Lexical Transfer: Between a Source Rock and a Hard Target

Alan

K. M E L B Y

D e p a r t m e n t of L i n g u i s t i c s Brigham Young University Prove, U t a h 84602 USA

Abstract L e x i c a l t r a n s f e r is the p o i n t of t r a n s i t i o n b e t w e e n an u n c h a n g e a b l e s o u r c e text (a rook) a n d an i n f i n i t e a r r a y of t a r g e t t e x t s (a h a r d p l a c e to f i n d an a c c e p t a b l e one). The a u t h o r ' s C o l i n g 8 6 p a p e r (pp. 104-106) d e s c r i b e d a n e w m e t h o d o l o g y for t e s t i n g l e x i c a l t r a n s f e r in m a c h i n e t r a n s l a t i o n . This p a p e r r e p o r t s on the a p p l i c a t i o n of that m e t h o d o l o g y to a t e s t of the DLT s y s t e m and describes a synchronized bilingual data base by-product. F u r t h e r u s e of the m e t h o d o l o g y is e n c o u r a g e d .

Topic: E v a l u a t i o n translation A d d i t i o n a l Topic:

of m a c h i n e Text data bases

I. D E F I N I T I O N

OF L E X I C A L

TRANSFER

A l t h o u g h the t e r m l e x i c a l t r a n s f e r a p p l i e s m o s t d i r e c t l y to m a c h i n e t r a n s l a t i o n s y s t e m s b a s e d on a l i n g u i s t i c m o d e l of a n a l y s i s , t r a n s f e r , a n d g e n e r a t i o n , it c a n a l s o be a p p l i e d to s y s t e m s in w h i c h t h e r e is no d i r e c t c o r r e s p o n d a n c e b e t w e e n s o u r c e and t a r g e t w o r d s (f~uch as i n t e r l i n g u a l s y s t e m s ) by d e f i n i n g l e x i e a l t r a n s f e r as the p o i n t in p r o c e s s i n g w h e r e the t a r g e t l e x i c a l forms first appear. It is the c r u c i a l p o i n t aL w h i c h all the i n f o r m a t i o n a v a i l a b l e in the s y s t e m m u s t be b r o u g h t to b e a r on the p r o b l e m of c h o o s i n g l e x i c a l forms. The e h o i e e s m u s t be a p p r o p r i a t e , or all the s o p h i s t i c a t i o n of the s y s t e m in o t h e r a r e a s s u c h as w o r d o r d e r a n d d i s c o u r s e m a r k e r s w i l l be of no a v a i l in p r o d u c i n g a c c e p t a b l e output.

L e x i c a l t r a n s f e r m u s t be b a s e d on the s o u r c e text, w h i c h is g e n e r a l l y a given. T h a t is, one c a n n o t c o m e b a c k d u r i n g the e v a l u a t i o n of the o u t p u t and s u g g e s t t h a t the s o u r c e text be c h a n g e d to b e t t e r m a t c h the t a r g e t text. Thus the s o u r c e t e x t can be c o m p a r e d to a reck or to a text c a r v e d in stone. The t a r g e t text, on the other' hand, is s u p p o s e d l y s o m e w h e r e in an i n f i n i t e c o l l e c t i o n of t e x t s c o m p o s e d of m e m b e r s of an i n f i n i t e set of s e n t e n c e s g e n e r a t e d f r o m a l a r g e f i n i t e set of l e x i c a l items. E v e n if one c o u l d list in a d v a n c e all the p o s s i b l e t r a n s l a t i o n s for each word, w h i c h one cannot, the task is d a u n t i n g . T h u s the s p a c e of all t a r g e t l a n g u a g e t e x t s is a h a r d p l a c e in w h i c h to find an a p p r o p r i a t e one. For t h o s e r e a d e r s not f a m i l i a r w i t h the s a y i n g " b e t w e e n a rock a n d a h a r d p l a c e " on w h i c h the t i t l e of this p a p e r is based, I m e n t i o n that it r e f e r s to a d i f f i c u l t s i t u a t i o n in w h i c h all apparent options present problems. Of c o u r s e , the t i t l e t w i s t s the s a y i n g in s e v e r a l ways. Its p u r p o s e is both to e m p h a s i z e the d i f f i c u l t y of lexica] t r a n s f e r and to i l l u s t r a t e it. The i l l u s t r a t i o n is this: P l e a s e t r a n s l a t e the t i t l e into some o t h e r l a n g u a g e , b a s i n g it on an e q u i v a l e n t t a r g e t l a n g u a g e s a y i n g a d j u s t e d to d e s c r i b e lexical transfer. It is novel s i t u a t i o n s that m a k e ] e x i c a ] t r a n s f e r t r u l y d i f f i c u l t to p r o g r a m for.

2.

BACKGROUND

At C O L I N G 8 6 , the a u t h o r p r e s e n t e d a m e t h o d o l o g y for t e s t i n g the ] e x i c a l t r a n s f e r m e c h a n i s m of a m a c h i n e t r a n s l a t i o n system. S i n c e then, the p r o p o s e d test has b e e n p e r f o r m e d on an e a r l y v e r s i o n of the D L T ( D i s t r i b u t e d Language Translation) machine translation system. This paper will d e s c r i b e the r e s u l t s of the test. In the c o u r s e of p e r f o r m i n g the test, a b i l i n g u a l d a t a b a s e of F r e n c h a n d E n g l i s h t e x t s was p r o d u c e d . This data b a s e c o n s i s t s of p a i r e d d o c u m e n t s w i t h

411

synchronized paragraphs. This paper will also describe the data base, which has been edited and is now available to qualified researchers for a small fee.

b e t t e r to f i n d out w i t h d i c t i o n a r i e s of one t h o u s a n d w o r d s a n d all t h e i r c o l l o c a t i o n s t h a n a f t e r the d i c t i o n a r i e s c o n t a i n t h i r t y t h o u s a n d words.

3.

TUNING

DISTORTION

5.

SOME R E S U L T S

FROM THE DLT T E S T

A g o o d m e t h o d o l o g y for t e s t i n g l e x i c a l t r a n s f e r m u s t a v o i d the t r a p of "tuning distortion". Tuning distortion r e f e r s to the m i s l e a d i n g ( d i s t o r t e d ) results obtained from a machine t r a n s l a t i o n s y s t e m w h e n its d i c t i o n a r i e s a n d a l g o r i t h m s are a d j u s t e d (tuned) to a p a r t i c u l a r text. Almost any machine translation system can produce brilliant r e s u l t s w h e n the s a m e text is run t h r o u g h it a g a i n a n d a g a i n w i t h s u c c e s s i v e tuning. The p o w e r of t u n i n g is ~ e ] l - k n o w n a n d has b e e n g i v e n a n a m e in AI research, namely, defining a mieroworld. Corresponding to this power is the well-known difficulty of expanding a microworld system to function intelligently in a macroworld. In a machine translation system, difficulties arise when a tuned system Js applied to a new text.

The D L T m a c h i n e t r a n s l a t i o n p r o j e c t is a v e n t u r e of the B S O c o m p a n y in Utrecht. T h e w o r d list a p p r o a c h w a s u s e d to t e s t its l e x i c a l t r a n s f e r p h a s e e v e n b e f o r e the s y n t a c t i c a n a l y s i s p h a s e was c o m p l e t e . T h i s was d o n e b y m a n u a l l y a n a l y z i n g the t e s t s e n t e n c e s . The four test p a s s a g e s i n c l u d e d o v e r 2000 w o r d t o k e n s w h i c h r e d u c e d to a b o u t 600 c o n t e n t w o r d types, to w h i c h w e r e a d d e d a b o u t 200 m i s l e a d i n g w o r d s . The w o r d l i s t of a b o u t 800 w o r d s w a s u s e d to build dictionaries containing thousands of e n t r i e s . A f t e r the t e x t s w e r e t r a n s l a t e d by the DLT s y s t e m f r o m E n g l i s h to F r e n c h ( d u r i n g the f i r s t q u a r t e r of 1987), t h e y w e r e c o m p a r e d w i t h o f f i c i a l v e r s i o n s of the t e x t s p r e p a r e d by p r o f e s s i o n a l h u m a n t r a n s l a t o r s at the CEC. This comparison r e v e a l e d t h a t m a n y w o r d s m a t c h e d the official language versions, some were a c c e p t a b l e s y n o n y m n s and, as e x p e c t e d , some words were translated inappropriately. The D L T p r o j e c t is be c o n g r a t u l a t e d on the o v e r a l l s u c c e s s of the e x p e r i m e n t . The p r o b l e m w o r d s to be d i s c u s s e d in the p a p e r are not i n t e n t e d to be s i m p l y a c r i t i c i s m of DL'r b u t r a t h e r o b s e r v a t i o n s that m a y be of i n t e r e s t to all m a c h i n e t r a n s l a t i o n researchers. Some inappropriate t r a n s l a t i o n s w o u l d be e a s i l y c o r r e c t e d by d e t e c t i n g p r e d i c t a b l e c o l l o c a t i o n s . In the D L T test, the c o l l o c a t i o n s o f t w a r e was not o p e r a t i o n a l . For e x a m p l e , computer-assisted r e q u i r e s a p a r t i c u l a r t r a n s l a t i o n of assisted. A n o t h e r p r o b l e m is bring, w h i c h can s o m e t i m e s be t r a n s l a t e d as faire venir b u t w h i c h is n o r m a l l y t r a n s l a t e d as prendre in the c o n t e x t of the e x p r e s s i o n bring x to y's consciousness. This r e q u i r e s s y n t a c t i c t r a n s f e r of a t y p e the D L T p r o j e c t c a l l s m e t a t a x i s a n d w h i c h w a s not i m p l e m e n t e d for the test. In a r e c e n t i s s u e of Language Monthly ( D e c e m b e r 1987, p. 7), it was r e p o r t e d t h a t P e t e r Lau, of the E u r o t r a p r o j e c t , said, at the 1987 A s l i b c o n f e r e n c e , t h a t the real p r o b l e m of m a c h i n e t r a n s l a t i o n is not the " r e d u c t i o n of s t r u c t u r a l d i f f e r e n c e s " b ~ t r a t h e r the " d i s a m b i g u a t i o n of l e x i c a l e n t r i e s " . The D L T t e s t f o c u s s e d on s u c h l e x i c a l transfer problems. Some are: words of i n t e r e s t from the test

4.

THE

WORD

LIST

APPROACH

To a v o i d t u n i n g d i s t o r t i o n in a t e s t of l e x i e a l t r a n s f e r , one c a n b u i l d a d i c t i o n a r y f r o m a w o r d list w i t h o u t k n o w i n g w h a t t e x t w i l l be s u p p l i e d later, e x c e p t t h a t it w i l l c o n s i s t of w o r d s f r o m the w o r d list. This approach has s i g n i f i c a n t a d v a n t a g e s o v e r s u p p l y i n g an a r b i t r a r y text a n d u p g r a d i n g the d i c t i o n a r i e s to h a n d l e the text, b e c a u s e t h e r e is a c o n s c i o u s or u n c o n s c i o u s t u n i n g of the d i c t i o n a r y e n t r i e s d u r i n g the u p g r a d e p r o c e s s so long as the t e x t is a v a i l a b l e . In the w o r d list a p p r o a c h , all the w o r d s of the text are c o m b i n e d w i t h a n u m b e r of m i s l e a d i n g w o r d s w h i c h m a k e it d i f f i c u l t to tell w h a t is the s u b j e c t f i e l d of the text. T h e n the c o m b i n e d w o r d s are s o r t e d i n t o a l p h a b e t i c a l o r d e r a n d r e d u c e d to t h e i r b a s i c forms. The a l p h a b e t i c w o r d list is s u p p l i e d to the machine translation dictionary updaters a n d the d i c t i o n a r i e s are s t a b i l i z e d . T h e n the t e x t is p r o v i d e d a n d immediately translated without any u p d a t e s to the s y s t e m a n d w i t h o u t a n y w o r d s m i s s i n g f r o m the d i c t i o n a r i e s . If one a r g u e s t h a t t h i s m e t h o d f o r c e s the d i c t i o n a r y u p d a t e r s to c o n s i d e r too m a n y p o s s i b l e c o l l o c a t i o n s of e a c h w o r d in the list, one is s i m p l y e o m p l a i n i n g a b o u t the d i f f i c u l t y of h a n d l i n g real text. At l e a s t t h i s m e t h o d a l l o w s r e a l i s t i c t e s t i n g of a s y s t e m B E F O R E its d i c t i o n a r i e s h a v e r e a c h e d full size. If t h e r e is a p r o b l e m in the s y s t e m d e s i g n , it is

hardware, area, sheet, pratice, giving, perform, produce, schedule, concern, field, application, induced, lead, benefit, covers bachelor, courses duty, a n d form.

412

For e a c h of the a b o v e w o r d s , the DLT s y s t e m p r o d u c e d a t r a n s l a t i o n w h i c h was n o t a p p r o p r i a t e to the c o n t e x t . These w e r e n o t the o n l y m i s t a k e s , b u t on the o t h e r hand, the D L T s y s t e m t r a n s l a t e d the m a j o r i t y of the w o r d s (60 p e r c e n t ) a o c e p t ~ b l y , w h i l e a f o u r t h (25 p e r c e n t ) w e r e p r o b l e m s for one r e a s o n or a n o t h e r , w i t h I|~ p e r c e n t in the g r a y a r e a b e t w e e n aceept~bility a n d u n a c c e p t a b i l i t y . The r e a d e r is i n v i t e d to c o n s i d e r h o w t h e s e w o r d s w o u l d be h a n d l e d in his or her s y s t e m , be it m a c h i n e t r a n s l ~ t i o n , c o n t e n t a n a l y s i s , or o t h e r natural language processing system. How w o u l d the p r o p e r d i s t i n c t i o n s be m a d e or a n a p p r o p r i a t e t r a n s l a t i o n for t h e s e w o r d s be f o u n d w i t h o u t b e i n g t u n e d to a p a r t i c u l a r t e x t or s u b l a n g u a g e ? N o t s u r p r i s i n g l y , the w o r d hardware needs ~ special translation when r e f e r r i n g to c o m p u t e r h a r d w a r e . But in today'E~ t e c h n i c a l d o c u m e n t s , t h e r e c a n be r e f e r e n c e to c o m p u t e r s b u t a l s o to h a r d w a r e in a m o r e g e n e r a l s e n s e or in r e f e r e n c e to t o o l s or w e a p o n s . How can the a p p r o p r i a t e s e l e c t i o n be m a d e w i t h o u t an e n o r m o u s w o r l d m o d e l a n d a s y s t e m w h i c h t r u l y u n d e r s t a n d s the t e x t ? ( S h a d e s of B a r - H i l l e l ) Another example is the w o r d area, w h i c h c a n be t r a n s l a t e d r~gion or pattie. However, t h e s e two o p t i o n s a r e n o t i n t e r c h a n g e a b l e a n d the d i s t i n c t i o n is s u b t l e a n d not d e p e n d e n t on p r e d i c t a b l e collocations. A sheet c a n be a drap (on a bed), a feuille (of p a p e r ) , or a lame, d e p e n d i n g on c o n t e x t . U n f o r t u n a t e l y for lexica] t r a n s f e r , the w o r d sheet w i l l n o t a l w a y s be f o l l o w e d by a p r e p o s i t i o n a l p h r a s e i n d i c a t i n g the c o m p o s i t i o n of the sheet.

e l e c t r o m a g n e t i c p r e s s u r e ) , lead (a w i r e or a s a l e s c o n t a c t ) , benefit ( a d v a n t a g e or g o v e r n m e n t p a y m e n t ) , cover (lid or a b s t r a c t limit), duty ( o b l i g a t i o n or i m p o r t tax), a n d course (path or aoadmemic class). Two of the w o r d s in the list i n v o l v e an e l e m e n t of p o e t i c j u s t i c e . Katz and F o d o r d i s t i n g u i s h e d the a c a d e m i c d e g r e e a n d u n m a r r i e d m a n r e a d i n g s of bachelor w i t h m a r k e r s , but d i d not tell DLT h o w to d i s t i n g u i s h b e t w e e n t h e m w h e n the w o r d is e n c o u n t e r e d in text. A n d the t r a n s l a t i o n of form d e p e n d s on its (~ontent.

6. T H E B I L I N G U A L D A T A B A S E P r e l i m i n a r y to the D L T test, a c o r p u s of t e x t s was g a t h e r e d to a s s i s t in d i c t i o n a r y d e v e l o p m e n t . A p o r t i o n of the c o r p u s was k e p t s e c r e t a n d the t e s t p a s s a g e s w e r e c h o s e n f r o m this p o r t i o n . A l a r g e r p o r t i o n w a s m a d e a v a i l a b l e to the D L T p r o j e c t for l e x i c a l studies. The W a t e r l o o c o n c o r d a n c e s y s t e m was U s e d to g e n e r a t e K W I C l i s t i n g s for the l e x i c o g r a p h e r s to o b s e r v e the v a r i o u s u s e s of w o r d s in a c t u a l texts. The b i l i n g u a l d a t a b a s e u s e d in the t e s t was d e r i v e d f r o m p u b l i c d o m a i n d o c u m e n t s of the C E C a n d the U n i t e d N a t i o n s , to a v o i d c o p y r i g h t p r o b l e m s . It c o n s i s t s of t w e n t y d o c u m e n t s , each w i t h an E n g l i s h and a F r e n c h v e r s i o n , on s u b j e c t s r a n g i n g f r o m m i g r a n t w o r k e r s to the E S A s p a c e l a b to the a u t o m o b i l e i n d u s t r y to a g r i c u l t u r e . The documents were first scanned using a Kurzweil OCR device. T h e n the d i s k f i l e s w e r e h a n d e d i t e d into 400 s m a l l s y n c h r o n i z e d files, 200 E n g l i s h a n d 200 French, r e p r e s e n t i n g a t o t a l of a b o u t e i g h t m e g a b y t e s of d a t a (i.e. o v e r one m i l l i o n words). As of this w r i t i n g , the small f i l e s are b e i n g f u r t h e r p r o o f e d a g a i n s t the o r i g i n a l d o c u m e n t s , a n d the p a r a g r a p h s or o t h e r l o g i c a l u n i t s of the t e x t s are b e i n g s y n c h r o n i z e d by e d i t i n g in s e g m e n t n u m b e r m a r k s . These marks are u s e d by a s i m p l e p r e p r o c e s s i n g p r o g r a m to p r o d u c e s y n c h r o n i z e d twoc o l u m n b i l i n g u a l o u t p u t for i n d e x i n g by WordCruncher, a new dynamic concordance s y s t e m w h i c h has b e c o m e a v a i l a b l e s i n c e the p r o j e c t b e g a n . This two-column f o r m a t w i l l f a c i l i t a t e the s t u d y of all o c c u r r e n c e s of w o r d s or e x p r e s s i o n s w i t h the o t h e r l a n g u a g e s e g m e n t a u t o m a t i c a l l y d i s p l a y e d to a l l o w the r e s e a r c h e r to q u i c k l y see h o w t h a t w o r d or e x p r e s s i o n was t r a n s l a t e d in the c o r p u s . T h e e d i t e d c o r p u s is a v a i l a b l e w i t h p e r m i s s i o n of B S O for a m o d e s t fee to qualified researchers. It is h o p e d that the c o r p u s , the w o r d list m e t h o d o l o g y , a n d the r e s u l t s of the DLT t e s t w i l l be of use to o t h e r s in the m a c h i n e translation community.

A practice c a n be w h a t a m e d i c a l doctor does when treating people, what m u s i c i a n d o e s to g e t r e a d y for a c o n c e r t , or w h a t is n o r m a l l y d o n e in some endeavor. T h e s e t h r e e m a y be translated differently.

T h e v e r b f o r m giving can r e f e r to a t r a n s f e r of an o b j e c t to s o m e o n e or to a r e s u l t ("one p l u s t h r e e g i v e s four"). To perform can r e f e r to o n e ' s n o r m a l d u t y or to a s t a g e p e r f o r m a n c e a n d m a y be t r a n s l a t e d d i f f e r e n t l y . L i k e w i s e , to produce c a n t r a n s l a t e d i f f e r e n t l y , d e p e n d i n g on w h e t h e r one is t a l k i n g a b o u t a p l i y Or f a c t o r y . T h e r e a d e r c a n use a s t a n d a r d d i c t i o n a r y to see the d i f f i c u l t i e s in the f o l l o w i n g w o r d s : sohedule (time t a b l e or p r i c e list), c o n e e x ~ ( i n t e r e s t or a n x i e t y ) , field ( l i t e r a l a r e a of t e r r a i n or f i g u r a t i v e f i e l d of i n t e r e s t ) , application ( t r e a t m e n t or level of effort), induced ( s o c i a l or

413