The treatment of Scope and Negation 
in Ro~etta * 
Etly van Munster 
Philips Research Laboratories 
P.O. Box 80.000, 5600 JA Eindhoven, The Netherlands 
A bst, ract 
This paper deals with the treatment of Scope and, in 
particular, Negation in Rosetta, a Machine Translation 
:~ystem which translates between Dutch, English and 
Spmfish (Spanish only as a target language). It will be 
~u-gued ~hat the SOV- versus SVO-character of a lan~ 
guage has important consequences for its possibilities 
of reflecting scope through word order. A description 
will be given of the problems that arise translating from 
one type of language (the SOV-language Dutch) to the 
other (the SVOo.lar~guages English and Spanish). The 
extent to wtfich these problems can be solved will be 
out;lined. 
The paper has been divided into two main sections. 
In section one the phenomena are described linguistio 
cally, in section two a general idea is given of how these 
phenomena are dealt with in Rosetta. 
~. Linguistic phenomena 
1o:~ :Expression of Scope 
3cope beaxing elements ea~l be divided into two classes: 
1) NEG, containing both the adverbs niet(Dutch)/ 
not(English)~ noCSpanish)and quazltifiers with mor- 
phologically incorporated negation, such as nicts(D.)/ ~o,hi,g(n.)/nada(S.). 
2) NPs and adverbials containing a quantifier (from 
now on Q-elements)', like reel kinderen(D.)/ many chil- 
drea(E.)/ touches nir~os($.), een vis(D.)/ a fish(E.)/ 
.n pe~(s.), ,a~k(D.)/ o/ten(nO~ m,,chas ~ee~,(S.), in 
~o,,,~i~e ge,omnCD.)/ in ,omc ca,es(S.)/ ~n olg~o, casos(S.), 
etc. 
The question I am concerned with is how the scope or- 
der of Q-elements and NEG can be determined. In 
a Montague GraJnmar of the PTQ type, (1) would 
have two interpretations, one with Every man having 
wide scope mid one with two women having wide scope 
/Montague 19~3/, /Dowry 1981/: 
(1) Every man loves two women. 
~This paper condenses part of the content of the author's 
~,heai~, /Van Munster 1985/I extended with the results of 
further research. This research was partially sponsered by 
Nehem (Nederiandse Herstructureringsmaatschapplj). 
XIn thin paper 1 limit myself to Q-NPs~ although a par- 
allel can be drawn between Q-NPs and Q-Adverbials (in- " 
eluding Q~PPs). Furthermore only extensional contexts are 
taken into consideration; intensional contexts are ignored. 
However, the interpretation with Every man llavirLg 
wide scope is far more natural. Therefore, in actor-. 
dance with Jackendoff's principle~ /Jackendoff 1972/~ 
I make the simplifying assumption that sur/ac~ order 
in principle represents the most plausible sFope order 
of Q-.elementa azld NEG in the following sense: 
"A Q/NEG element has scope over ~he Q/NEG el~me~d, 
on its right and is itself iuside the scope of the Q/NEG 
elements on its lefl". 
Stating ii'om this principle implies that other impor~ 
taut factors are not taken into consideration, such as: 
o Intonation, which is not visible in a written teXto 
Therefore sentences are considered raider neutral 
stress and intonation. 
Context. Currently Qnly isolated sentences ~ce 
taken into consideration. 
Essential for translation is that, even if we assume that 
(1) is ambiguous between two scope readings, the se~l- 
tence in the target language will have this same amble 
guity as long as the Q-elements have the same surface 
order as in the source language. ~n'thermore, both 
sentences will have the same 'most plausible reading'° 
\]\[.2 SVO versus SOV 
Rosetta translates between two types of languages, 
namely the SOV-type (Dutch) and the SVOotype 
(Spanish and English). This SOV- versm~ SVOo 
character has important consequences for the expres° 
sion of scope. I claim that in both types of languages 
the position of NEG is as close to the left-hand side 
of the verb as possible: it only precedes possible Q~ 
elements that are wlthin the scope of NEG/Van Muno 
ster 1985/. 
Consider the following scheme: 
Dutch: S O-(O) NEG-V 
English/Spanish: S NEG-V 0-(0) 
In an SOV-language the verb (in basic position) is in 
sentence-final position, while in an SVOdangaage the 
verb is in second position. Consequently, in a~a SVO.. 
language only two elements (one in subject po~ition 
and one in 'zhift-position~, i.e. the position to the left of 
the subject) s can precede NEG; the objects are to the 
right of NEG in basic position° ~L an SOV-language 
like Dutch, however, the objects axe to the left of NEG 
in basic position. In principle there is no restriction to 
the number of elements that can appear to the left of 
NEG. 
In genera\[ it can be said that, especially if the sentence 
442 
contains a NEG, an SOV-language is more 'suited' to 
express scope through word order than is an SVO- 
language. 
This basic difference between Dutch and Span- 
ish/English can cause problems when translating from 
one type of language into the other. Consider e.g. (2), 
where (2)a cannot be translated into (2)b since the 
relative order of NEG and the Q-NP is not the same: 
(2) a De kinderen aten reel snoepjes niet op. 
'The-children-at e-many-sweeten.not ' 
b The children didn*t eat many sweets. 
~u the :English sentence the object has to be topicaUzed 
in urde~ *o get the correct scope relations: 
(2) c lvfany sweets the children didn't eat. 
)L2i Subdivision of Quantifiers 
l: argue that the following subdivision of NPa can be 
made: 
IA) Qo)~Ps oenuttive to scope, i.e. the surface order 
of Q-NPs mid NEG is crucial for the interpretation. 
(Dutch: iemand ('everybody'), ten N ('a N'), reel N 
('many N'), alle N ('all N'), twee N ('two N'); English: 
many Ill; meet; Spanish: muehos ('many'), dos N ('two 
~'), etc.). 
E.g. (3)a does not mean the same as (3)b: 
(3) a Niet reel mensen houden van vis. 
~Not-many-people-like-fish' 
b Veel menses houden niet van vis. 
~Many-people--llke~not -fish ' 
IS) Q.:,NPs not sensttlve to scope, i.e. the sur- 
face order of Q-NPs and NEG is not crucial tot the 
interpre~,ation; these NPs always have wide scope, ir- 
respective of their position. (e.g. sommige N ('saline'), 
most N, alguien ('someone'), something, etc.) Since 
these NPs do contain a quantifier, however, there is a 
strong preference for a surface order which reflects the 
scope. Therefore, (4)b is a much more natural word 
order than (4)a, although both sentences have in fact 
the same meaning. (NB. For some speakers (4)a is even 
out). 
(4) a )Niemand gelooft sommige opmerkingen. 
~Nobody-believes-some-remaxks ' 
b 3ommige opmerkingen gelooft niemand. 
~Some-remarks-believes-nobody' 
2) deflxdte iqPs (e.g. Jan ('John'), her boek ('the 
book*), the maay linguias, etc.). Surface order is ir- 
relevant for scope-interpretation. If in (4)a and (4)b 
2Another term would be *topicaliv.ation-posltion'. How- 
ever, thin term can cause confusion since in Rosetta a dis- 
~inction is made between 'scope-shift' (treated in this paper; 
the sentence still ha~ a neutral intonation), and 'topicaliza- 
tion ~ (th0~ sentence has a non-neutral intonation; not the 
surface order but the original position of the topiealized 
Q-elemettt reflects the scope.} Both types of shift go to 
aM#-po~'tfo~ Scopeoshift is a transformation, Topicalization 
rule. (For these terms ave section 2.1). 
,ommige is rephced by the de.re d~, (~) ~nd (b) 
not only have the same meaning but there is no dif- 
ference in naturalness either. Definite NPs have the 
feat.~e \[-q\]. 
1A and 1B together are the NPs traditionally called 
quantifiers. I will indicate tile two types with resp. 
the features \[+S\] and \[-S\]. 
Note that the subdivision of Q-NPs is language spe° 
tide, e.g. iemand in Dutch is scope-sensitive, while 
the Spanish and English equivalents (resp. alguien axial 
someone) are not. 
Spanish and English seem to have much more \[-S\] ele~ 
ments than Dutch does, which, as I claim, relates to the 
fact that they are less suited to express scope through 
word order (see section 1.2). 
1.4 Two approaches to scope 
In TG-oriented theories a distinction is often made be-. 
tween different types of negation: S~negation and V}?~ 
negation (e.g./Jackendoff 1972/, /Las~ik 19'n/, a.o. L 
with the speciM addition of TVP- and Vonegation for 
Dutch (e.g./Hoekstra 1985/) s. To illustrate briefly 
what is meant by the different types of' negation~ let 
me give an example of each: 
(5) a Jan heeft geen boek gelezen. 
'John--has-no-book-read' 
(S-negation) 
b Niet reel menses, hebben een boek geleze~° 
'Not-many-people-have-a-b ook-read' 
(S-negation) 
c Iemand heeft niet gereisd. 
'Somebody-has-not-travelled' 
(VP~negation) 
d Veel menses hebben geen boek gelezen. 
'Many-people--have-no-bo ok-r cad' 
(TVP-negation) 
e Jan heeft veel mensen lets niet verteld. ~ 
'John-has-many-people-something-not-t old' 
(V-negation) 
It is assumed that the constituents to the right of 1NEG, 
including the verb, are within the scope of NEG. In (a) 
NEG follows the subject, but since Jan is definite, S- 
negation is equivalent to VP-negation: as a general rule 
the position of NEG is after a definite (unless the sen- 
tence is eontrastive). Both (a) and (b), however, cart 
be paraphrased by 'it is not the case that ...' which is 
a proof of S-negatlon /Jackendoff 1972/. In (c) NEG 
has scope over the VP containing an intransitive verb, 
in (d) over the VP containing a transitive verb plus di- 
rect object and in (e) merely over the transitive verb. 
In logical terms, however, these sentences merely diI- 
fer in the relative scope order of NEG and Qoelement~t 
(i.e. of scope operators). Ill a semantic, Montagueo 
like theory a verb (unless it is a modal) is not a scope 
SFor a treatment of ~cope-ambiguity in TGoframework 
see/May 1977/. 
Sin the sample sentences the perfect ~ense i~ u~ed since 
this renders a word order with the main verb in basic, i.e. 
sentence final, position. 
4/!,3 
p.~ et:-t~,hm (,f ~;he a~.'~ .i;e*~(,. :\],'i~<~ k~,.ci; i;h..t i{}utch~ b~_t x~.()~ 
eat (oo~mq~.lc:,.*<~i (,f the Ja.cL tive;t IN~);(~ ~x~a~ p4:pea'c ht 
more Fm,'fa.ce o::~.e_r po~d,~;h,:~x~; :m/)~d;, h (belog ~ g()V- 
la:o.g-~ ~, :e ). 
e~y.~.,tv.ctic d; ,ely d;.~ t_,'~ ,e~ (~},;e ~eci.io~,. 2l). 
\]~* ith,se¢i;a (~,ompoMt;.icm;d G.c~:nomar,~ of the M:o_,,taguc 
J;ype :m'~ u.sed, This me; ~a gh;~,'L scltte.(mea axe ht~ihJ rip 
~';;,a,.vfhxg tro:~l }-a,'fi; e:.,)n:e;-mi,~ by applying ,~y~H;actic 
r; d:~; wh{ ':L..~x%~lc:dbe..1 o:v bJ.gger (.,* \])~e ..... ~ ~.io'w ~* <a~.,~ be 
CO* i ;~I( ;t~, t~ Jt'O~f* ,qiO K!,\]\]ev o:~les. '-J~Je gr ",o:u t~'~'* J.~av~: 
h> obey Lb.e (:o~eq)o:dtLo:**MJty f'j'i.ac(fde: eve\]:y :rtth; 
an,.! e'vcx 2 .,a-.~ ~," e~.pve~4td(m ha.*; a ',~w:\]\]-deihted me,ruing. 
':Fbi~ de>i,mi, io.,,, p).oeet~,'~ ,':a,~ be ::lhow,~ ht a, ho-ealled 
8.y:~ta:acg~c dc£;!v~?.i, aoa, t;~-c.co Go.uald.c,, e j,;.. ~..~t¢; btrmtgly 
siml)ii~h~d ~),**{;a:,.¢:t~.(: ,leriva_LJo,~. t,ee of (~i)a which co!!- 
*,aim; L'wo Q-. iq P;o 
00{ ~i; , (G) (,¢dereea !,cost re,, ~e ~ 
/ 
I?.!:', ii i~ ai;, ~:J 
/ ',. 
J?,gvai,;;'~;, X:; J.e(~(.'z'¢! ¢~,. 
/ ~,, 
l%a'i; a'c~ F! i ,~:~ t; b O ('.' !~ -'.; ~). 
~tc2;eyt :~J. "g2 
'.{}h~: *, (,~aex~ of ,mt*`~Jte~io:a of ~ Q.NP indicates i~;~; 
gmot)e-domahL lit (6) icdere¢;~is mtb,~tituted latin: thm~ 
~ it he.~ ~, wider ~copeo ht other v:ord,% 
io:c ~,,t ~: most p\].r.u~,fible readi.~g of a 8e~_lteliee we W&llt 
;..b~: Q~elements to be m~butito.ted f_ro:,a x'ight to left in 
~eate',.'a¢i(m (r}l;ht-h'£1; gene.ra~iox O. ,qince the Sub~)titw 
i;ioSl rll!eu apply freely in prindple, i{ itt easeiliial that 
Lhe conditions (m the applicability of Substi~aiion- aiid 
N egatio~x rules force tlhiu order, ba,~ie;~lly in ~lle foflow- 
iug way: 
(a) An argume.nt-aubutit*d;io(t rule (.f~aubst,~) o:aly ap- 
plies if there are .ao free variables (from now on VAY{~), 
to {;he righ~ of the variabl.e to be mfl)stituted. 
(b) A ~egation role (l~r~eg) only ::y)plies it' there are no 
free W'AI{a Lo the right oi Um poai~i(m where NEG ie. 
inser~;ed. 
\])'or (6) thh~ mea,nu that ~he o~p~lL of the nfle~ ia; au 
Follom: (deLaih~ omil;ted): 
}bm})s~b ~.2 : >rTi !we:: }mekmt \]ee~t *f 
_....*, ,ore.:.,. t o lede.rec~r~ twee boel~:m~ Ice,at 
( .A.' V'e:d ~;4 .~;!:~,:aafor:m.:af, io:~, ~ fn*te ~,he "vm"b i:~ ~mcond po 
~I{;~ m: .- i'cde:,:...'.:ex~, ie~.~i; '.'.wee boeke.n) 
~;}leLween .tL~targ m(td ~gsv_bsa/~ tile Time-.~rules apply, 
('~ 4// 
\]q~/;::~ i&aL 2g, sv.bst, u;g camtt)~; apply before tgeub~,z2 m> 
d,~r ;e(,(a~emeu~ (a,) ~ta~¢;d above~ since x2, a h'ee "VAle. 
Lo ;\]~: ~:'i~.,,},~ of ~I, "ha_~t hog boext mfl),~tltuted for yet. ht 
otb.e:r word~b ~uba~,zi iu blocke.d0 
.A'~ a, dva*!.~aa;e o* the de~'iva~,ion-t~'ee method is that 
t},~ea~ ~.xe.e~ r.ax~ ~epr~.~sen~ ~he gcope or(let of Qoelemex~Ls 
~,.>,~ Iq>;C4 ~* :,z se*tte-~t(:e ht a ~atm'al way (as they ca~ 
be a~mig~md a~t h,teq)):eg;#.io~ h~. Le*wu of h)gical ex- 
• ,,:celia%m J). 
'rr;~ala,f;iox~ i,,~ (t.m~e by meaa~ of Jzomoyp.hJc gr~t.,ml~t;% 
<:.,;bi.d~. me;:m:~ that the e, raamHara o~' ~he lm~lp~.ages co*~-. 
~.,v.:<Jted a'we atJ;,aned h) e~.,.ch o~her h~ the fo\]Iowh~.g w~,y: 
f(~x' e~.~ch ha,h; ext)reazio** ia one language there must be 
a{; ieaut one co:rre'~pondi:ag e*~pressio*~ it, the othe:,: !amo 
,,x~a.ge w~,h the s~m*`e -mea~rdx, g. For each. ~y:uta(;tk; }*,tie 
iit o~o ta-ogr~gc thog~ ~o*isJ; l.)e a'~ _h:a..8~ o~'t.': cox~.'esIm~d.- 
Ja~ sy*-~i;aaeth; :<*,fie i~* the. oU~e~' hmgna, g~ with ~\]Je sa.o~m 
mea".~q~ ope~'ationo Two ue~.te(~.ees are & h'ax, alafioxt (d' 
e~-,.cb, uther if ~hey .,re derived flxm:,, con'e,apm,ding ba. 
~Jh: expre~J~ion~ by applies,!ion of cox:t°espondi:0-g rtde,'~o 
grope ea~'*, be :alaJ~,aine4 i~t trunalatiou if ht SI, aad 
TG the S~butit*ttio~> and NegaLhm ~-~fles t.re applied ix.t 
the ~*me. order° (,%nsider ~tow the F, ng\]iuh derlvatiort 
teee oi" (7)~ eoxre,,;po~tdi~g {o tile Dutch. o~m: 
(7) Evexybody rea&~ two book~o 
) 
ituwbut., x~. 
ltuabul, 0 X2 gve~'ybody / '\ 
l#,ul;ari; ~vo bo 0k~; 
/ \] \ 
y~md xJ. x2 
i{ataxL: xl ~'ea4 x2 
_~-absi;,x:~: xl reada two books 
}L:mbsGxl: Everybody :reads tWO books 
\]n g.hia example there ia no proble:m making an isomo~'o 
phlc de~'iva.,i;J.o~t for English (m: ,(!panic,h), becanue ht ~he 
English lmn~;lation the order of Q~NTPs is the sa~ne as 
h't D a~(:ho 
.fit }b~e.~a a disti~u;~io~. ~a made between r~des~ wtfich 
axe :mca~fi.gfal and reJ.ewu~ for Sranslation, aud ~ranso. 
formations, which are lang~tage-speeific~ meaningle~'m 
~:ad. ~mt releva~ for ~ra).lslation. Since in the derivation 
bee 9~dy ~he sules are represented~ the correspmtdi~g 
U'ee.~ h~c¢~: exa.e~ly ~,he scum geomet):% 
i{iC ~he ~;e:Y~e~.ce corttMnu ~z ~eg~tion, thi~ ~mgation ia 
~n~ted ~d. fhe prqiec~hm pM:h ~ mt aente,tce level~ ioeo 
no~ co~¢ikw~i-i~ttemal, whereever poaslbleo 'r 
e ogo h~ (8) iJ~e l,(mifio_~, og NEG il~ ay:atactically inside 
{he Y'O? 'bai; will be Im~ i*~ ~\]du position (ge~terativety) 
y :meaLm of a b:acmforma~iOXto (ghe de:dvaLio*~ tree h~: 
atJ'or Udu notion see/.Appeto e¢ at/et aL 
7To give au example era ~eugoace where ~he NEG i~a 
conag~'ae,~t (m~,mely ADV'°) h~¢e~.,ta\], i~, (i) not ha~ oMy 
and ~,_o~ aaaSie, aa~ follows: 
(i} Not v¢i.L~.o~x{; <¢ re~uoi¢~ ~ ~aniuhed ~o'me, one. 
(8) a Niet iedere man loopt. 
:'Not-every-man-walks' 
(a) / 
Itneg 
/ 
Rsubat. xl / \ 
Rstart ledere man 
/ \ 
1open xl 
Rneg applies at sentence-leveL s 
(For more theoretical details about the Rosetta frame- 
work se,~/Appelo 1987 et al./). 
2.2 'IYanslatlng Scope 
Now, there may be various reasons why the right-left 
substitution order causes problems, both within one 
language and in translating from one language to an- 
other, in the subsections 1 and 2 the problems will 
be sketched, in 3 a general strategy for a solution in 
Rosetta will be given. 
2.2.1 Switch of arguments 
Problems within one language arise if the arguments 
have be~n switched with respect to the order of the 
verbpatteru (i.e. the argument structure of the verb), 
in order to express the correct scope relations in the 
sentence. Consider e.g. (9): 
(9) Veel boeken leest iedereen. 
'Mmly-bo oks-reads-everyb ody' 
Recall that in analysis the Q-arguments are substituted 
from left to right (cf. section 2.2), i.e. reel boeken (= 
x2) before iedereen (= xl). Now, the output of the 
generative rules is as follows: 
Rst~trt: xl x2 lezen 
Rsubst,xl: (blocked) 
Rsubst,:rl has to apply first but is blocked since there 
is a free VAR (x2) to the right. 
This type of switch also occurs in translating from one 
languagv into the other, namely if the verb in the TL 
has a different order of arguments than the verb in the 
SL. Consider e.g. the following verbpatterns: 
Sp~aish: xl dar x2 x3 
Dutch: xl x3 x2 geven 
Again a~suming that surface-order reflects scope order, 
(10)a aud (10)b are not a correct translation of each 
other: 
(10) a Jan geeft iedereen een boek. 
' John-gives-everybody-a( ='some')-book' 
b Juan da un libro a todo el mundo. 
' John-glves-a('cer t aln')-book-to-everybody' 
The order of Q-NPs in the Spanish sentence has to be 
awitr~_mmchaw~ 
SNEG is introduced syncategorematically, although it 
could have been a basic expression as well. 
2.2.2 SOV/SVO problems 
If a sentence containing a NEG-element has to be 
translated from an SOV-language (like Dutch) into 
an SVO-language (like English/Spanish) problems may 
arise. Recall that the position of NEG is closely related 
to the position of the verb (cf. scheme in section 2). 
In principle no problems arise if NEG does not follow 
a Q-object in Dutch. (ll)a and (12)a can simply be 
translated into (ll)b and (f2)b respectively: 
(11) a Niet iedereen komt. 
b Not everybody comes. 
(12) a Veel mensen krijgen geen kado. 
b Many people don't get a present. 
However, as I explained in section 1.2, in Dutch (an 
SOV-language), NEG may occur to the right of a non- 
topicalized Q-object, as in (13): 
(13) Wij stelden veel vragen niet. 
'We-asked-many-questions-not' 
The corresponding syntactic derivation tree is as fol- 
lows (since the Dutch and English trees are isomorphic, 
I only give the English or target one): 
/ 
Rsubst, xl 
/ \ 
Rsubst, x2 We 
/ \ 
Rneg many quest ±ons / 
Rstart 
/ I \ 
ask xl x2 
Schematically, the output of the rules is: 
Dutch: 
Rstart: xl x2 stellen 
1L~eg: xl x2 niet stellen 
Rsubst,x2: xl veel vragen niet stellen 
Rsubst,xl: Wij veel vragen niet stellen 
English: 
Rstart: xl ask x2 
P..neg: (blocked) 
Since x2 is to the right of the verb (and thus of the 
NEG-position), Rneg is blocked. Note that this block- 
ing is justified: without blocking the result of applying 
the English rules would be (14), which is not a correct 
translation of (13): 
(14) We didn't ask many questions. 
In other words, the wrong output is blocked but how 
can a correct translation be obtained? 
445 
2.2.3 General strategy for a solution 
In \]I{osetta two types of VAR~ are distinguished, 
namely \[+Q\] and f-Q\]. Later on a Substitution rule 
can only substitute a \[+Q\] NP for a \[+QI VAR and a 
f--Q\] NP for a \[-QI VAR'9 The following general strat- 
egy is followed: 
Ira the Mffft-transformations, where VARs are shifted, 
\[+Q\] VARs are shifted to shift-position under certain 
condltions, x° Since there is only one shift-position, 
only one VAR can be shifted at a time. These trans- 
formations precede the substitution-rules. 
The shift-transformations can be subdivided into two 
Ca, ses: 
1. A VAR is shifted over a \[+Q\] VAR. 
For (9), (in which both arguments are \[q-q\]), this 
means that the shift-transformations render two 
surface orders of VARa: 
(path i) xl x2 lezen 
(path ii) shift/x2 xl lezen 
Later on, in the Substitution rules, only (ii) of- 
fers the correct input for a succesful application 
of Substitution rules, since VAR1 has to be sub- 
stituted first (right-left generation). This type of 
shift is only meant to get the correct scope rela- 
tions in a sentence, both when translating Dutch- 
Dutch and Dutch-English/Spanish (see also note 
2). 
2. A VAR. is shifted over a f-Q\] VAR, under one of 
the following conditions: 
a. There are two \[+Q\] VARs in the VP. The left 
one can shift over the f-Q\] NP (subject). 
b. There is one \[+Q\] VAR in the VP; a negation- 
rule has to follow. 
Although the conditions for both Dutch and Span- 
ish/English are the same, the motivation for the shift 
over a f-Q\] NP in both types of languages (i.e. SOV 
vs. SVO) is different: 
For SVO-languages this shift is necessary in order to 
put a \[+Q\] argument in a position to the left of NEG, 
i.e. to get the correct scope relations. 
Consider again the output of the English rules for (13) 
(V = 'to ask'): 
Rule Output 
Rstart: xl V x2 
Tshift: (i) xl V x2 
(ii) shift/x2 xl V 
Rneg: (i) (blocked) 
(ii) shift/x2 xl not V 
Rsu,x2: (ii) many (IS xl not V 
Rsu,xl: (ii) many qs we not -V 
(Final result: - Mazly questions we did not ask) 
o This is the theoretical approach. In order to avoid many 
wrong paths in the derivation process, the implementation 
iB slightly different: it is possible to extract information 
about the substituent from the derivation tree and assign 
the correct Q-value to the VAR, before the generative rules 
apply. 
1°The shift-transformations also shift WH-elements andI 
relatives. 
4/46; 
For Dutch, being an SOVqanguage, this type of shift 
is not necessary for scope, since all Q-NPs can pre- 
cede NEG without shift (see section 1.2). However, 
this type of shift should be done anyway in order to 
generate both (15)a and (15)b as paraphrases of each 
other: 
(15) a Jan geeft reel kinderen een snoepje. 
'Jolm-gives-many-children-a-sweet' 
b Veel kiuderen geeft Jan een snoepje. 
'Many-chilclren-gives- John-a-swe et ' 
(16)a and (16)b are also paraphrases of each other, but 
(17)a and (l~)b are not, considering condition 2b TM 
(16) a Jan heeft veel boeken niet gelezen. 
' Jotm-has-many-books-not-read' 
b Veel boeken heeft Jan ulet gelezen. 
'Many-bo oks-has-Jolm-not-read' 
(17) a Jan heeft veel boeken gelezen. 
' J olm-has-many-b o ok s-re ad ' 
b Veel boeken heeft Jan gelezen. 
'Many-books-has-John-read' 
2.2.4 Loosening conditions 
As I explained earlier (sect.ion i.3), in the SOV- 
language Dutch it is easier to express scope through 
word order than in English and Spanish, especially if 
the sentence contains a negation. In this section I will 
explain how the conditions (a) and (b), stated in seco 
tion 2.1 can be loosened in order to be able to translate 
a Dutch sentence with more than two VARs to the left 
of NEG into Spanlsh/Euglish. 
In general it can be stated that Rneg and Rsubst can 
apply freely even if there is a free VA.R. to the right, if 
this VAR is \[-S\]. Now there are two possibilities: 
- The VAI~ is definite. The rules apply without 
restrictions, x2 
- The VAR is a \[-S I Q-VAR. In this case the surface 
order which rejects scope order is pre/erred (of. (4)). 
Now, this preference will be handled in Rosetta by 
means of a so-called bonus system. Every output of 
a rule has a bonus 0. Application of a rule can change 
this value. If there is more than one output, the bonus 
merely determines the order in which the output sen- 
tences appear. 
XXThe reason behind these facts is that only an NP that is 
not the focus, can topicalize without changing the meaning 
(or the theme/rheme relations~ /Suffer 1982/) of the sen- 
tence. However, since a theory about focus is not available 
yet in Rosetta, this is a way of either avoiding or making 
certain paraphrases of a sentence. In fact, (17)b is a cor- 
rect paraphrase of (17}a in case vee/bo~ke~ is not focus, but 
not in all cases. In short, under the conditions given only a 
paraphrase is given if the topicalized argument has to be a 
non-focus element, namely in case the negation is the focus 
((16)b as paraphrase of (16)a) and in case another argu- 
ment as the one topicalized is focus ((15)b as paraphrase of 
(l~)a). 
x2R-L gsnsration in case of deflnitea is merely done for 
efficiency, in order to avoid unwanted ambiguities. 
Let me illustrate this process with an example. Con- 
sider (181 and its Spanish derivation (V = 'entender'): 
(18) \]k begrijp iemand hist. 
'I-understand-somebody-not' 
Output 
01 xl 
Rule bonus 
Rstart: V x2 0 
T~hift: (i) xl V x2 0 
(ii) sh/x2 xl "V 0 
Rneg: (i) xl no V x2 -1 
(ii) sh/x2 xl no V 0 
Rsu,x2: (i) xl no V alg. -1 
(ii) alg. xl no V 0 
Rsu,xI: (i) yo no ~d" alg. -1 
(ii) alg. yo no V 0 
Application of Rnegwith a free \[-S l VAR to the right 
lowers the bonus with one. Consequently~ the order of 
output ~cntences is: 
(I) A alguien no entiendo. (output o/path i 0 
(2) No entiendo a alguien, zs (output of path 0 
Furthermore, if the Dutch sentence has more than two 
\[+S\] VARs to the left of NEG, I tentatively propose to 
deviate from the conditions in the following way: 
- Application of Rnegwith a free \[+S\] VAR to the right 
lowers the bonus with 2. 
- Application of Rsubst.~ with a free \[+S\] VAR to the 
right lowers the bonus with 3. 
In other words, a deviation in the order of NEG- 
\[+S\]/\[+S\]-NEG is preferred to a deviation in the order 
of \[+S\]-elements mutually. 
Now con.sider the Spanish derivation of (191 (isomor- 
phic to the Dutch one): 
(191 Twee kinderen aten veel snoepjes niet op. 
~Two-children-ate-many-sweets-not' 
/ 
Rsubst, xl 
/ \ 
Rsubst ,x2 dos ninos 
/ \ 
Rneg muchas dulces 
/ 
Rstart 
/ I \ 
comer xl x2 
The output of the rules is (V = 'comer'): 
bonus 
Rstart: xl V x2 0 
Tshift: (i) xl ~q" x2 0 
(ii) sh/x2 xl V 0 
l~mg: (i) xl no V x2 -2 
(ii) sh/x2 xl no Y 0 
Rsu,x2: (i) xl no ~t r m.d. -2 
(ii) m.d. no xl no V -3 
Rsu,xI: (i) d.n. no V m.d. 0 
(ii) m.d. d.n. no V -3 
lSThis ~entence is marginal for many Spanish speakers. 
Compare Dutch, cf.(4), with surface order NEG + I-S\] NP. 
For a fairly extended description of Spanish data w.r.t. 
negation see/Bosque 1980/. 
The output sentences are: 
(i) Dos .~os no han co~do muchos d~ces. (path 0 
(2) Muchos dulces dos amos no ha~ comido."(path .) 
Note that in neither of the output sentences the scope 
order is the same as in Dutch. The limit of scope~ 
translation has been reached. 
3 Conclusions 
In this paper I showed how in Rosetta translation probo 
lems with respect to scope can be solved to a certain 
extent by means of shift-transformations. Since a solu.. 
tion is not 'always possible if we want to strictly main~ 
rain the surface order of Q-elements (and since on the 
other hand sentences are ambiguous anyway, particu- 
larly in SVO-languages like English and Spanish), rule~ 
which break scope-order prescriptions apply anyway 
but assigu a lower bonus. This bonus has influence on 
the output order of sentences. 
Acknowledgement s 
I would like to thank the members of the Rosetta team 
who ga~'e their constructive comments on earlier ver~ 
sions of this paper. 
Z4An obvious consequence of this approach is that in anal-. 
ysis the Spanish sentences will be ambiguous, too. 

References 

A.ppelo~ L. and Fellinger, C. and Landsbergen, J. (1987), 
~ubgrammars, Rule Classes and Control in the Rosetta Trandation 
System, In: ACL-Proceedings, 1-3 april 1987, pp.118-133, 
University of Copenhagen, Copenhagen, Denmark. 

Bosque, I. (1980}~ ~obre la Negaci6n, Ed. C£tedra, S.A. 
Madrid. 

Dowry, D. and Wall, R. and Peters, S. (1981)~ Introduc. 
tlon to Montague Semantics, volume 11, D. Reidel Publishing 
Company, Dordrecht. 

Hoekstra, H. (1985), The scope of Negation withb~ a frame- 
work of or* MT system, Doctoral Thesis, State University of 

Utrecht. Jackendoff, R. (1972), 
Semantic Interpretation in Generative 
Grammar, M.I.T. Press, Cambridge, Mass. 

Landsbergen, J. (1987)~ 
Isomorphic grammar8 and their 
u~e in the Rosetta translation system, In M. King (ed), Ma- 
chine Translation today, pp.351-372. Edinburgh University 
Press, Edinburgh. 

Lasnlk, H. (1972), 
Analyses o\] Negation in English, lvLI.T. 
Dissertation. 

Ma~'~ R. (1977), The 
Grammar of Quantification~ M.I.T. Dis- 
sertation. 

Munster~ E. van, (1985), The 
treatment of Scope and Negatio~ 
it* ROSETTA: a Datch-Spanlsh view, Doctoral Thesis, State 
University of Utrecht. 

Sttfier, M., (1982), Syntax and Semantics of Spanish Presen~ 
tatlonal Sentence-types, Georgetown University Press, Wash- 
ington, D.C. 
