search

keymanapp / keyman

Keyman cross platform input methods system running on Android, iOS, Linux, macOS, Windows and mobile and desktop web
https://keyman.com/
Other
402 stars 112 forks source link

bug(web): autocorrect initial feedback #12648

Open mcdurdin opened 2 weeks ago

mcdurdin commented 2 weeks ago

I've been using Keyman 18 alpha on my phone now for a while, with autocorrect switched on. Here are some initial experience issues:

  1. Sometimes space inserts two spaces; I cannot pick exactly what triggers this but once it starts, space will always insert two spaces until I move the insertion point manually
  2. Pressing period, comma, etc, does not accept an automatic suggestion -- only spacebar
  3. When typing rapidly, one often does not notice if the autocorrect chose wrongly. Backspacing to undo autocorrect only seems to work immediately after the spacebar is pressed; once any other keys are pressed, you have to delete the entire word and retype.
  4. Because we don't have learning, wrong autocorrects will always go wrong -- which is super frustrating (e.g. see 23 -> 2nd below). This may be a dealbreaker.

The following may all be related to nrc.en.mtnt. The limitations in the wordlist become more significant when we have autocorrect, e.g.:

  1. Numbers get autocorrected, e.g. 23 -> 2nd.
  2. We have many words missing in our English dictionary, so rarer words are frequently autocorrected. It will be essential to have 1 and 2 letter words in the dictionary.
  3. Curiously we have many words where a longer form is present but a shorter form is not, e.g. we have authorized but not authorize.
  4. Many longer forms are missing, e.g. we have dance, dances, dancing, dancer, dancers, but not danced.
  5. It is a bit androgynous when it comes to en-US vs en-UK -- it has some UK forms but not all, e.g. authorize but not authorise, whereas we have both behavior and behaviour (and behavioral but not behavioural). We may need to split en-US and en-UK into separate models for better outcomes.
  6. i is not corrected to I, in English (not sure if one letter words can be resolved by updating the wordlist, or if it will be blocked by lexical model implementation also).

All this said, generally, I find the autocorrect a positive experience. Most aspects work as I would hope and expect, but we need to iron these rough edges out before 18.0 release.