When you want to process a sequence of vectors while incorporating information from a second sequence, you use attention. Attention means you create a weighted sum of all vectors in the second sequence for each vector in the first sequence. You basically add a customized summary vector of the second sequence to each vector of the first.