Flashcards: Remember Previously Drawn Character Images

Hello Mike,

as I was studying in Flashcards with Fill-in-the-blanks, Prompt for Characters, I had the following thought:

About 3-5% of the time I draw a character in Fill-in-the-blanks, the handwriting recognizer misrecognizes the character because of sloppy handwriting, at which point I either start over drawing the character/undo some strokes and redraw them until it is recognized correctly, or I just tap it with two fingers, which results in an incorrect answer, and from the next card, go back with the Back arrow and change the previous result to "Correct". In the midst of a learning session, I might begin to doubt whether the previously drawn character was really correct, so that is where my feature suggestion comes in:

Pleco could remember a picture of the handwriting strokes for each character that was previously entered, in such a way that when going back, one can verify by tapping on the 字 button what was actually drawn previously, and then decide if the answer was or was not correct. This would only apply to previously entered cards, not to previous characters in cards being entered at that moment. (This ought to be empty, because one usually just wants to redraw another character rather than undo individual strokes.)

I think this would fit in neatly with the current interface—no extra buttons, just an additional, logical behavior. Right now, if I tap on the 字 button in a previously answered card, I get an empty field with a blue border around it. With the new feature, the blue border could remain, just with the character looking exactly the way it was entered. If the memory requirements for this feature turn out to be too steep, perhaps only the last 10 or the last 3 card's character drawings could be kept in memory?

Here's an example of the way it looks now:

Hanzi button.jpg

Many thanks, Regards,

Last edited:


Staff member
Makes sense, though I think I'd probably also want to do something about this workflow if it's a concern - e.g. a mode to review all of your overridden cards at the end of the test to confirm you were happy with your overrides.
Oh yes, it would probably work both ways; both right after a misentry and at the end would be excellent. I would probably only use the feature immediately after a misentry, but others might do it differently. Thanks!
Last edited:
interesting. I was thinking something similar because it's a mistake I often make and have to correct.

Is the recognizer "trainable"? Because it sure seems like I am getting 目 instead of 日 a lot and I think I may have.

Having a picture to go back to (even 1) would allow me to verify that "yes, I really did write in what I thought I did" or say "yup, screwed that up"
Oh yes, exactly! I think with algorithms for speech recognition, for example, they used to be trainable, but as they improved, training wasn't necessary anymore to achieve good results. So, modern speech recognition systems don't need training. It could be similar with Handwriting recognition.


Staff member
Pretty much, yeah. It would be hard to attempt a trainable handwriting algorithm with Chinese anyway since there are so many templates to match against - both with handwriting and OCR the usual model-training AI techniques are less-than-optimal.