The biggest problem in Japanese input is the conversion step. Then why don't we input kanji directly, without conversion?
One possibility is to use the character code itself to specify a kanji
character. For example, you type 3962 for
and 3636 for
.
Of course this idea is ridiculous. You have to remember the whole
character code table and you have to type 4 keys (in hexadecimal) just
for one character. It is not for a lazy man. I don't like it.
Now the long introduction is over, and it is the time to tell you what
I am using for Japanese input; it is the T-code input method. This
method assigns a sequence of two keys to each Japanese character. To
input a kanji, all you have to do is type only two keys. For example,
to input
you type "ri"; to input
,
you type "lt" (assuming that you are
using a DVORAK keyboard). No conversion, no ambiguity, no Roman
transcription. Of course touch typing is possible.
See the complete list of the mapping table of T-code for DVORAK keyboard and its QWERTY version if you are interested..
The sequence arrangement policy of T-code is simple; the more frequently a character is used, the easier it should be to type. Therefore there is no semantic relation between the key sequences and the inputted Japanese characters. (For this reason T-code is classified as a non-associative input method.) You may think it is impossible to remember the correspondence between the two key sequences and kanji characters. But remembering a two key sequence is much easier than writing that character by hand. And you do not have to remember every sequence; the number of frequently used kanji is limited. You have to learn only what you want to write in kanji. Use hiragana for other words.
In addition to that, the T-code package for Emacs provides you various supplemental utilities. Even in the case that you do not know the sequence for a certain kanji, you can input it by temporary invoking a kana-kanji conversion program from the T-code environment. The T-code package also allows you to specify a kanji in a manner like "the character I want to input consists of the left part of the character X and the right part of the character Y". These features are very convenient.
You can download related programs and data from the FTP site for T-code.
The assignment of T-code strokes is based on the character frequency in newspaper articles. Therefore keystrokes are not allocated to those characters that appear rarely in newspapers. So I have extended the T-code table for DVORAK so that I can, with two strokes, type all the kanji that the Japanese pupils learn in elementary school and the ones used in Japanese prefecture names. (Here is the QWERTY version.) I also replaced two symbols with more frequently used one. I am now trying its usability. Primary impressions are fairly nice.
T-code is not the only method that allows you to input kanji directly.
For example TUT-code is another one. It is similar to T-code, but would be easier to start learning because the layout of the hiragara is highly regular.
There is yet another input method called NIK-code. Its basic idea is to input kanji by their basic components. Each basic component is assigned to a key. The user can then input a kanji by specifying its components sequentially as if he were writing on paper. The NIK-code homepage has disappeared.