What is natural language generation?
Natural language generation (NLG) is the production of readable text by software from data, instructions, or other text. It is the output half of natural language processing, paired with natural language understanding on the input side. Early NLG systems filled templates: a weather service inserted forecast values into fixed sentences. Modern systems use large language models that compose new sentences word by word, which is why NLG is now usually discussed under generative AI.
Natural language generation in practice
A property and casualty insurer uses NLG to turn structured claim data into the first draft of a settlement letter. The model receives the policy number, the covered amounts, and the adjuster’s coded decision, and produces a letter in plain language that a reviewer edits and signs. A logistics company generates daily shipment summaries for each customer from tracking data, replacing a spreadsheet nobody read. In both cases the facts come from a system of record and the model supplies only the wording.
What to watch for
- Keep the facts in structured data and let the model write around them, because a model that invents a figure produces a hallucination that reads as confidently as the truth.
- Decide who reviews generated text before it reaches a customer, a regulator, or a court.
- Test for tone and consistency across many outputs, since models drift in style from one generation to the next.
- Record which model and prompt produced each document, so a disputed letter can be traced later.