Language generativity can be described as the ability to produce sentences never before said, and to understand sentences never before heard. One process often cited as underlying language generativity is response generalization. However, though the latter seems to promise a technical understanding of the former at a process level, an investigation of definitions and approaches to the term ''response generalization'' that appear in the literature suggests that it does not do so. We argue that a more promising candidate for the role of key process underlying language generativity is derived relational responding. We introduce the latter concept and describe empirical research showing its connection with language. We subsequently present a relational frame theory (RFT) conceptualization of derived relations as contextually controlled generalized relational responding. We then review a series of recent studies on derived manding in developmentally delayed children and adults that arguably demonstrate the applied utility of a derived relations-based approach with respect to the phenomenon of generative language.