We examine the nature of phonological and semantic similarity in early language learning. We consider how the use of this information might change over the course of development. To this end, we represent the lexicon as either a phonological or semantic network and model the growth of this network. Constructing normative vocabularies from the Communicative Development Inventory norms, we utilize a preferential attachment growth algorithm. We predict and quantify the words which will be learned next, comparing the two network representations. We consider the effect of age, total vocabulary size and language ability as measured through CDI percentile. Our findings suggest that the semantic representation does not outperform the baseline bag-of-words model, whereas the phonological representation conditionally does. More generally, we show that the network representation influences the ability of a model to capture vocabulary growth. We further offer a method of analysis for testing representational assumptions in network models.