Template:R:GNV/documentation

Use this template to link to, showing time-dependent graph of word form or spelling frequencies.

Parameters
The following parameters are used by this template:
 * 1
 * The terms or terms to be graphed.


 * 2
 * A display override for the term or terms.


 * corpus
 * The index of the corpus to be shown, see available corpora. Defaults to 26 i.e. English.


 * startyear, start
 * The year to begin the graph at. Defaults to 1800.


 * endyear, end
 * The year to end the graph at. Defaults to the newest available (see available corpora).


 * caseinsensitive
 * Whether to search with case insensitivity on or not. Any value taken to mean yes. Defaults to no.

Examples
Here are some:



























Available corpora
A list (with descriptions) is also available at https://books.google.com/ngrams/info.

Limitations
Google Ngram Viewer suffers from some limitations: 1) scanning errors (scannos); 2) corpus increasingly biased toward academic publications with passage of time; 3) each book has the same weight regardless of popularity; 4) wrong assignment of year of publication. Some of the problems are covered below. The scanno problem does not seem to completely invalidate the results, especially for English and longer words. The severity of the problems depends on what we want to measure, whether cultural change over time or relative frequencies of word forms.

Bias toward academic publication
reveals the problem: capitalized Figure rises to the top during 20th century, suggestive of use in captions of academic literature. When we restrict the corpus to English Fiction, the problem disappears:.

Long s vs. f
shows the problem: there is no way there were so many instances of "fuck" before 1800; rather, these are likely scannos of "suck" caused by long s (ſ). On the other hand, this problem does not occur after 1820.

Dropping hyphens
and show the problem: scanning sometimes drops the hyphen. There is no way there are so many occurrences of "antiAmerican" and the Google Books search confirms that. Other examples:,.

Some hyphens are dropped when used within an unbroken line, other are dropped at a line break, which is ambiguous as for the presence of hyphen.

Dropping spaces
and show the problem: the space was dropped and the result is as common as the legitimate nonchocolate. On the other hand, shows this happens relatively rarely.

Joining different columns
shows the scanning problem: there are very few occurrences of "misargument" and some of the found items result from joining parts from different columns in multi-column publications. This one example does not make it into GNV statistics, though. It is unclear this could significantly impact frequencies of common words, though.

Changes in capitalization
There is no reason to think there are spurious changes in capitalization. looks plausible, unlike.

Links

 * The Pitfalls of Using Google Ngram to Study Language, 2015, wired.com
 * Guideline for improving the reliability of Google Ngram studies: Evidence from religious terms, 2019
 * Google Ngram Viewer

Hyphens
As of Oct 2022:
 * To search for hyphenated phrases, do one of the following:
 * Hope that GNV will continue working like before, e.g.
 * Make sure to enter spaces around the hyphen and use [] around the term, e.g..
 * Google will often pick hyphenated phrase as non-hyphenated, whether in the middle of the text or at a line break.
 * Thus, comparisons like show results much more favorable to exteacher than reality. One needs to check in Google Books what is actually found on the scanned pages. It still shows convincingly exteacher is rare; it is in fact much rarer.
 * shows too many hits for antiAmerican. A similar result is for.
 * You can plot frequency ratio: nonstandard/[non - standard]:.