A method for efficient storage and rapid application of context-sensitive phonological rules for automatic speech recognition