[Bug] keyboard predictions/dictionaries (language-specific?)

On my phone he suggests heute, je, ne, hr, hd - I assume that these abbreviations are actually in the dictionary and the system considers on of the two characters being mistyped (the suggested letters are all near h and e in the German keyboard layout. Even though this algorithm works fine for actual words, it is not an ideal behavior for two characters, I think.

It's not logical for it to suggest entire words for just one letter, and only abbreviations for two letters. the second letter should just narrow down the number of suggestions and go by probability in their sequence. why would a bunch of abbreviations be prioritized before complete words? It's pretty obvious that it's a bug and not intended to be this way. otherwise, the abbreviations would be suggested at 1 letter as well.

@bennypr0fane I totally agree. This might work well in languages that actually have a lot of two-letter words, but not in German. I think that typo-suggestions generally should not appear prominent in the list as long as less than 75% (or so) of the letters actually match. This would put complete-suggestions before typo-suggestions, too.

Even though I agree it is not logical to priorize abbreviations, I still think that it works as designed, though. Just noticed that the keyboard seems to use XT9 for autocomplete. Obviously XT9 simply looks up potential typo-words from the dictionary with word length matching input length, hence suggesting two-letter-abbreviations when having two characters input. I assume that there is nothing like single-letter-abbreviations in the dictionary ;-)

I still don't believe it's even abbreviations that are looked up by XT9 or whatever is at work here. If you look at the two-letter combinations in my example, at least four of them aren't actually abbreviations for anything (or if they do exist, I've never heard of them). These are just strings that shouldn't be in the dictionary, and I think in fact they aren't, but completely random, as produced by the wrong algorithm. In short, this is a different kind of bug than you guys are suggesting :-)