On Tue, Jun 24, 2003, Diego Iastrubni wrote about "Re: kde":
> it guessed right only ???©?????¢, even ???????????¨ it did not guess (the closest I 
> got 
> was ?????????¨) . All the other I got were, really too far away guesses.
> 
> ???????? ?©?????©??, 24 ?????????? 2003, 22:52, Meni Livne ???×??:

Hspell's current correction algorithm is very simplistic: it tries to correct
problematic Hebrew spelling mistakes - not typos (which anybody can correct on
their own). Those mistakes include writing and extra vav or yud or missing
one, replacing chet with chaf and all the similar mistakes, and so on. It
also assumes only one mistake per word when correcting, so don't expect it
to find a word that is off by two letters.

This was not only laziness on our part - there's a reason behind this
madness. The problem is that in Hebrew, more than in English, many seemingly
random words turn out to be correct words. If you take a mispelled word
and look for all the words that have any 2 letters changed, you might end up
with dozens of useless suggestions.

Anyway, it should be very easy to modify hspell to use a different correction
algorithm. If anybody can suggest other things the corrector should try,
I'll gladly consider them. If anybody can point me to existing research
on this topic, it would also be nice.

By the way, further into the forseeable future, Hspell will be smarter
about recognizing valid uses of prefixes, and you'll no longer see some of
the silly suggestions you saw like בסידר (ב+סידר, a non-sensical prefix for
this verb).

> I must admit that it is kinda slow, but it does work, even here in kmail 
> 3.1.2-texstar. You can try it in kedit where it works out of the box if you 
> specify the spell checker correctly.

The announcement I'm going to make now is a little premature, but I'll
make it anyway: the next version of Hspell (0.6) will include a new
"hspell -a" front-end written in C. This new front-end is already working
quite well and I now started testing it and completing the last missing
features before we can send it to some of you people for alpha-testing.

The performance difference between the new "hspell -a" and the old one
is staggering: Startup time is about 30 (!) times faster (down to only 1/3
second on my slow Pentium 500) and memory consumption is about 6 times
smaller (an hspell process now takes less memory than an "xterm" process,
on my system). The Hspell installation remains a tiny 100K :)

The new version will hopefully be released sometime next month, but no
firm release date has been set yet.

> Koffice 1.2.1 cannot use hspell, btw.

Why? Can it use an "ispell -a"-like program for doing spell-checking?
If it can use "ispell -a" and not "hspell -a", maybe "hspell -a" is
still missing necessary features. I'd be happy to know what these are.

Note that Dan and I (the Hspell developers) do not use all the countless
editors, word-processors, mailers, and other systems that could use Hebrew
spell-checking, so we very much appreciate when somebody takes it upon
themselves to make sure that some application could make use of Hspell -
like Meni Livne has been doing with KDE.

> > It currently doesn't work with koffice 1.2.x apps, but it should be in
> > koffice 1.3.
> We have our hopes high... I hope that it will not go down...

Not if I and Dan (on the Hspell side) and Meni (on the KDE side) can help
it :)


-- 
Nadav Har'El                        |    Wednesday, Jun 25 2003, 25 Sivan 5763
[EMAIL PROTECTED]             |-----------------------------------------
Phone: +972-53-245868, ICQ 13349191 |A language is a dialect with an army.
http://nadav.harel.org.il           |

================================================================To unsubscribe, send 
mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Reply via email to