DMI Webinar "Based on Billions of Words on the Internet, People = Men"

Image of DMI Webinar "Based on Billions of Words on the Internet, People = Men"
Zoom Meeting



"Based on Billions of Words on the Internet, People = Men"

24 May 2021, 17:30 - 18:30 CEST




Is the concept of “people” centered on men? To test this, we asked whether words for people (e.g., humanity) are used more like words for men (e.g., male) as opposed to those for women (e.g., female). By measuring the similarity between their static word embeddings, we found that man-denoting words are generally more similar to human-denoting words than to woman-denoting words. We also found a consistent asymmetry for gender stereotypes, whereby word embeddings for men are similar to a wide range of traits and actions—even those stereotypic of women. In contrast, word embeddings for women are only similar to traits and actions that are specifically more stereotypic of women than of men (e.g., communal vs. agentic). Because static word embeddings' usage in context reflects word meanings, the fact that words for people, traits, and actions are more similar to words for men than to words for women indicates that, at least in English, the concept of “people” generally centers on men.




Adina Williams is a Research Scientist at Facebook AI Research in NYC. Previously, she earned her PhD at New York University in the Department of Linguistics, where she investigated the brain basis of syntactic and semantic processing. Her main research goal is to strengthen connections between linguistics, cognitive science, and natural language processing. Towards that end, she brings insights about human language to bear on training, evaluating, and debiasing ML-based NLP systems, and applies tools from NLP to uncover new facts about human language.


The talks will be held online. If you would like to participate, please fill this form.


For more information, write to

Remote video URL
Based on Billions of Words on the Internet , People = Men