Previous attempts to a utomatically determine m ulti-words as the basic unit for language modeling h ave been successful for extending bigram models 10, 9, 2, 8 to improve t he perplexity o f t he language model and or the w ord accuracy of the speech d ecoder. However, none o f t hese techniques gave improvements over the trigram model so far, except for the rather controlled ATIS task 8 . We t herefore propose an algorithm, that minimizes the perplexity improvement o f a bigram model directly. The n ew algorithm is able to reduce the trigram perplexity a n d also achieves word accuracy improvements in the V erbmobil task. It is the n atural counterpart of successful word classi cation algorithms for language modeling 4, 7 that minimize the leaving-one-out bigram perplexity. W e also give some d etails on the usage of class nding t echniques and m-gram models, which can be crucial to s u ccessful applications of this technique.
Due t o r o b u s t n e s s , l e a r n a b ilitya n d e a s e o f i n t e g r a t iono f e r e n t i n f o r m a t ionsources, c o n n e c t ionist p a r s ings y s t e m s r p a r s ing s p o k e n l a n g u a g e ,
We describe and experimentally evaluate a system, FeasPar, that learns parsing spontaneous speech. To train and run FeasPar (Feature Structure Parser), only limited handmodeled knowledge is required. The FeasPar architecture consists of neural networks and a search. The networks spilt the incoming sentence into chunks, which are labeled with feature values and chunk relations. Then, the search finds the most probable and consistent feature structure. FeasPar is trained, tested and evaluated with the Spontaneous Schednling Task, and compared with a handmodeled LRparser. The handmodeling effort for Fea-sPar is 2 weeks. The handmodeling effort for the LR-parser was 4 months. FeasPar performed better than the LRparser in all six comparisons that are made.
e) Institu t f u r P rogram m stru k tu renu n d D a ten organ i s a tion Uni v ersit at K a rlsru h e, W{7500 Karlsru h e, Germ any Appeare d in the pr oce e di ngs of t he Ni nt h IEEE C o n f e r e n c e o n A r t i c i a l I n t e l l i g e n c e f o r A p p l i c a t i o n s Or l ando, Fl or i da, Mar ch [1][2][3][4][5] 1993 A general approach is presented f o r bui l di n g t r a n sport a bl e nat u r a l l a n guage i nt e r fa c es f or q u est i on a n swering syst ems based o n a KL-ONE-l i ke knowl e d ge represent a t io n . A n exampl e s y stem ,Y A K R, i s d escri bed: T h e Y A K S k n o w l e d ge represent a t io n o f c o n c ept sa n d r el at io n s i s a n n o t a t e d w i t hm i n i m a l s y n t a c t i ci n f o rm at io n t o generat eas e m a n t icc a sefram e g r a m m a r w i t h i n h eritance o f c a ses. T he generat e d g r a m m a r d i r e cts a case f r a m e p a rser, whi c hp r o c esses w r it t e nin p u t i n t oinstant ia t e d c a sefram es. These i nstant ia t ia t io n s a r e e asily translat e d i n t ok n o w l e d ge base queri es. The s a m e m et hod i s a p p l i c a bl e to ot h er object-orient e d k n o w l e d ge bases and o t h er parsing t e chni q u es.The o ri gi nal cont r ib u t io n o f t h i sw o rk is to show a n a p p r o a ch wi t h w h i c h n a t u r a l l a n guage i nt e r fa c es can w i t h l o weort be adapt e d t o w o rk wi t ha n y n ew knowl e d ge base: Whi l e m o st ot h er system s r e qui r e a c o m p l e t e m o d el of t h e d om a i nf o r t h e n a t u r a l l a n guage i nt e r fa c e k n o w l e d ge represent a t io n , w e d eri ve most of t h i si n f o rm at io n f r o m t h e a p p l i c a t io n 's knowl e d ge base. T hi st e chni q u e r e d uc es t he a m o u n t o f w o rk needed t o c r e a t et h e i n t e r fa c e t oa b o u t a d d i t i o n a l 1 5 p ercent a f ter bui l d i n g t h e k n o wl e d ge base for t h e a p p l i c a t io n kernel .AI topic: knowledge acquisition, k n o w le d g e representation, n a tural l a n g u a g e i n terface Language/Tool: Unix ,C ++ Status: im p le m entation c o m p le t e ,e v a lu a tion i n p rogress Eort: about 4 p erson y ears Im p act: quic k d evelopment o f restrictednatural l a n g u a g e in terface for certain classes of k n o w le d g e-basedapplic a tions ( 1 5 % a d d it io n a l k n o w le d g e a cquis it io n w o rk).
IntroductionThe r e haveb e e ns e v e r a la t t e m p t st oc r e a t en a t u r a l l a n g u a g e i n terfaces for databases [15, 10, 5 ] of these natural language int e r f a c e st t r a n s p o r t a b l e ,t h a ti s ,t h e y domai ns by changi ng de pe nde nt knowl e nal dat ab e
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.