> That's a nice theory; the problem is that autocorrect is often stunningly bad at that It gets even worse for some non-English languages. The standard way to

> That's a nice theory; the problem is that autocorrect is often stunningly bad at that It gets even worse for some non-English languages. The standard way to do autocorrect seems to be roughly "pick the the word I know that has the lowest Levenshtein distance from whatever that was typed". The problem with this is that in, for example, Finnish, every word has a few dozen conjugated forms that have a very short Levenshtein distance to the original word, each of those has a few dozen more conjugated forms that have a short distance to them, etc etc until each Finnish base noun has roughly ~2000 conjugated forms and every verb has ~12000 of them. No autocorrect system in existence actually knows the rules on how to form those words, and none of them actually have anywhere near all of them stored, either. The result is that the minute you output anything even remotely complex it gets autoincorrected to something else. Despite this, most devices and software ships with autocorrect on. The first thing most Finns do is turn it off. The irony of it is that even if you had perfect autocorrect, it really would not be that helpful near the endings of words, simply because the possibility space is so full of correct words that a typo has a good chance of hitting one.