[Announce] Mana 0.0.2

[Announce] Mana 0.0.2

[Yamagata Yoriyuki]

I'm pleased to announce mana-0.0.2, a Japanese input engine based on
a hidden Markovian model.  It is more accurate than previously used
engines, in particular for the case of long input.

Mana is written C and OCaml.  Currently, Emacs interface using Egg is
provided.

The project page is
http://sourceforge.jp/projects/shinji/
(Unfortunately, English page is still missing.)

Any opinions, suggestiona and contributions are appreciated.
--
Yamagata Yoriyuki

Re: [Caml-list] [Announce] Mana 0.0.2

[Matt Gushee]

Yamagata Yoriyuki wrote:

 > I'm pleased to announce mana-0.0.2, a Japanese input engine based on
 > a hidden Markovian model.  It is more accurate than previously used
 > engines, in particular for the case of long input.


Interesting. But is "input engine" the right term for this? From your 
description and the fact that it works with Egg, I'm guessing it's a 
kanji conversion server, much like Canna or Wnn. By the way, when you 
say it is more accurate than previously used engines, are you referring 
to all of them (including, say, Wnn7) or only the open source ones?

 > http://sourceforge.jp/projects/shinji/
 > (Unfortunately, English page is still missing.)


Looks like there's not much in Japanese either ;-) I was thinking I 
might offer to translate the documentation, but I guess I'll have to 
wait a bit for that. Oh, well, it's a new project, so that's okay.

Thanks for doing this!

-- 
Matt Gushee
Englewood, CO, USA

Re: [Caml-list] [Announce] Mana 0.0.2

[Yamagata Yoriyuki]

From: Matt Gushee <mgushee@havenrock.com>
Subject: Re: [Caml-list] [Announce] Mana 0.0.2
Date: Sun, 22 May 2005 21:11:00 -0600

> Interesting. But is "input engine" the right term for this? From your 
> description and the fact that it works with Egg, I'm guessing it's a 
> kanji conversion server, much like Canna or Wnn. 

Not a server.  For each Emacs process, Egg spawns a different Mana
process.  Hence I call it an engine.

For input methods, client/server models are out of favor today.
Modern conversion engines (Anthy, Prime) are using the same
configuration as Mana, or completely subsumed to applications (SKK,
Sumibi). The one of the reason for client/server models is resource
sharing.  but Mana mmaps dictonary files, hence large amount of the
memory is shared among Mana processes.

>By the way, when you say it is more accurate than previously used
>engines, are you referring to all of them (including, say, Wnn7) or
>only the open source ones?

Only the open source ones.  Note my comparison is just an impression.

>  > http://sourceforge.jp/projects/shinji/
>  > (Unfortunately, English page is still missing.)
> 
> 
> Looks like there's not much in Japanese either ;-) I was thinking I 
> might offer to translate the documentation, but I guess I'll have to 
> wait a bit for that. Oh, well, it's a new project, so that's okay.
> 
> Thanks for doing this!

Thank you for the interest!  Anyone interested in Mana is invited to the
Mana development mailing-list.

http://lists.sourceforge.jp/mailman/listinfo/shinji-devel
--
Yamagata Yoriyuki