Natural language generation

Natural language generation (NLG) is the computational production of human-readable language from data, structured representations, or other input. It ranges from selecting words for fixed templates to generating open-ended text with neural models.

Content determination: The process of selecting which information an output should express. It decides which input facts belong in the generated text.
Natural language processing: The computational study and processing of human language. NLG is the language-producing side of this broader field.
Automated journalism: The use of software to produce news stories from structured information. News organizations use generation to report routine, data-rich events.
Rule-based system: A system that follows explicitly encoded rules to transform inputs into outputs. Rule-driven generators are interpretable but costly to expand across language and tasks.
Hallucination (artificial intelligence): A generated claim that is unsupported by, or conflicts with, available evidence. Fluent output can contain invented details absent from its input.
Document planning: The organization of selected information into a coherent document structure. It arranges chosen content before sentence-level wording begins.
Data-to-text generation: The generation of natural-language descriptions directly from structured data. It is a common NLG setting with explicit input records or measurements.
Data storytelling: The communication of insights from data through narrative and visualization. Generated prose can explain patterns that charts alone leave implicit.
Statistical natural language generation: Language generation that uses statistical models learned from data. It replaced many hand-authored decisions with patterns estimated from examples.
Controllable text generation: The generation of text under specified constraints such as style, topic, or form. Controlling output attributes remains difficult without sacrificing fluency.

Linked from 1 page