4. M a c h i n e Tr,ansUation
M a r t i n K a y , C h a i r p e r s o n
Xerox Corporation
Palo Alto, CA 94304
Panelists
M a r g a r e t King, I S S C O
J a c k L e h r b e r g e r , Universit6 de M o n t r e a l Alan Melby, B r i g h a m Y o u n g University J o n a t h a n Slocum, University of T e x a s
A M a n h a t t a n p r o j e c t could p r o d u c e an a t o m i c b o m b , and the heroic e f f o r t s of the sixties could put a m a n on the m o o n , but e v e n an all-out e f f o r t on the scale of these would p r o b a b l y not solve the translation problem. In one sense, this almost goes w i t h o u t say- ing. T h e first ninety p e r c e n t of the w o r k that was invested in reaching the m o o n did not get the astro- nauts n i n e - t e n t h s of the w a y there and the heat gener- ated b y the b o m b did not increase steadily as the date of the first e x p l o s i o n a p p r o a c h e d . T h e t r a n s l a t i o n p r o b l e m is one whose solution must b e r e a c h e d incre- mentally. T h e r e will be no d r a m a t i c event to signal the end of the search and there is no single b r e a k - t h r o u g h that would assure success.
T h e translation p r o b l e m is real and will in fact r a p - idly r e a c h crisis p r o p o r t i o n s unless s o m e action is t a k - en. T h e p r o b l e m c a n n o t be alleviated b y b e t t e r lan- guage teaching, g r e a t e r incentives for t r a n s l a t o r s , or i m p r o v e d a d m i n i s t r a t i v e p r o c e d u r e s , w o r t h y t h o u g h these goals u n d o u b t e d l y are. T h e o n l y h o p e f o r a t h o r o u g h g o i n g solution seems to lie with technology. But this is not to say that there is only o n e solution, n a m e l y m a c h i n e translation, in the classical sense of a fully a u t o m a t i c p r o c e d u r e that carries a text f r o m o n e language to a n o t h e r with h u m a n i n t e r v e n t i o n only in the final revision. T h e r e is, in fact, a c o n t i n u u m of w a y s in which t e c h n o l o g y could be b r o u g h t to bear, with fully a u t o m a t i c translation at o n e e x t r e m e , and w o r d - p r o c e s s i n g e q u i p m e n t and dictating m a c h i n e s at the other.
T h e p r o d u c t i v i t y of the p r o f e s s i o n a l t r a n s l a t o r could almost certainly be greatly increased b y t e c h n o - logical aids which, though straightforward, are not all obvious. P o w e r f u l machine aids to translators could quickly be available and m a y be the best w a y to allevi- ate the translation p r o b l e m in the short run. T h e y will not arise as a natural b y - p r o d u c t of w o r k on fully au- t o m a t i c t r a n s l a t i o n b e c a u s e , f o r the m o s t p a r t , t h e y address such issues as c o m m u n i c a t i o n a m o n g transla- tors, identification of r e l e v a n t s e c o n d a r y material, and special editing devices, r a t h e r t h a n issues of syntactie
analysis, p r o n o m i n a l r e f e r e n c e a n d q u a n t i f i e r scope. T h e m o s t valuable resources that a t r a n s l a t o r has for solving difficult p r o b l e m s are the text he is w o r k i n g on, o t h e r texts like it in the target as well as the source language, and his colleagues. At present, ac- cess to these r e s o u r c e s is h a p h a z a r d at best. But, i m p r o v i n g it i m m e a s u r a b l y is well within the s c o p e of existing technology.
A n o t h e r easily identifiable point on the c o n t i n u u m is occupied b y h u m a n - a i d e d m a c h i n e translation. This could be a v e r y different kind of enterprise b o t h f r o m fully a u t o m a t i c and m a c h i n e - a i d e d t r a n s l a t i o n , By h u m a n - a i d e d m a c h i n e translation, we m e a n to r e f e r to systems in which the m a c h i n e , while retaining the initi- ative, works with a h u m a n consultant, w h o n e e d not be a translator. O n c e again, the subtleties in the de- sign of the s y s t e m would n o t reside so m u c h in basic linguistic q u e s t i o n s as in h o w to r e c o g n i z e reliably w h e n a difficulty of a certain t y p e h a d arisen and h o w to c o m m u n i c a t e the n a t u r e of the difficulty to the c o n s u l t a n t in such a w a y as to elicit a quick and u n a m - b i g u o u s r e s p o n s e . E s p e c i a l l y in the early stages, a h u m a n - a i d e d m a c h i n e - t r a n s l a t i o n s y s t e m i n t e n d e d to p r o d u c e o u t p u t of high quality might well require at least as m u c h w o r k on the p a r t of the consultant as a trained translator would t a k e to do the job in the tra- ditional way. H o w e v e r , two facts can be set against this. First, the c o n s u l t a n t would n o t h a v e to be a translator and could quite possibly be d r a w n f r o m a m u c h larger s e g m e n t of the l a b o r pool. S e c o n d l y , while the labor involved in translating a text grows in direct p r o p o r t i o n to the n u m b e r of l a n g u a g e s into which it must be rendered, the w o r k required of the c o n s u l t a n t in such a m a n - m a c h i n e t e a m w o u l d g r o w m u c h m o r e slowly. I n d e e d , if those languages were closely related, it could be e x p e c t e d to fall o f f sharply as s o o n as that n u m b e r e x c e e d e d one.
A s u b s t a n t i a l p r o p o r t i o n of w h a t follows will be d e v o t e d to upholding the p a n e l ' s view t h a t it is i m p o r - t a n t for w o r k to p r o c e e d in parallel on a n u m b e r of different fronts. While fully a u t o m a t i c translation is the m o s t a d v e n t u r o u s , it is f r o m this t h a t we stand to learn m o s t a b o u t language in general, and translation in particular. If fully a u t o m a t i c s y s t e m s c a n b e built whose p e r f o r m a n c e e x c e e d s that of p r e s e n t s y s t e m s b y e v e n a m o d e s t a m o u n t , we should profit greatly as well
f r o m their practical utility as f r o m the theoretical les- sons e n s h r i n e d in them. M a c h i n e - a i d e d t r a n s l a t i o n can e n h a n c e the t r a n s l a t o r ' s productivity, though we have yet to discover h o w m u c h e n h a n c e m e n t is possi- ble in this way. It could also be a source of invaluable i n f o r m a t i o n on h o w translators work. H u m a n - a i d e d m a c h i n e t r a n s l a t i o n can be e x p e c t e d to give b e t t e r results than could be achieved with the fully a u t o m a t i c m e t h o d , since the h u m a n c o n s u l t a n t c a n be called u p o n to resolve otherwise unresolvable p r o b l e m s , but at an u n k n o w n cost. H o w e v e r , there are i m p o r t a n t applications, n o t a b l y where one text must be translat- ed into several languages, w h e r e the gains m a y b e substantial.
We have b e e n at pains to m a k e it clear that the three m e t h o d s of involving machines in language trans- lation are not essentially different in kind but lie on a continuum. E x c e p t in a few special applications, some of which we will m e n t i o n shortly, we do not f o r e s e e a time w h e n translations of a n y quality will be p r o d u c e d without any h u m a n intervention w h a t s o e v e r . In the so-called fully a u t o m a t i c m e t h o d , the h u m a n plays the role of an editor, or revisor. His i n v o l v e m e n t begins only after an initial draft in the target language exists and it is for this r e a s o n that we remain c o n t e n t with the t e r m "fully a u t o m a t i c " . The other two m e t h o d s involve him earlier so that he influences e v e n that first draft.
T h e m e t h o d s t h e r e f o r e differ as to h o w the p e r s o n is involved. T h e y also differ in the extent of his in- volvement. If the h u m a n p a r t n e r can influence all the decisions that are made, he m a y be in a position to forestall sequences of errors, each resulting f r o m the one before, thus reducing the total a m o u n t of his c o n - tribution. On the o t h e r hand, if a conservative s y s t e m insists on having him c o n f i r m e v e n those choices for which its o w n decision m e t h o d s are substantially ade- quate, then the overall extent of his i n v o l v e m e n t m a y be increased. In any case, the utility of a given s y s t e m in a particular situation c a n n o t be assessed b y a simple equation. The a p p r o p r i a t e utility function involves at least the h u m a n cost, the machine cost, the quality of the result, and the nature of the consumer9s require- ments.
T h e c o n s u m e r ' s r e q u i r e m e n t s are, of course, cru- cial. T h e various types and degrees of a u t o m a t i o n in translation are, as we have seen, positioned along one d i m e n s i o n in a s p a c e of possible a p p r o a c h e s to the overall p r o b l e m . The different types of text and their c o n s u m e r s are a n o t h e r dimension and, not surprisingly, the two dimensions are far f r o m independent. T h e type of t e c h n o l o g y a p p r o p r i a t e to a p r o b l e m , and the benefits to be e x p e c t e d f r o m it, differ greatly with the type of the text to be translated and the use to which the result will be put. In the intelligence services, a great deal of translation is done for p u r p o s e s of cur-
rent awareness. T h e first priority is to k n o w the sub- ject m a t t e r of the d o c u m e n t . It is also helpful to be able to discern the gist of the a r g u m e n t so as to dis- c o v e r w h e t h e r it t o u c h e s on certain k e y questions. A rough and r e a d y t r a n s l a t i o n , especially if it can be done quickly and cheaply, m a y give an excellent basis on which to decide which parts of a d o c u m e n t , if any, need to be translated m o r e carefully. Fully a u t o m a t i c translation, even of quite inferior quality, has already p r o v e n v e r y valuable in this role.
Fully a u t o m a t i c translation, or some close relative of it, has also p r o v e d useful in recent years in situa- tions w h e r e a s u b l a n g u a g e has c o m e to be used, or where one can be readily imposed. C a n a d i a n w e a t h e r reports are routinely translated by such a system. The system itself determines w h e t h e r each translation unit
- a p p r o x i m a t e l y a sentence - is within its capabilities. If it is, then it p r o d u c e s a translation, which is the o n e that will be used without h u m a n revision. If not, it presents the translation unit to a h u m a n collaborator, who m a k e s the translation. We p r e f e r to classify this with fully a u t o m a t i c t r a n s l a t i o n b e c a u s e , t h o u g h the machine does not translate everything that is translat- ed, the t r a n s l a t i o n it does is d o n e entirely w i t h o u t h u m a n i n v o l v e m e n t e v e n at a post-editing stage. The machine in fact translates eighty per cent of all trans- lation units and readers of the reports p r o v e unable to discern which parts were t r a n s l a t e d by m a c h i n e and which by a human.
The success of this METEO system comes f r o m the f a c t t h a t m e t e o r o l o g i s t s n a t u r a l l y write in a highly constrained subset of English. Fully a u t o m a t i c trans- lation has also b e e n successfully applied to the task of translating m a i n t e n a n c e manuals for machines. The success of this does not rest on the existence of a nat- urally occurring sublanguage. In this case, the techni- cal writers who p r e p a r e the manuals learn to follow a set of rules intended to ensure that their p r o d u c t s will automatically translatable b y simple means. The rules are s t r a i g h t f o r w a r d and can be learnt in a t w o - w e e k course. T h e m a c h i n e translates the whole text without outside assistance a n d p r e l i m i n a r y results e n c o u r a g e the belief that little or no editing will be required.
The f e a t u r e s of a sublanguage that m a k e it suitable for fully a u t o m a t i c m a c h i n e translation are (1) restrict- ed v o c a b u l a r y , with c o n s e q u e n t reduction in the n u m - b e r of words with m o r e t h a n one g r a m m a t i c a l c a t e g o - ry, (2) small n u m b e r of senses for each word in a giv- en c a t e g o r y due to the restricted semantic domain, and (3) restricted syntax resulting f r o m the p u r p o s e of the text, e.g., instruction manuals m a y contain only i m p e r - ative sentences a n d w e a t h e r r e p o r t s only declaratives. It should not be thought, h o w e v e r , that a sublanguage is simply a s u b s e t of the s e n t e n c e s of the s t a n d a r d language. T h e s y n t a x of a s u b l a n g u a g e m a y differ radically f r o m that of the standard language so that a
Martin Kay Machine Translation
grammar of the latter would not cover the construc- tions of the former. Thus "Fair tommorrow" and "Winds from the northeast" are "sentences" in a weather bulletin. There are also closely related do- mains in which texts have a common syntax, and differ only in vocabulary. An extensive study of sublanguag- es, their restrictions and interrelations, will be impor-
tant for determining the range of applications of fully automatic translation. The question of is a complex subject to which the report of another panel is devot- ed.
The following table summarizes our view of the three most interesting points on the continuum.
Information Acquisition
Denotative Translation
Connotative Translations
Fully A u t o m a t i c M a c h i n e T r a n s l a t i o n
can be quite cheap (revision excluded)
requires effort and experience to read
technical material; possibly other material
applications exist, greatly improvable
faster and cheaper than human 1st pass
human revision required
technical material
technology coming of age; applications exist (METEO); very large intermediate and long-term payoff
N / A
H u m a n A s s i s t e d M a c h i n e T r a n s l a t i o n
N / A
very high-quality especially multi- lingual
possibly high cost
technical material
few or no existing prototypes; FAMT spinoffs possible in near term with suitable funding
M a c h i n e A s s i s t e d H u m a n T r a n s l a t i o n
increased human efficiency
more expensive and slower then FAMT
almost any material
technology exists; might use FAMT to select candidates
increased human efficiency
high minimum costs
almost any material
commercial systems exist (e.g., ALPS); FAMT spinoffs could reduce costs in near term
advantages
disadvantages
text types
status and prospects
advantages
disadvantages
text types
status and prospects
increased human advantages
efficiency
necessarily costly
N / A legal and religious
texts; literature?
technology exists; greatly improvable
disadvantages
text types
status and prospects
Legend
Fully A u t o m a t i c M a c h i n e T r a n s l a t i o n (FAMT) refers to translation wherein the p r o g r a m s run in " b a t c h " m o d e ( o f f - l i n e ) and p r o d u c e t r a n s l a t i o n s w i t h o u t h u m a n i n t e r v e n t i o n ; a f t e r w a r d s , h u m a n revision (post-editing) m a y be p e r f o r m e d with a text editing p r o g r a m or via o t h e r means, if desired.
M a c h i n e - A s s i s t e d H u m a n T r a n s l a t i o n (MAHT) r e f e r s to translation wherein the p r o g r a m is a f a n c y edit- ing and dictionary c o n c o r d a n c e tool which the hu- m a n t r a n s l a t o r uses to i n c r e a s e his e f f i c i e n c y by a u t o m a t i n g his access to word definitions and ter- m i n o l o g y c o r r e s p o n d e n c e s . All initiative resides with the human, unlike FAMT and HAMT.
I n f o r m a t i o n Acquistion refers to a situation in which t r a n s l a t i o n is being p e r f o r m e d f o r " c u r r e n t a w a r e n e s s " or " s c r e e n i n g " p u r p o s e s where a quick- a n d - d i r t y a p p r o a c h m a y be sufficient, at least to d e t e r m i n e if a m o r e careful translation is justified. No h u m a n revision (post-editing) is assumed. D e n o t a t i v e Translation refers to an i n f o r m a t i o n disse-
mination situation in which the e v e r y d a y and tech- nical definitions of the words are m e a n t , and where subtle nuances of a word choice are unjustified or even undesirable. This is typical of technical texts. C o n n o t a t i v e Translation refers to an i n f o r m a t i o n dis-
s e m i n a t i o n situation in which subtle n u a n c e s of word choice are very important, if not critical, in conveying the intended meaning of the text. This is typical of, f o r e x a m p l e , religious, and literary texts.
At the opposite end of the s p e c t r u m f r o m current- awareness services are such delicate enterprises as the translation of legal statutes and political speeches. A lawyer in Finland can base his a r g u m e n t s either on the Finnish or the Swedish version of the applicable law, a c c o r d i n g to which he considers will m o s t f a v o r his client's case. All statutes must be translated and the translators must be at great pains to ensure that there is no construction, h o w e v e r perverse, that can be put on one version but not on the other. We do not fore- see a time w h e n any p a r t of this job could be usefully consigned to a fully a u t o m a t i c or e v e n a h u m a n - a i d e d , system. On the other hand, it would be a prime candi- date for m a c h i n e - a i d e d methods. T h e r e is c o n s t a n t need to c o m p a r e one p a r t of the text with others, and with other legal texts, to ensure consistency, and this is where these m e t h o d s c o m e into their own.
If our o p t i m i s m a b o u t the future of m e c h a n i c a l m e t h o d s in translation has increased during the t w e n t y years during which it has b e e n seriously p u r s u e d , it must be largely b e c a u s e of i m p o r t a n t advances that we perceive in theoretical and c o m p u t a t i o n a l linguistics as well as c o m p u t e r science. A d v a n c e s in c o m p u t e r sci- ence are the least c o n t e n t i o u s of these. The construc- tion of m o s t large internal m e m o r i e s was not available
and external m e m o r y could be accessed only in a serial m a n n e r . T h e c o n s e q u e n t inefficiency in the p r o g r a m s t h a t w e r e w r i t t e n is less i m p o r t a n t t h a n the u n d u e a m o u n t of e f f o r t that was required to m a k e t h e m w o r k at all. A m a c h i n e - t r a n s l a t i o n p r o g r a m was large, e v e n by t o d a y ' s standards, and each o n e p r o d u c e d in the sixties was a p r o g r a m m i n g tour de force. The achieve- m e n t s are e v e n m o r e impressive f o r the fact that they were m a d e without the aid of the compilers, editors and o t h e r p a r a p h e r n a l i a t h a t p r o g r a m m e r s n o w take for g r a n t e d , and b e f o r e the great value of certain pro- g r a m m i n g practices and disciplines h a d b e e n recogniz- ed.
M a n y of the i m p o r t a n t a d v a n c e s m a d e in c o m p u t a - tional linguistics during the same period also tend to- wards the easier c o n s t r u c t i o n of m o r e robust s y s t e m s that can be m o r e readily maintained. T h e m o s t obvi- ous examples c o m e f r o m the d o m a i n of syntactic anal- ysis which is n o w universally t h o u g h t of as a job to be done by a fairly general parsing p r o g r a m , coupled with a g r a m m a r . T h e parser e m b o d i e s the necessary strate- gies and techniques while all k n o w l e d g e of the particu- lar language resides in a static data structure, n a m e l y the g r a m m a r . Associated with the g r a m m a r is a formal language in which a linguist writes rules f r o m which the d a t a s t r u c t u r e is o b t a i n e d a u t o m a t i c a l l y . This f o r m a l language is specially designed to facilitate the s t a t e m e n t of linguistic facts and is largely d e c o u p l e d f r o m the g r a m m a r itself a n d f r o m the m e t h o d s that will be used to process it. This greatly increases the p o w e r that the linguist can bring to the j o b and his ability to m o d i f y the s y s t e m in the light of experience.
In the s a m e period we h a v e c o m e to understand, not just h o w a g e n e r a l - p u r p o s e p a r s e r can be c o n - structed, but h o w to m a k e these parsers m o r e effective by the application of some v e r y general principles. In particular, we have c o m e to appreciate the value of the n o t i o n s of c o m p l e t e parsing and of n o n d e t e r m i n i s m . By c o m p l e t e parsing, we m e a n simply the r e q u i r e m e n t that nothing shall c o u n t as part of the final result ex- cept inasmuch as it is part of an analysis of an entire sentence. The practical value of this a p p a r e n t l y obvi- ous restriction would be difficult to overestimate. A parser that i n c o r p o r a t e s it largely releases the g r a m m a r writer f r o m c o n c e r n for w h e n it would be incorrect to allow an analysis t h a t would be c o r r e c t in a n o t h e r e n v i r o n m e n t , a c o n c e r n which c o n s u m e d m u c h time on the part of the designers of early translation systems.
G e n e r a l m e t h o d s for i m p l e m e n t i n g n o n d e t e r m i n i s m go t o g e t h e r with this and h a v e an equally liberating effect. These m e t h o d s are useful in situations where, at a n y given stage of the process, a n u m b e r of c o n - flicting possibilities are o p e n , a n y n u m b e r of which could lead to useful results. In particular, a c o m p l e t e parser, in the p r e s e n t sense, m u s t pursue lines of at- tack that seem r e a s o n a b l e on the basis of local eviden-
Martin Kay Machine Translation
ce, but which m a y or m a y not lead to a c o m p l e t e anal- ysis. A general m e t h o d for handling n o n d e t e r m i n a c y releases the p r o g r a m m e r f r o m all c o n c e r n for h o w the machine will contrive to follow up on all possibilities; how and w h e n it will return to the choice point and restore the exact state as of that m o m e n t , h o w it will follow all possibilities once, but n o n e m o r e t h a n once, and so forth. T o the extent that early translation sys- tems faced these p r o b l e m s at all, they did so on a case b y case basis, and at great cost in h u m a n labor.
T h e panel was also i m p r e s s e d b y the a d v a n c e s that have b e e n m a d e in general linguistics and our overall u n d e r s t a n d i n g of the workings of h u m a n language in recent years. Various classical p r o b l e m s - - n o u n - n o u n c o m p o u n d s in English, ambiguities of p r e p o s i t i o n a l a t t a c h m e n t , c o n j u n c t i o n , p r o n o m i n a l r e f e r e n c e , and m a n y o t h e r s - - h a v e b e e n m a d e the o b j e c t of intensive research with results that are direct relevance in the construction of translation systems.
The panel t h o u g h t it h a d e v e r y r e a s o n to assume that progress on the relevant fronts would go f o r w a r d at least as quickly as in the past. M o s t of the m e m - bers confidently expect to see some m a j o r new fully a u t o m a t i c systems in use during that period. In partic- ular, it is h o p e d t h a t EUROTRA, a v e r y l a r g e - s c a l e collaborative E u r o p e a n effort, will result in a w o r k i n g p r o t o t y p e . In addition, there is w o r k in progress in a n u m b e r of places on c o m p u t e r - b a s e d w o r k stations for use by translators. It is not clear w h a t these will in- c o r p o r a t e , but it is likely that they will explore some new parts of the large space of possibilities t h a t exist in m a c h i n e - a i d e d translation.
T h e p a n e l m a d e no p r e d i c t i o n a b o u t just which areas of research were likely to fall b e f o r e the inexora- ble a d v a n c e of theoretical linguistics but felt that the future of m o r e technological areas was easier to f o r e - see. T h e r e will, in all p r o b a b i l i t y be m o r e flexible m e t h o d s of syntactic analysis, c a p a b l e of relaxing re- quirements in the face of constructions that would not m e e t with a p u n d i t ' s approval. Either t h e y will fall b a c k on m o r e permissive rules or t h e y will m o d i f y the s e n t e n c e to " c o r r e c t " the a p p a r e n t " e r r o r " . It is also e x p e c t e d that wider use will be m a d e of parsing de- vices which, while allowing for n o n d e t e r m i n i s m in a general way, will be able to m a k e fairly a c c u r a t e j u d g e m e n t s a b o u t the paths that are m o s t likely to lead to a successful solution, and so c o n c e n t r a t e on these first.
T h e panel was in a g r e e m e n t on the a c h i e v e m e n t s of the p a s t and on the desirability of following a n u m b e r of parallel p a t h s in the future. T h e d i s a g r e e m e n t s c o n c e r n e d the extent of the o p t i m i s m in future suc- ce,;ses that those past a c h i e v e m e n t s warrant. S o m e me, m b e r s t o o k the view t h a t a d v a n c e s in c o m p u t e r science and c o m p u t a t i o n a l linguistics, i m p o r t a n t though they are, do not go to the heart of the p r o b - lem; they m a k e easier w h a t once was hard b u t they m a k e nothing possible that once was impossible. Lin- guistics has m a d e a d v a n c e s of which it c a n feel justly p r o u d b u t which, while t h e y m a y i n d e e d go to the heart of the m a t t e r , barely scratch the surface of w h a t needs to be done.
All agree that it would be unjustifiable to d e v o t e an excessive p r o p o r t i o n of the available resources to fully a u t o m a t i c s y s t e m s while neglecting c h e a p e r and m o r e m o d e s t a p p r o a c h e s with m o r e certain s h o r t - t e r m p a y - off. Unless c a u t i o n is exercised, b o t h in p r o m i s e s m a d e and policies followed, there is a high risk that t a x p a y e r s and a d m i n i s t r a t o r s will call all too s o o n for a second ALPAC r e p o r t whose e f f e c t on the entire field of c o m p u t a t i o n a l linguistics will b e a l t o g e t h e r m o r e d e v a s t a t i n g t h a n the first.
S o m e m e m b e r s believe t h a t s p o n s o r s h a v e g r o w n m o r e realistic in their e x p e c t a t i o n s , t h a t so long as t h e y are involved in a continual dialog a b o u t the prog- ress of the w o r k t h e y can be m a d e to u n d e r s t a n d the p r o b l e m s , and t h a t t h e y h a v e the f o r t i t u d e to with- stand u n r e a s o n a b l e pressure f r o m their superiors and their electors. T h e y no longer, for e x a m p l e , e x p e c t full translations of arbitrary texts, but are c o n t e n t with texts f r o m suitably restricted domains.
It is claimed earlier in this r e p o r t , and a g r e e d u p o n b y all the panel m e m b e r s , that fully a u t o m a t i c transla- tion is the line of a t t a c k w h o s e benefits, if realized, would be g r e a t e s t . F u t h e r m o r e , its success would c o n t r i b u t e g r e a t l y to the successes of all o t h e r a p - proaches. T h e subscribers to this view are i m p r e s s e d b y the extent to which the designers of early s y s t e m s were o v e r c o m e b y the sheer c o m p l e x i t y of the design a n d p r o g r a m m i n g t a s k t h a t t h e y h a d u n d e r t a k e n so t h a t the s y s t e m s they built c a n n o t be t a k e n as a m e a s - ure of the t e c h n o l o g y that linguistics, e v e n the linguis- tics of that day, could support.