A team is designing a Watson Assistant solution and wants to separate natural language understanding from the dialog flow so the same language model can be reused across multiple customer experiences. Which Watson Assistant component should they focus on reusing?