This touches on one of my interests, the solving of simple substitution cryptograms. In my solver, one of the available actions
is to click on an unsolved word and get a list of possible candidates taking account any known letters, and the patterns of the unknown letters.
In my case, I don't use an ordinary dictionary, because it would suggest uncommon words. I want a list of likely words in order of likeliness. To do this, I've fed my program a steady diet of cryptograms for many years. Every solved word is added to the dictionary, and it keeps track of how often each word shows up. I start by building a list using only the known letters and letter pattern of the unsolved word, and then I try to whittle down the list by seeing what selecting a word as a proposed solution would do to other words within the puzzle, specifically whether proposed letters in one word will create odd letter pairings within other words. I call this the "funny looking word" test. Whatever is left after the filtering is presented sorted by number of previous appearances.
My goal here isn't to solve the cryptograms efficiently. If you had a perfect dictionary, it's a fairly simple exercise to build candidate lists for various words (starting with the long and short ones, if you really want to be efficient) and cross reference the solution letter sets until you find a common solution. Instead I'm trying to emulate the way a human (me) solves the puzzles.
BTW, I got that one in 3 tries. Started with THINK. So far best I've done is 2.
