How to use the ground truth answers from the previous time step for the decoders

Question

I found this on an article here:

Decoder takes the hidden state of the last Encoder RNN cell as the initial state of its first RNN cell along with the t