Comparing Semantic Networks of Early Vocabulary Across Languages
Date:
Network-based approaches have provided important insights into language processing in adults, as well as the organization and acquisition of early vocabulary in children. The present study aimed to compare the organization of early vocabulary networks across three different languages: English, French, and Cantonese.
Previous studies often use the CHILDES corpus or adult behavioral data to infer links in early vocabulary networks. We explored the construction of semantic networks using responses from the MacArthur-Bates Communicative Development Inventory: Words and Sentences (MB-CDI: W&S) in the Wordbank database. Specifically, Wordbank contains response matrices, where each column represents a unique word in MB-CDI: W&S, each row represents a child (16 to 30 months), and each cell denotes whether the child produce the word (“1”) or not (“0”). Based on the findings that children are more likely to acquire words that share semantic associations in the learning environment, we inferred network links by calculating word-pair cosine similarities from Wordbank response matrices.
Our preliminary analysis showed similar structural properties, such as small-worldness, across the three language networks, supporting cross-linguistic consistency in early vocabulary. An analysis of the nodes with the highest cosine similarity values revealed higher proportions of verbs and adjectives in the Cantonese network than in English or French networks. Our initial results demonstrate the usefulness of MB-CDI data in constructing early vocabulary networks, which may be used to compare semantic organization in children with developmental language disorder.