Children Semantic Network Growth: A Graph Theory Analysis

In this study, we aim to investigate how children's language develops. To do so, we apply the network model of language and examine the graph-theoretic properties of Word2Vec semantic networks of children through development. The networks are made of words children learn prior to the age of 30 months as the nodes. The links in the word-embedding networks are built from the cosine vector similarity of words normatively acquired by children prior to 2½ years of age. By exploiting some graph measures such as the clustering coefficient and path length, the growth pattern of these semantic networks will be revealed. The small-world property allows for high amounts of local structure combined with global access. Within these semantic networks, there is a considerable local structure in the form of clusters of words. For global structure, some nodes act like bridges. They are actually the hubs of the network and connect the clusters which are semantically far-away. We explore the small-world property of these semantic networks and their changes through language development. The results demonstrate that the Word2Vec semantic networks of children show the small-world property from the early age of several months.