GLR-parsing of word lattices using a beam search method

The process of understanding spoken language re quires the e cient processing of ambiguities that arise by the nature of speech This paper presents an ap proach that allows the e cient incremental integra tion of speech recognition and language understand ing using Tomita s generalized LR parsing algorithm For this purpose the GLR lattice parsing algorithm is revised so that an agenda mechanism can be used to control the ow of computation of the parsing process Subsequently the HMM evaluations of the word models are combined with a stochastical lan guage model to do a beam search similar to where chartparsers are used to do the job