To reduce failed recognitions and computation, current speech-recognition technology requires a list of valid sentences that can be spoken. However, such a list could include thousands or even a million entries. A context-free grammar scheme allows an engine to reduce the possible entries to a list of several hundred rules.