Language modelling is an area of speech recognition research which has received considerable attention in recent years. Probabilistic context-free grammars provide a relatively powerful model, but the computational requirements of the application are stringent. This paper examines the computational requirements of a probabilistic generalised LR parser, both in terms of the size of the tables required to drive the parsing algorithm and the run time data structures involved. An efficient general method for generating the n best parses of uncertain input is also described.
Bibliographic reference. Wrigley, E. N. / Wright, J. H. (1991): "Computational requirements of probabilistic LR parsing for speech recognition using a natural language grammar", In EUROSPEECH-1991, 761-764.