ref: 951ab17262c61dca249baf5aca1810b55117e684
dir: /sys/src/cmd/ktrans/README.kenji/
This version of ktrans works with pipefile by Rob Pike, which replaces /dev/kbd of
2nd Edition Plan 9 by Dennis Ritchie. By this new idea, we can input Japanese
anytime onto any windows. I included his post to 9fans mailing-list here (pipefile
file).
This is a distribution as a derived work license of Plan 9, and I included the
Plan 9 License term here. Please refer to it on the details of this license.
Main features are as follows:
1) There is a "local" dictionary file for translation from kana to kanji, which can
easily be edited by the user. The default file name is $home/lib/ktrans-jisho.
If you want to use another dictionary file, set the KTJISHO environment
variable to point to that file. This dictionary comprises many lines of
edittable text strings.
2) Capital romaji input is used for words such as verbs or adjectives with okurigana,
which follows the idea of the SKK system by Masahiko Sato of Kyoto Univ.
(masahiko@kuis.kyoto-u.ac.jp). If you want to get the kanji string (runes)
"動かす", which is a verb, you may input "ugoKasu" from the keyboard.
Note here the Kasu's k is a capital (important). You will see hiragana
runes "うごかす", and then the kanji runes of "動かす", when you type
"Shift+Space".
If you are satisfied by that translation, continue to input the next word.
If you are not pleased with that candidate, hit 'Shift+Space' once more
to see more candiates for that hiragana input. When there are no more candidates
registered in your dictionary, you will see the initial hiragana input.
3) For Japanese "joshi", a post-positioned short word after a noun, you can use another
method that I developed for this work. If you want the kanji string (runes) "私は",
then type "watashiHA" from the keyboard. Note that the sound of "wa(ha)" is expressed
as capitalized "HA". You will see a hiragana string of "わたしは", and then "私は"
after `Shift+Space'.
4) A control sequence of 'ctl-l' is introduced to leave input hiragana runes unchanged.
This is occasionally neccessary.
5) A simple leaning mechanism has been implemented for the in-memory hashing
dictinary, in which the most recently used kanji runes (candidate) moves to the top
of the list of candidates. This is valid only during the session you called ktrans.
It is done this way intentionally, because the present learning method is ..well...
naive. ^_^ I know this; however, I believe you can solve it by making a good
dictionary best fitted to your purposes on your own.
6) 'ctl-x' re-reads the new kana-kanji translation dictionary when you have edited your
kana-kanji translation, and want to incorporate it into your current in-memory translation
dictionary. The kana-kanji translation dictionary is usually read only once, at the
beginning of the ktrans session. I believe this simplification is not a serious constraint,
because you can edit the dictionary anytime by co-working with acme and using this ctl-x
mechanism.
7) A mode change to kana-input is triggered by 'ctl-n', ctl-k for katakana mode,
ctl-g for Greek mode, and ctl-r for Russian mode.
8) As the starting $home/lib/ktrans-jisho, you may re-format the SKK-JISYO.S (66.9KB) of the
SKK system, which can be retrieved from ftp.kuis.kyoto-u.ac.jp. The next three lines
show the short sed filter to transform from an SKK type dictionary to Plan 9. Before
this, you should change the kanji code of the SKK dictionary from ujis(euc) to UTF-8 by
the tcs utility, of course.
s/\// /g
s/ / /g
s/ $//g
The header items are sorted in a strange order in the original SKK dictionary.
This implementation does not care about the order, therefore, you can change it on
your own.
9) SKK jisho, such as SKK-JISYO.S, is composed of two parts, okuri-ari and okuri-nashi
entries. This greatly depends on the Japanese grammer, and okuri-ari may represent
verb/adjective etc., i.e., not noun. These two parts work differently in the original
SKK system, however, I did not employ that method; rather, I took a simple approarch
as described in (2) and (3). Here, we make no difference between these two parts,
and the reason why I left the two-part structure is just to make it easier to read for
editting. Of course, you can change it without any side-effects.
10) The essence of this Japanese input method is to convert every one word by one key
triggering. This may cause some cumbersome feelings for Nihongo users who are accustomed
to, say, Windows. I know this. However, I intended to keep the codes as compact as
possible as a first step towards developing a Nihongo input system on Plan 9.
Furthermore, I've never seen the latter work perfectly. I think the conversion failed
essentially when we see more than, say, five/six candidates for one set of hiragana runes.
Finaly, I'd like to add that I have no problem to write long Japanese documents by
this version.
11) The translation trigger key has been changed from ^t to Shift+Space, because
we experienced butting of the trigger key sequence sometime. For this reason,
you have to edit the key-binding table, such as kbtabshift[0x39] in /sys/src/9/pc/kbd.c
or keymapshift[0x79] in /sys/src/9/ss/screen.c to assign Shift+Space to '^\'.
I use here '^\' as the trigger key. Therefore, you can trigger translation
from kana to kanji by Shit+Space or just '^\'.
12) A usage example: If you want to make the Japanese text as below:
私は毎日35分以上歩いて、 更に10分電車に乗って学校に通います。
健康の維持にも役だっていますが、 なかなかたのしいものです。
your keyboard typing stream should be:
watashiHA[^t]mainichi[^t]35[^l]fun[^t]ijou[^t]aruIte, [^t]saraNI[^t]
10[^l]fun[^t]denshaNI[^t]noTte[^t]gakkouNI[^t]kayoImasu.[^t]
kenkouNO[^t]ijiNImo[^t]yakuDAtteimasuga, [^t]nakanaka[^l]tanoshiI[^t]
monodesu.[^l]
where [^t], [^l] indicates 'Shift+Space' and 'ctl-l', respectively.
Kenji Okamoto August 14, 2000