Biases We Research By

Biases We Research By is a tutorial on detecting and mitigating understudied forms of bias in NLP research. Its aim is to address existing gaps by providing a comprehensive overview of approaches for conceptualizing and uncovering bias, integrating insights from social sciences, actionable methodologies, and perspectives from data labelers.

The tutorial focuses on three main objectives:

• Understanding bias in the social sciences. We introduce foundational theories such as implicit and explicit bias, the stereotype content model, and intersectionality, showing how cognitive biases emerge and influence technological systems.
• Reducing underrepresentation in NLP. We present methods to detect and mitigate the underrepresentation of vulnerable communities across the NLP pipeline, from data selection to model development.
• Highlighting the role of human labor. We examine the often invisible contributions of data labelers and annotators, and how structural inequalities and labor conditions introduce additional layers of bias.

Detailed Program

TBD