Quote:Maybe it would be better if you re-did the algorithm and take jouyou kanji with no onyomi into account this time (but not non-jouyou kanji with no onyomi).

How many "perfect" groups would there be then?

Should I include キョ/コ for 裾?

Vempele Wrote:Should I include キョ/コ for 裾?

I think you should maybe assign a fake reading of "X" to all jouyou kanji (or all kanji) that lack an onyomi reading (or the reading is missing from JMdict) and re-run the script.

Why is 映 separate from 英? They share the same reading, and 映 has the wrong signal, because 央 is read Ou. I too found the following relations that could be joined into one single group:

戈+戋 (right side of 浅)-我-戠-蔵 = sen (I might have overlooked an exception here)
臣+戈 + 蔵 = sou/zou (same here)
夹 +夾 = kyou
夆 + 丰 + 奉 = hou
古+固+胡-啇-居-克 = ko

Also, 志 is in the list, but 士 and 仕 aren't at all and share the same ON reading.
瀬 is also missing from the 頼 (rai) group, which includes 嬾 and shouldn't.

184 groups
419 perfect Jouyou kanji
587 pure Jouyou kanji
930 total matching kanji in the above groups
494 exceptions. (ouch!)
661 pure kanji in total. (over 2000 if we don't prune the readings)

士 is a component in 131 kanji. 5 of them have the reading シ, which is less than the minimum 9%. Also, the list of conditions would probably end up being longer than 5.

I now make an exception to the 9% rule for the primitive's own readings as a kanji.

Quote:Why is 映 separate from 英?

Because all they have in common is 央?

Quote:瀬 is also missing from the 頼 (rai) group, which includes 嬾 and shouldn't.

See reading_counts.tsv for the list of kanji/readings that are included at all (other than the ones at http://ja.wikipedia.org/wiki/常用漢字一覧). 瀬 isn't in there.

Why only 嬾? 頼 and 瀬 seem to be the only kanji that don't have the 刀 variant.

I think I've found a big group

一+口 (MOUTH with ONE on top): 同,荅,豆 (-司 shi, -合 kou): トウ tou (21 elements)


It also explains why the exceptions -壴-喜-吉 don't share the reading, as they have 士 instead of 一.

Edit： Also found that 務 group is a redundant subgroup of 矛 group.

Vempele Wrote:182 groups
429 perfect Jouyou kanji
593 pure Jouyou kanji

Vempele Wrote:184 groups
419 perfect Jouyou kanji
587 pure Jouyou kanji

I was expecting a larger difference. Will you update the zip?

Here's what happens if you remove the 9% rule altogether:

Phonetics.zip

198 groups
460 perfect Jouyou kanji
630 pure Jouyou kanji
995 kanji
162 exceptions.
1524 pure kanji in total.

StrictPhonetics.zip

206 groups
458 perfect Jouyou kanji
632 pure Jouyou kanji
995 kanji
609 exceptions.
796 pure kanji in total.

New column: Word coverage is sum(instances of matching readings in JMDict) / sum(all instances of matching+exception kanji on'yomi in JMDict). Note the absence of kun'yomi and words that ryuujouji was unable to make sense of.

Huh. I thought there was a bug with word coverage, but it turned out Kanjigen had ゼン セン for 全 and ryuujouji happily used the latter reading for all 521 instances of 全 in JMDict. Now changing ryuujouji so that it'll test in ぱばは order.

I don't know why there are more pure Jouyou kanji in StrictPhonetics.

I might have spotted a well-hidden group. That would be

王-玉-主-全-呈-?(left side of 理) = オウ = 7 kanji

It looks like 王 doesn't work as a phonetic component when it's the component at the left side of the kanji (the one usually carrying the meaning), but works as one in most other cases (except when it's part of another phonetic component, as in the exceptions above).

I've finished introducing the components on an Anki deck. The deck can be found on https://ankiweb.net/shared/info/3283034296. There's also a LITE version made by another person, with includes only the 95 most common patterns https://ankiweb.net/shared/info/2079428463.

The decks seem to make almost no mention of exceptions. That's bad. In particular, 東 トウ only works for three of six of the Jouyou kanji it covers. Also, I'd prefer hiragana instead of romaji.

I use a deck that randomly picks a matching kanji from the list (anyone know a better way to do this than separate note types for 2,3,4,...kanji?). I also memorize exceptions separately.

There's a field for exceptions. I've tried to include the ones that tax less the mind. I might have added some bad ON readings or groups when I added Heisig's RTK2 as a source to identify ON groups. I can update the deck later, with fixes and adding/removing cards.

I guess hiragana is an option, but as katakana is already there, I don't know if it's worth the extra effort. Romaji is only there because I picked the card HTML style format from another ANKI deck, actually. :p

Here's another group (a pretty big one, up to 28 kanji)

小+少(not at right side)+肖+尚-(一 above 小: 示,京,尗,寮,賓...): shou

when on the right side, 少 has 50% of being byou, 尚 is tou for (堂,党,non-Tōyō)

Thought I'd just update my progress.

School's been back for about 2 weeks now, and that and a family emergency have set my progress way back. I'm still at it though, and have less than 200 onyomis to go^^. I'm hoping to have finished by the end of the weekend (one big final push!), and then just reviews until all cards are 'mature'...

And then on to kunyomis I guess, though it looks like I might have to make my own deck...

I added a few more groups

? (top-left side of 有,友,右,...)-左-若-布-迶 = ユウ yuu (7-9 kanji)
龺 (left side of 乾) -朝 = カン kan (5 kanji)

Can you recommend some fonts for the radicals? I can't see a lot of them.

Oh, and thank you for the deck! I plan to make good use of it.

EDIT:
Ok, I just read the description, no need to answer my question.

killua Wrote:Can you recommend some fonts for the radicals? I can't see a lot of them.

What's your OS? If it's Windows, the main Japanese fonts are MS Gothic, MS Mincho and Meiryo; Meiryo is the only one that can't render them, at least on Windows 7.

I'm on Linux but I use Meiryo.

I just installed some Chinese fonts and I'm able to see some more...
I can't see 龺 for example.

Unfortunately, some characters are on the extended (and even extended 2) CJK unicode lists, which means that most fonts (even specialized ones) won't display them. You can paste these missing characters on http://glyphwiki.org to see them graphically (and copy and paste them into ANKI). ANKI lets you paste pictures, but it looks like these decks cannot be shared as easily as I would like (instead of pictures, you find broken image links).

I'm still proofreading the list, adding some entries and removing less useful ones as I dig through dictionaries. There might be some errors, and some exceptions are missing.

One thing I want to do, is to mark the less useful primitives so that people can easily suspend these cards.

killua Wrote:I'm on Linux but I use Meiryo.

I just installed some Chinese fonts and I'm able to see some more...
I can't see 龺 for example.

I suggest the Hanazono fonts (HanaMinA and HanaMinB) — it's one of the most complete free ones you can ever find for Kanji, with almost 90,000 Unicode characters, covering up to CJK Extension D:
http://sourceforge.jp/projects/hanazono-font/releases/

gdaxeman Wrote:I suggest the Hanazono fonts (HanaMinA and HanaMinB) — it's one of the most complete free ones you can ever find for Kanji, with almost 90,000 Unicode characters, covering up to CJK Extension D:
http://sourceforge.jp/projects/hanazono-font/releases/

Thanks, exactly what I was looking for!

DrJones Wrote:I'm still proofreading the list, adding some entries and removing less useful ones as I dig through dictionaries. There might be some errors, and some exceptions are missing.

I'm looking forward to the final version then.

Have a look at this:
http://sdsu-dspace.calstate.edu/bitstrea...Hiroko.pdf

There are some interesting tables at the end!

I'm already using that source, some of the ones listed (ex. 立) have too many exceptions to be useful, which is why I spent some time correcting them.

Another group:

厶 + enclosing radical (広,公,勾,厷)-公(on a side)-翁: コウ
足 (not on left side): ソク

I've added a few more groups. I found (but didn't include) this pattern

亻+ any character : (often) same reading as the character

The existence of this pattern means that many of the groups found by the algorithm are "fake" groups, such as 依, 仁, 傾, 働, 優... so you don't get any advantage by studying the component as a group.

Anyways, I updated the ANKI deck so that now it has 300 cards + 30 "extra" cards (marked cards). I consider the "extra" cards either "uncommon" (not many kanji or number of words in reading_counts.txt, or "advanced level" (many exceptions, readings, rules). The idea is to learn these 30 after you know the rest. You can download it now. I don't think I'll make any more substantial edits to it in a while.

Thank you for your hard work, DrJones.

So, 3 weeks later than I intended, I've now finished the 5th edition onyomis!^^
(still reviewing of course)

Now I just have to add in the new ones T_T Hopefully that doesn't take too long...

Oops! I hadn't noticed that 止 and 此 are redundant. I've merged them into one single card, and added another primitive ( 布 fu ) to the deck to make it a round number. I've also fixed some small mistakes.