Experimental research on the environmental conditions that promote generative language learning is reviewed. Recombinative generalization is introduced as a process that enables individuals to express and to comprehend novel utterances. This review focuses on the use of a miniature linguistic system paradigm to explore how recombinative generalization can be efficiently established. How the necessary stimulus conditions vary depending upon the linguistic repertoires of individuals are discussed. Examples of learners who have no knowledge, partial knowledge, and near complete knowledge of the lexical constituents included in language systems are used to illustrate differences in the necessary conditions that can yield generative language learning. Also mentioned are implications for understanding normal language acquisition and for developing efficient language intervention programs. (Author)