Training generative repertoires within agent-action-object miniature linguistic systems with children.

This study investigated processes responsible for generative language acquisition through the use of a miniature linguistic system. The miniature linguistic system consisted of nonsense syllables and concrete-enactive, agent-action referents. The purpose of Experiment 1 was to determine (a) whether children would recombine agent and action constituents to produce novel utterances and (b) whether children would generate further extensions of the linguistic system (e.g., agent-action-object sentences) following training of a novel syntactic construction. Four children (aged 8:8, 7:4, 4:9, and 4:5) produced novel utterances to describe untrained agent-action referents. They also progressed from agent-action learning to producing agent-action-object sentences after training on only one or two examples of this sentence type with the appropriate referents. Experiment 2 explored conditions more likely to facilitate recombinative generalization among preschoolers. In particular, how a history of lexical learning affects subsequent language learning was investigated with seven 4-year-olds. Results indicated that a history of lexical learning greatly enhanced generative production of untrained agent-action utterances. In addition, all seven children learned new syntactic rules to generate three-word utterances, regardless of the orderings of agent, action, and object words. Implications for developing efficient language remediation programs are discussed.