[libvoikko] Mozvoikko 2.2

[Timo Jyrinki]

2016-01-05 15:07 GMT+02:00 Sjur Moshagen <sjurnm at mac.com>:
> I would like that to happen as well, but the first thing to do is to enhance Enchant to support the newest libvoikko version, ie with multilingual support, and with support for hfst and vfst turned on by default. As the voikko integration in Enchant is now, it is hard-coded to Finnish only.

Right, that would need to be done.

>> it's probably not going to happen easily, and I don't want to be the
>> one trying to file and push those bugs forward to the
>> English/Indo-European language browser developers.
>
> The Divvun group is very much working on adding support for the non-English/non-indoeuropean (minority) languages, especially for languages not very well suited to Hunspell. If we can help, we will try to do that. Please tell us how :)

I'm happy when other minority languages have common interests :) Well
as a first thought I was thinking about bugs about the usability
issues against Firefox and Chromium regarding the default behavior in
text areas when using languages for the browser that don't have a
spell-checker/dictionary.

That is, at least the following usability problems (to be also verified):
- Firstly, a bug about spell checking being enabled by default in
general, meaning that most words in all text areas are underlined with
red for people who use languages that don't have spell-checker support
- even if the browser is installed in such a language. A better would
be disabled spell-checker for default when the browser is installed in
those languages, since most ordinary users use their native language
pages.
- Secondly, as a refinement, browsers could improve by not disabling
spell-checking by default unconditionally, but instead adjusting the
spell checker functionality based on the page language if known. On
English pages, one would then get the English spell-checker still.

To file a bug, it'd be worthwhile testing on Windows and Mac OS too
using a clean profile, which I cannot do. This was another reason I
decided not to file such bugs myself.


Then... separately there could a long term bug about switching to
using Enchant as abstraction layer (forget about GtkSpell, it's GPL
only and not suitable to Mozilla's policies) for at least Linux and
Mac OS X where Enchant already supports system spell checkers. That
is, it would allow continuing to use of My/Hunspell for currently
supported languages as before.

Regarding Enchant, I did inquire a bit from a Finnish Mozilla
developer: https://twitter.com/timojyrinki/status/684265338503081984 -
Enchant might be within possibilities. I'm not sure how it would work
in practice, since the Voikko library surely won't get included in
Mozilla's source trees - can anything included inside Firefox default
builds, like Enchant, utilize system plugins and libraries? For
example, if the system would provide the Voikko Enchant plugin and
Voikko libraries. Or just Voikko libraries, if we would get
Firefox/Chrome developers to enable Voikko Enchant plugin build by
default.

There's some food for thought.

> On the major mobile platforms the speller is tied to the keyboard. This is true for Android, iOS and Windows Phone - the rest I don’t know. The Divvun group has released a keyboard app targeted at minority languages (Sámi in the first release, but easily extendable to other languages), and we are now working on the next version with speller support. The speller used is not voikko-based, instead using hfst as the engine, mainly due to speed differences (we tested libvoikko in an early prototype, but it was too slow at least on lower-end 2015 phones). Using hfst-ospell should not be a major issue, as both it and voikko are now fst-based, and conversion from vfst to hfst should be trivial. Presently we only support Android and iOS, but the system is extensible, and adding support for other platforms should be within reasonable reach.

Right, that's great, and useful to be keep on mind. I think that might
also the case of maliit-framework users (Jolla, Ubuntu).

-Timo