Tales and Tropes: Gender Roles from Word Embeddings in a Century of Children’s Books

Proceedings of the 29th International Conference on Computational Linguistics, pages 3086–3097 October 12–17, 2022. 3086 Tales and Tropes: Gender Roles from Word Embeddings in a Century of Children’s Books∗ Anjali Adukia1, Patricia Chiril1, Callista Christ1, Anjali Das2 Alex Eble2, Emileigh Harrison1, Hakizumwami Birali Runesha1 (1) University of Chicago {adukia, pchiril, callistac, harrisone, runesha}@uchicago.edu (2) Columbia University ad3888@columbia.edu, eble@tc.columbia.edu Abstract The manner in which gender is portrayed in materials used to teach children conveys mes- sages about people’s roles in society. In this paper, we measure the gendered depiction of central domains of social life in 100 years of highly influential children’s books. We make two main contributions: (1) we find that the portrayal of gender in these books repro- duces traditional gender norms in society, and (2) we publish StoryWords 1.0, the first word embeddings trained on such a large body of children’s literature. We find that, relative to males, females are more likely to be rep- resented in relation to their appearance than in relation to their competence; second, they are more likely to be represented in relation to their role in the family than their role in business. Finally, we find that non-binary or gender-fluid individuals are rarely mentioned. Our analysis advances understanding of the different messages contained in content com- monly used to teach children, with immediate applications for practice, policy, and research. 1 Introduction Educators and parents use books to teach children messages about society, conduct, and the world. These messages may be encoded in how different identities are, and are not, represented. If there are systematically different associations between spe- cific identities and depictions, such messages can shape how children view the roles that they them- selves, as well as others, can occupy in society. In this paper, we apply Natural Language Processing (NLP) tools to analyze the gendered association of different attributes (e.g., traits, occupations, phys- ical characteristics) to measure how females and males are portrayed in children’s books.1 ∗All the authors contributed equally to this work. 1As gender is not a binary construct, we wished to include characters who identified as non-binary or gender fluid, in ad- dition to those who identified as female or male. In an analysis To measure portrayal, we use word embeddings, a prediction-based method for analyzing the se- mantic meaning of words in context using high- dimensional vectors. We supplement our analysis by training a model to detect individual sentences containing stereotypes in order to gain a deeper understanding of the implicit and explicit messages conveyed to children by the books they read. This awareness can, in turn, also help inform content- selection decisions of educators and caregivers. Messages about gender-specific abilities and roles may influence children’s beliefs and career paths (Leslie et al., 2015; Riley, 2017; Bian et al., 2017, 2018). Gender representation in children’s content has traditionally been measured by manual content analysis, in which one or multiple annota- tors slowly read through the text of written content to classify the messages within one or multiple …

Content truncated to 500 words