Base Architectures for NLP

Tom Mahieu, Stefan Raeymaekers, Karel De Vlaminck, Gerda Janssens, Wouter Joosen 
 

Our goal is to make an OO framework for natural language processing (NLP).
With this framework it should be possible to create a variety of
applications ranging from simple spelling checkers to complex translation
systems, just by plugging different components (e.g. morphological lookup
component, chart parser component, etc.) in the framework.
This paper compares 2 base architectures that can form the core of such 
an NLP system.  The first one considers NLP from a processing point of view
(a text gets processed), the second one from a text point of view (a text
processes itself).
 
Some important considerations are made concerning the configurability and
openness of the system.  Considerations to make the system concurrent and
distributed to obtain better performance are also discussed.  Finally, the
impact on processing quality of this approach is contemplated.
 
 
Contact Info: 

Tom Mahieu 

K.U. Leuven 

Departement Computerwetenschappen 

Celestijnenlaan 200A 

B-3001 Heverlee 

Phone: +32 16 327575 

E-mail: Tom.Mahieu@cs.kuleuven.ac.be