[jrockway]

That doesn't really solve the problem. Homophones exist, and entering the phonetic representation in some native character set would still require an additional conversion step to disambiguate the homophone. (i.e. in Japanese you can type in hiragana, and many phone IMEs work like that, but you still have to convert that input to kanji; nobody wants to read your long stream of hiragana.) The input methods add value based on how frequently they suggest the correct conversion as the first candidate (based on context).

Imagine typing English by speaking. You can say "flower", but that might get written out as "flour". Some intelligence has to be implemented that picks the right one based on context, or give you the ability to correct it. That is where the complexity in east asian input methods come from.

(Yes, as English-speaking computer users we are very lucky. The exact sort of symbols that readers of English expect map 1:1 to our keyboards. Still kind of a pain on a phone, though!)