How I Built Lexicle – The Word Game Based on Meaning
I’ve always loved daily semantic games, but found them too hard to play because of how unexpectedly the words relate. Most rely on something called word2vec, which measures how close words appear next to each other in text, but not in meaning. This creates maddening moments (like in a recent Proximity puzzle) where the target word was “firefly” and the closest words to it were “owl” and “frog” rather than “insect,” “bug,” or even “fly.”
My partner commented that if someone could solve that problem and make the game more fun to play, it could be really cool…
So I got to work. π
A New Way to Relate
After ruling out word2vec for good, I found my way to Sentence-BERT, a framework that is great at understanding semantic relationships between sentences. I tweaked it to process individual words – specifically every 5-letter word in English minus offensive and unusable words. This created pairings that were most of the way there, and I optimistically dove into manual review to complete the process.
This was… not my most efficient idea π€¦π»ββοΈ
Secret Sauce
After realizing the pain of manual review, I remembered AI exists and built a script to send each word pair, one by one, to an API endpoint for a semantic judgement call. The full prompt took many iterations, and the result was reviewed and game-tested.
Putting It All Together
With words now fully paired together, I quickly realized from play-testing that the game was too difficult to solve from just guessing words alone. After iterating on hint mechanics I stumbled upon a formula that helps the vast majority of players guess the word by the 11th guess, while still rewarding the puzzle diehards.
Also, all Lexicle puzzles are unique, so if you’ve played the archive, you can rule out future words. π
If you find a word that is missing or needs to be recalibrated, please reach out via the contact page or in the comments below. The game is only as good as its words after all.
Happy Playing,
Matt