Describe-and-Dissect: Interpreting Neurons in Vision Networks with Language Models