Re: [tlug] CJK Input Methods

[Raymond Wan <rwan.kyoto@example.com>]

I have a feeling that "AI" in the 1980s was far from what we now know
as "AI".  I wouldn't be surprised if a user-defined look-up dictionary
was pre-populated and gradually filled up over time.  And companies
back then called that "AI".  Perhaps Mozc's dictionary can be
improved?

I don't use Japanese very often any more, but when I do, it seems fine
on Ubuntu.  Sure, not as good as Windows' IME -- I do get freezes and
"Huh?  What happened to the IME?" moments...but I put up with it since
I don't use it so often.

If you're eager, instead of re-doing it from scratch, perhaps you can
make contact with the maintainers of Mozc and help them fix it?  The
many software options out there is great, but sometimes, I wonder if
each is only a little buggy, wouldn't it be better to pool resources
and make one or two work better.  Having said that, "re-inventing the
wheel" is basically my day job...  :-P

Ray


On Tue, Jul 16, 2024 at 4:34 PM Yasuaki Kudo <yasu@example.com> wrote:
>
> Ok I can answer this! 😄
>
> * Mozc stalls all the time - I switch between English Russian and Japanese and while English and Russian are reliable, Mozc very often stops working - it says Japanese, but the actual mode is either English or Russian
>
> * Mozc conversion is so poor - it feels to me even my first PC8801 series computer in 1980s had better conversion.   I remember back in the days NEC would advertise how their AI could convert tricky sentences  involving 3 different senses of the word あめ.
>
> But the article I quoted in original post suggests that the free version (not the one that Google uses in the proprietary sibling) just has a very inferior dictionary or something?
>
> Just like the word 'Linux' usually means the whole package, not just some technical definition of Kernel this and that, by saying  'Mozc', I am probably referring to a much wider range of things not just the core whatever that is given that name.  I don't know the exact boundaries but the observable fact is that the overall Japanese input is just very poor in Ubuntu😅.
>
>
> > On Jul 16, 2024, at 18:10, Darren Cook <darren@example.com> wrote:
> >
> > 
> >>
> >>> The approach I am thinking is to understand how Linux input works and create a simple IME ...
> >> I've used, attempted to use, and eventually gave up trying to use,
> >> Japanese input methods in Linux over the last 30 years or so.
> >
> > I'm still not clear on what the problem is that Yasuaki is trying to fix.
> >
> > IMEs are now easy enough to install, at least on Ubuntu and derivatives, following the instructions I've posted a couple of times.
> >
> > And they work well in all apps that I know of. I wondered if it is because I mainly use webapps in browsers and electron apps (vscode, slack, etc), but I just tested in the terminal, and using vi, and again it still works fine (including being able to switch between ascii, Japanese, pinyin and Russian). I also tested creating Japanese filenames over ssh.
> >
> > There was a tangible improvement moving to fcitx5, particularly with Chrome/Electron apps.
> >
> > Also I was using unicode lookup for ² and ³ for a long time, until I discovered the AltGr+2 and 3 key combinations work for that (again, out of the box in recent Ubuntu and friends - I don't think it used to, though?).
> >
> > Darren
> >
> >
>