Quite often, I can guess characters without being able to visualize them. I just randomly paint a few strokes, Skritter modifies most of them into the correct form. In these cases, although I wouldn't be able to actually write the character, Skritter returns statements like "Perfect".
Similarly, albeit not as often, the reverse happens as well. I clearly know how to write a specific stroke, but due to a ridiculously small difference (that I may even fail to see) Skritter does not accept my way of writing it.

Stroke order recognition is as strict as the configuration allows it to be, but still I'd like to make Skritter even stricter. Are you working on more sophisticated ways of modifying Skritter's "strictness"?

I'm not currently working on making it stricter. We used to have a slider for that, but it didn't work very well and was confusing so we got rid of it. We are mulling an idea for adding an optional time limit to prompts, which would make it harder to do the futzing around trick.

I periodically make improvements to strokes that don't recognize when they should, though. So that may help with the reversed problem.

My suggestions would be to try not to randomly paint a few strokes or to mark yourself wrong when you do this and you didn't know it.

Thanks for your quick reply, very helpful.

Do you have a help page that explains how the script determines whether a character is right or wrong?
Is the script able to quantify the difference (e.g. distance) between the wrong stroke and the "actual" stroke? At first I thought a character is always counted as wrong as soons as the script suggests the next stroke, but apparently this is not always the case.

I always wonder how many wrong strokes are "allowed", and whether this depends on the total number of strokes or the "graveness" of the mistake.
Is a mistake counted as less serious if a stroke is right but written at the wrong time (wrong stroke order)?

If a stroke is accepted, there's no penalty--if it's out of order, it will just pulse the stroke that it should have been. All accepted strokes are the same and all rejected ones are the same, in terms of the character's correctness.

You're able to get n / 4 + 2 strokes wrong before a character is marked wrong for you, where n is the number of strokes in the character. Now that we have grade 2, it should probably switch to that before it goes all the way wrong, but I haven't done that yet.

Hope this helps!

Great answer, thanks!
I'm new to Skritter and really intrigued by the areas where Chinese and computer science intersect. The way your website works would have been unimaginable just a few years ago, and the future prospects are bright.
I can imagine that studying Chinese characters without tools like Scritter (especially on Smartphones, iPad etc.) will become a thing of the past during the next years. So keep working and make sure to retain your share of the market!

Something else I would suggest is: be stricter with yourself. If you couldn't recall the character, mark it wrong even if you can guess the strokes.

I think the optional time limit is a great idea. I would love a "blitz mode" where you have, say, ten seconds to write the character or else it counts as wrong.

I have to say that for me the right/wrong grading that Skritter provides is wrong about 50% of the time, which is to say it's about the same as flipping a coin.

The issue with strokes snapping into place and thus giving hints is a serious one. Many, many times I've been able to guess the rest of the character based on a completely random first stroke.

In the past people have suggested that we should have to draw the entire character before any of the strokes snaps into place; this could be accomplished by leaving the user's strokes on screen and hiding the snapped-into-place strokes until the character is complete.

James

I'm actually OK with Skritter's right/wrong grading as is.

I find that outside of clicking the 'Show' button for a character, Skritter rarely realizes that I got a character wrong, so I have been policing myself as I practice; I mark characters wrong when my mental image of the character is different from what it should be.

As jww1066 said, there was talk about not snapping the strokes into place to prevent hints, but I'm afraid that if that happens, Skritter won't be able to recognize many characters at all.

There are a lot of times when I adjust my stroke placement based on where Skritter moves the first character, and other cases where Skritter misreads stokes and recognizes them out of order.

I used to be concerned about snapping into place and learning because of that but I find that snapping into place over time helps to make my writing more standard (especially in terms of spacing).

I've also become somewhat more relaxed in terms of what self-marking means - a wrong means you want to study it more and a right means you want to study it less. This might mean Skritter says I've ‘learned’ fewer characters but to me that is not the end goal (though it can be a motivator to see it).

Skritter does a pretty lousy job of identifying when you've flubbed a character and it also picks up a fair share of false positives where it marks you wrong because the stroke isn't recognized. I think 50% is probably too pessimistic, but it's certainly bad enough that self-grading is crucial. But then, self-grading is already in... and it works fine.

I don't think there's any strong reason to tinker with this part of Skritter right now. It works well enough as it is, and there are so many other things (like full Pinyin practice) that are more important.

Before we had yellow grade, I was very concerned about Skritter prompting me with snap-to-position strokes. But now I figure, if I feel prompted, I didn't really know the character. Mark it yellow and keep going.

I agree with the statement that marking characters as "wrong" is essential. However, since the automatic recognition represents a/the core function of Skritter, I do think that the frequency of this issue should diminish over time. I would probably switch to a competitor as soon as I knew of a service that surpasses Skritter in this respect (combined with Skritter's outstanding usability).

Applying the objectivity and statistics that only a computer can provide to the practice of Chinese characters is Skritter's key strength. It is Skritter's competitive advantage and the single reason why I signed up (Pinyin practice etc. may be nice but not they are definitely not the "killer features" that would motivate me to pay for this service). The more subjectivity can be eliminated the better.

Part of the challenge is that reading arbitrary handwriting is still an unsolved research problem (if it were solved, those annoying CAPTCHAs where you have to write out two words to prove you are human wouldn't exist - spammers would just use OCR - text-to-character - software to defeat them). If I use Vista's character recognition or the iPhones or anybody else's I get a menu of several characters it thinks I wrote (even when I wrote a character 'perfectly') and I have to pick which one I want. So I would agree, that Skritter could improve (and I'm sure it will), I think that self-marking is always going to be slightly better than auto-marking.

-no real new insights here, rather an Amen to other entries -

I used to expect that Skritter would accurately perceive my knowledge of a character, that its recognition algorithms would read my brain and teach me wisely. Ahhh, if that were so! If there were an easy route to seeing my weaknesses and sharpening those skills.

Yes, Skritter will improve, the designers are capable and wise! But my own humility must enter in... I must grade myself - was that just lucky? or a clever reading of what "snaps" into place? Yes, we have become clever users of this tool. And we must, in order to get the benefit, not just the rewarding climbing count of 'characters learned', assess our own abilities.

quoting 2shanghai, "self-marking is always going to be better than auto-marking"

...and humility is a good instructor.

Yip, when I think about it, even if Skritter didn't snap into place and it waited until it thought you were finished you could still write 休 when you though you should write 体 and Skritter would be none the wiser unless you told it (of course if Skritter paused an annoying few seconds or required a "done" button it might figure this case out but that would be extremely annoying). I've made this particular mistake on occasion (and yes, I've been honest about reporting it to Skritter - if I wasn't I'd only really be fooling myself).

Another little problem is the difference between say 午 and 牛. The exact same strokes are valid for both characters as far as Skritter is concerned. I know it's easy to put the effort in and remember the difference yourself, but it would be cool if there was some way for Skritter to help. Maybe just a reminder to check that you got the difference right? Or a "similar characters" mode than could take you through all such characters in an explicit sequence to increase awareness of the differences.

I find it satisfactory. I like the way it moves my strokes into the right place - it helps me improve.

As an utter noob, I'm surprised how I could basically fudge through because of the snapping. It wasn't like they had to be even close sometimes, like the direction of a little 'drop' being completely wrong.

I'm of two minds. It's actually nice to make progress, like having a gentle reminder even when you're being utterly lame. It's helpful for learning but on the other hand... when I've got stuff absolutely wrong at a higher level or for characters I'm supposed to actually know, I feel it should be more strict.

In fact... it'd be kind of nifty if that's how the grading worked, it got more nazi on older stuff you should know already. If that was possible anyway.

I'm a little baffled by how the matching works. It seems entirely appropriate to completely resize/move first strokes but after that, if you're actually drawing something on an opposite direction (like a diagonal drop being the wrong way) it should just plain be wrong?

It should be wrong, but the recognition on those little hooks isn't very good--it is too lenient and usually accepts any direction. I've generally tried to fix recognition issues which are too strict and haven't improved the overgenerous ones.

I'm wondering how much effort should go into improving the handwriting recognition, hence the recent poll. There's a lot of stuff I could do if I made it a big project again. But perhaps going component-based on the recognition would be all that'd be needed.

@nick - Once you have component breakdown set up, how will it handle the case where you think the component is on the left, but it's on the right? Or if you think the component is the whole character? Those are pretty frequent problems for me now.

For example, I had a lot of trouble with 义 vs. 仪; if I'm supposed to write 仪 but start out by writing 义, it would presumably snap into place on the right and I would then see that it should be only a component of the whole character, right?

Another issue has to do with an issue that someone else already pointed out (can't find the comment right now), which is that if you think the character is 本 but you're supposed to write 木, Skritter will mark you correct; if you write the extra stroke quickly, then Skritter will even advance you to the next character.

James

If you wrote the component out of order, I think it would just reject the whole thing. At that point it usually isn't a stroke order problem. But we could try different things.

I don't think it'll ever be possible for Skritter to know you're wrong when you are intending to write 本 instead of 木. We would have to add a "submit" step for that to happen--not worth the slowdown.