What is Google Ngrams? 8 Ways to Track Word History and Cultural Trends

Google Books Ngram Viewer is a free research search engine that graphs the historical frequency of words and phrases across millions of digitized books printed between 1500 and 2019 (with updated corpora extending through 2024). Launched in 2010 by Google and Harvard researchers, the tool allows you to measure how language, ideas, and cultural references change over centuries.

Instead of counting how many total books contain a word, the Ngram Viewer calculates the percentage of total words in a given year that match your search. If you search for terms like "nursery school", "kindergarten", or "child care", the platform plots the exact usage trajectories across American English, British English, French, German, Spanish, Russian, Hebrew, or Chinese literature.

Click to enlarge

1. What an N-Gram Is: From 1-Grams to 7-Grams

In computational linguistics, an n-gram is a contiguous sequence of n items from a given sample of text. In Google Ngrams, an item is a word or token.

• 1-gram (Unigram): A single word, such as kindergarten, radio, or algorithm.
• 2-gram (Bigram): A two-word phrase, such as child care, machine learning, or nursery school.
• 3-gram (Trigram): A three-word sequence, such as University of Oxford or artificial neural network.
• 4-gram to 7-gram: Longer multi-word phrases, such as the United States of America. In the July 2024 update, Google expanded search support up to 7 words (previously capped at 5).

Tokenization splits punctuation and contractions based on language rules. In English, contractions split into separate tokens: they're becomes the bigram they 're, and don't normalizes to do not. This ensures consistent indexing across modern and historical texts.

2. How the Y-Axis Measures Normalized Frequency Percentage

A common point of confusion is what the numbers on the vertical axis represent. The Y-axis does not display raw book counts or raw word totals. It displays a normalized percentage frequency.

Because vastly more books were published in 2000 than in 1800, raw volume counts would create misleading spikes. By dividing the number of times your phrase appears by the total number of words scanned in that specific year, Google Books Ngram Viewer normalizes the data. If the phrase child care shows a frequency of 0.0008% in 1990, it means 8 out of every 1,000,000 bigrams printed in English in 1990 were child care.

3. Smoothing Controls: Raw Spikes vs Moving Averages

The Smoothing dropdown at the top of the viewer controls moving average calculations. Yearly publishing data can contain sharp annual spikes, particularly in earlier centuries when fewer books were printed.

Click to enlarge

• Smoothing = 0: Displays raw, unfiltered annual data. Every single year shows its exact percentage, revealing one-off historical spikes or sudden war-time publications.
• Smoothing = 3 (Default): Calculates a 7-year moving average window. For the year 1950, the viewer averages the values from 1947 through 1953 (3 years before, the target year, and 3 years after, divided by 7).
• Smoothing = 5 to 10: Averages 11 to 21 years of data, creating broad multi-decade curves that reveal long-term cultural and linguistic trends.

4. Wildcard Search (`*`) for Phrase Patterns

When you place an asterisk (*) in your query, Google Ngrams finds and plots the top 10 most common word substitutions that complete the phrase.

Click to enlarge

For example, querying University of * displays separate trend lines for University of California, University of Chicago, University of Michigan, University of Illinois, and University of Pennsylvania. You can also place wildcards inside phrases, such as president * Lincoln or cure for *.

5. Inflection Search (`_INF`) to Group Word Forms

Words frequently appear across multiple grammatical inflections, including past tense, gerunds, and plurals. Appending _INF to any word automatically aggregates all morphological variants.

Clicking on the resulting graph allows you to expand individual curves for each tense or view the combined sum. Only 1 _INF modifier is supported per individual query.

6. Part-of-Speech Tagging (`_NOUN`, `_VERB`, `_ADJ`, `_ADV`)

Many English words function as both nouns and verbs. Searching tackle without tags combines fishing gear (noun) with resolving problems or football maneuvers (verb). Appending Part-of-Speech (POS) tags isolates the exact syntactic category.

• _NOUN: Isolates nouns (e.g., water_NOUN vs water_VERB)
• _VERB: Isolates verbs (e.g., run_VERB vs run_NOUN)
• _ADJ: Isolates adjectives (e.g., light_ADJ vs light_NOUN)
• _ADV: Isolates adverbs (e.g., well_ADV vs well_NOUN)
• _PRON, _DET, _ADP, _NUM, _CONJ, _PRT: Identifies pronouns, determiners, adpositions, numbers, conjunctions, and particles.
• _START_ and _END_: Matches sentence boundary positions (e.g., words that start or end sentences).

According to research by Lin et al. (ACL 2012), automatic POS tag accuracy in modern English books exceeds 95%. POS tagging is available in the 2012, 2019, and 2024 datasets (the 2009 dataset does not support syntactic tagging).

7. Ngram Arithmetic (+, -, /, *) to Track Cultural Shifts

Google Ngrams supports five mathematical operators that combine, subtract, or scale different terms in a single graph:

If your search term contains literal slashes or hyphens (such as the word and/or or well-meaning), enclose the phrase in square brackets like [and/or] to prevent the parser from executing division.

8. Corpora Datasets and the 40-Book Minimum Threshold

Google Ngrams offers access to several distinct literary corpora:

• English (eng): All digitized English books regardless of publication country.
• American English (eng_us): Books published specifically in the United States.
• British English (eng_gb): Books published specifically in Great Britain.
• English Fiction (eng_fiction): Books categorized as fiction by libraries and publishers.
• English One Million (eng_1m): A balanced sample of 1,000,000 English books from 1500 to 2008 with a maximum of 6,000 books per single year.
• International Corpora: French (fre), German (ger), Spanish (spa), Russian (rus), Italian (ita), Hebrew (heb), and Simplified Chinese (chi_sim).

To maintain dataset quality and prevent individual obscure typos or OCR anomalies from dominating results, Google enforces a 40-book minimum cutoff. If an n-gram appears in fewer than 40 separate books across the entire corpus, it is excluded from the index, returning a 0% flatline.

Frequently Asked Questions