ISCA Archive Eurospeech 1993
ISCA Archive Eurospeech 1993

Integration of neural networks and robust parsers in natural language understanding

Ying Cheng, Yves Normandin, Paul Fortier

In this paper, we describe a natural language understanding system which focuses on spoken language and integrates a neural network classifier and robust parsers. With the help of classification, the application domain is divided into several subsets. Parsers are constructed for each subset and as a result, the complexity of grammar construction is much smaller than if a grammar for the whole application domain had to be constructed. Furthermore, by using a neural network, our goal was to take the advantage of its learning ability and robustness to noise. Although the system was implemented in a short time, it performed reasonably well in its first participation in a DARPA ATIS (Air Travel Information System) evaluation in Nov. 92. It was quite robust to speech recognition errors in particular.

Keywords: neural network, parser, natural language understanding.