Objective
The PALABRE project aimed to define and realise an acquisition and interrogation system for a large and evolving knowledge base. Information, interrogation and updating were performed using natural language. The input-outputs were to be either textual or vocal (speech understanding and synthesis). Common tools and methods were to be developed independently so that several languages (French, English, Italian) could be used. Requests were to be converted with the help of inference rules and sent to an intelligent retrieval process based on semantic knowledge representation. LISP and PROLOG on a LISP machine were used. The feasibility of such a system were to be demonstrated through a phone directory "Yellow Pages" prototype application in Italian, English and French. .
The project aimed to define and realise an acquisition and interrogation system for a large and evolving knowledge base. Information, interrogation and updating were performed using natural language. The input/outputs were either textual or vocal (speech understanding and synthesis). Common tools and methods were developed independently so that several languages (French, English, Italian) could be used. Requests were converted with the help of interference rules and sent to an intelligent retrieval process based on semantic knowledge representation. LISP and PROLOG on a LISP machine were used. Different parts of the system were identified and their interfaces defined. The general architecture was based on a blackboard approach to combine several sources of knowledge. Two different approaches were proposed to parse a query set by the user: a deterministic parser for written input, and an augmented transition network parser for spoken input. The knowledge was represented using KL-ONE. The deduction module allowing the interpretation of a user query by the knowledge base and the answer generation module were investigated.
Different parts of the system were identified and their interfaces defined. The general architecture was based on a blackboard approach to combine several sources of knowledge. Two different approaches were proposed to parse a query inputted by the user: a deterministic parser for written input, and an augmented transition network parser for spoken input. The knowledge was represented using KL-ONE. The deduction module allowing the interpretation of a user query by the knowledge-base and the answer generation module were investigated.
Exploitations
PALABRE served as a preliminary study for subsequent projects in the field of speech processing.
Fields of science
Programme(s)
Topic(s)
Data not availableCall for proposal
Data not availableFunding Scheme
Data not availableCoordinator
35018 RENNES
France