This is a solver for the New York Times Spelling Bee puzzle.
There were two goals to this:
- Efficiently identify possible solutions to a Spelling Bee puzzle
- As a subset to the above, identify the pangrams, words that use all of the puzzle letters.
There is a Shiny App hosted on ShinyApp.io that you can use to run the app. It is hosted on a free tier and it turns off if it has been idle for 15 minutes. You may need to wait a minute while the app restarts.
You can also run the solver through the SolverUI.ipynb notebook which is hosted on Binder.
The dictionary is broken down into 26 separate trees for each starting letter of a word. Each node in the tree represents an ordered lettering for a word. A simple example is below:
In the dictionaries, words with fewer than 4 letters are ignored and so "The" and "Tap" would actually not be marked as terminating words and would be excluded from the final solution.
Each puzzle has 7 letters and the solver looks for the 7 trees corresponding to these letters. It then performs a depth-first-search on these 7 trees for words made up of only these letters, ignoring paths with non-puzzle letters, and returns only words that have the center letter.
It then subsets these down to the pangrams by finding a matching set of letters between the puzzle and the solutions.
The New York Times Spelling Bee's answer words are chosen by moderators. The words include those that can be found in dictionaries like Merriam Webster but they can also include words not found in common dictionaries (this site has a quote that is no longer on the NYT website). As such this solver can miss uncommon words.
This project uses the words_alpha.txt dictionary from dwyl/english-words and the original copyright belongs to infochimps (archived). This dataset is a bit old, at least from 2013 at the time of archival and thus is more likely to be missing newer words.