How Can We Accelerate Progress Towards Human-like Linguistic Generalization?