>>6168436>googl ngram only an approximate indicative guide it can lead to flawed interpretations etcYou are correct Reptoid QM amusingly a good example of the phenomenon you described is just if you search the ngram word for "gay" lol obviously in the 1800s gay was a very common descriptive word for happy everyone was gay all the time gaily prancing and stroking and caressing each other with gay smiles on their gay faces but later on that word fell out of fashion only to be repurposed so these words can have a U type shape falling into decline and then being repurposed in meaning.
The conclusion with all these computational methods, don't just take the sophisticated looking graph at face value because it seems scientific complicated looking oohhh the algorithm etc you still have to ponder the flaws and robustness of the empirical methodology the meaning / reliability of the data inputs, sources and potential anomalies or overlooked bias in the methodology etc. You can make a chart show anything you want, this is vampire sorcery
When you mentioned
>"if people in the 19th century got really into writing about kinga and knights..." this is exactly what happened, I mentioned before IVANHOE by Sir Walter Scott nobody ever mentions this novel anymore but it was MASSIVELY influential to the extent that people named city streets and entire districts after all the characters, Tolkien's medieval department probably only originally existed and became established because Ivanhoe and Sir Walter Scott provoked an enduring revival in Victorian medievalism study, if you read 19th century authors from Dickens Balzac Tolstoy etc they are all evaluating their work in the context of his historical novels like Ivanhoe etc. so indeed you are correct there was a Victorian medievalism revival which lasted for decades and decades, it even extends into Celtic revival (early 20th century national myth stuff like WB Yeats his tarot occult mythological Irish legend poetry etc) I think in aggregate the ngrams probably do accommodate the broad trend shifts fairly accurately for closely associated words but from a statistical standpoint the greater issue is that the sample size is not constant over time (ie more books published as the decades roll on) there is not a time-invariant constant probabilistic occurrence "pool of words" to draw from etc the mean, variance etc estimators all would be biased I don't know how googl ngram calculates this or if they have some clever time varying statistical adjustment etc. A way to see how the ngrams capture poorly the very early data is to search for some really archaic words eg I tried "sithence" in English fiction (I have actually used this word in my quests lol, it is in a lot of Shakespeare, or Spenser) searching for these genuinely archaic ancient words the ngrams are just spikes and erratic nonsense, the sample size is probably too small to draw any meaningful inference
