Sign in Subscribe
Machine learning

Detecting Biased Language: A Hybrid System for Promoting Fairness and Inclusive Language

Strategic Insights

— 2 min read

Research by Wael Khreich and Jad Doughman at the American University of Beirut has led to the development of a novel data-centric system for detecting gender biases in language. This system, called Genderly, aims to identify and address the personal and societal consequences of biased language, including exclusion, labor force participation, stereotypes, and social inequalities. By leveraging machine learning and natural language processing techniques, Genderly can detect biases in linguistic levels, including implicit biases within nuanced language elements.

Key Takeaways:

  • The research highlighted the importance of detecting biased language, which has personal and societal consequences, including exclusion, labor force participation, and social inequalities.
  • The current machine learning and natural language processing solutions for gender bias face limitations, such as insufficient definitions, taxonomies, and annotated data, often relying on simplistic statistical analyses.
  • The proposed system, Genderly, is a hybrid and modular data-centric system for detecting gender biases in English across linguistic levels.
  • Genderly identifies gender-specific terms, analyzes idiomatic expressions, phrases, and metaphors perpetuating biases, and captures implicit biases within nuanced language elements like sarcasm and figurative speech.
  • The system includes a user feedback loop for ongoing improvement and has been experimentally evaluated using diverse preprocessing, feature engineering, and hyperparameter optimization methods.
  • The findings reveal that optimized ML models detect surface-level biases effectively, while LLMs excel in identifying biases embedded in semantics, with benevolent sexism being particularly challenging.
  • The research concluded that Genderly promotes fairness and inclusive language by helping users recognize biases, and aims for equitable language use and a more inclusive society.
  • The study was funded by the University Research Board, American University of Beirut, and the research was published in the journal Complex & Intelligent Systems.

Statistics:

  • 11(7):1-25 - The issue number and journal article reference for the published research in Complex & Intelligent Systems.
  • Genderly: a data-centric gender bias detection system - The title of the journal article published in Complex & Intelligent Systems.
  • 2025 - The year in which the research was published.
  • p 158 - The page number of the news report that covers the research findings.
  • 2 authors - The number of authors mentioned in the research, including Wael Khreich and Jad Doughman.

Sources:

  • NewsRx. Findings from American University of Beirut Advance Knowledge in Intelligent Systems (Genderly: a data-centric gender bias detection system). Robotics & Machine Learning. June 16, 2025; p 158.
  • Wael Khreich and Jad Doughman. Genderly: a data-centric gender bias detection system. Complex & Intelligent Systems, 2025,11(7):1-25.
  • https://doi-org.sdpl.idm.oclc.org/10.1007/s40747-025-01859-z - The DOI link to the free version of the journal article in Complex & Intelligent Systems.

Read more