The Colonial Impulse of Natural Language Processing - An Audit of Bengali Sentiment Analysis Tools

Explore the potential biases in Bengali sentiment analysis tools through this conference talk examining the perpetuation of colonial values in sociotechnical systems. Delve into an algorithmic audit focusing on gender, religion, and nationality - identity categories heavily impacted by colonialism in Bengali communities. Discover how these tools exhibit inconsistencies and biases when analyzing different identity expressions, despite similar semantic content. Gain insights into the implications of downstream bias in sentiment analysis tools and their connection to colonially shaped sociocultural structures. Learn about the importance of addressing these biases in natural language processing to avoid perpetuating colonial impulses in modern technology.