I remember there was recently a forum discussion about the scheduling of brand-new items vs. recently-added items, and apparently you guys have already changed this because I notice that recently-added items are now coming up a lot more quickly.

This is a huge help. Before, it would be a little frustrating because the recently-added items would be wrong at a very high rate. Now it feels like the progress is much smoother when dealing with new and recently-added items.

Anyway, I just wanted to give positive feedback. Keep up the good work!

James

Interesting perspective. I'm assuming this is with the new practice page? I think it's due to some sort of bug, because I didn't intentionally change anything with the scheduling (except to reduce the original wrong interval for new users, one of whom you aren't).

As I fix bugs with the Genius in the new code, please keep your eye out to see if this new behavior you describe changes.

Yeah I noticed that too... And I welcome it as well! :)

I thought I noticed something like this, on the old practice page. But I think it has more to do with how slowly a lot of my items become ready now. So I might have 100 items "to review" in the blue bar but 80+ of them are like 100-105% due. So after new items get added, if I keep practicing to try to get to 0 reviews, the new items will have a chance to come up several times before I finish. Or maybe I'm just crazy :)

OK, I found the old forum topic:

http://www.skritter.com/forum/topic?id=13255732

I don't know how to link to a specific comment; it's Nick's comment of 4:17 PM:

I'm thinking of changing the kitten
to prioritize overdue items before
really overdue (new) items. So maybe
you'd see those 500% - 150% items at
the start of your session before the
kitten decides it's profitable for you
to take a crack at those 999%+ ones.

So this hasn't been done yet? I have definitely noticed an improvement recently...

James

Yup, I haven't done that yet. It should still be showing the new ones and the "still learning" (readiness > 999%) ones first.

I noticed the same thing as James and had thought you made the change. Now the more I play with it, the more I think something might be a little screwy. I have words sticking around that I don't think I have gotten wrong in awhile (好运). Usually if I remembered a word overnight, I wouldn't see it again until tomorrow, now that doesn't seem to be the case.

I've also noticed that my retention rate is way way higher than it has ever been before. I have it set to 90%, usually it would be around 91, 92% right now it is creeping up past 96.1%, I don't think I have ever seen it that high.

It's a fuzzy enough process that I could easily be wrong, but I certainly get the sense that something has changed.

I also came seem to shake da4bao3yan3fu2 and you1dian3 no matter how many times I get them right.

I've noticed that the number of characters learned is dropping in an uncharacteristic fashion (usually only happens after a long break). Looking at the % prompt, I think that words I get correct are somehow being marked as incorrect. (skritter incorrectly thinks I got you1dian3 wrong 29 minutes ago).

I use the keyboard shortcuts a lot, not sure if it might be related to that.

I've also noticed that since the re-format, new words are also being added at about a tenth the rate they used to be (this might be related to the other problems though).

It is definitely incorrectly marking most of the items reviewed. Seems like it got quite bad all of a sudden. Not sure if you made a recent tweak or something like that.

I can see why the retention rate would be higher--items score as a 2 are counted "correct" for purposes of that calculation, and if you get a two-character word wrong, it will get a score of 2 instead of 1. I wonder if 2 should be counted as wrong, instead, or perhaps skipped in that calculation. Thoughts?

New word adding is very quirky in the new code; I have a lot of bugs to fix with that yet. The incorrect marking of items is weird, though. I'm not sure why it would be marking so much stuff wrong. A few basic questions:

Do you have the grade buttons on?
How do you use the keyboard shortcuts?
What's your normal process for reviewing a word? (Like, what do you write, what keys do you hit, what scores do you select, what color are the borders when you go to the next word, etc.?)

Thinking about it, I suspect the rising retention rate has more to do with not very many words being added. Although it seems odd that the retention rate is rising as the number of words known is dropping.

For the incorrect marking of stuff, eventually I started just hitting the space bar for everything in the word but it still somehow gets recorded as wrong. (last time da4bao3yan3fu2 came up [three hours ago] the only key I hit was space but the roll-over text now says "you forgot the writing of this word three hours ago...")

I do have the grade buttons on. I typically use some combination of the number keys and spacebar to skip words I know after writing enough to be confident I know it. (write, hit space or number, write tone, click arrow, move on to next word).

I believe the border is green, marking it as known, I tend to pay attention as I fairly frequently mark myself more harshly then the default. The only indication I can see that the word is being marked wrong is indirectly, i.e. the frequency that it shows up and the roll-over text saying I had got it wrong last time.

It does seem to be only happening with the writing and not the tones.

Hope that helps,

Thanks for all the information. I've identified a case where if you skip past all the characters in a word, it will mark it wrong instead of correct (even though the border is green). I've fixed that.

Is it possible that all of the words you so noticed being marked wrong were those for which you only used the space bar/next button without using the 1-4 keys, toggling the correct button, or writing a character? If so, then I've fixed it. If not, then I may have fixed it, or it could still be broken. Keep your eye out.

That's very possible. It would explain why it didn't seem to happen with new words, just words I already knew well (and skipped through). I'll look out for it happening again though.

On a separate note you asked "if you get a two-character word wrong, it will get a score of 2 instead of 1. I wonder if 2 should be counted as wrong, instead, or perhaps skipped in that calculation. Thoughts?"

I think it probably should be counted as wrong in the retention rate, there are lots of two letter words that I can get half of no problem, but I would still say that I got them wrong. Just my thought.

Thanks for looking into it! Having you guys be so responsive is one of the most impressive features of the site (aside from the characters and all that.)

I think you fixed 99% of the problem, but there might still be something a little screwy. A couple of three character words have a yellow check next to them when I finish, even though all the writings were correct and showed up green-(I checked using the back button. The tones have a green check though. An example is :烧烤酱 - for this character I spaced through the first two and then wrote the third character (correctly).

Unless the yellow check takes something else into account besides right wrong?

Nope, it should have been green. Got this one fixed, too--will upload shortly.

Looks like you got it, thanks!

Re: "Although it seems odd that the retention rate is rising as the number of words known is dropping. "

I think that the retention rate calculation omits words that are not "known" so that if you take ones you only-sorta know out the rest will be ones you always get right, thus a high retention rate. No?

We've changed it such that the score of 2 is still counted as missing it for retention rate calculations and such. That should keep retention in line with how it was previously calculated.

I've noticed something else that seems different in the new setup. Once again it is hard to say for sure, but there are a couple of 2 character words that I never get right, but I don't think the interval is dropping any. They tend to be words I know half of really well, but half I don't know at all. (肌肉, 清晰, 附近) I think this might be what you were talking about earlier:

"and if you get a two-character word wrong, it will get a score of 2 instead of 1".

If I'm reading that correctly, that the words I get half right will be scored 2, and if the interval for the word is based on this, that could be problematic. Like for 肌肉, I always get 肉 correct, but I don't think I have ever gotten 肌 correct. With that word always scoring a 2, the interval never drops to less than a day, and I don't seem to ever learn it. I don't see 肌 independently or in any other context, so I don't seem to learn it.

It seemed to work better for me when the whole word was graded and repeated, based on your performance on the worst character.

I might be mistaken, it's hard to remember what happened on the word a day ago, but that is the impression I am getting.

Another quick example: 受伤, I know both characters, but not together (can never remember 受 belongs in this word). I just got it at 22 hours, got 受 wrong, 伤 right, and it is still at 22 hours. I think I have done this a bunch of times.

For me, the old way seemed to make it much easier to learn.

I think it would make more sense to have two-character words with a score of two stay at the same interval if the individual characters were always activated. That is, you would see 受 when it was appropriate, instead of it not ever coming up on its own since it wasn't added by itself.

Might have to try it and see, or get more input on whether to activate individual characters by default.

I agree, adding the individual characters by default would fix a lot of the problem. I think if this were an option, I would probably want it.

However, I would still like to make the case for two character words where you got half wrong being scored as one by default rather than 2. The analogy I would make is with the first few strokes of a character. Often, after seeing the first 2 strokes I can remember the rest, but I would still consider this not knowing the character as I couldn't produce it on my own. Same thing with two character words, often I know the second character after seeing the first, but wouldn't have been able to produce either of them on my own. Personally, in this situation I would want the interval dropped until I could produce them both without any extra info. Even for words where i know both characters, one will often continue to elude me in that context. For me, I would still want this marked wrong, as the only way I will ever get it is if the interval is dropped to the point where the combination and context is drilled into my head.

I think part of the reason for the change is to avoid having to write characters that are well known over and over again in various words and contexts? I would argue that you have already addressed this with the skip arrow and the spacebar, you don't have to write any characters you don't want to. For me this isn't so much of an issue, I like having to actively decide if I know a character well enough to skip it. It forces me to pay attention to it, and if it is a common character in a new word, it reinforces that context and connection.

I think the change you made makes more sense to me for some of the longer 4 or 5 character phrases (often from chinesepod). I don't think there is much value in repeating the whole phrase at shorter and shorter intervals for one character that isn't well known. For me though, 2 character words are a different story and need to be known completely.

I might be in the minority on this but I wanted to at least make my case. I can always manually mark the word as wrong, (but it does take some time to cycle through to wrong from so-so). Sorry about the long post!

The more I think about it, the more I like activating individual characters with the new system.

I think that compared to learning characters themselves, learning which characters go into a multiple-character word is much easier, especially if you're studying those characters in between word reviews. So I'm not sure that a score of 1 for the word is needed if you miss one of the characters. If the word stays at the same interval, how hard is it going to be to get it the next time?

That's true, it is much easier. I guess I just find that for me, some words refuse to stick even after seeing them multiple times at a day or two apart, even though I know the characters. I'm sure they would eventually stick though.

One question though, does a word with a score of 2 return at the scheduled interval or at the reviewed interval? i.e. 受伤 is meant to be reviewed at 1 day, but I don't review it until 2 days and I get it half wrong again (score of 2), does it then return at 1 day (scheduled interval) or at 2 days (reviewed interval)? If it is the reviewed interval, it seems to me like the interval will actually have a tendency to grow, even if I never do better at the character. If it is the scheduled interval, I'm not nearly as concerned because the interval will stay at a day or so until I eventually get it.

It uses the scheduled interval, so it won't grow if you keep 2ing it. Scott looked into it and he's fixing an edge case where this wasn't the case.

There's still a little bit of random variation for all scheduling, so that items don't clump together as much.

Sounds good to me!

One other little thing on the new interface - the word level check/score indicator for writing seems to only show up if you write out the last character not if you hit space or the right arrow (only really applicable when you are doing both tones and writing on a word).

Aha, good catch! I see the problem and I think I've fixed it. Let me know if you see further problems with that.

I must say I like the idea of any character in a word wrong makes a score of one. I have "zhuyi" 注意 and 主意 in my queue and I always get 意 right and half the time I pick the wrong "zhu" and with the scheduling algorithm I think the spacing just gets farther out over time (meaning I never actually learn it).

I think it's different if you pull in an 8 character "word"　from ChinesePod but for a 2 character word I'd rather it just be marked wrong otherwise it'll never get into my memory (well, unless I manually override the "2" or use good old fashioned pen and paper reviews but it's that automaticness of Skritter that I love).

I'd still vote for that as well, but won't cry too hard if it doesn't happen...

If it doesn't happen, an easier way to override the 2 would be nice. I've started doing that with some regularity.

One other example of why I think this is important - all the 子 words (like 李子, 鞋子，etc..) or the 果 words like（水果，坚果，。。。）the second half is usually easy but knowing it doesn't really reflect any additional knowledge of the word that would reflect a score of 2....

Good points. I might try something that scales based on the length of the word.

Looks to me like you just switched it back. Huzzah!

Yes, it now scores it as a 1 if you get one of two characters wrong, and also if you get two characters wrong no matter how long the word is. Let me know how that scheduling works out as you practice some more.