In the SpeechMaster speech recognition system the grammar serves a two-fold purpose, (1) as the primary source for the creation of the language model, and (2) as the basis for semantic interpretation rules used to derive the task-specific semantics in a syntax-directed manner. After outlining the baseline approach some extensions are motivated by an overview of selected applications. The enhancements then described keep the simplicity and flexibility of the grammar based-approach, while allowing for. a more modular composition of a production language model from predefined parts.
Bibliographic reference. Bergmann, Henning / Hamer, Hans-Hermann / Noll, Andreas / Paeseler, Annedore / Tomaschewski, Horst (1995): "Modularization in task-specific language modelling", In EUROSPEECH-1995, 161-164.