A Similarity-Driven Transfer System 
Hideo W~t~n~})c 
IBM Research~ Tokyo Research Laboratory 
5-19 Sitnbancbo, Chiyoda-ku, Tokyo 102 Japan 
e-math watanabe@trl.vnet.ibm.cmn 
Abstract 
The transfer phase in machine translation (MT) sys- 
tems has been considered to be more complicated 
titan analysis and generation~ since it is inherently a 
conglomeration of individual lexical rules. Currently 
some attempts are being made to use case-based rea- 
soning in machine translation, that is, to make deci- 
sions on the basis of translation examples at appro- 
priate pohtts in MT. This paper proposes a new type 
of transfer system, called a Similarity-driven Trans- 
fer' System (SimTi'ao), for use in such case-based MT 
(CBMT). 
1 Introduction 
The transfer process in macbine translatiou systems 
is, in general, more complicated than the processes of 
analysis and generatimt. One reasmt for this is that 
it relies heavily on human heuristic knowledge or the 
linguistic intuition of the rule writers. Unfortunately, 
linguistic intuition tends to be unable to control the 
process properly for a wide variety of inputs, because 
of the huge amount of data and the huge number of 
situations that need to be considered. However, rule 
writers must rely on their linguistic intuition to some 
extent, because there is no linguistic theory on lexieal 
transfer \[7\]. Another reason \[81113 \] is that tile transfer 
task is inherently a conglomeration of individual lex- 
ical rules. Therefore, the transfer process can be said 
to fall into a class of problem that cannot easily be 
controlled by the linguistic intuition of rule writers. 
In accordance with these observations, various at- 
tempts have been made to overcome the problems 
of transfer; they include knowledge-based MT \[12\], 
bilingual signs \[13\], and Tags for MT\[1\]. One such 
approacb is case-based or example-based MT \[4\] \[9\] 
\[10\] \[11\]. The essential idea behind all case-based MT 
(CBMT) methods is that tile system chooses the case 
(or example) most similar to tile given input from the 
case base, and applies the knowledge attached to the 
chosen case to the input. 1 
Supposing that there is a corpus of parsed transla- 
tion examples in which corresponding parts are linked 
to each other~ we can regard those parsed transla- 
1 This approach can be regarded as an application of case- 
baaed tea.sorting \[3\] to ntttural language translation. 
tion examples as translation rules. A promising ~rp- 
proach is therefore to make a transfi~r process that 
(1) chooses a set of translation examples, each source 
part of which is similar to a part of the input~ attd 
all source parts of which overlap the whole input~ 
and (2) constructs an output by combining the target 
parts of those translation examples chosen. However, 
this does \]tot mean that existing transfer knowledge 
should be abandoned. Rather, such transfer knowl- 
edge should be used ms a fail-safe mechanism if there 
are no appropriate examples. In the similarity-dr~iven 
t,unsfer system (Simlmn) we have developed, both 
translation examples and existing transfer knowledge 
are treated uniformly as trauslation pattern% and are 
called translation rules. 
In Figure 1, for example, (a) is tile parsed depen- 
dency structure of an inpnt Japanese sentence, "kare 
ga kusuri wo numu." Suppose that (b) is selected as 
the most similar translation rule for the part "kare ga 
... nomu" frmn the translation rule-base, and that (c) 
is selected as the most similar translation rule for the 
part "kusuri wo nomu~" even though there are several 
translation candidates for the Japanese verb "nomu." 
This figure illustrates what we would like to do; that 
is, to construct (d), the translated structure by com- 
bining the target structures of the selected translation 
rules. 
To develop this kind of system, we must consider the 
following issues: 
(a) a metric for similarity, 
(b) a mecbanism for combining target parts of rules, 
and 
(c) correspondence between the source part anti the 
target part of a rule. 
To handle the last two issues, I developed a 
model called Rules Combination Transfer (RUT) \[14\]. 
SimTran is RCT coupled with a similarity calculation 
method. In tbis paper, I will introduce RCT and the 
similarity calculation method used in SimTran. 
The next section defines the data structure for graphs, 
aud the format of a translation rule. Section 3 
presents a method for calculating the similarity be- 
tween an input and the source part of a translation 
rule. Section 4 describes the flow of the transfer pro- 
cess in RCT. Section 5 gives examples of translation 
using SimTran, and Section 6 discusses related work. 
Some concluding remarks bring the paper to an end. 
AcrEs DE COLING-92, NAN2T~, 23-28 AOI~T 1992 7 7 0 PROC. OF COLING-92, NANTES, AUG. 23-2fl, 1992 
.....-"...... .... ........ .. 
l:igure 1: Sample Japanestv4o.English tr~u,s\[ation 
2 Translation Rules 
A basic type ,ff gra.ph used in this paper is a labeled 
directed graph, or art Ida. 2 At, ldg G consists of a set 
of nodes N, and a set of arcs A. Further, each node 
and art: has a label, ht particular, node labels are 
unique. Each node consists tff features, each of which 
is a pair of a feature name attd a feature v~lue. 
If an ldg lta.~ only one root node, then it is called ~n 
rldg, and if an Ida has no cyclic pr~th, then it is called 
an idag. s Therefore~ an ridag denotes an Ida that 
h~-s only one root node and no cyclic path. 
A translation rnle 4 r consists of the folk,wing three 
corrtpo,leots: 
r = (G,,,,M,G~) 
where Gm is a matching gr~rph, G~ is a construction 
graph, e.nd M is a set of mappings between Gm and 
A matching graph G',,, and a construction graph G~ 
must be at lea.st an rldag. 5 Further, nodes in (~,, 
must be labeled uniqnely; that is, each node in G,,, 
mnst hz~ve only one unique label, and the l~bel of the 
node n~ in G~ is determined to be the label of the 
~The term qabeled' means that nodes and arcs are labeled, 
and the term ~directed' means that each arc has a direction. 
Further, an Ida in this paper refers to a connected graph unless 
otherwise specified. 
ZThe term dag is often used in the NLP world, and usu- 
ally denotes a rooted connected labeled (as functional) directed 
graph. But in this paper, dag denotes a direct,:d acycllc graph 
that may have multiple toots, is not necessarily a connected 
graph, and does not necessarily itave labels. 
4In this paper, the term rule does not mean a procedure, 
but rather a pattern of translation knowledge. 
bSudl graphs are sufficient to express almost MI lingu~atlc 
strsct ures. 
Figure 2: Samph. rule for translation between 
Japanese ~tnd English 
node nm in G,. such that n:. = M(nm). 
Mat)ping between (:.,~ and G~ designates tile cor-. 
respondences be,wee. ,,\[}des in G,. and (;.. l'})r 
instance, in Figure 2, tim Japanese word "nagai" 
("tong") should c.rrespond to both of the English 
words "have" and ~\[(lll~111 bl!cal,se if am),her word 
g.ow~rn.~ the word "nagai" then its English ,re,rela- 
tion should be connected to the word "h~Lve." On the 
other hand, if the Japanese word "to,elan" ("very") 
modifies "nagai" then its English translation "very" 
should be connected to "long." This shows tllat fi)r 
node in ~ source languag% two kinds of connection 
point, for translations of both governing structures 
attd governed structures of the node, are needed in its 
translated structure. This implies that there shouht 
be two kinds of correspondence between G',, and (7~, 
namely, (I) a mapping from a G,, node n,, to a G~ 
node nc that is to be a node connected to translations 
ACq'ES DE COLING-92, NANqT!S, 23-28 AOUq" 1992 7 7 1 PP.OC. OF COLING-92, NAtWrES, AUG. 23-28, 1992 
of structures governing nm, and (2) a mapping from 
n,, to a G~ node n'~ that is to be a node connected 
to translations of structures governed by n,~. We call 
the former an upward mapping and the latter a 
downward mapping, and denote these twn kinds 
of mapping as follows: 
where M T is upward mapping, and M ~ is downward 
mapping. 
Not all kinds of mapping should be permitted as M \[ 
and M 1. A translation rule r=( Gm,M,Gc ) must 
satisfy the following conditions: 
(1)M T and M I are both injections, 
(2) there are no two distinct nodes x aml y in G.~ 
such that M(x)=M(y), e and 
(3) M l(root(G,,,)) .... t(a~). 
Condition (1) ensures that there is only one c()n- 
nection point in G~ for each translation of gow~rn 
ing structures and governed structures, coudition (2) 
ensures that the label of a G'~ node is determined 
uniquely, and condition (3) ensures that the result of 
this transfer model becomes a rooted graph (see \[15\] 
for details). A rule sat.isying these conditions is said 
to be sound. 
3 Similarity Calculation 
This section desribes how a similarity is calcuhm~d. 
3.1 Graph Distance 
The shnilarity between a Gm and an input graph Gi,, 
is defined as the inverse of the graph distance 7 be- 
tween thenL First, the simple graph distance D; be- 
tween Gi,, and G~ is given ;ks follows: 
D',(G~, a..) = o=(n~., R.,) 
+ E,,, min(D'a(VS(Ri ..... ),GS(t~,, .... ))) 
where R/, and /~ are roots of Gi~ and Gm, respec- 
tlvely~ D,, is a node distance, a= is an arc in G,n such 
that its source node is R.m, and GS(n~ a) denotes a 
subgraph that is related to an arc a from n. 
Briefly, a simple distance is the sum of the node dis- 
tance between two roots and the sum of the minimal 
simple distances between Gin subgraphs and Gm sub- 
graphs that, far each arc a outgoing from the GmmOt 
node, are related to the all arcs a from the root nodes. 
~This means that either M ~(x) or M l(x) is equal to either 
M T(Y) or M .~(y) 
rDistltnces defined in this section are not actual distances 
in the mathematical sense. 
However, the larger Gm is, the larger this simple dis- 
tance becomes. Therefore~ when normalized by the 
number of nodes in G,,,, the graph distance Dg is 
given as follows: 
D;(Gin,G,,,) Dg(Gin, am) -- N 
where N is the number of nodes in G~. 
3.2 Node Distance 
When considering the distance between two words 
(nodes), we usually think of their semantic distance 
in a semantic hierarchy. In general, no matter what 
semantic hierarchy we use, it is inevitable that there 
will be some sort of distortion. Further, ,as stated be> 
fi)re, a node consists of several features and may not 
have a lexica\[ form that is a pointer to a semantic hi- 
erarchy. Therefore, a promising approach to calculat- 
ing distances between nodes is to use both a semantic 
hierarchy and syntactic features~ that is, to use syn- 
tactic features to correct the distortion contained in 
the semantic hierarchy to some extent. 
The node distance between a Gin node n i and a G,,, 
node nm is detined ms follows: 
Dn (hi, nm ) D/+ D, * 6, 
N S +a. 
where DI is a feature node distance, D, is a semantic 
no(h." distance, N I is the number of features in nm for 
DI, and 6, is the weight of a semantic distance. 
The semantic distance D, between a Gi,~ word wi,~ 
and a G,, word wm is given by the following equation. 
In SimTran, Bunrul Goi Hyou \[5\] code (or bghcode s) 
is used for calculating the smnantlc distance between 
Japanese words. 
Do(wln, wm) = 
0 Win ~ Wm 
0.5 wiT~ or wm is unknown 
1 win and w,,, are unknown 
I~°h(~')-@h(~')l+~ otherwise bghmax-F~ 
where bgh(w) is the fraction part of the bghcode of 
w, bghmax is the mammal difference between two 
bghcode fraction parts, and 6b is a penalty incurred 
if two words are not identlcM. 
The feature distance l)f between a Gi~ node hi,, and 
a Gm nmle nm is given ms follows: 
D:(n~ ........ ) = E:~., d:(n.,, f) 
df(nin, fn : fv) = 
1 fi~(fnin : fvi,,) whose fni,~ = fn, and 
fv is consistent with fVln 
0 otherwise 
s A bgheode is a fraction of number. Its integer part roughly 
corresponds to a syntactic c~tegory, and therefore, only its frac- 
tion part is used. 
ACRES DE COLING-92, NANTES, 23-28 AOm" 1992 7 7 2 PROC. OF COLING-92, NANTES, AUG. 23-28, 1992 
.//" 
Each matching pivot in ~t simibtr i-cover rule set must 
have M I or M 1, to ensure that tim Gcs of the i 
cover rllle set pr(lduce a t:ounected graph a~s a result. 
If there atre rules in the given i-cover rule set that do 
not s~ttlsfy this condition, they are renloved from the 
set of ruh, camlidates~ and the cover search method 
is executed until an i cover rule set th~.t satisfies this 
conditinn is found. Such as, i-cover rule set is called 
a proper rule set. 
Next, for each projection nf the given i-cover, we nmst 
make ;t copy of its origin rule~ m" rule instance, be> 
C;-LUSe one ride IEay make lllort • thgn oue project(tin 
un (~in ' 
Figure 3: An isomorphic cover 
In the ~bove equatiolb tile consistency checking de 
pends on a feature. 
4 Rules Combination Transl~r 
In this section, I present tile tlow of the transduction 
process by using RCT formalism. 
4.1 Rule Selection 
A transfer process rnust first find a set of rules whose 
Gins' matching parts (called projections) totally 
overlap all input structure, and which is the most 
similar to the intmt. We call a uuimi of projections 
a cover, and a cower identical to the input an iso- 
morphic cover (or i-cover). In or(her words, wha'~ 
we want here is the i-cover th;~t is the most similar 
to the input. Further, if a G., make ~L llrojection pj 
on a Gi~, then tile G,a is called the origin graph 
of the pj. A pivot is a node of (;~,~ that has more 
than one origin graph, attd a matching pivot is the 
origin node of a pivot. For instance, in Figure 3~ A 
and D are pivots. 
There may be some methods for tinding such an i- 
cover rule set. One method is to pick up a rule whose 
projection does not have any arc ow~rlapped by 
cover by other selected rules until there ~tre no un- 
covered arc% if it is desirable that a rule set should 
}lave few overlaps as possible. We h;tve Klso developed 
auotlmr method using dynamic programmiug: which 
can choose the most similar rule set from cttndidate 
rule sets. Briefly, it stores the most similar rule set 
for each combination of arcs of each node from Ice.yes 
up to the root~ and the most similar rule set stored in 
the root node is tile one for the input structure (see 
\[6\] for details), 
4.2 Prc-Lexicalization 
It may It~qqlen that ~ lexit:al-hIrm of a 6'~ in the given 
rub! iust~tnce is lint ~t \[uuldldat~! translation word of 
its correspoudiltg word in the input, because a lexica\] 
form in a. l,~tci,iug node it, its G,. is not necessarily 
the same as the input word. hl this (:~e, such a node 
is lexlcMized by c~L.dida.te tr~tnslation words. 
4.3 Node Labeling 
The label of a (d,,, node becomes tit(." I~bel of its 
mateillng nude in (;~,,. Since (;i,, nodes are labeled 
uniquely, (¢,..odes are idso I~}mled uniquely. On the 
uther h;md, the label of a (7,: nude n~ becomes the ttt- 
bel of a (,',,, node (n,,~) such that ~'z~ = M T(nm) or 
'\['here nlay 1 h(lWeVl~r I be twn nodes ill (Jc ill ;¢ rule 
inst~ulce that are mapped by ;t node in (;,~ with M \] 
~.nd M ~, respectiwdy. In the succeeding process, (1~ 
nodes with the same bLbel are merged into one node in 
order to gener~.te an mltpul structure, lu this phase, 
tim transferred hdmls of these two nodes shoulcl be dif 
ferent~ becnuse the two (lodes should not be merged 
f.r this rule. We must therefore relabel G~nodes of 
rule it|stances as follows: 
G~ Node Relabeling: for any label l i,, G~, if l is 
distrilmted t\[) twt) distinct uoch!s of (;~ by troth M \[ 
and M ~ fronl a node (,f (;,,,, then a I~bel l iu a G~ 
tulde, which is mallped only by M \], or is mapped 
by both M \[ ~tnd M .{ ~tnd has no descendants, is 
Cil\[tUg{!d to I ' I 
4.4 Gluing 
Unificatior~ is ~t well-known c(unput~tiuual tool for 
c(mm.cting gra.phs, and is widely used in natural lan- 
guage l)rocessing. Usually, unitlcation uses two func- 
AcrEs DE COLING-92. NANTES. 23-28 Ao~'rr 1992 7 7 3 P~oc. oF COLING-92. NANTI~S. AUG. 23-28. 1992 
y 
(a) 
( 
x 
( 
(b) (e) Gluing of (a) and (b) 
Figure 4: Example of gluing 
lionel rldags as data and unifies them front the root 
node down to the leaves. In RCT, however, we want 
to merge those nodes of two graphs that have the 
same labels, even if their root nodes are different and 
they are not functiona L as shown in Figure 4. Unifi: 
cation, however, cannot proceed in this manner, be- 
cause it unifies two nodes that occupy the same p+ 
sition, and always starts from the root node. For 
instance, in Figure 4, even if unification starts from 
node B then it fails, since it tries to unify node D of 
(a) and node C of (b) for arc y. 
In Graph Grammars, this method of connecting two 
graphs is called gluing \[2\]. The ghfing used in 
Graph Grammars is not concerned with the con- 
tent of a node, so it must be extended in order to 
check the consistency among the nodes to be glued. 
in SiraTi'an, if two features conflict then the feature 
whose rule is more simi\[ar to the input is taken. 
Briefly, gluing is performed as followsg: ICivst, nodes 
with the same label are me~yed if they are consistent. 
If arty nodes fail to be merged , then the ghdn 9 also 
fails. If all the me~ges succeed, all ares are reat- 
lached to the original nodes, which may or may not 
be me~yed. As a result, some ares with the same la- 
bels and attached to the same nodes may be me~ed, 
if they are consistent. 
A glued graph is not nece~arily a cmmeeted, rooted, 
or acyclic graph, but we usually need a connected 
rldag iu natural language processing. Several con- 
stralnts satisfying such requirements are described in 
previous papers \[14\]\[15\]. 
After the G~s have been labeled and relabeled, the 
target structure is built by gluing the G~s. 
ODetMls of tire algorithm are given iu previous papers 
\[141115\]. 
4.5 Post-Lexicalization 
The constructed target structure is still bnperfect; 
there might be a G~ node thai. has no lexical-form, be- 
cause there are some rules made froul transfer knowl- 
edge that have no lexlcal-forms. Therefore, as in the 
pre-lexicalizatiou phase, non-lexical G: nodes are lex: 
icalized. 
5 Examples 
This sectimt gives examples of translation by 
SimTcan. Figure 5 shows how the Japanese sentmme 
"Kauojo no me ga totemo kireina no wo sitteiru" is 
translated blto the English sentence "(1) know that 
she has very beautiful eyes." In this figure, (a) is 
an input sentence structure, (b),(c), and (d) are rules 
(precisely, rule instances), and (e) is the output struc- 
ture produced. In these rules, a mapping line not 
marked M ~ and M ~ has both M ~ and M ~. Dotted 
lines designate matching or gluing correspondences 
between rule nodes and input or output nodes, re- 
spectively. I:'urther, numbers prefixed by '*' denote 
node labels. In this example, we assume type hierar- 
chies in which, for instance, 'yougen(predicate)' is a 
super-category of 'keiynu(axlj)', and "kaut6o(she)" is 
an instance of :hnmau'. Note that the node labels of 
both "have" in rule instance (c) and lower 'pred' in 
rule instance (b) are changed from that of the corre- 
sponding Japanese word "kirei(beautiful)" by the G¢ 
node relabeling procedure. 
Another example is shown in Figure 6, which shows 
how the Japanese sentence "US ga ... wo fusegu tame 
ni buhit/ul kanzei wo kakeru" is translated into the 
English sentence "US imposes tax on parts in order 
to blockade .... " In this example, (a) is an input 
structure, (b), (c) and (d) are matched rules, and 
(e) is the output structure produced. The Japanese 
verb "kakeru" has several trauslation candidates as 
sociated with different governing words, as shown in 
the following +~able: 
Similarity dapaues+Eng/ish 
5.988 (meishi) ni zeikiu wo kakeru 
impose tax on (noun) 
3,077 (meishi) wo salban ni kakeru 
take (noun) to court 
2.717 (meishi) wo mado ni kakeru 
hang (noun) in window 
2.545 (meishi) wo sutoobu ui kakeru 
put (noun) on stove 
haukati ui kousui wo kakeru 2.040 
spray perfume on handkerchief 
This table lists the top live similar rules for the part 
"buhin ni kanzei wo kakeru" of the input. As shown 
ACTI~ DE COLING-92. NANTES, 23-28 AOt~q" 1992 7 7 4 PROC. OF COLING-92, NANTES, AUG. 23-28, 1992 
., £+..--- --.. "r'+ 
*3 lit "LI' hive ~' 
~I !-\o, ,,T",. ,~/<.~ ...... >~.---7'-~xt ..~£.i Z-~ "~',,. ,--L-/')-% -" ~.+-,~L<, 
I \,t.--? tz-'.9" t_7.9 ",'U ,,I "i~:...>. ~" ,~r ~ 
t'3..-,7_L.,. ........... ~: ........... +_"~ "~I.~ 
\ "~I I "*~ i" .......... ('~ 
~....:.,..~-r~, fSb,7. ....... 
Figure 5: Exami~le 1 of translation by SimTi'~n 
,.-"\[" "2 ~.z ~_d'2 " .... 
///" \[ (b) ~',,.\. ",..,. 
I,} "/" " 
'2 / '5 ", 'l \ "6 
t ! t+ 4 _MI ~ " -" *3 / t '4 \[ / '3 / ~ 4 mort 'x ', ..~" / ~.." : 
/ : '~'e '.. >~ .. " F ' / ",/-" .. .. ........ .-."~ ! 
i / ' ,/" ",....-.:<i ................ "., / / i i \--~" .,. ....... :::~':(L~ f~ / 
" / / "" /" 'x" , B';'~") ~ pr }pr,~l ,! " 
i '{l(,,t,~L..j .............. /..d.,F..~, i ", .5" " '~ ,' 
(d) 
(c) 
Figure 6: Example 2 of transl;~tiou by Sim!l'~n 
A~l.:s BE COLING-92, Nnlqi~;~;, 23-28 ^Ot~T 1992 7 7 5 PROC. OF COLING-92, NANTES, AUG, 23-28, 1992 
in this table, rule (c) is the most similar one. Note 
that this similarity calculation was done for all rules, 
including non-lexical translation rules. There were 
no appropriate example rules for the part "US ga 
kakeru," and a non-lexical rule (b) was timrefore se- 
lected. Further, note that the lexical forms in *3 
nodes of (c) and (el are different, and that *4 node 
of (el has no lexical form other than a preposition, 
whereas "4 node of (el has a lexical form. The for- 
met was obtained by pre-lexicalization, and the latter 
by post-lexicaiizatiml. 
6 Related Work 
Although there were several early experimental 
projects on CBMT \[4\]\[9\]\[11\], MWF-H \[10\] is the first 
working prototype of a case-based transfer systern~ 
and demonstrates the promise of the CBMT alr- 
proadL It uses Japanese-to-English translation ex- 
anlples as translation rules: chooses the source trees of 
examples that are most similar to the iuput tree from 
the root node down to the leaves, and assembles those 
target trees to produce an output tree, With respect 
to the transducing mechanism, MBT-II is a tree-to- 
tree transducer adopting one--to-one correspondeuce. 
MT by LTAGs \[1\], although it is not an attempt of 
CI3MT, proposed a similar mechanism to RCT de- 
scribed in this paper. It uses paired derivation trees 
of English and French as translation rules. An input 
sentence is parsed by the source grammar, and at 
the same time, its output tree is generated by deriva- 
tion pairs of trees used in the parsing. As a traus- 
dueer~ this mechanism is also a tree-to-tree transducer 
adopting one-to-one correspondence. 
In contrast, the RCT employed in SimTran is a rldag- 
to-rldag transducer adopting upward and downward 
correspondences. These extended correspondences 
are desirable for expressing the structural discrepan- 
cies that often occur in translation. Moreover, this 
transducing model is a parallel production system \[2\] 
that Call produce an output structure in one execu- 
tion of gluing if all the G~s required to produce an 
output are supplied, 
7 Conclusion 
In this paper: 1 described a cas~based transfer sys- 
tem, SimTran, which combines I~CT with a similarity 
calculation method. RCT has powerful correspon- 
dences between the source structure and the target 
structure of a translation rule, which can express 
most structural discrepancies between two languages. 
As a transducing mechanism, RCT is a parallel non- 
destructive rldag-to-rklag transducing system. I also 
propose a similarity calculation method for graphs 
whose nodes consist of syntactic and semantic fea- 
tures, and show that a translation rule th~tt has no 
\[exical forms can he used ms a default rule, that is, 
that such rules can provide a fail-sMe mechanism if 
there are no appropriate translation examples. 
References 
\[1\] Aheilld, A., Schabes, Y., and Joshi, A. K., "Using 
l,exiealized Tags f,~r Machine qYanslation," Prec. of 
Coling 90, 1990. 
\[2\] Ehrig, 11., "Introduction to the Algebraic Theory of 
Graph Grammars," Prec. of Inle,'national Workshop 
on Graph Grammars, LNCS 73, 1-69, 1979. 
\[3\] l(olodner, J. and ILiesbeck, C., "Cmse-Bmscd Reason- 
ings" tutorial textbook of llth IJCAI, 1989. 
\[41 Nagao, M., "A Framework of a MechanicM Transla- 
tion between Japanese and English by Analogy Prin- 
ciple," Elittmrn, A. and Banerji, R. (eds.) : Arlificisl 
and Human lntelhgence, NATO 1984. 
\[5\] National Language ltesoarch Institute: Bunrui Got 
Hyou (in Japanese), Syuuei Syuppan, 1964. 
\[6\] Marnyama, ft. and Watanabe, l-l., "Tree Cover Search 
Algorithm for Example-Based Translation," Prec. of 
4th 1st. Conf. on Theoretical and Melhodological 1s- 
sues in Machine Translation, 1992. 
\[7\] Melhy, A. K., "l,exical Transfer: A Missing Element 
in Linguistic Theories," Prec. of Coting 86, pp. 104 - 
106, 1986. 
\[8\] Nitro, Y., "Idiosyncratic Gap: A Tough Problem to 
Structure-hound Machine q~'anslation," Prec. of Col- 
in~ 86, pp. 107--111, 1986. 
\[9\] Sadler, V., "Working with Analogical Semantics," 
Forts Publications, 1989. 
\[10\] Sate, S. and Nagao, M., "Toward Memory-based 
Translations" Cohn9 90, 1990. 
\[11\] Sumita, E., lida, 11., attd Kohyama, 11., "Translating 
with Examples: A New Approach to Machine 'l~'ans- 
lotion," l'roc, of lnfo Japan 90, 1990. 
\[12\] Tomita, M. and Carbonell, J.G., "Another 
Stride Towards Knowledge-Ba~qed Machine Transla- 
tion," Prec. of Coling 86, 1986 
\[13\] Tsujii, J. and Fujita, K., "bexical ~lYansfer bmsed on 
bilingual signs: Towards interaction during transfer," 
Prec. of Seoul Int Conf. on NLP, 1990. 
\[14\] Watanabe~ ll., "A Model of a Transfer t'roccss Using 
Combinations of Translation Rules," Proc. of Pacific 
Rim of Int. Conf. on AI '90, 1990. 
\[15\] Watanabe, tt., "A Formal Model of Transfer Using 
Rules Combination," submitted to Machine Transla- 
tion, 1991. 
AcrEs DE COLlNG-92, NANTEs, 23-28 AOt'Zr 1992 7 7 6 PROC. OF COLlNG-92, NANTES, AUG. 23-28, 1992 
