Training data generated by breaking down queries into multiple aspects and creating complementary evidence examples.