Singularity on the Plane

I’m in holiday now… Probably this blog also…

(It’s still possible that I will blog randomly)

じゃぁぁ…

Here’s my way of memorizing kanji. It is based on my experience of learning Japanese writing for around 1 year. I always tweak it from time to time.

The first is about picking which kanji to learn:

1. Choose a batch of kanji to memorize. For example, I’m now memorizing grade 3 jouyou kanji (200 kanji). Some other examples are all kanji in a specific Japanese Wikipedia article and all kanji in a Japanese song. Making a list reduces the sense of being overwhelmed by the gigantic amount of kanji one needs to master Japanese, while at the same time creating a challenging target to accomplish.
2. If the list is too large (like my current case), break it into more managable chunks. For example, I pick at least 10 kanji (from the total 200) to learn everyday.

Now to the individual kanji. For each kanji to memorize:

1. Pick one reading and meaning from the Japanese to English dictionary EDICT. Dictionary programs like Wakan and JquickTrans use EDICT for their data. At its core, EDICT is just a text file (SHIFT-JIS encoded) and browsing the contents using Firefox is possible. I use my own console program EdictReader which displays homophones and homonyms for a search result (also supports grade filter and super-limited regex):

   

   To give a concrete example, let’s suppose I’m rying to memorize 光. For 光 I can choose the reading ひかり (hikari) with the meaning “light”. Another possibility is 光る which reads ひかる (hikaru) and means “to shine”. Note that some kanji cannot stand by itself, which means that it needs okurigana (hiragana suffixes) like 暑い (あつい, atsui, hot) or appears in a compound like the 曜 in 日曜日 (にちようび, nichiyoubi, Sunday).

   When the kanji only appears in compound and I doubt about its meaning, I take a look at the kanji dictionary KANJIDIC. KANJIDIC has information about associated meanings for a kanji. Again, dictionaries like Wakan and JquickTrans use KANJIDIC for its kanji data. I eat my own dogfood, using SharpJiten to do it. For example, the 曜 mentioned before has the meaning “weekday” associated with it.

   PS: As a side note, “weekday” means any day of the week except Saturday and Sunday. However, the kanji 曜 is used in both Saturday (土曜日) and Sunday (日曜日).

2. After that, I make a mnemonic to associate its shape to its meaning. For the kanji 光, I separate its shape into subshapes:

   

   and assign the mnemonic “he walks (儿) with a hat (亠) to protect himself from the LIGHT of the sun ray (\ /)”. This is a creative process which is limited only by imagination (see other examples here). The weirder, funnier, and more personal the mnemonic is, the easier it will be to remember.

   After creating a mnemonic, I write it in a computer file for future reference. This is obviously useful when I forget the mnemonic.

   When you have memorized lots of kanji and radical (shapes occuring in various kanji, like 广), memorizing new kanji will be a lot easier. This is because many kanji are just a mix and match of other kanji and/or radical. Some examples are the kanji 明 (bright) (mix and match of 日 (sun) and 月 (moon)) and 線 (line) (mix and match of 糸 (thread), 白 (white), and 水 (water)). Making a mnemonic will then be only about making a sentence out of those well-known shapes. An example is “the sun (日) and the moon (月) are both visible on the sky so it is very BRIGHT (明) today”.

3. Next is to make a mnemonic to associate the sound to its meaning. Here’s an example for ひかり (hikari, light). In English, ひ (hi) is read like “he” and かり (kari) sounds somewhat like “curry”. So one possible mnemonic is (he (HI) cooks curry (KARI) by exposing it to sun LIGHT).

   I also archive the mnemonic in a computer file.

   After this step, the kanji, its reading, and its meaning are interwoven nicely with 2 mnemonics:

   kanji (光) -> mnemonic -> meaning (light) <- mnemonic <- reading (ひかり)

   For the sound-to-meaning mnemonic, I freely make use of Indonesian, English, Japanese, and any other language I could think of (Javanese, for example). I sometimes use my program WordFinder to find Indonesian words for the mnemonic ingredient (data from the Indonesian dictionary for OpenOffice.org, kindly provided by Benitius Brevoort):

   

4. Then I see see the stroke order using my program StrokeReplayer (data from Taka and KanjiCafe). Then I scribe the correct stroke order in my stroke order reference book (the usual tree-killing book, not a a digital thing). I then try to "feel" the flow of the strokes by writing the kanji several times in my practice book.

5. The finishing is to put two entries in Mnemosyne so that there will be computer-scheduled optimally-spaced review. One entry is to ask the reading and its meaning from its kanji, for example:

   Q: 光
   A: ひかり (light)

   Here’s an example entry for a kanji that has multiple meanings (note that multiple meanings is indicated in the question):

   Q: 体 (2)
   A: からだ (body, health)

   And here’s an example entry for a word that has more that one reading and meaning:

   Q: 一日 (2) (2) (1)
   A: いちにち (one day, first day of the month), いちじつ (one day, first day of the month), ついたち (first day of the month)

   PS: “(2) (2) (1)” means that there are 3 readings, two of them having 2 meanings and the other having 1 meaning

   The other entry is to test your writing, like so:

   Q: ひかり - light
   A: 光
   (grade info in category, for example “Japanese - Writing - Kanji - Grade 2″)

   When I encounter a writing question, I write it on the canvas of StrokeReplayer and compare it with the correct answer there also.

That’s all there is to it.

Is all the mnemonics really that useful? I can ascertain that it works great! When I can’t instantly recall a shape, word, or meaning, mnemonic allows me to rediscover it. When I’ve become familiar enough with the item, I won’t consciously use the mnemonic so it certainly isn’t a hindrance to instant recall.

goproblems.com is a wonderful community site containing thousands of… well… go problems. Problems are presented and solved using a Java applet. For me, the best thing about the site is that it is downloadable.

The downloaded problems can be navigated using the “previous” and “next” button. However, it iterates through problem id, which for me isn’t very useful.

My idea is to traverse the problems from the easiest to the hardest. My first solution was to copy-paste the contents of the “all problems” index into a spreadsheet and sort them there. Because the links in the speadsheet is clickable, I would navigate the problems from the spreadsheet.

My second solution was to use the sorted spreadsheet data to generate a set of htmls which allows me to traverse the problems without leaving the browser. Because OpenOffice.org Calc uses XML, parsing it was quite doable (I didn’t use System.XML, just some ad hoc parsing). The result was a frame based solution. The original goproblems.com page is contained in a frame and there is another frame for navigating backwards/forward. Bad usability because there were two sets of navigational buttons and the difference wasn’t obvious. Not to mention that frames are ugly.

Some days ago I downloaded a the latest goproblems.com snapshot (around 1000 new problems from my last one). I then finally forced myself to make a final solution. It was essentially a complete rewrite. I made the program read the difficulty directly from the html files, removing the need for manual sorting inside OOo Calc. I also made the program modify the URL of the already-existing navigational buttons, removing the need for frames.

So, here’s the result:

The program is here and the source code is here. It is made in C# so Mono or Microsoft .NET needs to be installed. To run it, just type:

gpsorter [input-directory] [output-directory]

or using Mono:

mono gpsorter.exe [input-directory] [output-directory]

Where [input-directory] is the directory of the unpacked and orginal goproblems.com.

30 new kanji I’ve learned:

度庫庭式役待急息悪悲想意感所打投拾持指放整旅族昔昭暑暗曲有服

Here is the associated word dump:

Word count is 1609.

I’ve finished writing the stroke order of all grade 1 and 2 kanji on my new reference book. I’ve also added the corresponding StrokeReplayer entries on Mnemosyne. I’m now I’m memorizing grade 3 kanji with the new infrastructure.

Here are the 60 kanji that I’ve memorized:

丁世両主乗予事仕他代住使係倍全具写列助勉動勝化区医去反取受号向君味命和品員商問坂央始委守安定実客宮宿寒対局屋岸島州帳平幸

And here are new words I’ve memorized (most of them come from the new kanji):

The word count is now 1579.

The font MS Gothic and MS Mincho differs in how to draw some strokes. Here are 2 examples:

The red strokes in MS Gothic are concave up, while in MS Mincho it is concave down. It seems that MS Gothic favors symmetry more than MS Mincho. Here are some more examples:

In MS Gothic, the blue stroke is a horizontal mirror of the red one.

However there are some notable exceptions that I found:

In the image above, MS Gothic and MS Mincho agree on stroke concavity.

All those examples suggest that concavity is up to the preference of the writer. MS Mincho’s style can be said to showcase wabi sabi more (wabi sabi is a Japanese view that states beauty is found in asymmetry and imperfection).

However, even on the same font there are variations for some shape. I’ll illustrate MS Gothic’s case.

The first is regarding the red stroke in the 木 shape. In the top row, all of them are concave up. In the second row, where the shape appears on the half left of a “horizontal flow layout”, it is concave down and slightly displaced below.

A horizontal flow layout is a layout where items are arranged from left to right. Constrast it with a vertical flow layout. In 数 the 木 shape is inside a vertical flow layout so it is drawn like the normal 木. This illustration shows the difference between a vertical and horizontal flow layout:

Next is the shape of 立:

Appearing by itself, the blue stroke is unconnected to any other stroke and the red stroke is connected to the lowermost stroke. For the 2 kanji to the right of the first one, the stroke corresponding to the red one is connected to the top horizontal stroke also. For the last kanji, all of them are connected to the top and bottom horizontal strokes.

The shape of 父 also varies:

In the last kanji, the red stroke is concave down, different from the first two.

The last example is 土:

In the top row, the lowermost horizontal stroke is… well… horizontal. However, in the bottom row, where 土 appears in the left side of a horizontal flow layout, the lowermost stroke has a positive slope.

There are many more examples that I found. Because some variation rules can be found, it casts doubt that the stroke detail (concavity, connectivity, slope, etc) is completely up to the writer. Whatever the case is, all those variations makes the task of memorizing the shape harder for anyone who want to follow the style of MS Gothic (or Mincho) as close as possible.

Using the combination of Drago, Mnemosyne, and KifuReviewer, I’ve memorized 6 Yi Se-tol games (first 50 moves). Here are them, from the order in which they are memorized:

All the results are Yi+R. No, Yi Se-tol does not always win by resignation. The games are deliberately chosen as such so that I can easily memorize the result.

I can’t say that those games improved my own play. It does broaden my joseki knowledge, especially on the komoku ones. I also found some surprising moves, such as approaching a hoshi enclosure from below and invading a double enclosed hoshi. However I won’t write more about it primarily because I just took note of them and haven’t bothered to study the positions in depth.

The openings are especially hectic. In most of the games, a complex fight starts early from one of the corners and spreads towards the center. When a fight like that occurs, other corners and sides will be untouched.

My purpose of memorizing the games is just to warm myself up for the real improvement activity: tsumego. If you can’t do life and death and solve tactical problems, you won’t get far.

I’ve started doing the Korean Problem Academy sets from gobase. For book 1 (200 problems, rated 25k-15k) I got all of them correct. Not surprising since I’m now 8k YIC and because I’ve been through book 1 quite a lot. For book 2 (200 problems, rated 15k-5k), my score is 96%. The mistakes are getting seki while a better solution exists. I’m currently redoing it again.

I’ll probably do daily goproblems.com (downloadable) soon. 30 minutes a day should be enough. I also need to study more advanced life and death concepts by reading Sensei’s Library (also downloadable). In the past, my theoretical study of eyeshapes (farmer’s hat, bulky 5, rectangular 4, etc) and elementary L&D techniques (nakade, throw in, oshitubushi) really helped me. There’s still very much to learn, from notchers to the dreadful carpenter’s square.

Though memorizing games isn’t my main priority now, I’ll add my collection from time to time.

Google now searches source code! The URL is http://www.google.com/codesearch. Other than indexing easily accessible source files (.cpp, .cs, .java, etc stored in a web server), it also searches inside a compressed file (like .zip) and code repositories (like cvs).

Anyway, I got 2 hits on Google code search :)…

I learned all grade 1 and 2 kanji some time ago. I could recognize the shape, know at least one reading and its meaning, and draw it properly. However, because lack of usage, I now forget how to write many of the kanji.

To not let it happen again, I’m going to mnemosyne all of them. Here’s how it looks in Mnemosyne:

When faced with kanji writing questions, I must switch to a program I developed for this purpose, StrokeReplayer:

I then draw the kanji in the top left canvas by holding the left mouse button. Right clicking will delete the last stroke and middle clicking will delete all the strokes. The stroke order and direction can be replayed:

How do I know whether I answered correctly? Just enter the kanji or its JIS code in the textbox and the references will appear:

The default is Taka which uses Java applet. To see the reference from KanjiCafe which is an animated GIF image, just click its link:

The last reference is just a still text which shows how the character is rendered in two fonts (MS Gothic and MS Mincho):

I’ll probably add other kanji stroke order database when I download them.

After the review, I must go back to Mnemosyne to grade myself. The answer section of Mnemosyne shows the correct character. This is so that I do not accidentally test the wrong character, for example 生 instead of 活 which both sounds and mean the same (ignoring nuances). The difference is in its grade which is shown in the category, but I can picture myself overlooking it or forgetting which is which.

Yes, this process is more complicated than it should be. Mnemosyne seriously needs plugin support or at least hyperlink and applet support.

Anyway, I’ve finished drilling my first 1500 words. I’m too lazy to give an accurate count, but I seem to score above 95% in my first drill (both the Japanese to English and the reverse version). Many mistakes are because of string inequality, for example answering “left” instead of “left hand side” and “to think” insead of “to think about”. After that I just repeat the stuffs I got wrong. I then repeat the whole thing again and again until I decided that it was enough.

I’m now writing the stroke order of grade 2 kanji in my reference book, moving them from the mixed and messy “reference & practice” book. After that I’ll add them to Mnemosyne. I’ve done the same thing for grade 1 some days ago.

After the grade 2 kanji is done I’ll finally advance to learn something new: grade 3 kanji!

PS: The stroke replayer is pretty cool. Since the example used a really dull character here’s another one: