January 30, 2004

Sasha Aikhenvald on Inuit snow words: a clarification

Oh, dear. It had to happen. People are so convinced that language is all about words. The New Scientist's interview with Alexandra Aikhenvald about working with endangered languages, cited recently by Mark Liberman, even got assigned "For want of a word" as its headline -- the familiar nonsense about language being a question of how many words you've got. Aikhenvald (known as Sasha to her friends, i.e., just about everybody who's ever met her) has done most of her fascinating work on grammar (and some sociolinguistics), not lexicography. So faced with a question about a favorite difference between languages she picked evidentials (required sentence marking of the evidential basis for the statement made). But the interviewer, Adrian Barnett, knew about (and probably shares) the general public's lust for word lore, so of course he forced vocabulary into the conversation: "And what about different types of vocabulary?" And so it was that, knowing what was expected of her, Sasha dutifully commented on the Eskimoan languages:

The story about Inuit words for snow is completely wrong. That language group uses multiple suffixes, so you can derive not 50, but 150 words for snow.

Sasha speaks fast; sometimes too fast. I think I see what she might have meant, but what she said here (or what Barnett scribbled down in his notes, perhaps) is highly misleading at best: it actually suggests there is an answer to the perennial question, namely 150. Not so.

Here's a replacement answer that she could have given. It's a bit closer to the extremely complex truth (for which you should consult a proper Eskimologist; I have merely an interested onlooker's acquaintance with this topic, but I've done a little reading in widely available sources like the Comparative Eskimo Dictionary).

The story about Inuit (or Inuktitut, or Yup'ik, or more generally, Eskimo) words for snow is completely wrong. People say that speakers of these languages have 23, or 42, or 50, or 100 words for snow --- the numbers often seem to have been picked at random. The spread of the myth was tracked in a paper by Laura Martin (American Anthropologist 88 (1986), 418-423), and publicized more widely by a later humorous embroidering of the theme by G. K. Pullum (reprinted as chapter 19 of his 1991 book of essays The Great Eskimo Vocabulary Hoax). But the Eskimoan language group uses an extraordinary system of multiple, recursively addable derivational suffixes for word formation called postbases. The list of snow-referring roots to stick them on isn't that long: qani- for a snowflake, api- for snow considered as stuff lying on the ground and covering things up, a root meaning "slush", a root meaning "blizzard", a root meaning "drift", and a few others -- very roughly the same number of roots as in English. Nonetheless, the number of distinct words you can derive from them is not 50, or 150, or 1500, or a million, but simply unbounded. Only stamina sets a limit.

That does not mean there are huge numbers of unrelated basic terms for huge numbers of finely differentiated snow types. It means that the notion of fixing a number of snow words, or even a definition of what a word for snow would be, is meaningless for these languages. You could write down not just thousands but millions of words built from roots that refer to snow if you had the time. But they would all be derivatives of a fairly small number of roots. And you could write down just as many derivatives of any other root: fish, or coffee, or excrement.

And the derivatives wouldn't all be nouns. If you wanted to say "They were wandering around gathering up lots of stuff that looked like snowflakes" (or fish, or coffee), you could do that with one word, very roughly as follows. You would take the "snowflake" root qani- (or the "fish" root or whatever); add a visual similarity postbase to get a stem meaning "looking like ____"; add a quantity postbase to get a stem meaning "stuff looking like ____"; add an augmentative postbase to get a stem meaning "lots of stuff looking like ____"; add another postbase to get a stem meaning "gathering lots of stuff looking like ____"; add yet another postbase to get a stem meaning "peripatetically gathering up lots of stuff looking like ____"; and then inflect the whole thing as a verb in the 3rd-person plural subject 3rd-person singular object past tense form; and you're done. Astounding. One word to express a whole sentence. But even if you choose qani- as your root, what you get could hardly be called a word for snow. It's a verb with an understood subject pronoun.

Of course, you can make lots of noun derivatives too. But although various lists of supposed snow words are passed around (public libraries in Alaska compile them, Canadadian Indian affairs bureaux hand them out, skiing magazines publish them, that sort of thing), they fail to back up the familiar myth. These lists tend to cite multiple derivatives of the qani- root; they usually have a bunch of derivatives of the api- root; they often include a word for a sort of rain-pockmarked snow that looks like herring scales, only that word is visibly based on the root meaning "herring"; they include a word for soft snow that is clearly based on the root meaning "soft"; and so on.

So, Eskimoan languages are really extraordinary in their productive word-building capability, for any root you might pick. But that very fact makes them exactly the wrong sort of language to ask vocabulary-size questions about, because those questions are virtually meaningless -- unless you ask them about basic non-derived roots, in which case the answers aren't particularly newsworthy.

That's the sort of thing Sasha would probably have said in the interview if she'd had another few seconds.

[Thanks to Mark Seidenberg for a comment by email that enabled me to make this clearer.]

Posted by Geoffrey K. Pullum at January 30, 2004 01:14 PM