What is the recommended approach for handling proper nouns (names, places, organizations) in realtime captioning?