How do I search for hyphenated phrases in Google Ngrams?

Question

If you type a hyphenated phrase into Google Ngrams, it inserts spaces around the hyphen and then searches as normal. However, its help page clearly states:

Why am I not seeing the results I expect?

[snip]

• Your phrase has a comma, plus sign, hyphen, asterisk, colon, or forward slash in it. Those have special meanings to the Ngram Viewer; see Advanced Usage. Try enclosing the phrase in square brackets (although this won't help with commas).

Also, when describing how to use the hyphen for subtraction, it says this:

subtracts the expression on the right from the expression on the left, giving you a way to measure one ngram relative to another. Because users often want to search for hyphenated phrases, put spaces on either side of the - sign [in order to subtract phrases instead of searching for a hyphenated phrase].

This seemingly contradictory behavior tied up this answerer on EL&U in so many knots that they eventually wrote:

My conclusion, after multiple highly unsatisfactory experiences with Ngram's handling of words and phrases containing hyphens, is that Ngram programmers never got around to making the program's "search for hyphenated phrases" feature functional. Instead, they seem to have instructed Ngram to transform requests for plots of hyphenated words into requests for plots of advanced Ngram comparisons of the component words or phrases within the originally specified hyphenated word or phrase. The upshot of all this is that I still haven't been able to find a way to get Ngram to generate meaningful line graphs of hyphenated words or phrases of the type that Kevin wanted to create.

They tried, among other things, using square brackets as the first quote suggests, to no avail (it came up with no results).

• Does Ngrams behave "correctly" if you put a hyphenated phrase into its search box?
• If not, how do you search for hyphenated phrases?

While researching this question, I found How to search for asterisks and other special characters in Google NGrams, but that does not answer my question because it does not specifically discuss the behavior of automatically inserting spaces around a hyphen and does not address what to do if square brackets give you no results (for a query which should obviously return results)

Answer 1

I was just using Ngrams to compare the usage of "double take", "double-take" and "doubletake". As far as I can tell, you can just type the hyphenated phrase as normal. Alternatively, you must follow both instructions, putting square brackets around the phrase and spaces around the hyphen.

---

Here's the experiments that bring me to this conclusion.

With the plain query double take,double-take,doubletake, it gives me a notification, but the search seems to have worked correctly.

ⓘ Replaced double-take with [double - take] to match how we processed the books.

With square brackets, double take,[double-take],doubletake, it gives me a notification and two warnings; and as the first warning suggests, the hyphenated form is not searched for.

⚠ Ngrams not found: double-take, [double-take]

ⓘ The Ngram Viewer is case sensitive. Check your capitalization.

⚠ The characters +, -, *, / require parentheses to be interpreted as a composition.

With spaces around the hyphen, double take,double - take,doubletake, it duplicates the hyphen! Naturally, there are very few results for "double -- take".

ⓘ Replaced double - take with double -- take to match how we processed the books.

And finally, with spaces and square brackets, double take,[double - take],doubletake, it gives no notifications or warnings. But the results are the same as the first search.

So while it doesn't seem that searching hyphenated phrases was left unimplemented, I reckon the implementers didn't talk to the documentation writers.