Q is worth ten points and E is worth one. That single fact hides a small theory of language, an argument about information, and a reason the game's tile set looks different in Spanish than in English.
The scoring rule you meet on your first turn is not an aesthetic choice. It is a measurement. In the early 1930s an out-of-work architect named Alfred Mosher Butts set out to build a word game that was neither pure luck nor pure vocabulary, and he needed a principled way to price each letter. So he did what an engineer does: he counted. Butts tallied how often each letter appeared across printed sources of his day, the front page of The New York Times chief among them, along with the Herald Tribune and The Saturday Evening Post.1
From those tallies he read off two numbers at once. The count, how many of each tile go into the bag, tracks how common the letter is in real writing. The value runs the other way: the rarer the letter, the more a play using it should pay. Common vowels earn a single point; the letters that surface only in loanwords and awkward corners of the dictionary earn the game's maximum. In the English game that maximum is ten points, and only Q and Z carry it; as the BBC once put it, Butts “calculated a value for each tile by measuring how frequently each letter appeared on the front page of the New York Times.”2
Line the letters up and the design reveals itself. Point value rises almost monotonically as real-world frequency falls. E is everywhere and cheap; Z and Q are almost nowhere and dear.
| Letter | Tiles in bag | Point value | Freq. in English |
|---|---|---|---|
| E | 12 | 1 | ≈ 12.7% |
| A | 9 | 1 | ≈ 8.2% |
| N | 6 | 1 | ≈ 6.7% |
| D | 4 | 2 | ≈ 4.3% |
| B | 2 | 3 | ≈ 1.5% |
| K | 1 | 5 | ≈ 0.8% |
| X | 1 | 8 | ≈ 0.15% |
| Q | 1 | 10 | ≈ 0.12% |
| Z | 1 | 10 | ≈ 0.07% |
Counts and values are the standard English set;3 frequency figures are the customary estimates for ordinary English text.4 WordChess uses point-valued tiles and a frequency-weighted tile set in this tradition (100 tiles in English, 98 letters and 2 blanks), though each player holds a complete set rather than drawing from a bag, measured from the game's design documents.
Here is where a clean idea meets a hard constraint. If value simply rewarded rarity, the ideal bag would be stuffed with Qs and Zs, enormous points waiting to be claimed. But a bag is also a hand you must actually play from. Draw seven high-value oddities and you sit frozen, unable to make a legal word. So the count pulls against the value: the bag must hold letters in roughly the proportion the language uses them, or the game stalls.
That is why there is exactly one Q, and why its near-inseparable partner U gets four tiles of its own: one Q, and four Us in the bag to give it a chance of finding one. The lone Q is precious because it is nearly unplayable, and playable at all only because the bag's other tiles were dealt in life's own proportions. Scarcity sets the reward; frequency keeps the game moving.
A tile's price is set by how rarely the letter appears, but the bag's contents are set by how often it does. The two forces are the same fact, read forwards and backwards.
If a fair tile set is a photograph of a language's letter frequencies, then changing the language must change the photograph. It does, visibly. WordChess runs in 22 languages, and its tile set is rebuilt for each, because a distribution tuned to English is simply wrong elsewhere.3
Spanish sets sold outside North America drop K and W entirely; those letters appear in Spanish only in imported words, so a faithful bag gives them no tiles at all.5 Italian goes further, omitting J, K, W, X and Y, letters that do not belong to native Italian spelling and turn up only in loanwords.5 Letters that the English bag rations to one or two tiles apiece are, in another tongue, not rare but absent. Rarity is not a property of a letter. It is a property of a letter in a language.
Butts was, without the vocabulary for it, measuring something a mathematician would formalize some fifteen years later. In 1948 Claude Shannon founded information theory on a precise claim: the less predictable a symbol, the more information it carries. A letter you could have guessed tells you little; a surprising one tells you a lot. Rarity is information, and information is measured in bits.6
In his 1951 paper Prediction and Entropy of Printed English, Shannon set out to weigh that quantity. Treating letters in isolation, English runs to roughly four bits per letter; but let a reader use context, the fact that a q all but promises a u, that few words end in v, and the true rate collapses. Shannon's prediction experiments put the entropy of ordinary English at somewhere between 0.6 and 1.3 bits per letter.6 Most of what we write is redundant; the surprising letters do the real work.
The same accounting had already shaped an earlier code. When Samuel Morse and Alfred Vail built their alphabet in the 1840s, Vail is said to have studied the type cases of a printer's shop to learn which letters were most used, then handed those letters the shortest signals, a single dot for E, a single dash for T.7 Short codes for frequent letters; cheap points for common tiles; few bits for the predictable. Morse minimized time on a wire, Butts balanced a game, Shannon named the underlying number, but all three were reading the same ledger. A letter's worth, it turns out, was always a measure of how much it could surprise you.