Blog
Guides and insights on AI voice agents, call centers, and calling at scale.
Glossary & Guides
What Is Natural Language Understanding?
Published
When you call a business and say "I want to change my delivery address" instead of pressing a number on a menu, and the system understands what you mean, natural language understanding is at work. It's one of the core technologies behind modern phone menus, AI voice agents, and many other tools that let people interact with computers using everyday language.
The Definition
Natural language understanding (NLU) is a branch of artificial intelligence focused on enabling computers to understand the meaning of human language, including what people want and the important details in what they say.
"Natural language" simply means the ordinary language people use, such as English, Spanish, or Urdu, as opposed to computer code or strict commands.
NLU is part of a broader field called natural language processing (NLP), which covers all kinds of computer work with human language, including understanding, translating, summarizing, and generating it.
Understanding vs Hearing
It's worth separating NLU from speech recognition:
- Speech recognition converts spoken words into text. It works out what words were said.
- Natural language understanding works out what those words mean.
On a phone call, speech recognition happens first, turning the caller's voice into text. NLU then interprets that text.
The Key Ideas in NLU
Intents
An intent is what the person wants to do. For example:
- "I'd like to pay my bill" → Make a payment
- "Where's my order?" → Check order status
- "Can I speak to someone about a problem?" → Speak to an agent
People express the same intent in countless ways. "Where's my order?", "My parcel hasn't arrived," and "I'm still waiting for my delivery" all point to similar intents. A key job of NLU is recognizing the intent regardless of the exact wording.
Entities
Entities are the specific details within what someone says. For example, in:
"I'd like to move my appointment to Thursday at 3pm."
The intent is reschedule appointment, and the entities are Thursday (date) and 3pm (time).
Other common entities include names, addresses, account numbers, product names, amounts, and locations.
Context
People often refer back to earlier parts of a conversation. If a caller says "Can I change it to the morning instead?", NLU needs context to understand that "it" refers to the appointment mentioned earlier. Understanding context is essential for natural conversations.
How NLU Has Developed
Rule-based systems
Early systems relied on hand-written rules and keyword lists. If a caller said "bill," the system might route them to billing. These systems were predictable but rigid and easily confused by unexpected wording.
Machine learning
Later systems were trained on examples of real phrases labeled with their intents. They learned to recognize intents from patterns in the data, handling more variety in how people speak.
Large language models
Today, many systems use large language models, AI trained on enormous amounts of text. These models can understand a wide range of phrasing, handle more complex requests, and follow context across a conversation much more effectively than earlier approaches.
Where NLU Is Used in Business Calling
Conversational phone menus. Callers describe what they need in their own words instead of navigating numbered options.
Call routing. Understanding what callers need and sending them to the right team.
AI voice agents. Understanding everything callers say throughout a conversation, so the agent can respond appropriately.
Agent assist tools. Understanding the conversation in real time to suggest relevant information.
Speech analytics. Identifying the topics and reasons for calls across large numbers of conversations.
Chatbots and messaging. Understanding written messages in chat and text channels.
The Challenges of Understanding Language
Human language is complicated, and NLU still faces difficulties.
Ambiguity. "I want to cancel" could mean canceling an order, a subscription, or an appointment. Context or a clarifying question may be needed.
Multiple requests. Callers often ask for more than one thing at once: "I need to update my address and check when my next payment is due."
Unusual phrasing. Slang, regional expressions, and incomplete sentences can be harder to interpret.
Speech recognition errors. If words are transcribed incorrectly, the meaning may be misunderstood.
Indirect language. People don't always say exactly what they mean. "I've been waiting a really long time for this refund" might be both a request for an update and an expression of frustration.
Names and specialist terms. Unfamiliar names or industry-specific terms can be hard to recognize.
Designing for Good Understanding
Systems that use NLU well typically:
- Ask clarifying questions when a request is ambiguous.
- Confirm important details, such as dates, amounts, and account numbers.
- Handle multiple requests in a single conversation.
- Offer a way to reach a person when understanding fails.
- Learn from real conversations, improving over time based on where misunderstandings happen.
Teaching Computers to Understand People
Natural language understanding is what allows people to speak to technology the way they speak to each other, rather than adapting to rigid menus and commands. In business calling, it makes phone systems more natural, routing more accurate, and AI conversations more useful. Language will always contain ambiguity and surprises, but NLU continues to narrow the gap between what people say and what computers understand.