Men Are Elected, Women Are Married: Events Gender Bias on Wikipedia

Jiao Sun, Nanyun Peng

Introduction

While Wikipedia serves as a global repository of knowledge, it is not immune to the implicit biases of its contributors. Traditional research into gender bias has often focused on word frequencies or sentiment analysis. However, this paper introduces a more structural lens: event-centric analysis. By examining specific occurrences—events—involving participants at certain times and places, the researchers uncover how the selection and placement of information can reinforce societal stereotypes, even when the language itself appears neutral.

Methodology

The researchers developed a multi-stage pipeline to quantify bias within Wikipedia biographies. The process began with Event Extraction, using state-of-the-art models to identify "trigger words" (typically verbs) signifying events. To determine which events were gender-distinct, they calculated Odds Ratios (OR) for male versus female biographies.

A critical technical contribution of this work is Strategic Calibration. To ensure that observed biases originated in the Wikipedia data rather than the NLP model itself, the authors created a synthetic dataset by swapping names and pronouns in real sentences (e.g., changing "he" to "she"). By measuring the model's performance on these balanced templates, they could "subtract" inherent model bias from the final results. Finally, they employed Word Embedding Association Test (WEAT) analysis to measure the strength of association between specific events and gendered attributes in a high-dimensional vector space.

The "Intermingling" Phenomenon

The study's most significant insight is the structural disparity in biography organization, termed the "intermingling" phenomenon. In biographies of women, personal life events—such as marriage, divorce, and birth—are frequently integrated into the "Career" section. In contrast, for men, these events are typically sequestered within a dedicated "Personal Life" section.

Comparison of Career vs. Personal Life event placement for Loretta Young and Grant Withers.

As shown in the example above, the same life event (a marriage and subsequent annulment) is categorized under "Career" for the female subject but under "Personal Life" for the male subject. This structural choice subtly suggests that a woman's professional identity is inextricably linked to her domestic status, whereas a man's professional achievements stand independently.

Key Results

The analysis of the Wikipedia Event Bias Corpus—comprising over 7,800 fragments from 10,412 celebrities across eight occupations—revealed systemic differences in event distribution.

In the "Career" sections of female writers and actresses, events like "marriage," "divorce," and "wedding" were among the most frequent. Conversely, male "Career" sections were dominated by professional or public-facing events such as "election," "protest," "arrest," and "war."

Frequency of top events in Career sections across different genders and occupations.

Quantitative WEAT scores confirmed these findings, showing strong negative associations for female-linked events and positive associations for male-linked events. Crucially, these biased event patterns were not rare outliers; they appeared within the top 10% of the most frequent events for each gender, indicating a widespread and systemic issue.

Conclusion and Future Work

The study demonstrates that events provide a high-signal lens for detecting bias, capturing the "skeleton" of societal stereotypes more effectively than raw text alone. The authors suggest that the next step is developing actionable tools for Wikipedia editors. Such tools could provide real-time feedback, flagging when a professional biography becomes overly focused on personal life events, thereby helping to maintain a truly neutral and professional point of view across all gender identities.

Summary updated

Abstract

Human activities can be seen as sequences of events, which are crucial to understanding societies. Disproportional event distribution for different demographic groups can manifest and amplify social stereotypes, and potentially jeopardize the ability of members in some groups to pursue certain goals. In this paper, we present the first event-centric study of gender biases in a Wikipedia corpus. To facilitate the study, we curate a corpus of career and personal life descriptions with demographic information consisting of 7,854 fragments from 10,412 celebrities. Then we detect events with a state-of-the-art event detection model, calibrate the results using strategically generated templates, and extract events that have asymmetric associations with genders. Our study discovers that the Wikipedia pages tend to intermingle personal life events with professional events for females but not for males, which calls for the awareness of the Wikipedia community to formalize guidelines and train the editors to mind the implicit biases that contributors carry. Our work also lays the foundation for future works on quantifying and discovering event biases at the corpus level.
BibTeX
@inproceedings{sun2021men,
  author = {Jiao Sun and Nanyun Peng},
  title = {{Men Are Elected, Women Are Married: Events Gender Bias on Wikipedia}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2021},
  doi = {10.18653/v1/2021.acl-short.45},
}
Figure