> Studying individual authors would be a good start. Or even taking a random sample of the ngrams to manually verify their usage hasn't changed over time, plus

> Studying individual authors would be a good start. Or even taking a random sample of the ngrams to manually verify their usage hasn't changed over time, plus a random sample of texts to check that depressive sentiments weren't previously expressed in different language. It's a great start, though.