Two student-led projects from the Rovira i Virgili University (URV) have launched open-source tools designed to quantify gender bias in digital media. By analyzing speaking time, conversational dynamics, and thematic roles in text, these tools provide empirical data on the systemic underrepresentation of women, currently estimated at just 23% in primary roles.
Quantifying Conversation Dynamics with EQUITalk
The Hive team, comprised of Sonia Oros, Gema Pérez, and María Mira, secured second place in the Women for Open AI challenge with their tool, EQUITalk. Developed at the Technical School of Engineering (ETSE), the software moves beyond simple word counts to audit the mechanics of dialogue. It tracks three specific metrics: total speaking time, conversational control, and the frequency of interruptions. By isolating these variables, the system identifies how power is distributed in recorded discussions, often revealing imbalances that remain invisible to the casual listener.
The significance of this approach lies in its ability to strip away subjective interpretation. When a conversation is analyzed through these indicators, the data often highlights a clear disparity in who sets the rhythm of a dialogue. The Hive team is scheduled to present these findings at the II Meeting of Women and Technology in Madrid on November 18.
Classifying Female Representation in Journalism
While EQUITalk focuses on audio, the team known as Las 3 neuronas—Sofia Boyko, Sofia Mirza, and Laura Lope—tackled the representation of women within press articles. Their tool utilizes artificial intelligence to categorize the roles assigned to women in text, moving past a binary “present or absent” count. The system sorts female subjects into four distinct categories: leaders or experts, secondary figures, companions, or victims.
The results from their analysis indicate a significant glass ceiling in media coverage: women occupy a primary role as leaders or experts in only 23% of the articles examined. This data provides a stark look at how language and framing can relegate women to secondary status even when they are present in a story.
Technical Requirements for Open AI Initiatives
The Women for Open AI challenge, which included the collaboration of institutions such as the Polytechnic University of Catalonia and OVHcloud, sought to develop open-source artificial intelligence tools to analyze the presence and prominence of women in digital content.
Tools Measure Audio Bias and Audit Journalistic Texts to Reveal Low Representation
- EQUITalk: Measures audio-based bias, specifically focusing on interruptions and control of dialogue.
- Las 3 neuronas: Audits journalistic text to distinguish between expert roles and secondary or victimized portrayals.
- Core Finding: Data confirms that women appear in main roles in only 23% of analyzed press articles.
- Next Steps: The Hive team will present their findings at the upcoming II Meeting of Women and Technology in Madrid.
The Persistence of Algorithmic Bias
Despite these advancements, the broader question of how these tools will be integrated into existing media workflows remains open.
The data confirms that simply increasing the number of women mentioned in articles is insufficient if their roles remain secondary.
