Basically, this program “teaches” itself words by generating a string of random characters, then trying to translate it to German, then from German to French, then from French to Spanish. If it is gibberish instead of a real word the translation fails, and it remains the same gibberish. By checking to see if the word remains the same between translations, we can see if it is a real word or not. As a further precaution all three versions are translated back into English, and we check to make sure they are all now the same. Duds occasionally slip through the cracks, but most are actually English words. The chances of any given random combo being a real word are only about 2.60% but it is trying about 83 combinations every second. The success rate will slowly decline the longer your list becomes, as the creation of a duplicate word will always fail.
Inspired by YourFriendHex, designed as a slightly faster, more optimized and more accurate version of his project English LLM. (Linked below) https://scratch.mit.edu/projects/1000353559/
