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

Jiao Sun, Nanyun Peng

Introduction

Wikipedia serves as a primary global knowledge repository, priding itself on neutrality and open collaboration. However, recent research suggests that the structural organization of its biographies may harbor systemic gender biases. While traditional natural language processing (NLP) research often focuses on word co-occurrences, this study shifts the focus to events—discrete occurrences involving specific participants, times, and locations. By analyzing how events are categorized in the biographies of men and women, researchers have uncovered a narrative disparity: women’s professional achievements are frequently framed alongside domestic milestones, whereas men’s professional sections remain strictly focused on career-related events.

Methodology: A Calibrated Event-Centric Approach

To quantify these biases, the researchers developed a multi-stage analytical pipeline. First, they utilized a state-of-the-art event detection model trained on the TB-Dense dataset to identify "trigger words" (typically verbs) that signify specific life events.

A significant challenge in this type of research is distinguishing between bias in the source data (Wikipedia) and bias inherent in the AI model itself. To address this, the authors introduced Strategic Calibration. They created a synthetic dataset by swapping names and pronouns in real sentences (e.g., changing "Mike" to "Emily"). By measuring how the model's performance shifted on these gender-swapped pairs, they could calibrate their findings to ensure they were measuring actual Wikipedia content bias rather than algorithmic error.

Finally, the team employed the Word Embedding Association Test (WEAT) to determine if extracted events were mathematically closer to male or female attributes within vector spaces like GloVe embeddings.

The Intermingling Phenomenon

The study's most significant finding is the "intermingling" of personal and professional life in female biographies. In Wikipedia's "Career" sections, women’s narratives are often porous, allowing events like marriage, divorce, and childbirth to appear alongside professional milestones. In contrast, these same events in men’s biographies are almost exclusively relegated to "Personal Life" sections.

Table 1: Qualitative comparison of sectioning for Loretta Young and Grant Withers.

As illustrated in the qualitative example above, the same event—the marriage and subsequent annulment between Loretta Young and Grant Withers—is categorized differently. For Young, it appears under "Career," while for Withers, it is placed under "Personal Life." This structural choice implicitly suggests that a woman's domestic status is a defining part of her professional identity, a standard not applied to her male counterparts.

Quantitative Disparities in Event Distribution

The researchers curated a "Wikipedia Celebrity Corpus" featuring over 10,000 celebrities across eight occupations. Quantitative analysis revealed stark differences in the types of events prioritized in the "Career" sections of different genders.

Figure 1: Distribution of top events by gender for Writers and Actors/Actresses.

For female writers, the most frequent events in career sections included "divorce," "marriage," and "wedding." Conversely, male writers’ sections were dominated by events like "argue," "election," "protest," and "shoot." This pattern persists across occupations; even in "Personal Life" sections, men’s narratives often include "achievement-oriented" events like winning awards, while women’s sections remain focused on domesticity. WEAT scores confirmed this, showing that Career sections for women are statistically "closer" to domestic concepts than those of men across most studied professions.

Conclusion and Future Directions

This research highlights how narrative choices on Wikipedia reinforce traditional gender stereotypes. By using events as a high-signal lens, the authors demonstrated that bias is not just about what is said, but where and how it is structured.

While the study is currently limited to binary gender classifications and English-language Wikipedia, it provides a robust framework for future audits of knowledge platforms. Future work may involve expanding this analysis to non-binary identities and other languages, as well as developing real-time tools to help editors identify and mitigate structural biases as they write.

Summary updated

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