Studies with adult participants and artificial languages in our lab suggest that two distributional characteristics of the Input Sample help the learner to master the highly complex structures of language: 1) Initial intensive training with simple constructions and 2) Semantic Biases. We studied these effects by manipulating distributional characteristics of small training samples of an artificial miniature language. I will argue that the effects found in the lab represent crucial aspects of natural language acquisition. Language is learned along with knowledge of the world, through the interaction with caregivers who tune their speech to the level of development of the learner. In that sense, these lab studies might help us understanding how young associative learners eventually grasp the non-linear patterns of language.