“So we threw out ‘New York’ and ‘Stockholm,’ and we threw out numbers like ‘1984’ and ‘1,000,000’, but we noted the usefulness of the topic Human Closeness (chapter 2), of the verbs ‘need’ and ‘want’ (chapter 5), and the contractions ‘I’m’ and ‘couldn’t’ (chapter 4). These significant features were brought to our attention thanks to machine-learning algorithms.”