6.3.1. Conversational Language Understanding (CLU)

💡 First Principle: CLU maps natural language to structured data your application can act on. When a user says "Book a flight to Paris next Tuesday," CLU identifies the intent (BookFlight) and extracts entities (destination: Paris, date: next Tuesday). The exam tests your understanding of intents, entities, training utterances, and the deployment workflow.

What breaks without proper CLU design:

• Missing "None" intent → Every utterance forces a match, even nonsense
• Too few training utterances → Poor generalization to real user phrasing
• Overlapping intents → Confusion about which intent matches

CLU vs. pre-built language features:

• Pre-built (NER, sentiment): Works immediately, general-purpose
• CLU: Requires training, domain-specific, custom intents/entities