I've only been using Skritter for about a week, and I think it's a fantastic system. I don't fully understand how the learning algorithms work, though, so I have sort of a newbie question: how and when are words from your selected vocabulary lists added to your active set of words/characters?

The vocabulary list page says "words in your Queue and words in the lists you choose will gradually be added as you practice". I'm wondering about the "gradually" part.

When I first started using Skritter with the HSK1 list, it quickly put the first three sections of the list (1 - 90) into my active set. This kind of overloaded me, so I used the "Pause" feature on the vocab page to stop the list. Once I got comfortable with the first 90 words, I un-paused the list. At this point, it quickly added another set of 30 words, which again overloaded me, causing me to pause the list again.

What I was expecting to happen is that once I got to a sufficiently high level of success with the characters in my active set, it would gradually start adding new characters, giving me a mixture of reviews for old characters combined with learning new characters. Instead I'm getting new characters in big chunks, causing me to pause the list because I'm afraid it's going to give me hundreds of characters I don't know.

Is this the right way for me to be using the system (pause until comfortable with most of the characters, briefly un-pause to add new characters, and then pause again)? Or should I just leave the list un-paused, trusting the system to introduce new characters at the "right" time?

You can use the bar at the top right to see how many words you have got left to review. Once or twice a day I go through my reviews and then stop using skritter.
When I feel my review queues are getting too short, I go ahead and add a bunch of new words by continuing even after my reviews are all done.

New words are only added after you have finished your review queue.

Yup, I'm a beginner too, and I definitely think there is a necessity to utilize the pause function often.

I pause after each chapter of a list. Then I review for a while (a few times a day) until I have a good grasp of most of them, then unpause and carry on (usually the next day).

Thanks for the input, ZachH and xiaosanyi. Your feedback prompts a somewhat-related question: What happens if I continue practicing after my review queue is empty?

From ZachH's response, it sounds like Skritter will at that point start introducing new characters from any unpaused lists. I think this might be what caused me to get dozens of new characters introduced in a short period of time. My preference is to get new characters/words in (for me) manageable doses - one or two or five or so per day. When I get 20+ new words in a short period of time, this overloads my processing capacity and makes it difficult for me to remember the new words.

It's worth emphasizing that I'm fairly new to Skritter, and so my review queue at any given point in time is only 20 or 30 words, and therefore it's pretty common for me be in a state of having finished my review queue, but still be wanting to continue practicing. So I guess part of what I'm looking for is advice on how to use this practice time effectively.

Since you are just starting now, it's normal that your review queue is fairly short. After a while using Skritter, it gets way bigger, and you don't use the pause function as much.
For example, right now I get more or less 100 reviews every 12 hours or something like that, which takes me about 15-25 minutes to go through.

So in a situation where you've got a backlog of reviews, is it the case that Skritter won't introduce any new characters until you finish the review queue?

Scott has just figured out a genius way to make it so that you can practice efficiently even after you've finished everything that's due, letting you overpractice when you want and still get maximum possible benefit. (Right now, it's pretty inefficient and often just keeps giving you the same set of words several times in a row).

Unfortunately, that genius way involves moving a lot of the word loading logic from Python to Flash, so that means I have to do it. Great!

jpo, what you can do for now is add a few new words every time you finish your review queue--don't use pause, but stop reviewing after the green bar goes up a little bit. Then you can pause it and use the currently suboptimal overpractice mode, if you like, but you're still getting new words in small chunks rather than a bunch at a time.

We're also planning on getting the kitten to add new stuff before you're quite done with your review queue, but we've got to be smart about it. I think that can be cool.

How can we use the pause fonction ? I don't find it

It's when you're practicing, under the "writing square". There you have bars showing your progress for each of your lists, and on the right of those bars you have a green pause button. If you can't find those bars, you might have minimized them, so just try clicking on the double black lines under the "writing square".

I have my HSK list paused, and deleted all characters passed the first 60, but it keeps adding characters beyond the first 60, completely overwhelming me. I posted a comment asking for help, but I imagine that since it's Sunday they might not get back soon.

Just wondering if anyone has any advice on how to completely avoid practicing any characters beyond the first 60. Thanks!

I still feel I don't get how scheduling works. Why is it efficient to add words after completing the review queue and not before? I.e. I can spend 3 hours to clear my review queue, then Skritter suddenly thinks it is time for new characters and I get a few characters. Wait another hour though and I have a bunch of characters to review, so no new ones. Shouldn't new characters be scheduled just as much as old characters, not only introduced when a review queue is empty? I.e. each character in all of my lists should have a next review date or an add date if I haven't looked at it.

It doesn't seem like whether or not I will remember a character is completely a function of time either, which makes me feel perhaps other factors need to be added, like more structured learning for characters. E.g. if I don't know how to write 青 then practising 请 will also be useful. That's another question though.

@ Andrew, I don't really understand what you mean by the "first 60". Is that the first 60 characters you've studied?
With Skritter, once you've added a word, it cannot be deleted from your reviewing, so it'll always come back (unless you erase your account :p).
If you mean you want your review list to always be under 60, then you need to add less characters at a time, and review often (say 2-3 times/day). Hope this helps.

@ximeng: I guess that if a character is due reviewing, it means reviewing it later would make it harder to remember. If you keep adding new characters when you still have some characters to review, you'll never get to the end of your review queue, and won't be able to memorize efficiently.
That's what I understand. I think the memorizing process has a lot to do with time, but I agree that mnemonic techniques help.

I see what you're saying Elie, but it seems to me that the cut-off between reviewing and adding new characters is quite abrupt. I suppose I'm wondering if it's really efficient to spend an hour only reviewing, followed by an 20 minutes almost entirely adding new characters, then repeat the next day. Wouldn't it make sense for those 20 new characters to be interspersed throughout the session rather than all at the end? It would still take about the same time to get to the end of the review queue.

I don't think I intuitively understand the workings of the SRS yet, so am still struggling to work out if it's really working efficiently and how it's doing its scheduling.

I'm sure time must have something to do with memorisation, but I also feel that there are other ways to memorise, and that Skritter will become more useful as it helps to bring out the connections between the structure of characters, or perhaps the pronunciation of characters in a way similar to how words bring out the semantic connections. For example, seeing characters with a similar sound-part together, or with similar components, or slightly different components. This would work much like the way tones are tested after the character is written. If this is difficult to schedule quickly (computationally speaking), then this could be done by putting completed characters into a scheduling queue and processing when there's time.

Actually I think the first part of what I said above is what Nick was referring to in his note.

Andrew, sorry for not getting back to you. I'm only up to Saturday night on feedback.

I think I know why you're still seeing extra HSK characters. We had it set up so that when you add a word, it adds the characters in that word, too. When you selected all the words and deleted them, it left some of the individual characters hanging out to be studied.

You can hit "view all" to get at those remaining characters to delete them. We're also changing it so that the individual characters don't get added in course, so that won't happen any more.

Ximeng, yes, we are planning on making it so that new words can come in during review, instead of waiting for review to end. It'll work better to mix them in because new words often have very short intervals, so you'll get more chances at them at first. It should also be more fun.

We just need a way of doing it smartly so that it's not adding too many new words when all you want to do is tread water and finish the review. I think the new loading code I'm working on will help with that. It's also going to make overpracticing really good, instead of broken.

But yeah, there's still going to be a lot of inefficiency in the SRS compared to when we can actually tune it to provide optimum intervals per user, per word, per part (writing, tone, reading, definition), etc. That's going to be sweet.

Eventually we might think of a way to illustrate small differences between characters so as to disambiguate them, but that'll probably be a while.

> We're also changing it so that the individual characters don't get added in course, so that won't happen any more.

Not sure I like that. My preference is pretty much the exact opposite. If I'm learning a word like 知道, I almost always want to learn what the individual characters like 知 and 道 mean. So I'd like to have it so that both the word and the individual characters get scheduled in the queue.

Part of the issue here is the overlapping learning modes that Skritter addresses. At one level, there's writing skill, which is almost entirely a character-level issue. This overlaps considerably, but not entirely, with learning character meanings and pronunciations. And then there's another level on top of that, which is learning the meaning of words composed from individual characters.

Skritter is quite good at all three of these things, although I would argue that its core strength is in teaching writing skill. Having the word-based recognition is useful also, but this is a domain that is also covered by many other flashcard-type systems. The word-based recognition is almost entirely separate from writing skills - if I know how to write 喜 and 歡 individually, then I'm almost definitely going to be able to write 喜歡, and vice versa.

My own preference is to focus my writing practice on the characters rather than the words, but I'm sure other people have different preference and learning techniques.

Hmm. What we've heard from most people is that learning to write characters is most helpful when they appear in context in the words.

How is the current approach working, jpo? Right now, the characters are added too, but almost never get practiced on their own because the words get marked wrong if any character is marked wrong, and the word usually takes priority, so it's almost as if the character level items aren't added.

We'll change this so that Skritter can estimate whether it's the word or the character that you didn't know when you got something wrong, which means they'll be scheduled independently. Then it will matter whether the characters are added as parts of words or not.

Would you prefer to not learn the multi-char words, or to schedule characters more often than the words (so you'll see them individually some times)? How would it work best for you?

> We'll change this so that Skritter can estimate whether it's the word or the character that you didn't know when you got something wrong, which means
they'll be scheduled independently.

This sounds perfect for my learning style. I absolutely do want to learn the multi-character words, but individual characters are important to me too.

In particular, as my practice list gets longer, I've been wondering how to deal with duplicate characters. For example, say I'm studying 想要 and 想法 and 想像 and 想念, and consistently nailing the 想 but missing the second character of the words. I'd like the system to schedule more repetitions of the individual characters to help solidify my knowledge of them, so I can recall them individually and also in the context of the words in which they appear.

Part of the issue is that I've only recently realized the value in skipping a character that you know well. So one way of addressing the example above would be to not write 想 each time, once I've actively recalled it in my head. I've been laboriously writing out characters I already know when they appear in combinations in words (什麼 and 怎麼 is a particular favorite, given that I know 麼 well), but what I probably should be doing is not writing out 麼 each time. This feels a little like cheating, though, so I've tended not to do this.

I'm in favor of more words, less individual characters, and I have no problem clicking through if I know the character--though like jpo, I had that cheating feeling in the beginning. Nick encouraged clicking a while back and my conscience was silenced. I don't need to prove anything to Skritter. if I know it, I know it. Click.

Unlike jpo, I find that if I know the words, it's annoying to be hit with individual characters. I think it may have to do with individual learning styles, but perhaps also the phase one is at in character learning. At this phase, the biggest concern for me is new words, not new characters--though I've plenty of them left to learn, it's mostly the case that I know the characters, but not the meaning of the word. So, for me, when I add new words to the cue and then get tested on their component characters, it's annoying.

Also, Nick, what you've heard it right. Research (at least what I'm familiar with) has consistently found that learners recognize characters better in meaningful contexts, rather than as atomized bits. Obviously, plenty of characters are meaningful on their own, but for those like 喜 and 歡 which one doesn't see except in multi-character words, it's more effective to learn them in said words. On the other hand, I don't know how much this has to do with remembering how to write, as opposed as read, characters.