Autocorrect for Slavic languages is a somewhat hard problem due to all the different conjugations (wrong word?) you can impose on a word. In English for instance, a noun can be in singular form, or in plural form. In Slavic languages, it can be in many forms - (singular, plural, dual) * (all the cases) * ("a", "the" -- if the language has the distinction). Adjectives then get all these, but times the three genders they need to agree with. Verbs can be conjugated in many ways, too. So a good autocorrect system needs to be grammar-aware to know that the word you're correcting is e.g. an adjective that's agreeing with a feminine singular noun in the accusative case. None of them do that from what I've seen. It's not all bad, since you mostly know what you're trying to write and just picking the closest word by levenstein distance does the job, but it can definitely be done better. One downside is you need to have every variant of every word in the dictionary which kind of makes it needlessly large in size, and I've sometimes found variants missing and had to add them.
Polish is the most popular Latin script Slavic language and has quite a nice market. After using Google keyboard for a few years I still daily add new words to it's dictionary. With English it is very rare.
Many times autocorrect has word only in present-masculine form. Or it doesn't have subjunctive form. Even for words I consider common. I can only imagine how it can be for less popular Slavic languages.
Nice clarification - I've been using declension/conjugation interchangeably for the last year or so and I'm glad to see it resolved pretty unambiguously!
I see, thank you for the insight. It seems to be a similar but much harder problem to the German dictionaries which don't know how to deal with compound words. Android is also missing a lot of forms for verbs in second person singular.
Bah yes, this drives me nuts. I don't get it, compound words aren't rocket science, and even some level of grammar awareness shouldn't be so hard considering the rather low expectations towards autocorrect when it comes to correctness.
