Love the UI, but the keyboard size and prediction is problematic.
Suggest removing hunspell from auto-correcting output, and letting (if possible) Presage kick out the word prediction instead, regardless of spelling correctly. Hunspell gets the word wrong 70% of the time--it's not great for the ambiguity of the touch keyboard.
For some reason, text2ngram doesn't seem to be part of the ubports build of presage? Ammending the English database with a few more complex English texts from Project Gutenberg which are somewhat modern (more Oscar Wilde, some longer Jack London, H. G. Welles "War of the Worlds", and for idioms the King James Bible [lots of English phrases and expressions stem from this text]) will aid immensely in more accurate contextual text prediction.