971066 - Auto correction is harder when you mistyped on a row on top or bottom of the letter you wanted

Status: RESOLVED DUPLICATE of bug 1048869

(Firefox OS Graveyard :: Gaia::Keyboard, defect)

Description

[Anthony Ricaud (:rik)]

If I type minures or minuyes, auto correction will find minutes.
If I type minuges, auto correction will not find minutes.

R and Y are on the same row as T. G is below T. But the key below is as easy to mistype as the keys on the side.

That makes it really hard for me to type quickly.

Comment 1

[Olof Wickström]

Considering the number of SMS sent, text input is one key function in the phone. It must be perfect.

Comment 2

[bhavana bajaj [:bajaj]]

[Tracking Requested - why for this release]:

(In reply to Olof Wickström from comment #1)
> Considering the number of SMS sent, text input is one key function in the
> phone. It must be perfect.

Olof, I understand your concern here, but this is not a recent regression and something we will stopship  2.1 for. We can ofcourse improve it in our next release hence I am setting the tracking flag here for teams to consider it. Also, i think it would help make a better case if you think there is a holistic problem around multiple such cases and these were 2.1 regressions.

Tim, marcia/I tried a couple of other cases in autocorrect including the examples above and the 2.2 results of words are much better. Has there been a lot of changes landing to improve that in 2.2 or is it something low risk that could be uplifted?

Comment 3

[Tim Guan-tin Chien [:timdream] (please needinfo)]

There weren't a lot of changes in the suggestions, and certainly not the part that directly affects prediction mentioned in comment 0. 

The way we decide weights of nearby keys remain unchanged since v1.0.1

https://github.com/mozilla-b2g/gaia/blob/master/apps/keyboard/js/imes/latin/predictions.js#L280-L292

Comment 4

[Rudy Lu [:rudyl] (inactive after 2015/7/3)]

We changed the nearby key algorithm in bug 1048869, which only affects v2.2.
We should evaluate the risk to uplift that patch if this is what we need for v2.1.

Comment 5

[bhavana bajaj [:bajaj]]

I think this is better fixed on 2.2. Uplifting this(1048869) on 2.1 at this point seems like a risky change opening up a big matrix for QA to test on. As Tim mentions, the behavior has been this way since 1.0.1 and will be improved upon in 2.2.