7/27/2019 C86-1084

1/3

L i n g u i s t i c B a s e s F o r M a c h i n e T r a n s l a t i o n 1)Christian Rohrer

Institut ff~r LinguistikUniversittlt StuttgartKeplerstraf3e 177000 Stuttgart 1

My aim in organizing this panel is to stimulate the discussionbetween researchers working on MT and linguists interested informal syntax and semantics. I am convinced that a closercooperation will be fruitful for both sides. I will be talkingabout experimental MT or MT as a research project and not as adevelopment project.[l ]A. The relation between MT and theoretical linguisticsResearchers in MT do not work with linguistic theories which are'on vogue' today. The two special issues on MT of the journalComputational Linguistics (CL 1985) contain eight contributionsof the leading teams. In the bibliography of these articles youdon't find names like Chomsky, Montague, Bresnan, Gazdar, Ka mp,Barwise, Perry etc.[2] Syntactic theories like GB, GPSG, LFG arenot mentioned (with one exception: R. Johnson et al. (1985 0.165)praise I.FG for its 'perspicuous notation', but do not (or notyet) incorporate ideas from LFG into their theory of MT). Therearc no references whatsoever to recent semantic theories.On the linguistic side one notices a similar lack of interest inMT or in the theory of translation in general. In the latestbooks on G/3, LFG or GPSG one will look in vain for any referencesto the theory of translation. Or, to give another example, lookat tile research goals of the recently created Center for theStudy of Language and I nforma tion (CSLI) at Stanford. The aim isto develop "integrated theories of language, information andcomputation". (CSLI Report No. 84-16, p.2). At CSLI philosophers,linguists, computer scientists and psychologists collaborateintensely on numerous projects, however MT is not among them.There is only one notable exception (M. Kay) which confirms therule. None of the other 40 staff members has ever been involvedin an MT project.This does not mean that there is no research in translation goingon at all, quite on the contrary. A growing number of countriesare creating chairs in the theory of translation, there aresymposia on the theory and practice of translation (L. Grlthls etal. 1978), and there exist innumerable papers on practicalproblems of translation, which are written primarily forinterpreters and teachers of a second language. The efforts intiffs area can be characteriz ed b y key words like contr astivelinguistics and literary theory. Within contrastive linguisticsthe emphasis lies on lexicology. (Typical examples which arediscussed are pairs like Engl. to know, Fr. savoir, connattre,Ital. scala, Germ. Leiter, Treppe.) Yet, there is little detailedformal work on eontrastive syntax. (For an exception see Hawkins(1986)). The re sults in eontra stive li nguistics ira general and incontrastive lexicology in particular will certainly be of greatvalue for writing transfer dictionaries and specific transferrules. The research undertaken in this area, however, has notproduced sufficient results to build a coherent formal theory oftranslation on. Finally there is the philosophical tradition,best illustrated by Quine's famous 'Meaning and Translation'(1959) and Montague 's 'Universa l Gramm ar' (1970). In this contexttranslation means translating from a natural language into alogical language. This sort of translation has had an enormousinfluence on semantic theories used in linguistics but againthere are very few connections with MT (for an exception seeLandsbergen (1982)).

(l )(2)

(3)

(4)

(5)

(6)

Why i s t h e r e s o l i tt l e i n t e r a c t i o n b e t w e e n M T a n d theoreticallinguistics?Most MT projects were set up as development projects. Therewas just not enough time or money to do basic research.MT projects were often directed by computer scientists whounderestimated the complexities of natural language.Analysis grammars for MT were often written in a formalismwhich made communicatio n with the non=initiated impossible.Theoretical linguists thought and still think that MT is toodifficult, that high quality MT is impossible. Some argue onthe basis of syntax: if we don't even have a satisfactorysyntax for comparatives in English, how can we translateEnglish comparatives into German. Some argue on the basis ofsemantics: satisfactory semantic analyses exist only for verysmall fragments of natural languages. How are we to translateGerman tense forms into French tense forms, if nobody has yetpresented an adequate semantic description for French tenses.Linguists seem to Imve a one-sided (lop-sided) perspective,working nearly exclusively on analysis. In LFG one parsessentences and builds up functional structures, in GPSGsentences are parsed and translated into formulas ofintensional logic, hardly anyone knows how to generate fromf-structures or from logical formulas.If there is little researc h on generation , there is even lesson transfer. Under the influence of Chomsky and (to a smallerextent) Moatague linguists started the search for UniversalGrammar. Transfer, however, presupposes he comparative studyof two (or more) langua~les. Linguists working with recentsyntactic theories don't seem to be interested in questionslike:[3] Is language A syntactically more ambiguous thanlanguage B? If a string S is ambiguous in language A, isthere a string S' in language B which presents the samenumber of ambiguities? Is it really easier to translatebetween two languages which are closely related (say Italianand French) than between distant languages (say Japanese andGerman)? Under what conditions is it possible to translate onthe basis of syntactic structure, and under what conditionsdo we have to use semantically based transfer rules?

C. What can we do t o s t i m u l a t e i n t e r a c t i o n b e t w e e n MT andt h e o r e t i c a l l i n g u i s t i c s ?

1 would like to argue that MT should be based on a linguistictheory which can be expressed in a formalism whose formalproperties are well understood, in other words, a formalism forwhich there exist results on decidability, generative capacityand recognition complexity. The linguistic theory itself,independent of its formalisation, should be well accepted withinthe linguistic community. Possible candidates for such theoriesare GB, GPSG, LFG etc.[4] What are the advan tages of worki ng withsuch a theory?

353

7/27/2019 C86-1084

2/3

(1) Advantages for the linguistThe MT linguists can use directly analyses which have alreadybeen proposed for the phenomena he is trying to describe. If alinguist writes a GPSG grammar for a fragment of English andwants to include coordination, he can just use the analysisproposed by I. Sag et al. (1984).He can discuss the insights expressed in his own analyses withlinguists outside the MT groups.Since the formalisms which are used for expressing linguistictheories are closely related ( Shieber (1985) has even sho wn thatthe formalisms, in which the different UGs are expressed, areinterredueible to a large extent), a particular analysis can betransferred from one theory into the other. The treatment oflong-distance dependencies ( w h - movement) in LFG for instancelooks formally very similar to that in GPSG.Linguists working outside of MT could use the implementation ofLFG, GPSG etc. as gra mmar testers.Theories like LFG, GPSG offer a well-defined interface tosemantics. To name just one particular problem we are working onat the moment, it thus becomes possible to test the adequacy ofan analysis of the French tenses and of the English tenses bywriting transfer rules which map a semantic structure, whichrepresents the temporal information contained in a small Frenchtext, into the corresponding semantic structure for English.In UGs the lexicon plays a central role. A lot of linguisticinformation is located in the dictionary entries. The richlexical inf orma tion collected by contrastiv e linguists could beincorporated into a transfer lexicon written according to theprinciples of UGs.(2) Advantages for the computer scientist and linguistIf linguists can write their grammars in a formalism whosemathematical properties are well understood, then the programmerwill have fewer problems implementing the grammar. Results onformal properties of grammar formalisms can guide the progra mmerin his search for appropriate algorithms. Furthermore, in thesame way in which a linguist working within LFG can benefit fromthe intuitions expressed in linguistic analyses within GPSG orGB, a programmer implementing LFG can benefit from insightscontained in implementations of related formalisms. It isthere fore not surp risi ng that F. Pereira (1981), W. Frey andU. Reyle (1983), L. K art tun en (1985) and J. Calder et al. (1986)all use "gap threading" for implementing long-distancedependencies. The authors are working with different theories(Extraposition Grammars, LFG, Helsinki Unification Grammar(HUG) and Unification Categorial Grammar (UCG)) and differentprogramming languages (Prolog and Lisp) and nevertheless thealgorithm they chose is the same. Let me give you a furtherexample. In the Prolog implementation of L FG by D6rre and Eisele(this volume) there are some problems with negative anddisjunctive constraints. Such constraints are linguistically wellmotivated but difficult to implement. Now if anybody within theUG community comes up with a good implementation of negativeand/or disjunctive feature specification, then this newimplementation can be incorporated.Results on recognition complexity may help to locate the sourceof the complexity and suggest ways to reduce complexity either bymodifying the linguistic theory or by showing that the "worstcase" is linguistically (empirically) irrelevant (in other words,the "worst case" will never arise in practice).A famous example, where the linguistic theory was changed afte r aproof of its generative power had been found, is Peters andRitchie's work (1973) on the generative power of TG and thesubsequent change of the theory (constraints on deletion rules).354

For GPSG Peters and Uszkorei t (1983) have exam ined the role ofmetarules and essential variables (variables which range over allstrings of terminal and nonterminal symbols). They proved thatunconstrained Metarule Phrase Structure Grammars have Turingmachine power. This result has led to intense research forconstraints that are linguistically motivated and descriptivelyadequate.For LFG there is a conje cture d result by Berwick: ce rtainLexical-Functional Grammars can generate languages whoserecognition time i s very likely computationally intractable, atleast according to our current understanding of algorithmiccomplexity." (Berwick 1982, p.98)Conclusion: Basically a ny arg umen t in favor of a declarative,simple, expressive, flexible linguistic formalism with a rigorousmathematical foundation can be adduced here.D. P o s s i b l e o b j e c t i o n s a g a i n s t using Uniflcatian Grammars o1'o t h e r r e c e n t l i n g u i s t i c t h e o r i e s f o r MTAre there linguistic phenomena, which make it impossible to useUGs for parsing and generation? Someone might argue that in MTthere exist specific problems, which have never been taken intoconsideration by theoretical linguists, and which cannot behandled with UGs. Personally I hope that whatever additionalproblems may arise can be handled in the transfer component. IfUGs provide us with the tools to write adequate grammars forindividual languages, why should it not be possible to use thesegrammars for MT?R.Johnson et el. consider the possibility of assuming a currentlinguistic theory and implementing it directly for EUROTRA. Theyreject this solution "because there is not sufficient practicalevidence of a single theory that encompasses translations betweenall pairs of the Commu nit y languages" (1985, p.165). Thisrejection can be interpreted at least in two ways: (a) there isno linguistic theory in which all the languages of the Communityhave been described, (b) recent linguistic theories have not beenused for (automatic) translation between all pairs of theCommunity languages. Of course, both interpretations are correct;however, on the basis of this argumentation one could reject anyexisting linguistic theory as insufficient for MT. Thisreasoning, however, would force us also to reject any newlinguistic theory because there would be even less "practicalevidence".So far only fairly small grammars have been written andimplemented in the UG formalism. What will happen if these toygrammars develop into big gr ammars with very large dictionaries?Won't we need special control mechanisms? Since the UG formalismis well understood, I can't see why one should have more problemsif the grammars get bigger than with any other formalism. We hopethat the contrary will be the case, namely that there will befewer problems.There is still another side to the problem of "toy grammars".Theoretical linguists have been working with a fairly narrow setof data. An ambitious young linguist today is more likely topropose the 127th version of the passive construction than towrite a detailed grammar for complex NPs like "die Eiufuhr vonJute aus Ll~ndern der dritten Welt zum Ausgleich des Nord-Stid-Gefalles". Will theoret ical lin guists ever sit down and a ctuallywrite big grammars? Why has there been no grammar writing projectsince the atte mpt by R. Stockwell et al. (1968) to collect all TGanalysesintoonegrammar?UGs have been used for analysis, but there are practically noresults for generation. This lack of experience in generation isa serious drawback for the whole field of computationallinguistics. Fortunately the situation is gradually changing. Theorganizers of the International Workshop on Language Generation(Stanford 1984) write: "More papers in this subfield [generation]have been published in the last four years than in its entireprevious history."

7/27/2019 C86-1084

3/3

h i M T g e n e r a t i o n h a s p l a y e d a m i n o r r o l e . I n a s y s t e m l i k e S U S Y( M a a s 1 9 8 5 ) f o r i n s t a n c e t h e r e i s n o e x p l i c i t g r a m m a r f o r t h eg e n e r a t io n o f G e r m a n . T h e J a p aB e se G o v e r n m e n t P r o j e c t f o r M T h a sa w e l l d e v e l o p p e d g e n e r a t i o n c o m p o n e n t ( N a g a o M . e t a l. (1 9 8 5 ) ) ,b u t i t i s d i f f i c u l t f o r t h e l i n g u i s t t o f i n d o u t h o w t h i sg e n e r a t io n g r a m m a r w o u l d l o o k i n a d e c l a r a t i v e f o r m a t . S o, e v e ni f t h e r e a r e n o r e s u l t s o n g e n e r a t i o n w i t h U G s w e a r e n o t m u c hw o r s e o f f t h a n a n y b o d y e l s e w o r k i n g w i t h a n o t h e r l i n g u is t i ct h e o r y .The most serious objection 1 see concerns the transf er component.I have been advocating the use of linguistic theories which canbe expressed in well-defined formalisms, bnt so far there doesn'tseem to exist a theory of t ransfer fo r UGs. Well this is notentirely correct. M.Kay claims that his Functional UnificationGranlmar (FUG) "provides an opportu nity to encompass within oneformalisna and co mputat ional system the parts of machinetranslation systems that have usually been treated separately,notably analysis, transfer and synthesis." (Kay, 1984, p.75).E . C o n c l u s l o z ~M T a s a r e s e a r c h p r o j e c t c u t s a c r o s s t r a d i t i o n a l d i s c i p l i n a r y a n di n s t i t u t i o n a l b o u n d a r i e s . T h e o r e t i c a l l i n g u i s t s , l e x i c o g r a p h e r sand computer scientists must cooperate in order to solve some ofthe problems of (machine) translation. GrammllrS and dictionariesused ill MT should be based on linguistic theories which arew i d e l y a c c e p t e d i n t h e l i n g u i s t i c c o m m u n i t y . T h e l i n g u is t i ct h e o r i e s s h o u l d b e e x p r e s s e d i n f o r n l a l i s m s t h a t f u l f i l l t h er e q u i r e m e n t s o f a g o o d p r o g r a m m i n g l an g u a g e. A n o p e n a n dt r a n s p a r e n t d i s c u s s i o n b e t w e e n l i n g u i s t s a n d c o m p u t e r s c i e n t i s t sw i l l l e a d t o a r e a l i s t i c a s s e s s m e n t o f t h e p o s s i b i l i t i e s a n dp r o s p e c t s f o r M T . W e s h o u l d a v o i d a r e i n t e r p r e t a t i o n o f t h e m o t t o' t r a d u t t o r e t r a d i t o r c ' .

F o o t n o t e s1 S u p p o r t f o r w o r k o n t h is p a p e r w , ~ p r o v i d e d b y a g r a n t f r o m t h e B M F T .2 T h e c h o i c e o f n a m e s a n d o f t h e o r i e s i s f o r e x e m p l i f l c a t o r y p u r p o s e s . W e d o n o tt r y t o b e e x h a u s t i v e i n a n y s e n s e . N o e v a l u a t i o n i a i m p l i e d b y t h e p r e s e n c e o r

a b s e n ce o f a n y p a r t i c u l a r n a m e , t h e o r y o r p u b l i c a t i o n .S F o r t u n a t e l y t h e ~ f i t ua t io n i s c h a n g i n g n o w . S i n c e C h o m e k y ' s P i s s L e c t u r e s t h e r ei s a g r o w i n g i n t e r e s t i n c o m p a r a t i v e s y n t a x . M a y b e s o m e d a y a t h e o r y o ft r a n s l a t i o n w i l l e m e r g e f r o m t h e t h e o r e t i c a l a n d p r a c t i c a l i n s i g h t s f r o m t h e s ec o m p a r a t i v e s t u d i e s w i t h i n t h e f r a m e w o r k o f G B . I t i s t o o e a r l y t o c o n s t r u c tt o d a y a t h e o r y o f t r a n s l a t i o n o n t h e b a s i s o f G B , u n l e s s t h is t h e o r y m a d er e f e r e n c e t o G B o n l y i n a m e t a p h o r i c a l s e n n e l i k e E . N i d a d i d i n h i s p r o p o s a lt o u s e t h e A s p e c t s m o d e l f o r t r a n s l a t i o n a n d t o d o t r a n s f e r o n " d e e p

s t r u c t u r e s " ( N i d r . ( 1 9 6 9 ) ) .4 I a t h e fo l l o w i n g ] w i l l u s e t h e t e r m U n i f i c a t i o n G r a m m a r ( U G ) a a h y p e r o n y m f o rG P S G , L F G , F U G , I I P S G e t c . , i . e , f o r g r a m m a r s w h i c h h a v e a c o n t e x t f r e e s k e l e t o na n d a n o p e r a t i o n o f u n i fi c a t i o n . I n g r a m m a r s o f t hi s t yp e~ s y n t a c t i c r u l e s a n dl e x l c al e n tr i e s c a n b e e x p r e s s e d a s s e ts o f a t t r i b u t e - v a l u e p a i r s . T h e v a l u eo f a n a t t r i b u t e c a n a g a i n b e a s e t o f a t t r i b u t e s a n d v a l u e s . L a t e r o n I w i l la r g u e f o r th e u s e o f U G a f o r M T . T h i a d o e s n o t m e a n t h a t G B o r J o s h i ' s T r e eA d j o i n i n g G r a m m a r s c o u l d o r s h o u l d n o t b e u s e d f o r e x p e r i m e n t a l M T . I j u sth a p p e n t o h a v e a p e r s o n a l p r e f e r e n c e f o r U G s .

R e f e r e n c e sB e r w i c ~ ( 19 82 }~ ' C o m p u t a t i o n a l C o m p l e x i t y a n d L e x i c a l F u n c t i o n a l G r a m m a r ' ,A J C L 8 , 3 - 4 , p p . 9 ? - 1 9 9 .C a l d e r _ _ J . e t a L ( 1 9 8 61 , P r o b l e m s o f D i a l o g u e P a r s i n g , M S . , C e n t e r f o rC o g n i t i v e S c i e n c e , E d i n b u r g h .

( 1 9 8 1 ) , L e c t u r e s o n G o v e r n m e n t a n d B i n d i n g : t i l e P l e a L e c t u re s ,D o r d r e c h t .F L ey ~ W . , R e y le ~ U ~ ( 1 9 8 S ) , ' A P r o l o g I m p l e m e n t a t i o n o f L e x i c a l F u n c t i o n a lG r a m m a r a s a B a s e f o r a N a t u r a l L a n g u a g e P r o c e s s i n g S y s t e m ' , P r o c e e d i n g s o f t h eF i r s t M e e t i n g o f t h e A s s o c i a t i o n f o ~' C o m p u t a t i o n a l L i n g u l s t ic e , P i s s , p p . 5 2 - 5 7 .G a z d a r J . e t a l : ( 1 9 8 5 ) , G e n e r a l i z e d P h r a s e S t u e t u r e G r a m m a r , O x f o r d .G r i i h s L . e t a L ~ ( 1 9 7 81 , T h e o r y a n d P r a c t i c e o f T r a n s l a l ; io n ,N o b e l S y m p o s i u m 3 9 , S t o c k h o lm .t I a w k i n s J . ( 1 9 8 6 1. A C o m p a r a t i v e T y p o l o g y o f E n g l i s h a n d G e r m a n U z d f y l n g t h eC o n t r a s t s , L o n d o n , S i d n e y .J o h n s o n R . e t a l . ( 1 9 8 5 ) , ' E U R O T R A : A M u l t i l i n g u a l S y s t e m u n d e r D e v e l o p m e n t ' ,C L l l , p p . 1 5 5 - 1 6 9 .~ A . K . L e v L . S . T a k a h a s h i M = ( 1 9 7 5 ) , ' T r e e A d j u n c t G r a m m a r s ' , J . G e t u p .a n d S y s . S c . 1 0, p p . 1 3 6 - 1 6 3 .~ r e s n a n J~ ( 19 82 1, ' Le x ic a l- F un c ti on a l G r a m ma r : A F o rm a l S ys te m f orG r a m m a t i c a l R e p r e s e n t a t i o n ' , i n : B r e a na n ~ J . ( ed . ) , T h e M e n t a l R e p r e s e n t a t i o n o fG r a m m a t i c a l R e l a t i o n s , C a m b r i d g e M a s s . , p p . 1 7 S - 2 8 1 .K a r t t u n e r h ~ L . ( 1 9 8 5 ) , t I U G : a D e v e l o p m e n t E n v i r o n m e n t f o r U G a , M S . , S t a n f o r d

( 1 9 8 4 1, * F u n c ti o n a l U n i f i c a t i o n G r a m m a r : A F o r m a l i s m f o r M a c h i n eT r a n s l a t i o n ' , C o l i n g 8 4 , p p . 7 5 - 7 S .L a n d s b e r e n J . ( 1 9 8 2 1 , 'M a c h i n e T r a n s l a t i o n B a s e d o n L o g i c a l l y I s o m o r p h i cM o n t a g u e G r a m m a r s ' , C o l i n g 8 2 , p p . 1 7 5 - 1 8 1 .M a s s I I . ( 1 98 5 1 , ' Z u r E n t w l c k l u n g y o n S U S Y - 1 1 ' , i n : U . K l e n k ( e d . ) K o n t e x t f r e i eS y n t a x e n u n d v e r w a n d t e S y s t e m e , L i n g u i s t i s c h e A r b e i t e n , T a b i n g e n , p p . 1 0 7 - l l T .

( 1 9 7 0 1, 'U n i v e r s a l G r a m m a r ' , T h e o r l a 3 6 , p p . 3 7 3 - 3 9 8 .~ ' d . ( 19 85 1, t Th e J ap a ne s e G o v er nm e n t P r oj ec t f or M T ' , C L 1 1, p p. 91 -1 1 0;N i d a F ,. ( 1 9 6 9 ) , ' S ci e n c e o f T r a n s l a t l o n ' , L a n g u a g e 4 5 , 3 , p p . 4 8 3 - 4 9 S .P e t e r s , P . S . , R i t ~ R . W . ( 1 9 7 3 ) , ' O n t h e G e n e r a t i v e P o w e r o f T r a n s f o r m a t i o n a lG r a m r a a r s ~, I n f . S o . 6 , p p . 4 9 - 8 3 .P e r e i r a F _: ( 1 9 8 1 1, 'E x t r a p o a i t i o n G r a m m a r ' , C L I , p p . 2 4 3 - 2 5 6 .

( 1 9 5 9 ) , ' M e a n i n g a n d " t r a n s l a t i o n ' , i n B r e w e r , 1 1 . ( e d . ) ,O n T r a n s l a t i o n , C a m b r i d g e M a s s . , p p . 1 48 - 1 ~ ' 2 .S h i e b e r S . ( 1 9 8 51 , ' S e p a r a t i n g L i n g u i s t i c A n a l y s e s f r o m L i n g u i s t i c T h e o r i e s ' ,M s . , S t a n f o r d .S t o c k w e l l R , e t a l . ( 1 9 6 8 1, I n t e g r a t i o n o f T r a n s f o r m a t i o n a l T h e o r i e s o n E n g l i s hS y n t a x , L o s A n g e l e s .U s z k o r e i t H . P e t e r ~ ( 1 9 8 3 1 , E s s e n t i a l V a r i a b l e s i n M e t a r u l e sp S R I T e c h .N o t e S 0 5 .

355