why is the input dim (hidden_dim x hidden_dim) not (output_dim x output_dim)

isn't the input a word as well?