I don't know about Kii (seemed pretty bad long time ago) nor TouchPal X, but although I love SwiftKey, I hate its model. I'm almost positive they use an n-gram

I don't know about Kii (seemed pretty bad long time ago) nor TouchPal X, but although I love SwiftKey, I hate its model. I'm almost positive they use an n-gram Markov model, which is pretty cool. But, and the big but (might be wrong but it's my feeling) is that they share the model across languages. I suspect so because I normally keep English/Spanish on and switch between Portuguese and French, and even when I'm writing Spanish in E/S/F mode I get Portuguese suggestions (because of identical and very similar words to Spanish). No, you should be way, way stricter to tell languages apart and use different models for each one, probably using another model to adjust the probability of each language. Example annoyance: After I type 3 words in Spanish that happen to be 2 words in Portuguese and 1 in English, I certainly don't want the English suggestion, even if I use that word more frequently in English.