Generatif

Generatif is an adjective used across disciplines to describe systems, theories, or methods that generate new content, structures, or data from a defined set of rules, parameters, or learned representations.

In linguistics, generatif is most often associated with generative grammar, a framework introduced by Noam Chomsky. It seeks to explain how speakers can create and understand an unlimited range of sentences with a finite set of rules and transformations applied to underlying representations, suggesting that syntax is governed by formal, rule-based structure.

In artificial intelligence and statistics, generative models learn the joint distribution of data and can sample new instances from that distribution. They contrast with discriminative models that predict labels without modeling how data is produced. Notable families include probabilistic graphical models, variational autoencoders, generative adversarial networks, and diffusion models. Applications include image, audio, and text generation, data augmentation, and simulation.

In art and design, generative approaches use algorithms, randomness, and procedural rules to produce works, often evolving over time or responding to input. Generative design applies similar ideas to products, graphics, and architecture.

The term generatif thus encompasses methods that synthesize novel content from an internal model. Its precise meaning depends on the field, but the common thread is generation from structured representations.