I’m using Skritter on a Surface Pro with a pen. Here are the biggest things that stood out to me:
I really want a “right handed mode”. Having the input area on the left means my hand blocks my view of the right panel unless I slow down and move my hand away for each exercise.
The input area needs a clear visual distinction between a tone quiz and introducing a new character. Both modes show a ghost of the character, and if I just start tracing (thinking I am seeing a new word) then I instantly get the wrong answer for the tone quiz. It doesn’t help to remember that you’ve seen a character (to assume it’s a tone quiz) because it’s often introducing new words or phrases that begin with characters you know.
It would be nice to have a little more emphasis on the specific form of the character that’s being prompted. It took me a number of “errors” to learn that I had to slow down for each “xi1n” and check which radical form it wanted.
Typing pinyin with a keyboard is very inconvenient on a tablet. Especially because typing numeric tone marks requires shifting to get numbers. It would be great to be able to hand write them using the same interface as the Chinese characters.
Even if I had to use the system keyboard to type the pinyin characters, it would be nice to be able to do a second step to specify the tones with a pen.
What are the tone numbers that it expects? It doesn’t seem to be 0-5 Mandarin tone numbers. Is it Cantonese? I never figured it out because Skritter is not trying to teach you the numbers, so you only ever input them and I just got them all wrong.
(Related note, if it’s going to want you to input tone numbers it should probably show them in addition while training)
It does not like spaces (including leading or trailing), and
It intrusively edits the textbox when you hit numbers, which interacts poorly with the Windows onscreen keyboard (and/or any attempts to delete/edit)
If I type with an actual keyboard it works as expected.
Another note: Some preferences save per browser (such as pinyin input disable) and some are per account (e.g. list settings). It could be more clear which is which.
A couple more minor annoyances I’ve hit after using it quite a bit for a month:
Strokes which end outside the canvas area (web) seem to cause an extra click. That click usually ends up advancing to the next character before I see the feedback and hear the audio.
There is not enough difference between the “you missed this stroke several times, here’s the hint” fade animation and the “you got the stroke right and I’m replacing it with my canned stroke” fade animation. Maybe I find this unreasonably annoying, but it’s because it always happens after Skritter has failed to recognize a stroke several times for no reason I can see (almost always a very simple stroke), and then each time it fails to recognize I think for a beat that it did recognize but no.
The large amount of “pop-in” on every “2.0” webpage. Each page loads with no data, and then numbers start changing at unpredictable intervals. It’s hard to be sure when it’s even done (because often when it stops it’s still wrong about things, especially if that thing is in any way dependent on the current day).
A couple areas that could use major improvement:
Sample sentences should be limited in complexity to match the word (or, if you’re really fancy, the user). For example, choose sentences that use only use words within some % of the frequency of the current word. In fact, hitting (i) and looking at the words that include a character is almost always more useful than the sample sentence, but it’s more clicks away.
Adding a single word or character has way too much overhead. I want it to be more like Pleco where there is a [+] button everywhere and hitting it adds it right now.
The biggest weakness by far (IMO) with the scheduling algorithm is the interaction of new words/characters with “established” words and characters. To learn a new character effectively really requires having it presented at the proper intervals so that it can stick. When Skritter adds a new character early in a study session (mixed with reviews), it often fails to revisit the new character fast enough to really make it stick. Instead it mixes in enough other words that my retention of the new character is poor. Then it fails to adapt and I end up getting it wrong over and over. This is especially obvious at the end of a session when the reviews have cleared out and suddenly Skritter is 10x more effective at teaching me new material because there are no old reviews to distort the timing. In particular, early in a session Skritter will often show me the stroke order hint for a new character, and then instead of revisiting that and having me draw it on my own within a 30s or a minute while it’s in my short-term memory, it will review several minutes of old material (whose review timing is on the order of days instead of seconds). Then the revisit will end up being me learning it again, followed by not seeing it until it’s out of my short-term memory again.
This is some great feedback. We’ll closely look at this to make sure any areas that can be addressed are addressed. Regarding the example sentences, and the complexity matching the word, we’re currently in the process of updating example sentences for both Chinese and Japanese to be just that. The first couple of HSK lists are done, and JLPT N5 for Japanese. The next update also will change the way the SRS system is presented, stay tuned!
I’ve given a test on strokes ending outside of the canvas (on the web), and it counting as an extra click. I haven’t been able to reproduce this however. Would you be able to let me know which browser and OS you’re using?
Firefox on Windows 10. I am pretty sure I’ve also seen it on Chrome on Windows 10.
I just did a short study session and tried to end every final stroke outside the canvas area. In each case when I exited the canvas it counted as a mouse-up and finished the stroke. But it didn’t advance to the next review in each case. Maybe it depends on which direction you exit the canvas? The third character reproduced the problem: 教. I drew the final stroke by clicking at the right place and moving down/right completely off the canvas. It finished the character and advanced immediately.
(you do have to make a valid final stroke to see the bug)
As my skill improves I’m starting to notice a subtle new issue: An increasing fraction of my study time is spent waiting to load a new set of reviews. This is especially noticeable the faster I am able to go. A 2-5 second pause to fetch a handful of new reviews feels like a big speed bump when I was whipping out answers in 1s on non-writing tasks.
It seems like adding new words is happening asynchronously, so I hope that ongoing study sessions can adopt that as well. (Of course making the server faster would also work!)
No, I don’t think my queue being near 0 affects it. I notice it more when doing tone/reading/definition reviews, but probably only because those are going so fast.
It might be due to me having so many words/lists? I have a ~8000 word list from Spoonfed Chinese (paused at the moment, but still reviewing) as well as a new 3000 word list where I’m experimenting with trying to get “Skritter 101”-like word/radical synergies with automatic sorting. However I only “know” about 1000 “items”.
Approaching month 3! I have spent most of the last month or so experimenting on myself with different lists focused on different strategies.
What’s become more clear to me is that the current scheduling algorithm is not working that well for me on “leeches” or soon-to-be-leeches. I think the scheduling algorithm has a combination of deliberate minimum review intervals and slippage (due to mixing of long-term reviews) that is just too long for my own personal worst case. I have a good memory, but I have the opposite of a “photographic” memory. I literally can’t remember images. You can show me a character, I can draw it, and after even a few other characters that displace the strokes from my short term memory, I will be unable to draw it or really remember anything about it. (I can probably recognize it, but I can’t produce it)
What’s especially frustrating is that I can even identify these soon-to-be-leeches as soon as I see them. If it’s a complex character that uses components I’ve never seen and have no “name” for, I just know that Skritter is not going to succeed at teaching me that character. It’s going to show it to me over and over, spaced far enough apart that I have utterly forgotten it each time. If I could have one more knob in the Skritter settings, it would be a way to dial the minimum review interval down, maybe even to zero.
What I’ve resorted to in the last few days is trying to write a script to extract the list of my own leeches (using the API, hence my frustration with previousSuccess being wrong). Of course what I would have to do with these is come up with some better way of learning them. Some kind of spaced… repetition. Just spaced closer together.
(Actually one idea I just had is to dump them one or two at a time into a special list which I can focus-study in the hopes that the limited pool of characters will prevent Skritter from mixing in too much other stuff. I just wonder if I end up having to create a lot of lists to do that as their effectiveness drops because of accumulation)
Strokes that are too hard. These are strokes where I have had to figure out a Skritter-specific motion to get a high recognition rate. Even then, these are strokes that fail to recognize more than others (I’m only showing the simplest example, but I’m referring to everywhere they appear as parts of characters):
The second stroke of 心 (particularly in radical form). Skritter is very particular about the hook on the end.
The second stroke of 匕 (common theme, the upward/back stroke at the end?)
The second stroke of 辶, both the form with 2 total strokes and 3.
The second stroke of 儿. Needs a lot of “up” flourish on the tail.
The second stroke of 讠. Another case where I put a huge up flourish in Skritter to be reliable.
The first two strokes of 火, especially the second. Skritter seems very particular about the direction. Also, this doesn’t show up as much so I haven’t had as much time to learn how to “game” it.
Strokes that are too easy. These are strokes that are recognized when I’m drawing the wrong thing. Particularly noticeable when they are the very first stroke.
The first stroke of 广 when I have forgotten if it’s 广 or 厂 or the first part of 斤. In general any kind of shape like 亠or 宀 where the first stroke is a dot and I make a wrong, large horizontal stroke and get the dian for free.
Placement of vertical strokes where crossing vs not crossing changes the character. Things like 力 vs 刀 or 天 vs 夫 or 王 vs 丰 or 元 vs 无. I don’t think I’d want every stroke to be super strict (allowing some slop is part of what makes the current system work), but in cases where it changes the character maybe it should matter more or have a visual hint (like accepting the stroke but hiliting where there was ambiguity).
The middle part of 聊 and similar complex “upward facing C shaped” strokes. Skritter will accept just a slash from upper right to lower left. Since I find these strokes difficult to remember anyway, I feel like it counts me as learning these too cheaply. Characters with this stroke I’m probably less likely to be able to reproduce with pencil and paper and no “Skritter assist”.
Strokes that make hints too easy. The value of each stroke as a hint varies wildly and makes hinting in general very uneven. Sometimes the hardest characters I end up learning by getting a hint several times and eventually learning “I’m going to get a hint. Wait, that hint is going to be a horizontal stroke right here. Oh, it’s part of 木…” This doesn’t work as well when the hints are dead giveaways, like:
