Gender Roles from Word Embeddings in a Century of Children’s Books

Anonymous

Abstract: When presenting content to children, educators and parents not only want to know whether characters of different backgrounds are represented; they also want to understand how these characters are depicted. In this paper, we measure the gender portrayal of central domains of social life as depicted in highly influential children's books using word co-occurrence and word embeddings. We find that females are more likely than males to be associated with words related both to family and appearance, while males are more associated with business-related words. The gender associations with appearance and business have endured over time, whereas family word associations have become more gender-neutral. We make two main contributions: one, we create a word embeddings data set, StoryWords 1.0, of 100 years of award-winning children's literature, and two, we show inequality in the portrayal of gender in this literature, which in turn may convey messages to children about differential roles in society. We include our code and models as supplemental data associated with this manuscript.