Word length

Why are the words people use most often the shortest ones in every language?

Look at the most-used words of English: the, of, and, a, to, is, you. Now look at words you might use once a year: nevertheless, photosynthesis, notwithstanding. The pattern is so ordinary it feels like nothing — but it holds in every language anyone has checked, in scripts and sound systems with nothing in common. Nobody arranged it. So what process, running quietly in millions of unconnected conversations, keeps arriving at the same answer?

Start by asking what a language would look like if the pattern were reversed — if the were six syllables and notwithstanding were one. Nothing about it would be unspeakable. It would simply cost more: the same conversations would take longer, for the same content. That is the first clue. The arrangement is not a rule of grammar but an efficiency, and efficiencies of that kind tend to be arrived at rather than designed.

George Zipf named the pattern the law of abbreviation and explained it by a principle of least effort. But whose effort? If only the speaker's mattered, the optimum is obvious and terrible: make every word one syllable. Then most words sound alike, and the listener cannot tell them apart. So there are two costs pulling against each other — the effort of producing a long word, and the risk of a short word being mistaken for another. Shortness gets spent where it can be afforded, and it can be afforded exactly where the listener has the most help in guessing.

Which words are those? Ask yourself how much a word actually tells you when it arrives. After of, or before a noun, the is nearly certain — you could have guessed it. A rare technical term is unguessable, and if you mishear it the sentence is lost. So the words that can safely be short are the ones carrying least information, and those are, by and large, the frequent ones.

That distinction turns out to matter. A study by Steven Piantadosi, Harry Tily and Edward Gibson found that word length is better predicted by a word's average information content in context — how surprising it is given the words around it — than by its raw frequency, across a range of European languages. Frequency was a good proxy all along, but predictability-in-context is the thing the system is actually tracking. The underlying idea is that speech runs best at a roughly steady rate of information, so highly expected material gets compressed and surprising material gets more signal.

Now the part that is easy to skip: no speaker computes any of this. So how does the outcome get built? Through ordinary sound change, biased by use. Frequent words are articulated more casually and reduce faster — God be with ye wears down to goodbye, going to to gonna. They also resist being restored by analogy, because everyone knows them too well to need regularising. And where reduction is not available, speakers clip: laboratory becomes lab, television becomes TV, and the clipped form survives if it is used enough. The result looks optimised because the wearing is proportional to the traffic.

The tempting misreading is that frequency is the cause — that using a word a lot is what shrinks it. Use is the mechanism of wear, but what determines how much shortness a word can afford is how predictable it is when it arrives. The two come apart in specific cases: a word that is rare overall but almost certain in its usual phrase can be shorter than its frequency alone would suggest.

Two honest caveats. First, the relationship is a tendency, not a law in the physical sense — it holds robustly on average while individual words scatter widely around it, and English a and I sit alongside plenty of short rare words. Second, causation is not fully settled. Frequent use plausibly shortens words, but short words are also easier to reach for, which would make them more frequent; historical cases of reduction support the first direction without ruling out the second.

• Why the frequency of words themselves follows Zipf's other law, the rank-frequency distribution.
• Whether the same abbreviation pattern appears in animal signalling, and how contested those findings are.

Every term the collection defines is gathered in the glossary.