Translation of Prepositions by Neural Networks

Translation of prepositions poses a very serious problem to machine translation because prepositions are highly ambigous. In theory prepo­ sitions can be disambiguated by a filter that excludes already generat­ ed representational objects with no selection restriction match between preposition and np, but it takes too long time in practice. A neurstl net­ work makes the disambiguation in fractions o f a second, because it is fast, robust and very powerful. 1 The Problem T h e tra n sla tion o f p r e p o s itio n s p o se s a v e ry seriou s p r o b le m t o m a ch in e tra n s­ la tio n b e ca u se p r e p o sitio n s are h ig h ly a m b ig io u s — ea ch o f th e m o s t 10 freq u en t p r e p o sitio n s in o n e o f th e 9 E U R O T R A la n g u a g e s is tra n sla te d in to 10 d ifferen t p re p o s itio n s in ea ch o f th e 8 o th e r la n g u a g es— a n d b e c a u s e p r e p o s itio n s a lw ays w ill g e n era te m a n y a tta ch m e n t p a tte rn s. T a k e th e e x a m p le : L en in w r o te th is n o te in his n o t e b o o k in 3 m in u tes in C o p e n h a g e n T h e r e are th e fla t s tru ctu re a n d th e d e e p s tru ctu re a n d 6 a tta ch m e n t p a tte rn s in betw een :