Pattern-Based Machine Translation 
I(oichi Ta,keda, 
Tokyo l:\[es(:a.r('.h La.1)(>ra.Lory, IBM R(~sea.r('h 
\] 623-:14 Shimol;suruma., Yama, I;o, Kasmgawa. 242, Japan 
Photo,: 81-462-73-4569, 811-462-73-7413 (FAX) 
takeda@trl, vnet. ibm. com 
Abstract 
In this 1)a,l)er, we (les(:ril)(~ a. "l)a.ttern-I)a.s(,d" ma.c\[fine 
transla.tion (MT) a.pproa.(:h tha.t we followed in designing 
a. l)(:rsona.1 tool for ilS(:rs who \[1,3,VO, ;t,C(:(~SS tO la.rg(: v()Ittlll('.s 
of text in la.ngua.ges other tha.n their owu, su(:h as WWW 
pa.gas. Sore(: of the critica.1 issues involv(:d it, th(: design 
of such a. tool in(:htd(: (:a.sy (:ustomiza.tion for div(:rse (h>- 
mMns, th(: (;tfi(:iency of th<: tra.nsla.tion a.lgorithm, a.nd 
sea.lability (in(:r(:menta.l improvcm(:nt in tra.nsla.tion qua.1 
ity through us(:r intera.ction). W(: a.lso describe how our 
pa.tterns fit into the (:ont(:xt-fr(:(: l)a.rsing aml g(:nera.tion 
a.lgorithms, a.n(1 h(-)w wo, intl)h:tn(:rtted a, prototyp(: tool. 
1 Introduction 
It wouhl b(: difficult, for a.nyon(: to dist)ute tit(: id(:a, tha.t 
th(: World-With: Web (WWW) has b(:(:tt tit(: most phe- 
rtom(:na.l invention of tit(: la.st d(:ca.de in the (:t)mlnH;ing 
(;nvironnwnt. It ha.s suddenly OltCncd up a. window to 
wtst a.mounts of da.ta, on the \[ntcrnet. Unfortunately fin' 
those, wit() axe not na.tiv(', English Sl)(;a.kers , textua.I da.ta. 
axe mort: often tha.n not written in a. foreign hmgua.ge. 
A doz(;n or so ma.(:hino tra.nsla.tion (MT) tools ha.w; 
recently bean put on the, ma.rket, to make such te, xtua.l 
da.ta, more a.ccessibh', but novice PC us(;rs will be simply 
a.ma.zed a.t the mea.g(:rness of their rewa.rd for th(: effort 
of building a. so-(:a.lh:d "user di(:tiona.ry." 
'\['lm main r(:a.sons tbr tiffs prol)h:m a,r(:: 
:1. Most MT systems do not employ a. l)ow(wful "lexi- 
ca.list" forma.lism. 
2. Most MT systems ca.n lm customized only by a.ddittg 
a. user dictiona.ry. 
Thero, for(~,, ,ls0,rs ca,It neither giva prefe, re.nct~,s Ol, i.divid- 
ua.1 prel)ositiona.l-1)hra.se a.tta.chments (e.g., to ol)ta.in in- 
forma.tion from a, server) nor deiinc tra.nsla.tions of spe 
tiff(: verb-object pa.irs (e.g, to take advantage of some- 
thing). 
Powerful gra.mma.r forma.lisms a.nd h;xica.l-sema.ntics 
forma\]isnts ttawe, I)een known for yea.rs (see l,F(~(Ka.l)la.n 
a.nd l/restore, 1982), 1 tPSG (Polla.rd a.nd Sa,g, 1987), a.nd 
Ge, ncra.tive l~exicon(Pustejovslcy, 199l), for example), 
bttt pra.ctica\] iml)h'~me.nta.tion of a.n M:I" system ha.s yet 
to tax:kle, the computa.tiona.l colnl)lcxity of pa.rsing a.lgo 
rithms fin' these formalisms a.nd the workl(m.d of building 
st. la.rgc sca.lc lexicon. 
I'~xaml)le-based MT(Sa.to a.nd Na.ga.o, 1990; Sumita. 
a.nd lida., 199l) a.nd sta.tistica.I MT(Brown et a.1., 1993) 
a.l'c both promising apln'oa.chcs tha.t genera.lly demon- 
sl;ra,te, incrementaJ iml)roveme, nl: in tra,nsla,tion a.ccura,cy 
a.s tile qua.lity of examples or tra.ining da.ta, grows. It 
is, however, a,n olmn qttestion whether these a.pl)roa.(:hes 
a, lone ca.n be used to crca,te a. fltll-fh;dged MT system; 
tha.t is, it is uncerta.in whether such a. system ca.n be 
used tbr wt,rit)us dotnains withottt showing sever(~ (h;gra.- 
dation in trans/a.l:ion accuracy, or if it has to 1)(' tb, d IW 
a. r(:as(ma.1)ly Ia.rge, set of (',xaml)les or tra.ining da.ta tk)r 
(~,Hch IIQ, W (IOlIl;l,il,. 
TAGlmsed MT(Abeilld et M., 1990) I a.nd pa.ttcrn- 
l)a.s(;d tra.nsla.tion(Ma.ruya.ma., 1993) shaye ma.ny intl)or- 
~See l/l'A(l(Sch;dms el; a.l., 1!)S8)(l,exicalized TAt',) a.nd 
ta.nt propertie.s fiw successful im ple, menta.tion in 1)racticM 
MT systetns, namely: 
• The existence of a, l)olynomia,l-tinm imrsing a, lgo-- 
rithnt 
• A Ca l)a.bility fl)r describing a la.rger do',t,(tirl, of lo(:(tl~ 
ity 
• ,%'nch'ror~,izatior~, of the source, a,nd target la.ngua.ge 
structures 
In this pa.pcr, wc show tha.t thc;rc exists a.n attra.ctivc 
way of crossing these apl)roa.(:hes, which wa ca.ll pattern- 
based MT. 'e In l;he tollowing two s(~x:l;ions, we intro- 
duce a. class of tra.nsla.tion "pa.tte, rns" ha.sad Oli (2otltoxt- 
i tl "~ Free Gramma.l (CI G), and a, pa.rsing a.lgorithm with 
O(\]G\]'2n ') worst-case time COml)lexity. Furthcrnlore, we 
show tha.t our fva, nmwork ca.n I)c (;xttmded to incorpora, te 
exa.mt)h;-lmsed MT a,nd a powerful le, a rning mox:ha.nisn,. 
2 'lYanslation Patterns 
A tra, nsla, tion pa, ttcrv, is defined as a, pair of CF(~' rules, 
a,nd 7,(;to or more syntactic head a,nd 5'n,k constra,ints for 
uontermina\] symbols. For example, tim I!;nglish-French 
tra,nsla,tion lmttern 3 
NP:\] miss:V:2 NP:3 -, S:2 S:2 ~-- NP:3 manquer:V:2 h NP:I 
essentia,lly &',scribes a sy'nch'ror~,izcd "1 pa,lr consisting of a, 
lct't-hand-side I,;nglish CFG rule (ca,lied a, smiter: ruh;) 
NII'VN\['- ~ S 
a.nd a right-ha.rid side French CFG rula (ca.lh;d ;t tawet 
rule) 
S ~- Nil' V ?t NP 
a.ccompa.nicd by the following constraints: 
1. Head coimt;i'aints: Tha nontermina.l symbol V ut 
the source, rule must have the verb "miss" a.s a, syn- 
ta.ctic head. The symbol V in the target rule must 
haw; the, w,rb "manquer" a.s a. syntactic hea.d. The 
hea.d of symbol S in the source (ta.rge, t) r,ll0, is idan- 
tica.l to the hea.d of symbol V in tlw source (t~trg0,t) 
rule, as they a.re co-indexed, llea.d constrahtts c~m 
lm specified in either or both si¢h;s of the l)a.tterns. 
2. l, ink constraints: Nontcrmina.I symbols in source 
a.nd ta.rgct CF(~ rules a.ve lirt.kcd if they a.re given the 
same index ": i". Thus, the first NI ~ (NP:I) in the 
souv(:e rule, corresponds to tit(; second NI' (NP:I) in 
the ta.rgct rule, the Vs in both ruh;s correspond to 
e,a.ch other, a.nd the second NP (NI':3) in the source, 
vuh, corresponds to the first; NI ~ (N1:':3) in the target 
I"lt~. 
STA(~(Sldeb,,r and Sch;d)es, :1990)(Synchro,dz,d '\['A(;) fin each 
member of the TAt1 ('lh'oe Adjoining (;,'a.nlnl;u') \[';unily. 
21{ecently, 'l'vee Insertiml (;vamm;u'(Sdud.~s a,,1 Waters, 1 !19:0 
ha.s 1ram1 introduced to show ;t similar possilAlity. ()m" approach, 
however, is more ira:lined toward the {'.Ffg fornlali~,nl. 
SAnd il.s il.\[lectional va,ri;tll|;s - We will (\]ist:~,lss ;Igl'I!Olllell\[; issues 
lal.el', in the "l'3xl;euded l'kn'tua.lism" sect:ion. 
4'l'he IIle;~ltilH~ o\[' I;}|e word "synchronized" hm'(~ is ex;u:t\[y l;lm 
same as in STA(l(.%hielmr and ,~chal)¢~s, 1990). 
1155 
The source a.nd target rules, that is, the CFG rules with 
no constraints, are called the CFG skdeton of the pat- 
terns. The notion of a syntactic hea.d is simila.r to that 
used in unification grammars, although the hea.ds in our 
pa.tterns are simply ene\[~ded a.s cha.racter strings ra.ther 
than as complex feature structures. A head is typica.lly 
introduced 5 in pretermina.1 rules such a.s 
lea~e ~ V V ~- partir 
where, two w~rbs, '"leave" a.nd "partir," are associated 
with the heads of the nonterminal symbol V. This is 
equiwdently expressed as 
lea~e:l --* V:I V:I +-- pa.rtir:l 
which is physica.lly implemented as a.n entry of a. lexicon. 
A set T of transla.tion patterns is said to accept an 
input s iff there is a. deriva.tion se, quence Q for s using 
the source CFG skeletons of T, a.nd every head constra.int 
a.ssoeia.ted with the CFG skeletons in Q is sa.tisfied. Sim- 
ilarly, T is sa.id to translate s iff there is a synchronized 
derivation sequence Q for s such tha.t T accepts s, and 
every hea.d and link constraint associa.ted with the source 
and ta.rget CFG skeletons in Q is satisfied. The deriw> 
lion Q then produces a transla.tion t a.s the, resulting se- 
quence of terminal symbols included ill the ta.rget CFG 
skeletons in Q. Transla.tion of an input string s essen- 
tially consists of the following thre, e steps: 
• Parsing s by using the source CFG skeletons 
• Propagating link eonstra.ints from source to target 
CFG skeletons to build a ta.rget CFG deriva.tion se- 
qllen(:e 
• Generating t from the target CFG deriva.tion se- 
quence 
The third step is trivia.1 a.s in the case. of STAG transla.- 
tion. 
Some imlnedia.te results follow from the a.bove defini- 
tions.(Takeda., 1996) 
1. Let a. CFG gramma.r (4 be a. set of source CFG skele- 
tons in T. Then, T accet)ts a. context-free, la.nguage 
(CFL), de.noted by L(1 ), such tha.t L('I ) L(G). 
2. Let a CFG grammar H be a. subset of source CFG 
skeletons in T snch tha.t a. source CF(\] skeleton k is 
in It iff k has no head constraints assoeia.ted with it. 
rl3 tl(} II , :\[~ ~1'(; (;(' I' t S '1, ~11 \['~et IJ(\]~) Of I \[l,'\[gl' }l,g(~ L(~\]) . 
3. L(T) is a proper subset of L(G) if, for exami)le , 
there exists a. pa.ttern p (C T) with a sonrce CFG 
rule X ~ Xi '"Xv such tha.t 6 
(a.) p has a. head constraint h:X for some nonter- 
minal symbol Xi (i = 1,2,..., h). 
(b) T ha.s a, deriva.tion sequence X --4 . .. -4 'w such 
tha.t X is assoeia,ted with a head g (h, ;/: g), 
and T has no se, quenee of nonterminal symbols 
~q...}~ that derives exactly the same set of 
strings a.s X does. 
5A nonterminal symbol X in a source or target CFG rule X -~ 
XI"" Xk can only be consl.rained to have one of the heads in 
the RHS .X1 " ' Xk. Thus, monotonicity of he~d cnnstraints holds 
throughout the parsing process. 
"This is not a necessary condition for L(T) C L(G'). It is prov- 
able that for any set T of patterns, there exists a (weakly) equiva- 
lent CFG grammar F, with possibly exponentially more grammar 
rules, such that L(T) = L(F). A decision problem of two Cl.'l,s, 
L(T) C L(G), is solwdJle if\[ L(b') = L(G). 'Fhis includes an un- 
decidable problem, L(F) = E*. Theret'ore, we can conclude that 
L(T) C L(G) is mtdecidable. Similar discussions ean be found in 
the literature on Generalized Phrase Structure Grammar(Gaz(lar 
et al., 1985). 
Although our "pa.tterns" have no more deseriptiw'~ 
power than CFG, they c, an provide considerably better 
descriptions of the domain of locality than ordinary CFG 
rules. For example, 
be:V:1 year:NP:2 old --, VP:I VP:I 4-- avoir:V:il au:NP:2 
can h~ndle sueh NP pairs as "one yea.r" and "un a.n," 
a.nd "more than two yea.rs" a.nd "l)hls que deux alIS," 
which would haare to be covered by a la.rge numl)er of 
plain CFG rules. TAGs, on the other ha.nd, are known 
to be "mildly context-sensitive" gra.mma.rs, and they ea.n 
ca.pture a wider ra.nge of synta.etic dependencies, such as 
cross-serial depe, ndencies. The computational complex- 
ity of pa.rsing fbr TAGs, however, is ()(\[G\]n°), which is 
t3.r greaW, r than tha.t of CFG parsing. Moreover, defin- 
ing a. new STAG rule is not a.s easy for the users as just 
adding an entry into a. dietiona.ry, beca.use ca.oh STAG 
rule ha.s to be speeifie, d a.s a. pair of synta.etic tree struc- 
tures. Our pa.tterns, on the other hand, ca,n be spe, cified 
as easily as 
to leave * -- de quitter * 
to l)e yea.r:* old = d'avoir an:* 
by the users, lie, re, the wildcard "*" stands for a.n NP by 
defa.nlt. The prepositions %o" a.nd "de" a.re merely ttsed 
to specify that these patterns are for VPs, and they a.re 
removed when compiled into interna.\[ forms so tha.t these 
pa.tterns axe a.pplica.ble to finite a.s well a.s infinite forms. 
Simila.rly, "to be" is used to show that the phrase is a 
be,-verb and its complement. The wiklea.rds ca.n be con- 
stra.ined with a. hea.d, a.s in "year:*" and %a:*". It, addi- 
tion, they ca.n be a.ssociated with a.n explMt nonterminal 
symbol such a.s "V:*" or "A\])JP:*" (e.g., '"leave:V:*"). 
By defining a. few such nota.tions, these, pa.tterns ca.n 1)e 
successfully conw~,rted into the forma.1 representations de- 
fined a.bow:. The notations a.re so simple tha.t even a. 
novice PC user should ha.re no trouble in writing our 
pa.tte, rns, a.s if it(; or site were lnaking a. voca.bula.ry list 
for English or French ex~mls. 
3 Pattern-Based Translation Algorithm 
A parsing a.lgorithm for translation patterns ca.n be any 
of the known CFG parsing algorithms, including CKY 
and Ea.rley a.lgorithms. It should be first noted, however, 
that CFG could produce exponentb~lly ambiguous parses 
for some input, in which ease we can only apply heuristic 
or stochastic measurement to select the most promising 
pa.rse. 
It is known tha.t an l!\]a.rley-ba.sed parsing a.lgo- 
rithm can be made to run in O(\](;\]Kn a) :ra.ther tha.n 
O(JaJ2n:'),(iVla.ruya.ma., 1993; Graham et al., 1980) 
where K is the number of distinct nonternfinal symbols 
in the gramma.r G. We ca.n expect a. very etfide.nt pa.rser 
tbr our pa.tterns, r The input string ca.n a.lso be scanned 
to reduce the number of relewmt gramma.r rules before 
pa.rsing, e The combined process is a.lso known as offline- 
parsing in LTAC,. 
Handling aml)iguous parses is a. difficult task. The ba- 
sic strategy for choosing a candida.te pa.rse during Eaxley- 
based pa.rsing ix a.s tbllows: 
1. Prei~;r a pa.ttern p with a source CFG skeleton X --~ 
Xt'" Xk over a.ny other pa.ttern q such that the 
source CFG ske, leton of q is X -4 X,...Xt:, and 
such tha.t Xi in p ha.s a head constraint h, if q has 
h. : Xi (i = 1,...,k). The pa.ttern p is said to be 
mort: specific tha, n q. This relation is similar to a. 
subsumt)tion rela.tionship(Pollard and Sag, 1987). 
rSchabes and Waters(Schabes and Waters, 1995) also discuss 
sewu'al techniques for optimizing parsing algoritlmm. 
SSuch scanning is essential for some languages with no explicit 
word bounda.ries (such as Japanese and Chinese). 
1156 
2. t'refhr a. 1)a,ttern p with a. source (,I~ ~ slw, leton over 
(me with D, wer termina,t syml)ols tha.n p. 
3. l)refhr a. pa,tt('rIl p tha,t d(le.s not viola,te a.ny hea.d 
constraint ov(',r one tha,t viola.tes a. head constraint. 
4. Prefer tile shortest deriwl.tion sequence for ea.ch in- 
put sul)string. A pa.ttern ~br a. la.rger doma.in of 
loca, lity tends to give a. shorter deriva.tion s(,,qu(,nce. 
Thus, our stra.te.gy fa.vors h'xi<:alizcd (or hea(I- 
(:onstra.ined) a.nd <:ollo<:(ttional pa.tterns, which is exactly 
what we axe going to a.chi('ove with pa.ttern-l)a.sed MT. 
Seh,,ction of t)a.tt(',rns in tit(', deriva.tion s('XlU(mc(~ accom 
l)aldeS th(; constru(:tion of a. l;a.rg(',t (h,riwt.tion se,(luen(:(', 
I,ink constra.ints are prol)aga.ted fronl SOlll'(;(2 t() ta.rget 
derivation trees. This is basically a. bottom-up I)rO(:t: 
(111 I'lL 
Silt(:(', the numl)er M of distinct pa.irs (X,w), for ;1. II()ll- 
t(!rminal symbol (or a. ::hart) X and a. sul)s(~quen(:(~ 'Ill of 
ini)ut string s, is bounded by h",. 2, th(;re a.r(, a.t m(/.~t 
h'n:" l)OSsibh~ tril)les (X,w,h}, such tha.t h is a. head of X. 
Thus, we ca.n COml)ute the 'm,-Scst choice of tra.rtsla.tion 
(:a.ndMa.tes \[n 0(\]7'\]\[(,,,") tim(:. I\[(;i'(',, /i is the nlllnl)el' 
()t distinct ll`otll;o, rnliIl`'t.lm symbols in T, a.nd 'n. is |:he size 
o\[ the input string. 
The reader shoMd note. critical diff'er0,nce, s between h'x- 
ica.lized gra.mma.r rules (ill. the s(;It,'-;o, of UI'AG) a.Ild tra.ns- 
\[ati(in pa.tterns when they a.re used for M'\['. 
Virstly, a. pa.ttern is not nec(;ssa.ri\[y lexica.lized. An 
e(:ononfica.1 way of orga.nizing tra,nsla.tion pa.tt(',rns is 1;o 
include, non--lexica.lized pa.tterns as "d(ffault" tra.nsla.ti(m 
yules. I:or exa.mple, the pa.ttern 
V:I NP:2 ) VI):I VP:I ( V:I NP:2 
is used a.s a. (hffa,ult tra.nsla.tion of "verb + dir(:(:t object" 
(,~xpressi()ns, but 
resemble:V:\] NIl:2 ~ Vlhl VP:I ~ res(ullblluuV:l it NIl:2 
is a.lways prelhrred over the default rule, I)(ma.use of (mr 
i)r(',fe, r(;nce stra.te, gy. Sitnila.rly, tho, pa.ttern 
please \:l':l ~ VI':I V\[':I ~ \:1':1 , s'il veins 1)\]a:d, 
should I)e liv(;foxred over a. h~xi<:alized t>a.l;t(n:n, if a.ny, 
AI)VP:I xxx:Vl':2 =~ VI':2 VI':2 +- AI)VP:I yyy:Vl>:2 
S(',c(mdly, lexica.liza.tion mighl; consido, ra.1)ly increase the 
stz(*, of ~ lAG gra.mma.rs (m pa.rticula.r, compositiona.l 
gra.mma.r rules such as A\]).IP NP -} NI)) when a. la.rge 
nulnb0,r of lexica.I items axe a.ssocia.ted with 1;hem. Since, 
it is not tlltllSlla,1 fol" a, ItOllll in a, SOllFC(; laIlg~tla,gj(? to ha,ve 
severa.l counterpa.rts in a. ta.rget la,ngua,ge, the number of 
tr(:e-pa.irs in STAG would grow much la.rgo, r tha.n tha.t 
of sour(:('. I2L'AG tre,(;s. Although in I:I'AG the gram- 
ma.r rules a.re (lifferentia.t(;d from their physica.l ol),jacts 
("pa.rsc'r rules"), a.nd "structure sha.ring"(Vijay-Sha.nker 
and Scha.bes, 1992) is propos(;d, this ambigMty r('ma.ins 
lit the pa.rser rllles~ too. 
Thirdly, a. tra.nsla.tion pa.ttern ca, n omit the tree stru(:- 
tur(: of a. (:olloca, tion, h,,a.ving it as just a. s0,(lU(',n(:e of 
termina, l symbols. }Pot" exa.ml)h',, 
See y(m later, NP:I , S S ~-- At, revoir, NP:I 
is perthctly a,c(:eptabh; a,s ;/, tra.nsla,tion pa.ttern. 
4 Extended Formalism 
Syntactic depend(umi(',s hi` na.tura.l \[a.ngua.ge s(~nt(',n(:o,s a.re 
so subth', tha.t ma.ny powerful gra,mmar forma.lisms ha.re 
I)e(;n l)roposed to account for them. The a.deqtmcy of 
CVG for des(:ribing na.tura.1 la.ngua.ge synta.× ha.s long 
l)eett questione, d, a.nd unifi(:a, tion gramma.rs, among oth- 
ers, ha.v(' been used to buihl a, pre(:ise theory of the, com- 
puta.tiona.l aspects of synta.ctic d('.t)(mdenci('.s , which are 
des(:ril)ed by tit(', notion of unifica.tion a, nd by fea.ture 
stru(:t ur('.s. 
Transla.tion pa.tt(;rns ca.n also 1)e ext(mded by m(;a.ns of 
unifi(:a.tion a.nd fea.tur(, structures. Such (',xtensh)ns lntlst 
be ca.refully a,l)t)lied so that they do not sax:rifice tit(', et u 
fici0,ncy of pa.rsing a.nd genera.tion a.lgorithins. Shi('J)(:r 
a.nd Schabes brMty dis(:uss the issu(',(Shiel)(~r a.nd Sch- 
abes, 1990). We can a.lso extend tra.nsta.tion l)a.tterns as 
fbllows: 
\[:',ach noilt(~rmirull node in a. pattern can be a.s 
socia.t0d with a. ti×ed length vc(:tor (if binary 
fcatu'rr:,s'. 
This will o, na.I)le, us to st)ecit~y such synta.ct, ic (h;po, ndencies 
as agreement and sulma.tegoriza.tion in 1)atterns. \[Jnifi- 
cation of Lina.ry featl,res, however, is much simphu': uni- 
fication of a. t'ea.ture value pair succeeds only when the 
imir is either ((),0) or(I,l). Since the. fl'at,H'e vector has 
a. fixed langth, unifica.tion of two t'eaturc vectors is per- 
formed in a consta.tlt time,. For o, xample, the pa,tterns 
V:\]:qTI{ANS NP:2 ~ VP:I VP:\] ~-- V:l: kTI{ANS NIh2 
V:I:+IN'ITRANS ~ VP:I VI':\[ *--- V:l:q INTHANS 
are unifiable, with tra.nsitiw; a.nd intra.nsitive verbs, re- 
spectively. We can also distinguish local and head fea.- 
tures, a.s postula.ted in I\[I)SG. Verb subca.tegoriza.tion is 
th(',II` encoded a.s 
VI':I:+TItANS-O|L\] NP:2 ~ VI':t:+OIL| 
VI':I:-F()ILI ~- VP:I:-k'\['RANS-OI~,J NIl:2 
where "-()ILl" is a. hma.1 fea.ture for hea.d Vl's in I,ItSs, 
while :' k()ILl" is a head featurt; for gl)s ilt the l{.\[\]Ss. 
\[Inifica.tion of a ht;ad fea.ture with q ()ILl succeeds when 
it is not bo'tmd. 
Another extension is to associa.te wo, ights with fleet- 
terns. It is then possilih', to ra.nk the ma.tching lmtterns 
ax:t:ording to a linea.r ordering of the weights ra.tho, r tha.n 
the pa.irwise pa.rtia.l ordering of pa.tterns described in the, 
previous section. Numeric weights for 1)a.tterns a.re ax- 
tr(,moly useful as a mea.ns of assigning higher priorities 
to us(:r-defined 1)a.ttevns. 
The final (;xttmsion of tra.nsla.tion 1)atterils iS int(,.gra 
lion of examl)h~s, or bilir~.:V,d cmpo',:t, int() our frame- 
work. It consi,~ts of the following steps. Imt :1' l)e a. set 
()f tra.nsla.tion pa.ttern,% \[~; a. bili~,gual corpus, a.nd (s,t) a. 
t)a,h' (If SOttFC(', ~lIl{l target ,,-;(;nt(;it(:es, 
1. If T can tra.nsla.te s into t, (lo nothing. 
2. If T can tra.nsla.te s into t' (t ~ t'), do the following: 
(a.) If the, r(; is a. pa.ired (leriwl.tion s(:(lll(;ll(;(; Q of 
(s,t) in T, crea.te a. new l)a.ttern p' tbr a. pa.ttern 
p used in Q such tha.t (',very nont(~rulina.1 syml)ol 
X in p with no head constraint is associa.to, d 
with h : X in q, where the, head h is instantia.ted 
in X of p. ekdd p* to T if it is not a.h'eady there,. 
(b) \[f there is no such pa,ired deriva.tion sequence, 
add the pah" to T (s,t} as a. tra.nsla.tion l)a.ttern. 
3. If Tca, nnot tra.nsla.te s, a.dd the, pa.ir (s,t) to T a.s a. 
tra.nsla.tion pa.ttern. 
The siml)lest wa N of integra,ting the corpus B into T is 
just to consider the sentence pair (s,t} as a translation 
pa.ttern. Some additiona.l steps a.re no, cessaa'y to achieve 
higher MT a.ccura(:y for a. slightly wider ra.nge of sen- 
tences tha.n those included in IL However, tit(', de, gree 
of hnprovement in MT a.ccura.{:y tha.t ca.n be, ax:hieved 
with this h;a.rning mechanism is opo.rt to question, since 
the a.ddition of tra.nsla.tion pa.tterns does not necessa.rily 
gua.ra.ntee a. monotonic improve, nwnt in MT a.ccuracy. 
1157 
5 hnplementation 
Our exl)erimental implementa.tion of a. pa.tto.rn-l)ased 
MT system consists of about 500 defa.ult-tra.nsla.tion t)a.t- 
terns, about 2400 idiomatic a.nd colloca.tiona.1 pa.tterns, 
a,n({ a,1)out 60,000 lexica.l items ff)r English-to-.Ia,pa,nese 
tra.nsla.tion. A sample run of the prototyl)e system is 
shown in Figure 1. tt shows one of the (l(;riva.tion se- 
quc;tiees for the input sentence 
John should he.a.r from Ma.ry M)ouf, the news if 
he re, turns home. 
Ea.ch lino. in the. deriva.tion sequence shows a.ii English 
source CFG rule of a. pattern used for the deriva.tion. 
For examt)le , the first line 
\[(0 13) S:*:/eFIN,ePRES,eSUBJ,ehUX/ 
-> Sl: l:+eFIN PUNCT:2 
in the deriw~.tion sequence shows tlia£ two nontermina.l 
symbols, $1 a, nd PUNCT, form a, sentenc(; S, tha,t S 
is coqnclex(xl with Sl, a,nd tha, t SI Inust have a, fi'n, itc 
form f(m.turc +(;FIN. The, curre, nt insta.nce of S ha.s four 
f'(;a.t,u'es finite, prcscr,,t (cPIH'TS), w/tA-.~u/t#ct (cS- 
UI\],\]), a.II(l with-a'mriliary-vcrh (cAUX) a.Ild it spa, ns 
the word l)ositions 0 to 13. o We ca, II Mso find several 
h('ad-cortstrained pa.tterns there. For examt)\]e, 
\[ ( :tO :1_2) VP : 1. :/eF\[N, e3SG, ePRES, e0BJ, eSAT/ -> 
VP"return" : 1 : -e0BJ NP"home" : 2 : +eChUS 
is a, l)a.ttern tbr tra.nsla.ting :'return:V tiome:NP". Tho, 
do, faxllt V+NP transla£ion pa,tteFIl will assign a, wrong 
Japanese, caso, mamker for this phra,se,. 
Our 1)rototyp(; took a,l)out 9 sec (ela,psed time) to 
transla,te this input s(mtence a,nd produce seve, n alter- 
ha, tire transla£ious. The deriva,tion shown in t\]le figure 
wa,s the first (i.e., the best), a,nd generates a, correct 
tra, nsla,tion. Therefore, colloca,tiona,l p~tterns a,nd de- 
f'~l,,lt patterns have, been a.pl)ropria,te, ly coral)it\]ell lui(le, r 
()Ill' pro, fe, ren(:(~ stra,f, egy. 
6 Conchlsions and Future Work 
hi this pa,per, we, ha,ve proposed a, pa,ttern-ba,s(~,d MT 
systein tha.t sa.tisfies three essentia.1 re, quir(;ments of the, 
current ina.rket: efficiency, scala.bility, and ea.se-ofuse. 
We are a~wm~ tha,t CFG-l)as(;d pa, tterns a,re lo.ss a,(h; 
qua.te for descril)ing synta.cti(: d(q)eridcnci(~,s tiia.n linguis- 
ti(:ally lnotiva,te, d gFa,Ill ln~/,r ~7)r\[nalisms Sllch ~lS TAGs a, ii(t 
III)S(7. To acid(we the best l)ossible average rllntinle and 
aC(:llrtl.(:y~ pe, rha.ps our t)a.ttern-based system shouhl be 
combined with lIlOrO, powe, rful ~ra,llllll&r forma, lisnls, V~/e, 
l)elieve tha,t the theory a,nd imph;mo, nta,tion of 1)a,ttern 
1)ased MT will contribut(~ to the realiza,tion of con\]puta,- 
tiona,1 linguistic theories. A corl)us integra,tion method 
to verify efficiency of tho, gra,mmar a,cquisition has yo, t to 
I)e, inlph;lnente, d. 
Soi/\]e of' tile assumt)tioils on 1)a,tt0,rns should be r0,- 
e×amine, d when we, extond the (l(ffinition of pa, tterns. The 
notion of Head constra,ints may havo, to hc ext(mdo, d into 
that of a set ln(md)(wshil) constraint if we need to ha,ndlc 
coor(lina, ted structures. Soine light verb phrases (:a,nnot 
1)e corre,(:tly tra,nsla,te, d without "excha,nging" sevo, ra,l tim- 
lure wflues betw(;en the verb a,n(l its object. A simila,r 
\])rol)h;nl has been fbund in l)e-ve, rb phra.scs. 
°Other Ihatures illclude nominative and t:ft'l/,8(tti'ts~ >. CaSQS, :{l'd- 
pal'sol/-Sill~/l\[;.I,r \[()1"HIs, alid c~q)ita\]ized words. Two f(!}lt/lrc)s, 
"*oAI{GS" and "*oAI{CV," {i,l'0 spec{;t\], ones for /'opres011ti/lg 
sub.leer.verb ~t~l'eo\[l\](4llL without sp\]ittitig :-t p;cttorn hito m/ oquiv- 
a.\[onl sel, ()F several lUtlterns for a, specific typo of ttgroolll(311t. This 
sol\]/"cc deriwd;ion so(\]llOl\]c(3 IS actually a(:compall\[O(l 1)y iLs .\] ai)anes~ 
com\]l;erl)arl , whicii was omitt;(!d due too the spaco luu\]la(;io;n. 
> John should hear from Mary about the news 
if he returns home, 
\[(0 13) S:I:/eFIN,ePRES,eSUBJ,eAUX/ -> SI:I:+eFIN PUNCT:2 
\[(0 i2) SI:2:/eFIN,ePRES,eSUBJ,e~UX/ -> NP:I:*ehGRS+eNSMI VP:9:*ehGRV+eFIN-eSUB~ 
\[(0 i) NP:I:/e3SG,eChP,eNOMI,eCAUS/ -> N~UN:I:-ePRO 
\[(0 i) NOUN:I:/e3SG,eChP/ -> NOUN"Jo~m"\]\] 
\[(1 12) VP:I:/eFIN,ePRES,eAUX/ -> VP:i SADJ:2 
\[(1 8) VP:I:/eFIN,ePRES,eAUX/ -> VP:I PP:2 
\[(i 5) VP:I:/eFIN,ePRES,eAUX/ -> VP:I: PP:2 
\[(i 3) VP:I:/eFIN,ePRES,ehUX/ -> 
AUX"should":-eNEG VP:I:+elNF-eSUBJ-ehUX 
\[(i 2) hUX:I:/eFIN/ -> hUX"should":l\] 
\[(2 3) VP:I:/eFIN,elNF/ -> VERB:I:-ePS 
\[(2 3) VERB:I:/eFIN,elNF/ -> VERB"hear":l\]\]\] 
\[(3 5) PP:I:/e3SG,eChP,eNflMI,eChUS/ -> "from" NP:I 
\[(4 5) NP:I:/e3SG,eChP,eNOMI,eChUS/ 
-> NOUN:I:-ePRO \[(4 5) NOUN:I:/eSG,eCAPi 
-> NOUN"Mary':I\]\]\]\] 
\[(5 8) PP:I:/eDEF,eNOMI,eCAUS,e3SG/ -> "about" NP:I 
\[(6 8) NP:I/eDEF,eN~MI,eChUS,e3S6/ 
-> "the" NP:I:-eDEF-eINDEF 
\[(7 8) NP:I:/eNOMI,eCAUS,e3SG/ -> NOUN:I:-ePRO 
\[(7 8) N~UN:I:/e3SG/ -> NOUN"news"\]\]\]\]\] 
\[(8 i2) SADJ:2:/eFIN,e3SG,ePRES,eSUBJ,eOBJ,eSAT/ -> 
"if" NP:i:*eAGRS+eN~MI VP:2:*eAGRV+eFIN-eSUBJ 
\[(9 i0) NP:i:/ePRO,eN~MI,e3SG,eI{I\]M/ -> PRON:i:-ePOSS 
\[(9 iO) PRON:I:/ePR~,~NSMI,e3SG,aHUM/ 
-> PRON"he":I\]\] \[(iO 12) VP:i:/eFIN,e3SG,ePRES,eOBJ,eSAT/ -> 
VP"return":i:-eOBJ NP"home":2:+eCAUS 
\[(I0 Ii) VP:i:/eFIN,e3SG,ePRES/ -> VERB:i:-ePS 
\[(I0 ii) VERB:I:/eFIN,e3SG,ePRES/ 
-> VERB"return":i\]\] 
\[(Ii 12) NP:I:/e3SG,eNOMI,eChUS/ -> NOUN:I:-ePRO 
\[(II 12) NOUN:I:/e3SG/ -> NOUN"heme":l\]\]\]\]\]\] 
\[(12 13) PUNCT:I -> PUNCT".":I\]\] 
= :,. ~ x % I;LI < t<-J:T~,:~. 
( .hJhn+SUB.l, if" he-t-SUIL/ h mm+(T()Ai, i',eturn, 
lle, WS-i OBJ tiea.r+should 
Fizure 1: Sample Pa.rsing 
References 
A. AI,ciH6. Y'. ScLlal}¢s: aud A, 1<. ,In:hi. t\[)gU. "(\[t~\[llJ( \[,,:xi':Miz~,\[ "\]bUgs tol ~/~ 
chin(, Translath, n". Ilk PI'(*<. ot Irht: \[3t, h \[tlt,*~rnai,{,Jll:d (JlJiilll',Hl,:,: ,)L/ (\]tJllLI)ll" 
la/i()ll;ll \[,ti/l~l\[\[~l,\[cs : vl*hll'llr ;I. i)~lgt:~ I (;. AIII~, 
P. I,'. lh'/*wn. ,'~;. A. l)/:lla Pi/t,':,, V. ,1. \[)ilia I'h:l, la. ~tl,l I/. I, M, tc.l. 1903. 
"'File Mat, hct~laiic~ .i ~-;t.al, i~ti,ml \]Vla,:lt{ltl '\]>r&n:l}tl, l{Jn: l'ar~m.:i, vic l~\]sthlYal, b,ll". 
G*Jtnl)Uf, lO, ionM I,lupini:l,h:s , 10(2):2(i/I ;ll I, \]lttll:, 
Ill. (\]\[~tldal', (;. K. \['ullum and \[. A. ~qap,. 19,%5, "G*m~:ralizc,\] Phl>~Lft: ~4h'ltCf, ltl'( 
(\]ratnlnar". Ilatvatd Univcr~if, y I'i',:::,, (::u,d)ridii,:. ~la,,~< 
S~. \[++ (~+l+lh;itll, ~\[, A. \[{+ll'litl+lt: :Llll\[ W, \[,, \[{tt~Zt). l{J~{\], "Jtl~ Ii+q)i',,vl,I (;+,ill<:Xr. 
tl'l:l Ill:col~niz<l". A(lik,i '\['l;~ll:~U~t.\]lJil~ Hll l~r*lg;ral*ailliltl., L~Hlj,,il;LI,I+ t tlid ~;ynt,* ut~. 
'2(3}:41fl 462 ,lulv. 
II. N;Hqatl and .i. \[h,snan. 1982. "l,~xi,:at I"un,:t,\]~,nal Gral.nml: A \],',umul Ny:,t,rln 
l-r (h n,~l'&liw:d (\]h',lHI.alh:M Illlu,w:lH,,ki,,a". \[n ,1. \]\]l'l~n&u .dlt,.. "'~\[*:nt4d 
\[I,l~rcwmi, aii.,n ol Ch'alnmalb:al ll~l~Dion,,", pai,;,:~ 173 281. iv\[\['F l'r+:,~. (Jam 
II. iWl~l,'ny¢~llla, l\[)f)~l. "\]'a\],l,c,. \[la~(,l 'Fl':~llt.laDi,lll; C\]ol/Lfxl \]"Fit '\['l'illl~lhlt t:l' ilJitl 
\[t~ Al)l)licai,i*.n:', i,o i)l&,:ti, al NLI'". ht lh(ic, i,I N;tI~II~I \[ ¢lli/,ll ll~J \[I;li\]\[ti(" \[lilll 
74711ipl~:dillll (N\[,P\[I£;: 93). i,ag, :; 232 237. \])l:,:. 
(;, Pollald ;uld |. 2*t. ;-l&,;. 1987. "All hlil)l'ilt;~iit~n-l\];tf~ld Syll£;tx ;Itl,\[ Y,~,l'll;tllt, icf,. 
V,,I.I I"lllld&lllrrltl.&lt"i (\]'74\[,\[ I,t:cLiir,: NlJl(t% ~.~,Ihtill:r 13, 
\], \['lt:dciov:dty, 19\[Ji. <"L'I/~ (;i IICl~liiv( \],l:xil:e*ll'. (~(*lllpll\[;llillll;l\] \],\]tljJlliM,ics. 
17(4):'t09 '14l: \[)<><:,:illl><l, 
+~. ~g~tf** &lid ~\[. N~lj,;~+I, IDDIJ. "Ttlw&t'c\[ iV\[llll~.r7 I.:u~l:d Tl';tl/:l&l, iNtl". \[tl \])till:. O\[ 
ih, 131h lnt,:lnati*ul&l (h,ltl,l',:llc, ,ul ('(~\[tq~lll,;il,\[l,ll;t\] \[,\]ill,ill tic~. I,&IX< ~ ;!t7 
252. lh Is\]nki. Aug. 
V. ,:~ch&l.~q an,l l{ (: W&IIt:~. lfJf)0 ""Fir\] hlhlllt, iOll (il~lilllll&l: '% (',ltl,\[C Thll, , 
l'alaald, \[e.l'l~ali,:la lh,M \[,cxi,:aliz,::: (:,lilt, %1 l?l',:l (3\[',~LIIIIIhLI w\]lh,ull (:hani<ilu~ 
t>\[t* 'i\'(,t Pl'ucluc, d'" (Jf,llll~Lll.&l, iNil II \].illplli~ilcs ~\](i):4Tf) flL~ \])¢ c, 
Y. ~;ch;d.:;,. A, Al.:ill,'. aud st K. Io>l~i 1988. "lD~rMng AIg.i\]ihl.~ wilh 'il×i,'a\[ 
tw,l: /:;\['#lllll\]l$1l'i~: Api)li, aihul i,. ill,: &dioinilll; ~l>~lllll'll~ll'~ "' . \[\[/ \[IFIJU, tit lh, t21, h 
\[nl*t>ll~d,i(,n~d (\]lHlil:l<n(< on (JOlnl)lO, cd,\[l,nal Linlguiat, ics , I);tj\]~(~ F,T,~ q,~:l. All/! :. 
{i. iV\]. ~glii~:l~ll' anl\[ Y. ~4/l/etl>lv,. /Dg(). "Tgylllhl'l~llOII7 "\]'l>l' Adi<, , ,g (;l'~,i\]lli,~,F~". 
\[ll \[>l'lJ(;. or" f, hl: 13th \[lit, clUlal, illn}~l (\]ontin'lnc\[ ()11 Cotllpuf,&iiOlt&\[ i,itl I libel, its. 
}1\]. ~;Ittllik~l ~till\[ \[1. l\],la. \]991. "'l~,Xl)¢:i'hlat:lilf, aild I'ro~l)tcts Nl b\]X&llipll: \]lar~c:d 
\[V\[;lf hin.r "Fl'~lll~i;lf, lon", \[n Pl'OC. (~'<" t,\[lt: 291,h Alln,i&\[ \]~\](:(rIIIIF~ lit fill: 1\~ fr,,:i~C\],i\[,ll 
tt,i' (3Olllpltlltf. itlllltl LinIoli,::l,t+:~ , pag,~ 1,~5 192. \]l(t'kcb:y, \]Hilt. 
I<. Talo, 1 9 \[ . ,, / , C ' × ~ ' ( " t! ,, h\[r h l k ' i~ I (Ill". ~11 PL'O(. Nt 1(: ' . I All\]lit& b((:, i n t i r. A(. ,. h &ll ~l ( I'IIZ, ( ~t t.. ~itta( . 
K. \:iiay ~Jh~lllkcF &lid Y. J,:hal,,: 1992. "~';IL'ILCIIIr, ;qh:tlil/l? ~ ill I., yi,mliz, ,I 'l'll:l 
Adi,dnill17: (Jl';tllllll;tl~ "* ill Ill\]l(:, i)1 \],\])It I<it\]l hll,lli&l, illll&\[ (\]tlllf,:l,tlrzr i,li (J()lll 
i>ul:di,,ual I,inyDti~di~:. i)it/Si, 2115 211. AIIj~. 
1158 
