Previous studies have demonstrated that the statistical properties of adult generated free associates can predict the order of early word learning in children. In this paper, we investigate the cause of this phenomenon. We propose that early word learning may be driven by the contextual diversity of words in child directed speech, which is in turn correlated with an underlying associative structure in adult language. We present evidence for this hypothesis by analysing the co-occurrence of words in the CHILDES corpus of child directed speech. We find that a words contextual diversityits number of unique neighborspredicts the order of early word learning and is highly correlated with a words associative diversity. Using longitudinal network analysis on developing early semantic networks from 15 to 30 months, we also find evidence for a specific growth process called preferential acquisition, in which words with more diversity in the learning environment are learned earlier than less diverse words. Only adjectives support preferential attachmenta process based on the structure of known wordsand this is consistent with the evidence that adjective learning is a strongly facilitatory process, in which the learning of an adjective is enhanced by comparison with other similar adjectives.