Google Books Ngram Datasets
How often every word and phrase appeared in millions of scanned books, year by year.
Counts of one- to five-word sequences in the Google Books scanning corpus, broken out by publication year and by language corpus. Each row gives an n-gram, a year, how many times it appeared, and in how many books. It powers the Ngram Viewer and a whole field of 'culturomics' studies of word use over time.
Access
- Cost
- FreeFree download of compressed tab-separated files.
- License
- Open licenseCreative Commons Attribution 3.0 Unported
- Sign-up
- No sign-up
- How you get it
- Bulk downloadTSV (gzip)
What's in it
- One row is
- One row per n-gram per publication year
- Covers
- Books in English (with American and British subsets) and several other languages
- Years
- Centuries of publication years; the 2020 release runs through 2019
- Updated
- Irregular
- Size
- Terabytes uncompressed across all n-gram orders; released in 2009, 2012, and 2020 versions
Quality
Enormous and consistently built, but the book sample changes over time and OCR errors are common in older years.
- Scientific and academic books grow as a share of the corpus over the 20th century
- OCR errors, such as long-s read as f, distort pre-1800 counts
- Only n-grams appearing more than 40 times are included
Connects to
- Project Gutenberg: Full texts of public-domain books for checking what the n-gram counts summarize
- Corpus of Contemporary American English (COCA): A balanced, genre-tagged corpus for recent usage, where Ngrams has only book text
Questions it could answer
- When did 'the United States is' overtake 'the United States are' in American books?
- How fast do new technology words spread into print?
- Which words declined fastest in English books between 1950 and 2019?