I’m grateful to fellow blogger Garden History Girl for alerting me to a splendid new way of wasting hours on the internet, while claiming to be researching important trends in landscape history.
A couple of days ago Google launched their Ngram Viewer. It is an oddly-uninspiring name for a nifty gadget that lets you trace the appearance of particular words or phrases in books over time. Garden History Girl writes amusingly of the 18th century fad for shrubbery, and uses the Ngram to show how for a while the word shrubbery was actually more popular than plain old shrubs.
Playing around with the gadget this morning, I have traced the dramatic rise and slow fall of the usage of picturesque and sublime, concepts painfully fashionable of course in the late eighteenth century; confirmed that the term jardin à la francaise is a nineteenth-century construct which would have had no meaning for Le Nôtre and his contemporaries; and discovered that parc de la Villette is the most discussed of the three great new parks created in Paris in the late twentieth century.
picturesque, sublime, in English 1600-2008
It is quickly clear that the gadget has all sorts of shortcomings: it is case and accent sensitive, so Jardin des Tuileries for instance will not find references to jardin des Tuileries, nor will Pere Lachaise find mentions of Père Lachaise; any pre-1800 results are pretty suspect, given the frequent lack of publication dates, non-uniform spelling and poor printing quality of many early books; and there is no way of verifying context, so mentions of Dan Kiley could equally refer to the master American landscape designer or to his pop-psychologist namesake.
Unlike Google Maps, there is no organised way of disseminating the results, except apparently as a ‘tweet’, and many of the graphs already appearing on blogs and online newspapers are blurry and difficult to read. It is also all too easy to search for the obvious, as I did, and get unsurprising results.
But the Ngram is great fun, and endlessly addictive. Its real value will come when unexpected patterns lead to fresh understandings and new avenues for research.
Postscript: A quick scout round the internet shows that I am far from the only one playing with Google’s new text miner. It is generating a lot of pretty unthinking coverage: see for example the gleeful US headline ‘Pants up, trousers down’ (non-American anglophones will quickly spot an omission in the writer’s analysis of the results).
But thoughtful and erudite posts are also appearing. This one explains the technical shortcomings of the Ngram, and argues that the gadget needs perfecting before it will be of any use. While this accepts its imperfections and welcomes the Ngram as another tool that can help us explore the past.
I remain on the fence. Try something seemingly simple like plotting the popularity of landscape designers over time. You can show that in British English in the twentieth century, Le Nôtre appeared more than Frederick Law Olmsted, and that poor old unfashionable Capability Brown barely merited a mention:
'Capability' Brown, Frederick Law Olmstead, Andre Le Notre, in British English, 1900-2000
Or you can show the opposite, with Brown easily more popular than Olmsted, and Le Nôtre regulated to a distant third:
Capability Brown, Frederick Law Olmsted, André Le Nôtre, in British English 1900-2000
If you look closely, you’ll see that the results are completely skewed by spelling variations. ‘Capability’ Brown yields no results at all in the first graph (the gadget doesn’t seem to like the quotation marks, even though they are commonly used). Without the quotation marks there are suddenly lots of results (but there is no obvious way to combine them with those for his real name Lancelot [sometimes Launcelot] Brown). Olmstead in the first graph is a surprisingly frequent misspelling of Olmsted, the correct spelling of which produced the second set of results. Trickier still in the English canon is André Le Nôtre, whose name most commonly seems to appear in English books without any accent marks (at least according to Google’s OCR), producing the results in the first graph. Add both diacritics and you get the second graph. Originally spelled Le Nostre, his name is still sometimes rendered that way, although none of those results show up here.
So it is almost impossible to trace any meaningful trend in the appearance of their names in books over time. Hmmm…
I do find it a useful tool, but before about 1820 it gets really tricky to use right now because of the f->s thing. If you try graphing picturefque and fublime you’ll get some interesting results.
Thanks for your comment. You’re right of course about the so-called ‘long s’ – which OCRs usually read as an ‘f’. I had not realised what a difference it makes to the Ngram results: it certainly explains why the dramatic rise in use of the word ‘sublime’ occurs in the graph above almost a century after I had expected: plot ‘fublime’ as well, and you can see how the word actually took off in the early 18th century with Shaftesbury and Burke. It is another example of how easy it is with the Ngram to produce a professional-looking graph that shows gibberish!
I found your blog thoughtful and fascinating, by the way, and would recommend it to anyone interested in the Ngram.