What dictionary is Wordle based on?

Question

The US Scrabble dictionary only has 9,000ish 5 letter words, of which a good portion are plurals, and then there's the list of "dead" words in the code and then most of these words are not "crossword" puzzle words. Logically you'd want Aunt Gertrude and all her friends to be able to play so the vocabulary bias would be to more common words. So I think finding out which dictionary this was based on is key to beating it without cheating it. Anybody figure this out yet?

Answer 1

I had this question myself, and did some analysis of my own.

My conclusion is that the original list used is CSW19.

The Wordle .js script has two source arrays that it uses for its final wordlist and for its match wordlist.

The final wordlist has 2315 words. The match wordlist 10657 words.

By taking the word lists from ENABLE/CSW/OSPD/etc., filtering them to five-letter words only, and sorting them, it's fairly trivial to review the results and see which match the best. Here's the result for CSW19 vs Wordle's match list:

CSW19:

(12972) ['aahed', 'aalii', 'aargh', 'aarti', 'abaca', 'abaci', 'aback', 'abacs', 'abaft', 'abaka', 'abamp', 'aband', 'abase', 'abash', 'abask', 'abate', 'abaya', 'abbas', 'abbed', 'abbes', 'abbey', 'abbot', 'abcee', 'abeam', 'abear', 'abele', 'abers', 'abets', 'abhor', 'abide', 'abies', 'abled', 'abler', 'ables', 'ablet', 'ablow', 'abmho', 'abode', 'abohm', 'aboil', 'aboma', 'aboon', 'abord', 'abore', 'abort', 'about', 'above', 'abram', 'abray', 'abrim', 'abrin', 'abris', 'absey', 'absit', 'abuna', 'abune', 'abuse', 'abuts', 'abuzz', 'abyes', 'abysm', 'abyss', 'acais', 'acari', 'accas', 'accoy', 'acerb', 'acers', 'aceta', 'achar', 'ached', 'aches', 'achoo', 'acids', 'acidy', 'acing', 'acini', 'ackee', 'acker', 'acmes', 'acmic', 'acned', 'acnes', 'acock', 'acold', 'acorn', 'acred', 'acres', 'acrid', 'acros', 'acted', 'actin', 'acton', 'actor', 'acute', 'acyls', 'adage', 'adapt', 'adaws', 'adays', …]

Wordle match list:

(10657) ['aahed', 'aalii', 'aargh', 'aarti', 'abaca', 'abaci', 'abacs', 'abaft', 'abaka', 'abamp', 'aband', 'abash', 'abask', 'abaya', 'abbas', 'abbed', 'abbes', 'abcee', 'abeam', 'abear', 'abele', 'abers', 'abets', 'abies', 'abler', 'ables', 'ablet', 'ablow', 'abmho', 'abohm', 'aboil', 'aboma', 'aboon', 'abord', 'abore', 'abram', 'abray', 'abrim', 'abrin', 'abris', 'absey', 'absit', 'abuna', 'abune', 'abuts', 'abuzz', 'abyes', 'abysm', 'acais', 'acari', 'accas', 'accoy', 'acerb', 'acers', 'aceta', 'achar', 'ached', 'aches', 'achoo', 'acids', 'acidy', 'acing', 'acini', 'ackee', 'acker', 'acmes', 'acmic', 'acned', 'acnes', 'acock', 'acold', 'acred', 'acres', 'acros', 'acted', 'actin', 'acton', 'acyls', 'adaws', 'adays', 'adbot', 'addax', 'added', 'adder', 'addio', 'addle', 'adeem', 'adhan', 'adieu', 'adios', 'adits', 'adman', 'admen', 'admix', 'adobo', 'adown', 'adoze', 'adrad', 'adred', 'adsum', …]

If you take the difference of the CSW19/Wordle match arrays, you get the Wordle list of 2315 words. This is because the match list and the final wordlist are exclusive. From the info in the NYT article, we know that the initial list of 12972 words was picked through by Wardle's partner, and separated into two arrays: the final wordlist, and the match list.

Your guesses are then checked against both the final wordlist and the match list.

You can confirm this by taking CSW19, removing all of the wordle match words, and removing all of the wordle solution words. This results in an empty array.

You then must do the reverse: take the wordle match/solution lists, and remove the CSW19 words. This results in an empty array.

So, to summarize:

1. Start with CSW19 (12972)
2. Separate the resulting list into two categories, the final wordlist (2315) and the additional match list (10657) for your guesses.

I agree that all this info probably won't help you much to solve the puzzles themselves. The best insight it grants is one you get from seeing the source arrays: just because your guess is accepted, doesn't mean it meets the commonality criteria of the solution. However, if you're looking for a more tactical approach, you might check out some of these blog articles, which get into the business of letter-counting and identifying the best starting words.

Personally, I prefer to ignore probability and dive in with a fresh new 5-letter word every day. You never know what you're gonna get!

Happy solving. 🙂

Answer 2

What follows is a frame challenge.

---

You request:

What I was hoping for was...someone who had taken the solves and ran them against the known dictionaries to find out which one was likely used.

This is fundamentally not a thing that can be done. Word lists (which is also what dictionaries become, after you strip the words away from their definitions) are primarily distinguished from each other by the unusual/rarer/new words they either include or exclude. Words where people disagree on their spelling or even their word-ness.

So, what of Wordle's word list? Well, here is Word Of Creator:

An initial list of all of the five-letter words in the English language — about 12,000 — contained a lot of obscure words that would have been near impossible to guess.

So he created another game for Ms. Shah: This time, she would sort through those 12,000 or so words, designating whether or not she knew them. That narrowed down the list of Wordle words to about 2,500

_{Archive.org copy}

The final word list (that is, the one used for possible solutions) is one of common-ish words. Words that a reasonably educated woman could reasonably know and recognize. Even among the slightly less common words, these are not ones which people disagree on word-ness. These are words that would be in any dictionary. Doing your proposed search using previous solutions would likely find Wordle's list to be a perfect subset of any you tried.

So it wouldn't work. And also wouldn't really help. Even if we could reach back in time and learn exactly where the original 12k word list came from, so many were eliminated from use that any sort of analysis would be seriously distorted.

---

Postscript:

There is no list of "known dictionaries". There are popular or well-known dictionaries (though which to include in those lists would be a subject of debate), but there are an Internet-powered practically infinite number of word lists and dictionaries out there. Some seek to fulfil specific purposes (e.g. a dictionary of medical terminology) and some seek to be general-purpose, covering English as a whole (though what counts as "English" is also up for debate, and is ever-changing as befits a living language). That's yet another problem with the "test all the dictionaries" plan.

Answer 3

There are two dictionaries. A short one of possible answers and a long one of allowed guesses.

The lists were created by the author of the game. Perhaps he'll provide an answer here as to where and how he formed them.

But I'll answer by guessing: the lists were designed to provide enjoyment to casual players (as opposed to hardened Scrabble or crossword enthusiasts). So, the list of possible answers is heavily culled to avoid obscure (and therefore not fun) words, and the list of possible guesses is heavily extended to avoid wrist-slapping guesses that are valid in the player's mind but not in this or that common dictionary (and therefore not fun).