Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

Stochastic Context-Free Language Modeling with Evolutional Grammars

Michael K. Brown, Stephen C. Glinski

AT&T Bell Laboratories, Murray Hill, NJ, USA

We describe an efficient implementation of a stochastic context-free language model suitable for real-time processing of speech and handwriting. Several similar prior efforts have some overlap with the methods we present, but in each case some shortcoming was overcome. This paper describes the first integration of a stochastic context-free language model with a stochastic acoustic model in a way that makes the language modeling overhead very small.

This is achieved through the use of an Evolutional Grammar (EG) consisting of both ephemeral (Markov) word models and grammatical rules which are dynamically expanded while processing speech input. The efficacy of this approach is tested with the ARPA Naval Resource Management speech application.

Full Paper

Bibliographic reference.  Brown, Michael K. / Glinski, Stephen C. (1994): "Stochastic context-free language modeling with evolutional grammars", In ICSLP-1994, 779-782.