each sentence is transformed into exactly one dot by a pretrained model — a neural network trained on enormous amounts of data.