This is bclsod mainly on tile soa- 
texi;llul structure of the analysis result. 
C. Ins(~rt tri() modifier in trlo position of tile truns- 
lati~)n marked by the niarker. 
For" exalllJilo,if a very is in the '-leg' form and tilere 
&s I:e time odvorbiol in the i~lput, the tease of tho 
contoxt are all progressive, then ignores trle time 
rHark. IF the predicate ore 'be going to', then trans- 
lates it as 'dashuang' ignoring trlo time mark. 
The world knowledge rules are defined in the same 
form as ihe semantic rules. "lhe application of these 
rules (Jr( to test tile context to find the semantic 
f'o(~ttlres Of the sltLIotIon end coIdpQre these to the 
world mo(iel definition dei"i~lc, d if~ the world knowledge 
rule to \[{el;ormiue the sit<lotion of" the utterance, and 
thO~l detormido the correch translation or exterlct the 
mounlllgS Of related words. 
Every semuntlc processing nleChOnlsm mentiorled 
.b<)ve co,~responds to u specific processing subrout.- 
irle. rtie;~.~ sabroatines are called ill the grammar one- 
lysls Ulid transformation processing to perform the 
related \[~{911tantic processing. (1) Is primarily asod in 
the case qnalysls, (2) and (~) are pr'Imarlly used in 
checking cbe analysis result and disembiguations in 
the analv:;Is and tronsfortllaLion,(4) is primarily used 
ill the tr,msformotlon. 
The gr~muner and word trnnsformatlon qlgor$thm is, 
(1) CHrrent=node <-- root of tile enalysls tree, 
(2))Z'£ the curront-InOdO is o loaf node, go (4), 
(~) The current--node is not a leaf node, the 
processin!j are as follow, 
a. iI' ~ll the elemerlts in the transformation 
e~:pression of the node c~re constant, go (5), 
b. Ji ~ till the variables le trio transformation 
expression of the node are substituted by 
c(~nstants, thee call semantic processing me- 
chanism to create suitable modifiers. Go (5). 
a. if' there are scale unsubstituted variables in 
the expression of the node. set these varla-. 
bias r~s current-node er)e by one, aed uses the 
results returned by each subnode to replace 
the vari(lbles. 
(l~) Whet! the current=node Js a leaf node,that is, 
it is n spec~flc word or arl ldlol~, then retrieves 
its defintt;lon l=l"Onl the IM'I'-,KB, cell the semantic 
i}re(:osslu\[I ,lecheli I Sill to determine on appropriate 
moaning far it according to the tree structure. 
(5) I|' Lhe eurrent-aode is root node,then returns 
trio curron{; fornl of the transfermatiou expreseion es 
the translation of the sentence. Otherwise, returns 
tile expr'ession to the parent uode,reeovers the parent 
nod~ rJs curreet node. Go (2). 
~.t, The modification of: the translation 
The objective of" the automatic modification of the 
tr'auslGtlo~i iS tO ililprove the roadability of the 
transl{~L:lorl,but tilts socrlflces part of the accuracy. 
It is more suitable for the non-sctelltiflc literature 
tdanslatlon. 
The main tasks comprises: 
a. Change the order of the phrases and words of 
the translation, 
b. Substitute some words which collocotion is not 
commonly used in the Chinese utterahce for the syno- i 
nymous words, 
c. lnsert some conjunctive words when necessary, 
d. E11m~.nate some redundant wards. 
The algorithm for these processing is, 
(I) According to the Chinese oollocotion rules 
defined in the IMT-KB, changes the words and phrases 
order of the tranelatlon which are not in accord with 
the collocat£on conventions in Chinese, such as, 
Budon .... Erchia .... 
(2) According to the co-occurrence rules of the 
Chinese words defined in the IMT-KB, check the uses 
of the Chinese words in the translation. If they are 
not in accord with the co-occurrence rules, then rep- 
laces these words with the Chinese synonymous words 
until they ape accord to the rules. If there is no 
suitable synonyms.then tries to extend the meaning of 
some words. The meaning extending rules are defined in 
the word entries. Its form is as follow, 
<word> :- <condition 1> <extension I> 
<condition 2> <extension 2> 
<condition n> <extension n> 
Here, <word> indicates the word appeared in the 
sentence, 
<condition> defines the extending conditions, 
<extension> is the utterances extended. 
i£ the word can not be replaced or extended, tilen 
just returns the source translation. 
(5) Check the translation to find the redundant 
words and eliminates them. The form of doletiorl rule 
is, 
X V X Z -> p (X), p (V) i X V Z 
such as, 'NP de NP de -> NP NP de'. 
Since the modification has no absolute standard 
and requires a large amount of world knowledge, it is 
rather dlfflcult to solve this problem in one day. In 
the ZMT/EC, we only deal with the most simple eases. 
More complex situations can be solved with the appli- 
cation and improvement of the system. Thus,the system 
is designed to be easily extended with the applica- 
tion. 
if the user needs high quality translation, he may 
call the post editing subroutine to modify the trans- 
lation by human beings or with the aid of human 
beings. At tile same time,we can also set the learning 
mechanisms in working status to trace the modifica- 
tion procedure oF human beings and produce some use- 
ful rules For the system. 
I~. Summary 
In conclusion, we hove lntroduce~ ~ translation 
processing procedure 9f the English-Chinese machine 
translation system IMT/EC, and describe its principal 
processing algorithms. 
Aknowledgement: We would like to thank Hang Xiong, 
Zharlg Yujie, Ye Yimln, Tong Jioxion, Zong Llyi, 
Zhong Zife, Chen Z1zong, Chen Zizeng and Fu Wei For 
their cooperation in th e implementation of IMT/EC, 

References

\[1 \] Axelbiewer, Christian Fenneyrol, Johannes Rltzke, 
ErWirl itegentrltt(1985), ASCOF-A modular multilevel 
system for French-German translation, OL, Voi.11, No. 
2-5, p'157-'154, 1985. 

\[2\] Bernard Vauquois and Christian ~\]oitet,Automated 
transletion at Grenoble university, CL, Voi.11, No.I, 
p28-36, 1985. 

\[5\] galena Henlsz-Dostert et el., Machine Transla- 
tion, Mouton publishers, Hague, Paris,New York, 1979. 

\[4\] Harry tennant, Natural Language processing, Pe- 
trocelli books, New York, 1981. 

\[5\] Hiroshl Uchida, Fujltru machine translation 
system: ATLAS, FGCS, Vol.2, No.2, p95-1~0,1986. 

\[6\] Jaime G. Carbonell and Masaru Tomita, New 
approaches to Machine Translation, TR-CMU-CS-85-143, 
Carnegie-Mellon University, 1985. 

\[7\] Jonathan Slocum, A survey of machine transla- 
tion: its hlstory,current status and future perspect- 
ives, CL, Vol. 11,No. 1, p1-17, 1985. 

\[8\] Kazunori Muraki, VENUS: Two-phrase machine 
translation system, FGCS, Vol.2, p121-124, 1986. 

\[9\] Martin Kay, The MIND system, Natural Language 
processing, edited by Rustin, Algorithmics press, New 
York, 1975, p155-189. 

\[10\] Makota Nagao, Current Status and future trends 
in machine translation, FGCS, Vol. 2, No. 2, p77-82, 
1986. 

\[11\] M.Nogao, J.Tsujil and J.Nakamura, Science and 
Technology agency's machine translation proJect,FGCS, 
Vol.2, No.2, p125-14~, 1986. 

\[12\] Murlel Vasconcellos and MarJorie Leon, SPANAM 
and ENGSPAN: machine translation at the PAN American 
health organization, CL,Vol.11,No.2-5, p122-156,1985. 

\[15\] Paul L.Garvin(ed.), Natural Language and the 
Computer, McGraw-Hill book, New York, London, 1979. 

\[14\] Perelra F. and Warren D., Definite clause 
grammar for language onalvsls, Artlf~clol Intelllgen- 
ce, Vol.15,p251-278,198~. 

\[15\] Pierre Isabelle and Laurent Bourbeau, TAUM- 
AVIATION:Its technlcal features and some experimental 
results, CL, Vol.11, No. 1, p18-27, 1985. 

\[16\] Richard E. CulZingford, Word-meaning selection 
in multiproces~language understanding, IEEE Trans. 
on Pattern analysis and m~chlne intelligence, Vol, 
PAMI-.6, No.4., July 1984,p493-509. 

\[17\] R.F.Simmons, Technologies for machine trans- 
lation, FGCS, Vol.2, No.2, p85-94, 1986. 

\[18\] Rod Johnson, Maghi Kinq, end Louis des Tombe, 
EUROTRA: A multilingual system under development, CL, 
V01.11, No.2-3, p155-169,1985. 

\[19\] Roger Sehank, The condeptuel analysis of 
natural language, Natural Language processing, edited 
by Randall Rustln, Algorithmics Press, New York,1975, 
p291-511. 

\[20\] Rozena Hennisz-Dostert, R.Ross Macdonald and 
Michael Zarechnak, Machine translotion, Mouton Publi- 
sher, Hugue, Paris, New York, 1979. ~. 

\[21\] S.Amano, The Toshiba machine translation sys- 
tem, FGCS, Vol.2, No.2, p121-124~11986. 

\[22\] Terry Wlnograd, Language as a Cognltlve(Vol 1) 
process, Addison-Wesley Publishing Company, 
California,London, 1985. 

\[25\] Vtnfield S.Bennett and Jonathan 61ocum, The 
LRC machine translation system, CL, Vo1.11, No.2-5, 
p1¢1-121, 1985. 

\[24\] Y. Wilks, The stanford machine translation 
project, Natural Language Processing, edited by 
Randall Rustin, Algorithmics Press, New York, 1973, 
p245-291. 

\[25\] Yoshlhiko Nltta, Problems of machine transla- 
tion systems-effect of cultural differences on sent- 
ence structure, FGCS, Vol.2, No.2, p117-120, 1986. 
