"The Inuit have a hundred words for snow." You have probably heard it, and maybe passed it on. It is a lovely line: a faraway people, a white world, a language that slices experience finer than ours. Because it is so lovely, it became the textbook specimen of a discipline getting something wrong.
Linguistics is good at correcting itself, and the errors it has corrected share a pattern: the wrong step was almost always treating the language closest to hand as the default for all languages. The three stories here go wrong in three places — in the telling, in the sample, and in a value judgment.
The source is the introduction to the anthropologist Franz Boas's 1911 handbook of North American Indian languages. His point was modest: languages carve the world into words differently. English has a string of separate words for water — lake, river, brook, rain, dew, wave, foam — with nothing in them that says "water"; the Eskimo language he described had a few separate roots for snow: aput for snow on the ground, qana for falling snow, piqsirpoq for drifting snow, qimuqsuq for a snowdrift. Four, an ordinary example, on a par with the English water words.
In 1940 Whorf retold it in an article for a general audience, raised the number of kinds a little, and suggested an Eskimo would find a single catch-all word for snow almost unthinkable. Then textbooks, lectures and newspapers retold it, each layer copying the one before, and the number crept upward — seven, dozens, until a 1984 New York Times editorial said one hundred. In 1986 the anthropologist Laura Martin traced the chain link by link; in 1989 Pullum wrote a sharp little essay calling it the Great Eskimo Vocabulary Hoax.
The deeper problem is that "how many words" cannot be counted here at all. Inuit languages are polysynthetic: a root can take a long chain of suffixes, so a single word often does the work of a whole English sentence. Count that way and English snow, snowing, snowfall and snowed in become several "words" too, and the total can grow without limit.
Then comes the twist. In 2016 Regier, Carstensen and Kemp looked at nearly three hundred languages: the warmer the region, the more often a language uses one word for both snow and ice. Where people need the distinction, they draw it; where they don't, the words merge. The myth was wrong in its exaggeration and its arithmetic, but the mild version — languages divide more finely where precise talk is needed — has cross-linguistic support, though the evidence is correlational. Skiers have powder, crust and slush for the same reason.
The snow story went wrong in the telling. The second mistake hides much deeper: in the sample. In 2022 Kidd and Garcia went through forty-five years of every paper in four major child-language journals. The research covered about 103 languages, roughly 1.5% of the world's. English alone accounted for 54% of the papers, other Indo-European languages for another 30%, and the remaining thousands of languages for 16% between them.
A narrow sample is not automatically wrong — unless the one you picked happens to be odd. In spelling, English is odd. Take though, through, tough, cough: the same four letters, four pronunciations. In 2003 Seymour, Aro and Erskine compared first-graders across 13 European writing systems. Children reading Finnish, Greek, German, Italian and Spanish could mostly read familiar words and made-up words accurately and fluently before the end of their first school year; French, Portuguese and Danish were slower; English was slowest, developing more than twice as slowly.
In 2008 the psychologist David Share wrote a long review whose title calls English an "outlier" orthography. His argument: because most reading science was done in English, English's peculiar difficulties were mistaken for the difficulties of reading itself. The field fixated on reading aloud accurately, while silent reading, comprehension and fluency got less attention, and even models of the stages of reading development took on an English shape. Evidence strength: the cross-orthography comparisons have been replicated by several studies; Share's critique is a review-level argument, highly influential, and exactly which theories it distorted, and how much, is still debated.
One distinction matters: an outlier spelling system does not make English a worse language. Spelling is a convention, not the language itself, and English spelling keeps word shapes stable — the silent g in sign is a clue that links it to signature, so related words look like family at a glance. The price is a slower start for beginners; the payoff is quicker recognition of word families for fluent readers.
The third mistake is the ugliest, and the most thoroughly corrected. Nineteenth-century European scholars sorted languages by how their words are built. Isolating languages, where a word is basically one unchanging unit — Chinese is the standard example. Agglutinating languages, where clear-cut affixes are stuck onto a root one after another — Turkish. Inflecting languages, where a single ending packs several meanings at once — Latin, Greek, Sanskrit. The sorting is still useful. The trouble was the next step: August Schleicher and others arranged the three types into an evolutionary ladder, with isolating as the most primitive stage and inflection as the summit. Chinese went to the bottom; at the top stood exactly the languages these scholars had memorized grammar tables for at school.
Two things dismantled the ladder. One was fieldwork: Boas and his students, recording Native American languages, found societies Europeans had called "primitive" speaking languages with enormously intricate word forms. The other was direction. English itself moved from the case endings of Old English toward letting word order carry meaning, with few word-form changes; by the ladder's logic that is "decline", yet nobody thinks Shakespeare's English inferior to Old English. Chinese, meanwhile, keeps growing new grammatical parts: in 吃了饭 (chī le fàn, "ate the meal"), the completion marker 了 (le) was worn down from the verb 了 (liǎo, "to finish") — the path traced in the topic on how language changes like a river.
In 1921 Sapir wrote a line quoted ever since: when it comes to linguistic form, Plato walks with the Macedonian swineherd, Confucius with the head-hunting savage of Assam. Grammatical complexity, in other words, has nothing to do with level of civilization. There is a curious modern echo. In 2010 Lupyan and Dale surveyed over two thousand languages and found that languages with more speakers, and more adult learners, tend to have simpler word inflection — the opposite direction from the old ladder. It is only a correlation, its causes are argued over, and it ranks nothing.
The myth is one sentence with a picture; the debunking first has to explain how words are counted. Perhaps a correction only spreads if it carries an equally memorable counter-picture — "English has seven words for water, too."
Candidates: sentence processing (English word order is unusually rigid), children's early vocabulary (English-learning children lean heavily on nouns, while children learning some other languages produce more early verbs), and speech perception (English has an unusually large vowel inventory).
At least half of it is empirical: no human community has been found whose language cannot express what that community needs to express. How to measure "complexity" in the first place is still argued over.
The difference is scale and visibility: a scholar's preferences sit in a book where you can see them; a model's sit in the mix of its training data, which its users may never know.