GOSt-MT: A Knowledge Graph for Occupation-related Gender Biases in Machine Translation

Machine translation systems reinforce harmful gender stereotypes in occupational contexts: The paper demonstrates that MT systems frequently associate specific occupations with particular genders, perpetuating traditional labour market stereotypes through automated translations that can influence how people perceive career opportunities.

Gender bias in translation is measurable and systematic: By creating the GOSt-MT Knowledge Graph, researchers developed a structured tool that can identify and track occupation-related gender biases across multiple languages (English, French, and Greek), showing that these biases are not random but follow detectable patterns.

Real-world labour data reveals misalignment with MT systems: The research integrates actual gender statistics from labour markets with data from MT training corpora, revealing gaps between how occupations are actually distributed by gender in real workplaces versus how they are represented in machine translation outputs.

Addressing MT gender bias requires systematic intervention: The paper suggests that understanding the specific areas where gender stereotypes persist in translation systems is a necessary first step toward building more equitable AI tools, implying that bias reduction requires targeted, evidence-based approaches rather than generic solutions.

Gender bias in translation has real societal impact: By focusing on the labour domain specifically, the research highlights how biased translations don't just produce inaccurate outputs—they can actively shape perceptions of which careers are appropriate for different genders, with potential consequences for career guidance, hiring, and opportunity access.