ChristopherFritz's Study Log

[Elizabella]

@ChristopherFritz

I’m saving so much money by having lifetime

True that. I’m of the mindset that this language is already challenging enough without needing to add speed to the equation. I follow the Kanken thread in the community and there’s a member who aced the highest level of the exam last year. Anyway, he joins a tiny group of non-Japanese who have done this but what’s interesting about his journey is how bumpy it was. And how many YEARS of studying it took for him to master the parts of the language that are on the exam.

He made a video if you’re interested:

https://www.youtube.com/watch?v=osWAemVGefY

I think there is something to be said for using an exam like this as a learning tool, meaning using each level as a benchmark for mastery, which appeals to me. I’m not sure that I want to study for an exam like this for real but it could be a cool bucket list line item…

[ChristopherFritz]

I think I’ve come up with a way forward for me with WaniKani.

Using the Flaming Durtles app, I’ve put together the following settings:

Lessons:

• Order: Level, Type
• Reverse Order

Lessons will come up vocabulary first (if any), then kanji first (if any), then radicals.

If I do a kanji lesson, then get it to Guru and unlock a vocabulary lesson, this puts the vocabulary lesson ahead of any other kanji and radical lessons.

Reviews:

• Order: SRS Stage
• Reverse Order
• Overdue items first*
• Priority: Current-level items first

Current level items come up before anything else, regardless of stage.

Next is reviewing by SRS stage, reversed. This allows me to do Enlightened reviews first, then Master, and finally Guru. Within each stage, overdue items should come up last, meaning the most recent items come up first.

* I only enabled the overdue items option after my last review session, so I’m waiting to see if it works how I expect it to.

Once I reach Apprentice reviews (which will be all leeches, as current level Apprentice are prioritized to the start of reviews), I keep doing reviews until either I reach an Apprentice item I don’t remember, or I reach the end of a session. (A session is 12 items.)

When I (eventually) level up, anything still on Apprentice for my current level will lose its priority. I’m hoping the combination of overdue items first and reverse order means my non-leech Apprentice will come up before my leech Apprentice.

I’m on day 100 of level 31. I haven’t done a single radical lesson, and I have 22 kanji lessons (plus any vocabulary lessons that unlock along the way) that I will prioritize before I start doing radical lessons. Gotta see if I can reach level 32 by June!

I look forward to upcoming manga-reading sessions filled with instances of, “Oh, I just started learning this kanji in WaniKani” again. It’s been a long while since the last one (due to lack of doing lessons).

[shuly]

@ChristopherFritz

“Oh, I just started learning this kanji in WaniKani” again. It’s been a long while since the last one (due to lack of doing lessons).

still…you’ll find your groove…I’m certain! keep it up! Took me 3 years and I’m just barely level 37…after lots of leechy death… found a way that’s working for now but level ups take 42-45 days…but at least it’s still forward progress without resetting

until the time comes wk people actually do something about the inherit deficit in their base application… (complete lack of leech management)… I’ll be looking forward to any fancy new tricks to manage the leech problem (I still love your script)…about once a week it saves me!

[ChristopherFritz]

@shuly

(I still love your script)…about once a week it saves me!

I may even be able to use it again myself one day! (But not while I have 90 leeches purposely piled up and growing.)

[Daisoujou]

You know I’ve been wondering but never actually asked, is the long time on this level about dealing with the aforementioned leeches before adding more cards? Just curious, if you don’t mind me asking.

Good luck anyway, hope the new plan pays off!

[ChristopherFritz]

@Daisoujou

is the long time on this level about dealing with the aforementioned leeches before adding more cards?

That’s exactly it.

Since I was prepping for, and then started, reading 「 かがみの孤城」, I turned my focus to leeches these past few months.

A few months ago, my Apprentice+Guru (total, not just leeches) was at around 370. After a month, that was down to about 330. A couple more months of time and effort put into my leeches, and it’d dropped all the way down to about 320 (not counting my recently-done lessons).

The problem is I spend so much time reviewing the same cards over and over, day after day. And if I do any lessons, they get mixed into the leeches.

I’m hoping these settings in Flaming Durtles allow me to start moving forward again.

Side-note: I have been adding cards to Anki, but at a very slow pace (laziness).

[ChristopherFritz]

@ChristopherFritz

I look forward to upcoming manga-reading sessions filled with instances of, “Oh, I just started learning this kanji in WaniKani” again.

And here we go.

Lesson done about an hour ago:

Didn’t figure I’d be seeing this one often.

Then in today’s manga reading:

[VikingSchism]

Ha, I also had a similar experience with that one which I just learned recently

Was watching a bit of Yuru Camp recently to see how it went (turns out a lot smoother than when I tried preciously) and 棒 showed up as you might expect when they were talking about a tent pole. Took me a little moment to look up the kanji because I couldn’t work it out from hearing it. Then I think the next day it shows up in WK. Needless to say that stood out to me when the kanji turned up in WK haha

[ChristopherFritz]

The best part is when they keep coming.

I also recently had 除 as a kanji lesson, which is one I’m familiar with from 除(のぞ)く. But I keep failing my reviews for it because I don’t know any word that uses 除 with the reading じょ.

Looks like I’ll be making a card from today’s reading:

I may have over 300 leeches among Apprentice and Guru, but I don’t care. I’m making forward progress again.

[ChristopherFritz]

Mid-month check-in: how is my reading going?

Two years ago, I aimed to do a little reading from multiple comics each day.

Last year, I aimed to do one chapter from one comic per day.

This year, I’m not imposing anything, but I’d like to try aiming for a certain amount of time spent per day (although I haven’t actually done any timeboxing yet).

Not counting book clubs, here’s how things are looking 16 days in:

I’d say this is fairly good progress so far.

Naturally, それでも歩は寄せてくる managed to be my first finished volume of the year. And just in time for the new volume to be out.

名探偵コナン is going a lot better than I expected it would. This may change later, but the series has always felt a bit heavy on words when I’ve considered reading it in Japanese before, and I’m not feeling it now that I’m actually reading. I especially thought it’d be bad as I’m using this as my intensive reading series, looking up every unknown word, and checking into every unknown grammar that I typically gloss over.

Not shown here, I finally pushed through the last few pages of かがみの孤城 week 7 today. I used to read it every morning, but some things happened at the start of the month that messed up my sleeping schedule, and I haven’t been able to get it back on track. This means I haven’t had that reading time before work when my Japanese reading ability is at its freshest.

I’m now just one and a half weeks behind the club. I need to get back to ensuring I read at least one page per day. That’s not enough to keep on schedule, but reading one page makes it easier to get in a second and maybe a third page. I need to average three Migaku Reader pages per day to keep pace with the club, so I need to double that to catch up…

[ChristopherFritz]

Slightly-Past-Mid-Month Check-In

Reading

Don’t you just hate it when you plan to do something, and then it simply doesn’t work out as planned?

Plan: Try and read at least three pages of 名探偵コナン each day, looking up every unknown word and grammar along the way. Maybe finish volume one before March?

Reality: I read 56 dialogue-filled pages of 名探偵コナン today, finishing up volume one. (And also looked up unknown vocabulary and grammar along the way.)

(Technically, I suppose that does match the plan. Just a bit aggressively so.)

That brings me up to five volumes complete this month:

Since I have a lot of different series I’m reading, I’m trying to avoid buying the next volume of anything I’ve finished until a new month arrives. But I happen to have the next volume of 名探偵コナン already, so I’ll continue it as my daily intensive(?) reading. And maybe I’ll remember to make some Anki card from it tomorrow, unlike today when I was just enjoying reading it (having read it in English about four years ago) and made only one card…

I have to admit, it’s nice to take a series that seemed like an insurmountable target and actually make good progress with it. Perhaps in the next year, I’ll be in a position to re-read the first volumes of 「GALS!」, 「ハヤテのごとく！」, and 「GOSICK―ゴシック―」?

Aside from that, I'm about on track to get back to reading かがみの孤城.

An update with the alpha Migaku Reader broke pagination in late December, so I didn’t get in any planned “reading ahead” while the club was on break. Then after that was fixed, New Year’s Eve firecrackers going off late into the night kept me up a few hours past my bedtime, throwing off my always-fragile sleeping schedule. As a result, I’ve been stuck falling asleep much later than my bedtime and waking up much later than my preferred wake time, leaving no time in the morning for reading. It’s taken me a few weeks, but my sleeping schedule’s just about back on track. This will re-open mornings for me to read かがみの孤城, during the time of day when my mind is freshest.

Wanikani

The instances of seeing new lesson kanji showing up in my reading have died down a bit, but there have been a few. (Not counting 探 and 偵 showing up while reading 名探偵コナン.)

Still, it’s great to be making progress again after three months of no progress!

Leeches:

• 171 Apprentice
• 153 Guru
• 106 Master
• 164 Enlightened

If I recognize any leeches in Guru, I’ll probably send them back to Apprentice (by purposefully making three mistakes). Since I’m only reviewing Apprentice for my current level, and I’m ignoring all other (leech) Apprentice reviews for now, knocking a few Guru leeches down would open a little bit more time for me to do new lesson reviews each day.

[ChristopherFritz]

In response to this thread by @rfindley, rather than sending an e-mail, I decided to add it to my study log. This entry is written as if specifically for rfindley (as the first parts I originally write into an e-mail).

---

I predict this e-mail to run long, so I plan to split it into two parts. Part one will be the primary requested information. Part two will detail how I do my reviews, as it will skew my leech results in an unexpected way (and my qualify my leeches as low-priority to review). And in part three I will list my thoughts on my leeches as I go through and review them. More on that as I get closer to it.

Part One

How many leeches? According to one of the userscripts I have installed:

• Apprentice: 221 (per level: 6 / 2 / 168 / 45)
• Guru: 163 (per level 36 / 79)
• Master: 115
• Enlightened: 152

Those numbers look about right to me, after I spent about three months recently really working at getting my Apprentice and Guru numbers so low.

What kind of leeches? I’ll cover this in part two, as I analyze some of my current Apprentice leeches in real-time.

Part Two

I’ve spent years working on some of my leeches.

I best learn words I’m having trouble with if they show up in manga I’m reading. But many of my leeches never show up in any manga I read!

Eventually, I “gave up” on my leeches. I was spending too much time on reviews and didn’t have time for lessons.

That’s when I wrote…the leech auto-pass script. What it does is it allows me to auto-mark my Apprentice leeches as correct without actually reviewing them. I’d still review them as Guru, but Apprentice leeches would auto-pass. The script is (purposefully) a bit cumbersome to run, so sometimes I’d just review the Apprentice leeches instead of running the script. I had two different periods of time when I used the script. The first was probably under a month, and the second under a week.

Recently, I spend three months focusing on reviews without doing lessons, but again I wanted to start making actual progress again. This time, I came up with a new approach: don’t do Apprentice leech reviews. Using the Flaming Durtles app’s review settings, I am able to arrange my reviews so I do my current-level reviews first, then my non-Apprentice reviews second. This leaves Apprentice for levels prior to my current level. Since I just spent three months on my current level, I know these Apprentices are all leeches, and I simply don’t review them anymore (starting this month).

So, that’s what shows up in my review history: a mix of “auto-passing” Apprentice leeches, and “don’t review” them.

Part Three

This part is mostly going to be for myself because it’s probably more information than you are interested in. For that reason, don’t feel bad about skipping over any of it. But if you get any kind of useful information out of it, great!

After doing my non-Apprentice leech reviews, I’m down to 188 Apprentice leeches waiting to be reviewed.

All reviews are done reading first, meaning second (but not necessarily back to back).

Note: Due to sorting, Apprentice IV leeches come up first.

Apprentice IV Leeches

Vocabulary 復習: Looking at this one, I recalled the reading. I’m thinking the meaning is…lesson? Review? Something along those lines. I settled on “review”, and found it was correct. This one’s moving out of Apprentice!

Kanji 往: I was able to type this one here before reviewing by writing 往復. This tells me the reading is おう and the meaning is “depart”. And upon review…that’s right and right. That’s another leaving Apprentice.

Kanji 絞: I can never remember if this is こう or かく, nor the meaning. Failed both reviews.

Now that I'm reading the Detective Conan manga, maybe I'll start to recognize it?

My newest Anki card I made just now. If I can link the kanji to “strangle”, then I can imagine こういち as the culprit every time someone gets strangled, and maybe I’ll recall the reading then.

Vocabulary 構う: I managed to recognize the reading and meaning for this one.

Kanji 販: I remembered the reading easily, but not the meaning. Looked like a kanji I’ve never seen before in my life. Failed meaning review. Twice. (I thought “buy” the second time around, but it’s “sell”.)

Kanji 担: Reading was easy to figure. Meaning is…bear a grudge, shoulder a burden? Turns out it’s “carry, bear”, so I’ll take as “correct”. (I’m using Anki mode on Flaming Durtles.)

Kanji 製: Another easy reading. System? Control? I’ll go with system. Reviewing…nope, manufacture. Failed meaning review. Checking mnemonic, it doesn’t help that I don’t recall the keyword for 衣 as “clothes”, and I burned that radical probably a long time ago.

Vocabulary 観客: Got both reading and meaning correct.

Kanji 態: Managed to think of the correct reading. Went with “ability” for meaning, but it’s “appearance”. I get this one wrong all the time. (We are talking leeches here, after all.) Failed meaning review.

Vocabulary 営む: Got reading and meaning right. I wonder how I get usually them wrong. I have no clue.

Vocabulary 資料: I managed to type it right on my first try here, but that was after running through ひりょう, しりょう, and ？りょう (I forget the other one) in my mind briefly. For meaning, I considered wealth (probably due to 資産家, “wealthy person”), but it’s “material, data, documents”. Failed meaning review.

Vocabulary 税金: I had to think for about ten seconds to remember the reading, but I got it right. Normally I fail it if I don’t remember it sooner than that. Once I remember the reading, the meaning is easy for this one.

Vocabulary 防止: Reading easy, meaning: I thought suspension, but it’s prevention. Failed meaning review.

Vocabulary 改正: Got both sides right. I’m not certain how I get this one wrong sometimes, as I know 改 from 「改造マリオ」. Maybe I mistake the meaning for “betterment” sometimes?

Kanji 容: A nemisis of mine. Managed to get the reading right this time. (Issue: よう? よく?) And the meaning is…valley? Bathe? Abundant? Nope, it’s “form, appearance, shape, figure”. Failed meaning review.

Kanji 僧: I managed the reading with uncertainty, but I thought maybe “increase” for meaning. (That’s 増, which I burned two months ago.) Failed meaning review.

Vocabulary 虚栄心: Failed to recall the reading of 虚, but knew meaning. Failed reading review.

Vocabulary 感覚: Struggle with if second kanji is かくor がく, but got it this time. Something about…feeling? Couldn’t recall that it’s “(the) senses”. Failed meaning review.

Kanji 典: Thought きょく, but it’s てん. That tells me it’s not music, but rather “rule, standard”. Failed reading review.

Kanji 府: Got the reading. Meaning is…something related to government or government office? Turns out it’s “government”, which I’m marking as wrong due to uncertainty. I think I have my mnemonic mixed with another “canopy” one (maybe that involves “street”?), but I don’t know which one.

Kanji 基: I happen to recognize the reading and meaning, but I know I get one or the other wrong a lot.

Kanji 格: I always know this one. Every single time. かく and “status”. So why is it Apprentice IV (now going to Guru)? I have no idea.

Vocabulary 骨折: I know why I get this one wrong! I think こっけつ, but this time I recognized that it’s こっせつ. Meaning is easy enough.

Kanji 頑: Reading I got. Meaning is…stubborn? Ah, that’s right. Yet another one going to Guru.

I think I’m discovering the secret to getting my leeches right. Stop doing leech reviews for a week or two, then type each one as I go through them, to force myself to think them through. Of course, that also means I review 24 cards in about 45 minutes.

Kanji 詩: じ? し? Poem? Ah, I got it right. Normally I go with じ and get it wrong, but this time I looked at it more carefully, and the 言 portion had me think poem, which led me to し.

Vocabulary 想定: Reading is easy, but I have no idea what the meaning can be. Failed meaning review.

Vocabulary 終電: Another easy reading. Meaning is…last train? But I think blackout sometimes.

Kanji 待: I thnk とく (特) and じ (持), but it’s たい. Worst part is I know 持 versus 待, but need to look left radical every time I see one without kanji and I need to specifically think of whether it’s a “hand” or “person loitering”. But that doesn’t help me with the reading, only the meaning. I hope this goes away after I (eventually) get to reading a lot of material without furigana.

Kanji 功: I know the reading this time. Meaning is…something about effect, or success? Or something that starts with an “a”? No, I’m quite certain it’s effect. (Turns out, it’s “achievement, accomplishment”.) Failed meaning review.

Kanji 命: めい, although sometimes I think れい. Order. Or life. I’ll go with order. Or was it pray? No, I’ll stick with order. (It’s fate.) Failed meaning review.

Kanji 算: Managed reading and meaning this time, but I know I get both wrong a lot. Most often a matter of “I have no idea what the answer is.”

Vocabulary 求める: Mistook reading as みとめる. I kind of have the meaning, request, so I try to think of someone requesting more and more (もっと) to get the reading as もとめる, but I always forget. Failed reading review.

Vocabulary 親しい: Got the reading this time. Meaning is…not “confidential”. Maybe “kindness”. Or “friendly”. I’ll go with friendly, because that one’s an adjective. Looks like I managed to get it right. This time.

Vocabulary 仮に: I thought ゆえ for the kanji, and got it wrong. Meaning wrong as well, thinking temporarily or tentatively. Failed both reviews.

Vocabulary 科目: Had to try sounding this one out a few times before I got the reading figured. And that lead me to the correct meaning as well.

Vocabulary 出社: One of the items recently added to WaniKani that I’ve had trouble with. Getting easier since I added it to Anki as well. Wait, I’m confusing this with…由来? Let’s move on. Failed meaning review.

Vocabulary 工作: I used to get this one wrong all the time, but now I know the reading and that it means “handicraft” when I see it. Somehow or other.

Over an hour in, I think that’s a good stopping point. My next review session’s going to be crazy with all these new Gurus added in… And I tend to get things wrong only once or twice in a review session, so they’ll probably be hanging out in Guru for a long time if they don’t move on up the ladder. Oh boy…

[ChristopherFritz]

WaniKani Update

My Flaming Durtles settings are now:

Lessons:

• Order: Level, Type
• Reverse Order

(Unchanged from before.)

Reviews:

• Order[1]: SRS Stage / Level
• Order[2]: Level / SRS Stage
• Reverse Order
• Overdue items first *
• Priority: Current-level items first

I do reviews with one Order, then I go into the Advanced Review Settings and change the order and do more reviews.

For non-Apprentice, I want to review everything (including leeches), and that’s what “SRS Stage / Level” allows me to do. I review until I reach an Apprentice item.

At this point, the only Apprentice I want to do is recent levels, regardless of SRS Stage. This is when I switch to “Level / SRS Stage”.

I’m making steady progress with lessons (at least five Lessons per week). My Apprentice leeches continue to be mostly ignored. (I have 253 Apprentice leeches.)

Reading Update

@ChristopherFritz

Aside from that, I’m about on track to get back to reading かがみの孤城.

Well, that didn’t happen.

Looking at my manga reading pace, I can read more volumes this year than I did last year if my budget allows it. I’ve reached a point where it’s my available cash for hobby spending that’s the constraint.

I should focus on novel reading to help continue my reading practice and progress while also keeping my expenses down. (Especially when I’m averaging completing two manga volumes per week. If I leave the furnace off for the weekend and eat rice or ramen, I can afford to buy another volume!)

I’m planning to try reworking my reading schedule as such:

1. Do not read comics until after doing book reading for the day.
2. Aim to complete one page of the かがみ per day. (At this pace, I would finish by mid-2023!)
3. Join in reading 夜カフェ and keep on pace.

My hope is that 夜カフェ being Beginner Book Club and かがみ being Intermediate Book Club translates to the former being easier to read multiple pages per day. Granted, club designation sometimes is more about the pace of reading than the difficulty of the material.

[ChristopherFritz]

I’ve finally happened upon a Manga OCR solution that actually shows some real promise.

GitHub - kha-white/manga-ocr: Optical character recognition for Japanese...

I haven’t followed machine learning too much. Just enough to get a sense of what is possible. From that, I’ve been wondering when it would find its way to OCR’ing manga. And it looks like we’re there now.

Here’s an example of a manga page that I ran through it (one panel at a time), and the output (with some mark-up I added after to highlight where I found mistakes:

 

I’d say the mistakes are completely understandable.

If I ran the manga pages through a program that finds boundaries around text in word balloons, then took those boundaries, cropped them, and fed those into Manga OCR, it should be fairly easy to get a decent automated OCR to a text file for a whole manga volume. There’d still need to be human review for minor cleanup for perfection, but the end result is a text file that can be used to analyze kanji and vocabulary in the manga with minimal manual steps.

Edit: Main issue there would be getting the word balloons in order. A predictable layout such as a 4koma would work best. But lines of dialogue being out of order isn’t an issue if it’s just for kanji and word analysis.

What’s really exciting here is that Manga OCR was able to read the banner in the artwork, even though it’s at a slight angle. It makes me wonder, what else can it read that I would not expect it to?

I ran it over Ayumu’s text on the right, then Urushi’s on the left:

田中歩高校１年

Just missing the 生 from the end.

I’ll admit, I tried this text three times (each with a slightly different bounding box), and this was the best of the three. I expect if there was a plain white background, it would have read correctly every time.

Everything below is the result of only one attempt at parsing.

八乙女うるし高校２年生

Very nice.

あれがかっこいいとか渡る大丈夫かねえ．．．あはは

It misread the name 凛子 as 渡る, but that’s probably understandable due to it being a name. The question mark came out as え as well, and it’s missing the ellipsis at the end. I wonder if I cropped off a little too much from the bottom of the bounding box.

お．．．驚いたよパトラッシュ．．．世の中にはこんなキレーな人がいるんだよ．．お前は大だからわかんないよう！！

Aside from a missing period, the final 「だろうーけど。」 managed to become 「よう！！」. Otherwise, the result is really good!

自分の危険は回避するのに

A lot of these results are images that traditional manga OCR would never be able to handle.

All right, one more:

見つからぬのか「幻の銀水晶」は

It seems that even the primary villains of Sailormoon can’t stop Manga OCR.

Yeah, I’m all around impressed.

I expect this will perform much better than Tesseract on Little Busters, so I might actually be one step closer to being able to play it. (I decided against it before because Tesseract had too many errors.)

[ChristopherFritz]

I’m starting to think this might actually work.

Original Image

Masked for text only.

Accidentally applying mask backward when trying to get text only.

Correct masking applied.

Text detection applied to find the boundaries of text, then bounds saved as separate images.

Manga OCR applied to the images.

えっ
ひょ
きょ教官ということだな
うれしそうですね
ドイン
よろしくねリゼちゃん
これは普通の女の子にはキツイよ
は
この顔のどこがそう見える！
上司に口を利くときは言葉の最後にサーをつけろ！
りょうかい
ねぇリゼちゃん
えっ
重い．．．
確かに重い
では早速この荷物をキッチンまで運んでください
あぁ
落ち着いてサー！
ほんとにバイトさんだった．．．
いいです
、
こうさんもしないことらずっと
彼女はここのバイトのリゼさんです
リゼさん先輩としてココアさんにいろいろと教えてあげてください
似合ってます

Minor issues.

1. A bounding box was found where it shouldn’t have been. Nothing text was extracted from it, so no harm there. But it’s something to keep an eye on.
   

2. There was no bounding box for the 「ふーん」 dialogue. This may be because I have a minimum size coded in for bounding boxes, and this one was a little too small. I may have to adjust that, but I risk adding in false positives.

3. This was not properly transcribed and that’s perfectly understandable:
   

4. The ぃっ in ひょぃっ didn’t make it in, but that’s fine as it’s only a sound effect.

5. A couple of graphical elements received bounding boxes, and were processed as text. However, these wouldn’t impact collecting stats on kanji or vocabulary.

[ChristopherFritz]

One week later and I’ve somewhat streamlined my process for picking kanji to learn from manga I’m reading.

The first half of this post is essentially the same as the prior post, only I’ve streamlined things a bit.

The problem: When learning kanji from reading manga, it’s time-consuming to determine which kanji and vocabulary are more common, and thus better to learn sooner.

The solution: Use OCR to find the most common kanji within the manga volume. For this use, the OCR doesn’t need to be 100% perfect. The more common a kanji is, the more likely it will be correctly recognized multiple times.

My current process is:

Step 1) Buy manga from Kobo, remove DRM, extract images, and copy into a folder to work from:

Image

Step 2) Run a Python script to pad images so the dimensions are a multiple of 32 (required for text segmentation process).

Step 3) Run a Python script to put the images through Manga-Text-Segmentation.

This takes a little time and uses up all my CPUs.

I’m glad I built a new computer a few months ago. My old one would probably take at least two hours for this, rather than about seven minutes for 130 images. (I wonder if I can run a Python script on my CPU’s integrated GPU instead.)

The end result is the mask for text.

Step 4) Run a Python script to apply the masks to the padded images.

Result

(Looks a lot like the mask image, but the mask is two-bit black and white, while the masked image is the actual grayscale or color text directly from the comic.)

Step 5) Run manga-ocr.

Step 6) Run a Python script to separate the text into separate images based on contours (real fast), which manga-ocr monitors and converts to text in realtime (which takes about six minutes for 130 pages).

manga-ocr also takes up all the CPUs, but neither instances of high CPU usage blocks me from using other software.

Images

Step 7) Run a Ruby script to:

1. Read the output from manga-ocr.
2. Extract the kanji.
3. Count the number of occurrences of each kanji.
4. Exclude any kanji that I already have kanji cards for or kanji marked as known in Migaku’s kanji add-on.
5. Exclude anything that shows up fewer times than a set threshold, as they are either uncommon kanji (within the volume) or misparsed kanji.

This gives me a list of potential kanji to learn: 俺 村 第 撮 知 繋 頑

Step 8) Check output to ensure kanji looks like it’s something showing up in vocabulary. For example, here 第 shows up in chapter titles, and 村 is part of a character’s name.

Sample search results

Step 9) Add kanji to Migaku’s kanji add-on.

Image

Step 10) Review kanji cards as part of normal daily Anki reviews. (Also look up WaniKani mnemonics if they may be helpful.)

Image

Step 11) When reading manga, I’m primed to see the kanji I’ve recently started reviewing. That makes them likely to stand out when I see them appear in a sentence. If there are no other unknown kanji or vocabulary words in that sentence, I’ll make an Anki card from it.

Image

Step 12) Review vocabulary as sentence cards as part of normal daily Anki reviews.

Images

So far this seems to be working out okay.

But I might need to cut back on WaniKani lessons again, as my number of daily reviews (not counting Apprentice leeches that I’m ignoring) has been getting a bit high again.

State of WaniKani lessons, reviews, and leeches...

Nearly 300 Apprentice leeches…

[Daisoujou]

I remain quite impressed by your methodology when it comes to this stuff. Mine is truly thoughtless by comparison; I’m as far away from you as possible. I come across words and go on pure, dumb, gut feeling. Do I like the look of this kanji? Do I want to learn this word right now? Do I think I’ll remember it, on the basis of almost literally nothing? I’ll probably know some words in the future, but no guaranteeing what sort of words they are…

[fallynleaf]

Same, haha! If I notice a kanji in something I’m reading isn’t in WK, I’ll add it (and the word that it’s in) to Anki, but I frequently don’t even notice. I figure I’ll be adding a lot more once I’m level 60 and anything I don’t recognize must necessarily be a non-WK kanji, but until then, I’m not bothering too much. I have so much vocab still left to learn, I just add anything that has kanji I already know, regardless of how common the word is. With the kind of stuff I read (at least, the only stuff I’ve been mining words from currently), the innocent corpus number and other frequency indicators are pretty much utterly useless, and I don’t exactly have a fixed body of text I can run calculations on, so I just grab up whatever I see and let the WK level be the hard cap on whether or not I add a word.

But my vocabulary is also much smaller than both of yours, which means almost every sentence has multiple unknown words in addition to unknown grammar, so I’m trying to just throw enough words into the Anki soup in the hopes that eventually sentences will make more sense, haha.

[ChristopherFritz]

It’s been a long journey trying to find what works for me.

Somehow, I reached a point where most of the manga I read, I only require minimal lookup for unknown words (when I can’t infer what their meaning may be, or when I want to verify).

I know that years of learning vocabulary via SRS was mostly a waste, as I wasn’t reading yet and thus I forgot most of the words and I never learned any kanji.

I know that WaniKani works for me, except for all the leeches that sap away vast amounts of time. I’m pretty much brute-forcing my way through what I can here.

I know that when I learn a new word from a manga, oftentimes I’ll continue to remember which manga and scene where I learned it. By “learn”, I mean actually learn, not “look up, read meaning, then forget moments later”.

This applies to new words picked up watching anime as well. I learned 降(お)りる and 落(お)ちる (the spoken words, not the kanji) watching the first several episodes of Ojamaj Doremi a decade or so ago.

None of the “what works” has been anything intentionally reproducible (outside of WaniKani). Ah, congrats on passing me by on WaniKani, @Daisoujou =D

Regardless of whether anything I try works or not, I enjoy the methods of trying to find different potential solutions, and I love the programming challenges.

But, really, here’s hoping something works out.

@fallynleaf

With the kind of stuff I read (at least, the only stuff I’ve been mining words from currently), the innocent corpus number and other frequency indicators are pretty much utterly useless,

Maybe a kind Japanese author will write a 20 volume light novel series titled, “Every time I go to sleep, I wake up in the body of a different pro-wrestler, and it’s messing with my love life”, which can then be used to develop a pro-wrestling frequency list from =D

[Daisoujou]

@ChristopherFritz

Ah, congrats on passing me by on WaniKani, @Daisoujou =D

I appreciate it! Though, funny timing, I just posted in my study log that I’m experimenting with learning kanji on my own from here and just doing my reviews on WK for now. Going to see if I’m ready to handle that well enough. So I’m stalling out on levels too, just stopped slightly after you, haha.

@ChristopherFritz

I know that when I learn a new word from a manga, oftentimes I’ll continue to remember which manga and scene where I learned it. By “learn”, I mean actually learn, not “look up, read meaning, then forget moments later”.

That right there basically being the reason, so that I’m not grinding WK words I don’t feel like I know super well. We’ll see if I’m truly good to learn kanji this way totally, though.

@fallynleaf

But my vocabulary is also much smaller than both of yours, which means almost every sentence has multiple unknown words in addition to unknown grammar, so I’m trying to just throw enough words into the Anki soup in the hopes that eventually sentences will make more sense, haha.

Haha I know that experience well; I struggled to find anything good to mine from when I started. Luckily it didn’t take too long for the situation to improve, but I was using Satori Reader at the time, so I probably benefitted from it being a little more handholdy than native material in that transition.

I know I have a lot of not useful things in my deck because often they just interest me. I love the 4 kanji compounds and proverbs and stuff. They’re often not really “worth the time,” but they keep me excited about the language, so I’m gonna mine them anyway. If nothing else, I figure since Japanese words have kanji readings, I can reinforce useful info even in “useless” items, so that’s nice.