Language Models Recover Human Word-Order Biases from Training Data That Deliberately Withholds the Evidence

Language Models Recover Human Word-Order Biases from Training Data That Deliberately Withholds the Evidence | BedrockNews