My personal reason is that I don't consider the output of speech recognition to be "internal concepts for reasoning". In a Spoken Dialog System, this task is typically performed by a subsequent component that does natural language understanding (rather than recognition).