Intersectional Male-Centric and White-Centric Biases in Collective Concepts

Area of Science:

• Social Psychology
• Natural Language Processing
• Sociolinguistics

Background:

• Fundamental concepts like 'person,' 'woman,' and 'man' should be equitable across genders and ethnicities.
• However, these concepts may implicitly favor certain demographic groups.
• Androcentrism theories suggest a male-centric bias in societal language and conceptualization.

Purpose of the Study:

• To investigate linguistic biases in the concepts of 'person,' 'woman,' and 'man.'
• To test the hypotheses that 'person' is more associated with men and 'woman' is more associated with women (i.e., women are more gendered).
• To examine how these biases vary across different racial and ethnic groups.

Main Methods:

• Utilized natural language processing (NLP) tools, specifically word embeddings.
• Analyzed a large-scale corpus of linguistic data (Common Crawl) representing millions of individuals' language output.
• Quantified associations between core concepts and demographic identifiers.

Main Results:

• Confirmed a significant linguistic bias where 'person' is more associated with 'man' than 'woman' (person = man).
• Found that 'woman' is more associated with women than 'man' is with men, indicating women are perceived as more gendered (gender = woman).
• Observed that these biases were more pronounced for Hispanic and White women and men compared to Asian individuals, and also revealed a general bias favoring White individuals across concepts.

Conclusions:

• Western societal language demonstrates a bias prioritizing men and White individuals as the default 'person.'
• Women are often 'othered' as 'people with gender,' highlighting a gendered conceptualization.
• These linguistic biases have significant implications for policy and decision-making, potentially affecting equity.