Thursday 25 September 2008

Pic of the day: Oh no, I broke the rules! What will happen now?

Second-guessing Mnemosyne

(Still talking about the memory-boosting software here. There are several other things called Mnemosyne, after a minor Greek deity.)

I have mentioned this deceptively simple-looking little program several times over the last month and a half, and rightly so: It gives me an unfair advantage when it comes to learning large masses of simple facts, like vocabulary. Without it, I would probably never have taken the step to learn to read Japanese when I will never get paid for it. But learning any simple data is trivially easy with this tool. Sort of. Today is about the "sort of" part.

***

I'm trying to take one kanji three times a day, spacing it out so I don't mix them up. Also, if the example words contain only signs I have already learned, I add the example to my vocabulary (a separate category in the same system). So far there are few such words, but I am sure that will change eventually. So far, so good.

Yesterday morning as I looked at the list I'm taking them from, I realized that I had forgotten many of the first kanji already. That is not strange, they are mostly meaningless glyphs and I'm still human. No, the strange part is that the program had not quizzed me on them just as I was about to forget them, as it should have done.

Quick reminder here on how this program is supposed to work, and why. Until recently, we believed that memory is one thing, but it is actually two. The actual memories are laid down separately from the "index" that lets us recall them. (This is eerily similar to hard disks with their "file allocation table", even though the technology is completely different.) After the memory is laid down, it actually grows stronger for a while. The index to it, however, starts fading very quickly. The association between a glyph I have never seen before and a sound I have never heard before stands a snowball's chance in a very hot place unless refreshed repeatedly.

The new discovery that gave rise to Spaced Repetition Software was: There is no benefit in repeating data during the period where they grow stronger anyway. Only when the index fades from sight does the main memory stop growing and start fading. If I repeat it 30 times during the first hour and then never again, I will have delayed forgetting it with approximately two hours, the same as if I had repeated it once after one hour. That's a horrible waste of time, and therefore popular with college students.

The best time to remind me, then, is just as I am about to forget or just after I have forgotten, but while it still looks so familiar that I could kick myself for not remembering. Not being omniscient, the program instead adjusts itself through trial and error - its trials and my errors. The practical goal is to make sure I recall 90% of the material on an average day. This may not sound ambitious, but it is. Most people forget almost everything they don't regularly use. "The only one who remembers anything after finals is the teacher."

I seem to have noticed that the system errs slightly on the downside: I seem to remember 80% more often than I remember 100%. But here was a long row of kanji that I had forgotten, all of them. Why had not Mnemosyne reminded me? Now I stared at them and knew that somewhere in my brain their meaning had started to fade. The only way to stop it was to look them up again. But if I did that and then got them right the next time, the program would reasonably assume that I had a "glue brain" and it could wait longer next time. Dilemma!

In the end, I looked them up and read their meaning. I did not read them out loud or meditate on them, just quickly checked the translation. Since you can grade your recall on a scale from 0 to 5, I could always give them a low grade, I reasoned.

This also came to pass. Today they were magically in my review box, as if the program had looked over my shoulder. I marked them as 2, "just barely remembered, the wait was much too long". Now that I think about it, I should probably have marked them as 1, "forgotten but looks familiar", because that's what they were yesterday. So in the end I may still have messed it up a little.

But it is the program's fault! It should have asked me the day BEFORE I forgot them, not the day after. What have I done to make it believe I was smarter than I am? Perhaps it has to do with me giving it more work now. After the first bulk of hiragana started to sort themselves out, I added one kanji each day. I aim for three now. And if one of them has example phrases that contain only known kanji (or kanji and hiragana), I memorize those as well. For instance the kanji "ni/futa" (two) is used in "nijuu" (twenty= two ten), "nigatsu" (february= two moon) and "futari" (alone together= two person). So yeah, I guess there is always some resistance when you accelerate. But I still do so very gently. By the time I am up to full speed (if ever) I will likely be reviewing more than 100 items a day, compared to around 20 now.