Click to add attention to a translation model. Instead of one summary of the French sentence, the decoder can look at every word directly, at every step.
Without attention the decoder sees only one summary vector.
