RedditBias: A Real-World Resource for Bias Evaluation and Debiasing of Conversational Language Models

Soumya Barikeri, Anne Lauscher, Ivan Vulić +1

Text representation models are prone to exhibit a range of societal biases, reflecting the non-controlled and biased nature of the underlying pretraining data, which consequently l…