Gender stereotypes embedded in natural language are stronger in more economically developed and individualistic countries

Abstract

Gender stereotypes contribute to gender imbalances, and analyzing their variations across countries is important for understanding and mitigating gender inequalities. However, measuring stereotypes is difficult, particularly in a cross-cultural context. Word embeddings are a recent useful tool in natural language processing permitting to measure the collective gender stereotypes embedded in a society. In this work, we used word embedding models pre-trained on large text corpora from more than 70 different countries to examine how gender stereotypes vary across countries. We considered stereotypes associating men with career and women with family as well as those associating men with math or science and women with arts or liberal arts. Relying on two different sources (Wikipedia and Common Crawl), we found that these gender stereotypes are all significantly more pronounced in the text corpora of more economically developed and more individualistic countries. Our analysis suggests that more economically developed countries, while being more gender equal along several dimensions, also have stronger gender stereotypes. Public policy aiming at mitigating gender imbalances in these countries should take this feature into account. Besides, our analysis sheds light on the “gender equality paradox,” i.e. on the fact that gender imbalances in a large number of domains are paradoxically stronger in more developed/gender equal/individualistic countries.

The Fill-Mask Association Test (FMAT): Measuring propositions in natural language.

“…By 2024, over 10,000 BERT model variants, among which over 1,000 were trained on English corpora, have been openly available on Hugging Face at https://huggingface.co/models?pipeli ne_tag=fill-mask. The vast diversity of these BERT models (covering more than 250 languages) offers an unprecedented opportunity to apply the FMAT to study psychology across multiple languages and cultures, together with other societal variables such as linguistic features and economic development (for similar work using word embeddings, see DeFranza et al, 2020;Napp, 2023). Another promising direction is to explore how modern LLMs can measure individual differences and analyze texts produced by specific samples (e.g., customer reviews), different geographical regions (e.g., states/provinces), and underrepresented social groups (e.g., ethnic minorities).…”

“…Using the WEAT, Caliskan et al (2017) replicated a spectrum of classic findings in social psychology originally obtained with the IAT (Greenwald et al, 1998), including attitudes (e.g., toward flowers vs. insects), social biases (e.g., toward European vs. African Americans), and stereotypes (e.g., the gender stereotype associating men with career and women with family); furthermore, they captured factual associations that can predict real gender distributions of occupations and first names. Since then, a rapidly growing number of studies have used the WEAT to assess social biases and stereotypes (e.g., Bailey et al, 2022; Charlesworth et al, 2022; DeFranza et al, 2020; Napp, 2023) and even to track changes in stereotypes (e.g., Garg et al, 2018) and cultural–psychological associations (e.g., Bao et al, 2022) using decade-specific word embeddings.…”