What is natural language understanding?
Natural language understanding (NLU) is the part of natural language processing that works out what a piece of text or speech means. It takes an utterance such as “I need to move my delivery to Thursday” and produces structured output: an intent (reschedule delivery), an entity (Thursday), and sometimes a sentiment or urgency score. Software can act on that output, while it cannot act on the raw sentence.
Classic NLU systems were trained on labeled examples for each intent and entity type, and they needed hundreds of phrasings per intent to cope with how people actually talk. Large language models now handle much of this with a prompt and a few examples. The job is the same either way: convert open-ended language into a fixed set of fields that downstream systems expect.
Why natural language understanding matters
NLU is the front door of every chatbot, virtual assistant, and natural language query tool. When it misreads intent, everything behind it fails, however good the rest of the system is. A distributor’s order-status bot that confuses “where is my order” with “cancel my order” creates work and angry customers.
The quality of NLU depends heavily on the vocabulary of the business. Part numbers, internal product names, and regional phrasing all trip up a general model. Teams get the best results by testing against real transcripts from their own customers rather than invented examples, and by keeping a clear path to a person when the model’s confidence is low.