4.2 Azure AI Language Service
Key Takeaways
- Azure AI Language is the unified NLP service combining many text capabilities under one endpoint; it is configured in Language Studio within the Foundry portal.
- Pre-configured features (sentiment, key phrases, NER, entity linking, language detection, PII, summarization, opinion mining) require NO training.
- Customizable features (CLU, custom text classification, custom NER, custom question answering) require you to train a model with labeled data.
- CLU replaced LUIS — the LUIS portal closed October 31, 2025 and runtime/authoring endpoints fully retired March 31, 2026; CLU extracts intents and entities from utterances.
- Custom question answering replaced QnA Maker and builds a knowledge base from FAQs, manuals, and web pages to answer natural-language questions.
Quick Answer: Azure AI Language is the unified NLP service providing both pre-configured features (sentiment analysis, NER, key phrases, language detection, PII detection, summarization) and customizable features (CLU for intent/entity extraction, custom question answering, custom text classification, custom NER). It replaces several older services including LUIS and QnA Maker.
What Is Azure AI Language?
Azure AI Language is Microsoft's comprehensive NLP service that provides a wide range of text-analysis capabilities under a single resource and endpoint. It is the go-to Azure service for text-related AI tasks, and it is configured and managed in Language Studio (now part of the Microsoft Foundry / Azure AI Foundry portal).
Pre-Configured Features (No Training Required)
These features work immediately — send text and get results:
| Feature | Description | Example |
|---|---|---|
| Sentiment analysis | Positive/negative/neutral with confidence scores | "Great product!" -> Positive (0.98) |
| Key phrase extraction | Main topics and concepts | "The hotel was clean and affordable" -> ["hotel", "clean", "affordable"] |
| Named entity recognition | Entities with categories | "John works at Microsoft" -> Person: John, Organization: Microsoft |
| Entity linking | Links entities to a Wikipedia knowledge base | "Paris" -> Paris, France (Wikipedia link) |
| Language detection | Identify text language | "Hola mundo" -> Spanish (es) |
| PII detection | Find and redact sensitive data | "Call 555-1234" -> "Call [PHONENUMBER]" |
| Text summarization | Extractive and abstractive summaries | Long article -> key-point summary |
| Opinion mining | Aspect-based sentiment | "The camera is great but battery is poor" -> camera: positive, battery: negative |
Customizable Features (Training Required)
These features require you to train a model with your own data:
| Feature | Description | Training Data |
|---|---|---|
| CLU (Conversational Language Understanding) | Understand intents and entities in utterances | Labeled utterances with intents/entities |
| Custom text classification | Classify text into your categories | Labeled documents |
| Custom NER | Extract your own entity types | Labeled text with custom entities |
| Custom question answering | Answer questions from your documents | FAQ pages, manuals, web pages |
Exam trap: if a question hinges on whether something works out of the box or requires labeled training data, the pre-configured vs. customizable split is the answer. Sentiment, key phrases, NER, language detection, PII, and summarization are pre-configured; the four "custom"/CLU items require training.
Conversational Language Understanding (CLU)
CLU lets you build models that understand the intent (what the user wants to do) and extract entities (key information) from natural-language utterances.
CLU Replaces LUIS
CLU is the successor to LUIS (Language Understanding Intelligent Service). The LUIS authoring portal closed on October 31, 2025, and the LUIS runtime and authoring endpoints were fully retired on March 31, 2026 — after that date all LUIS requests fail. If you see "LUIS" on the exam, the current replacement is CLU. CLU adds multilingual training (train in one language, predict in 99+), LLM-powered quick deploy, and orchestration workflow, which routes an utterance to the correct CLU or custom-question-answering project.
CLU Concepts
| Concept | Definition | Example |
|---|---|---|
| Utterance | The user's input text | "Book a flight to Paris next Friday" |
| Intent | What the user wants to do | BookFlight |
| Entity | Key information in the utterance | Destination: Paris, Date: next Friday |
How CLU Works
- Define intents — the actions users want (BookFlight, CancelFlight, CheckStatus)
- Define entities — the key information to extract (destination, date, passenger name)
- Add utterances — example sentences per intent with entities labeled
- Train the model — CLU learns patterns from your labeled examples
- Test and publish — test with new utterances, then deploy
- Use in applications — send user input, receive intent + entities
After training, a new utterance such as "Find me a flight to Berlin next week" returns Intent: BookFlight (0.95), Entities: {destination: Berlin, date: next week}.
On the Exam: CLU identifies INTENTS (what the user wants) and ENTITIES (the details). It requires training with labeled utterances. LUIS is retired — CLU is the replacement.
Custom Question Answering
Custom question answering creates a knowledge base that answers natural-language questions from your existing content. It is the successor to the retired QnA Maker.
How It Works
- Add sources — FAQ pages, product manuals, documents, or manual Q&A pairs
- Build the knowledge base — the service extracts question-answer pairs automatically
- Test — ask questions and verify answers
- Publish — deploy as an endpoint for chatbots and applications
Key features include multi-turn conversations (follow-up questions that reference prior context), chit-chat (built-in greetings and casual replies), synonyms (handle different phrasings), and confidence scores on every answer.
On the Exam: CLU is for understanding intent and entities; custom question answering is for finding specific answers in your own documents. A FAQ bot built from a handbook = custom question answering. A chatbot that must figure out what action the user wants = CLU.
When to Use Which Feature
| Scenario | Feature |
|---|---|
| Decide if a review is positive or negative | Sentiment analysis |
| Find the main topics in a document | Key phrase extraction |
| Identify people and organizations in an article | Named entity recognition |
| Detect and redact Social Security numbers | PII detection |
| Understand what a user wants in a chatbot | CLU (intent + entity extraction) |
| Build a FAQ bot from product documentation | Custom question answering |
| Classify support tickets into custom categories | Custom text classification |
| Extract custom data types from documents | Custom NER |
Which Azure AI Language feature understands user intents and extracts entities from natural language utterances?
Which retired Azure service did Conversational Language Understanding (CLU) replace?
A company wants to build a chatbot that answers employee questions about company policies using their existing HR handbook and FAQ documents. Which Azure AI Language feature should they use?
In CLU, what does an "intent" represent?
Which TWO of the following are pre-configured Azure AI Language features that require NO training? (Select two)
Select all that apply