Stimulus-to-Stimulus Learning in RNNs with Cortical Inductive Biases