predictive text works weird with accented words

[Julia Palandri]

I've seen this both in italian and spanish where there are accented words: if for example in spanish I write "mama", it doesn't suggest anything else. If, instead, I type "mam" (or if I go back and delete the last "a"), it does suggest "mamá" (which is the word I want but am too lazy to tap myself).

It's even more annoying in italian: writing "e" doesn't suggest "è" (which is the one I usually want) but in this case I cannot delete anything to get the suggestion offered and I need to long tap the e key everytime I want to type "è".

Could we make accented words "similar" to those same words without any accentuation so they get suggested?

[Jean-Baptiste Lallement (jibel)]

Thanks for your report. I agree, it's the same problem in French. When the user types a non accented letter, the keyboard should propose accented variants of the word.

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Julia Palandri (julia-palandri)]

Hi @michael-sheldon, I think these are two separate things (most surely related anyway, but different things: this bug is about not getting accentuated words suggested while writing when i write the same word without accents.
The other bug, as per what I understood, is about autocorrection, which might or might not be enabled in the phone at the same time as the suggestions I'm talking about here.

[Michael Sheldon (michael-sheldon)]

Hi Julia,

 The two actually have the same underlying cause, as the auto-correct simply replaces the user input with the first entry from the prediction engine (assuming it isn't too different from the original input). Unfortunately the current prediction engine (based on libpresage) treats accented forms of characters as being completely unrelated to the non-accented forms, so the predictions it generates (and which are then used for auto-correct) are often not very helpful.

 Thankfully however, we're currently in the process of improving the prediction engine through the use of libskeyer, which should drastically improve the quality of our predictions and part of that is handling accented characters correctly :)