NYU's crowdsourced questions probe extent of language model bias

A study finds that biases are more wide-ranging and pervasive in popular natural language processing models than originally thought.