This document discusses natural language generation (NLG), which generates natural language text or speech from a data source using logical rules like grammar. It describes simple NLG using text templates or filling in blanks, and more complex NLG involving 6 stages: determination, structure, aggregation, word choice, references, and realization. The document also discusses using artificial intelligence like machine learning to generate text without grammatical analysis, and examples of NLG uses such as product descriptions, chatbots, journalism, and reports. Finally, it covers Generative Pre-trained Transformer 3 (GPT-3), a powerful new NLG model from 2020 that can generate lengthy coherent text from only a few starting words.