Common language modeling metric that measures how well a model predicts a sequence.