How do linguistic conventions emerge among a population of individuals? A shared lexicon can self-organize at this level through local interactions between individuals, as this has been modelled in the Naming Games computational framework. However, the dynamics of the convergence process towards this shared convention can differ a lot, depending on the interaction scenario. Infants, who acquire social conventions really fast, control actively the complexity of what they learn, often following a developmental pathway. Adults also adapt the complexity of their linguistic input when speaking to language beginners. We show here that such active learning mechanism can improve considerably the speed of language formation in Naming Game models. We compare two scenarios for the interactions: the speaker exherts an active control, or the hearer does. The second scenario shows faster dynamics, with more robustness.