I currently have the same issue (and am also doing the MIT deeplearning course). The reason for the issue is (I think) because there's been a change to the Keras API. I am not sure what the alternative keyword arg would be, and I haven't found anything yet.

https://www.tensorflow.org/api_docs/python/tf/keras/layers/Embedding

The embedding layer doesn't have anything at all relating to batch_input_shape. Unfortunately, I have very little knowledge on the Keras API and deep learning in general. I will keep posted if I find a solution.