Jon A. Willits (UC Riverside): What Can Children Learn From Six Million Words?

 

What Can Children Learn From Six Million Words?

 

What could children learn about what words mean from how those words are used in child-directed speech? In addition, does the way that parents use words suggest anything about the possible organization and development of semantic memory? We addressed these questions by constructing a semantic model using the distributional statistics for 6,000,000 words of child-directed speech (from the CHILDES corpus), spoken to children under five years of age. First, we show that the statistics of speech to children is very useful for inferring nouns’ semantic category memberships.  Second, we show that the semantic space that emerges from how parents use words is strongly hierarchically organized. These analyses demonstrate that speech to children is useful for the construction of a semantic memory system, and that classical and valuable properties of a semantic system (such as taxonomic and hierarchical structure) ought to emerge naturally, given the structure of the input.