[NTLK] Handwriting Recognition > Options > Letter Shapes

[Larry Yaeger]

If you're referring to the stuff intended to let you fine tune the recognizer for your own handwriting, that only ever affected the Paragraph recognizer (the original recognizer, later the "Cursive" recognizer).  I doubt anything can be added to that.  And even if it could, it wouldn't affect the "Print" recognizer (Rosetta / Mondello).

If this is just something that's intended to show you how to draw your letters so they'll be better recognized, then I suppose better stroke patterns might teach you to do something the recognizer would be happier with, but it wouldn't affect what either recognizer is doing.

I've never seen anyone draw a lowercase 'y' from the bottom up.  Doesn't mean it doesn't happen.  And it's possible there are even a few samples in the training data.  But I doubt it's common enough to make it especially well recognized in the "Print" recognizer.  If Paragraph used it as one of their templates it could very well help them out to draw it like that.  But I'd be surprised if that was used as a template (or recommended to users).

- larryy

> On Sep 12, 2023, at 8:07 PM, Doug <ispinn at gmail.com> wrote:
> 
> I cringe every time I see the tiny letters retrace themselves, then doubly
> cringe when I see the weird "suggested letter constructions" they imagined.
> Does anybody really construct a lowercase letter "y" looking like a capital
> "Y," just smaller, but from the bottom upward? Really?
> 
> I think it would be important to be working with Larry Yaeger, asking him
> for input and suggestions on the necessity or the benefit of adding
> additional letter shapes to the existing letter shapes sets already
> defined. I'm curious to know if adding additional letter constructions
> would immediately be applied forward in the recognizer's accurate
> recognition of the handwriting, without affecting the recognizer's overall
> accuracy.
> 
> Another way of saying the same thing might be that I'm wondering if the
> existing sets of suggested letter constructions have been compiled, and
> therefore cannot be easily added to. I seem to remember asking this
> question once before. I'm not sure what the answer is. What I'm trying to
> explain is rather difficult.
> 
> Some of the existing animations are so difficult to see that I cannot
> really detect what path or pattern is, they are so small, and the
> animations happen so quickly. That's another thing that could be worth
> expanding on, literally and figuratively.
> 
> Doug
> 
> Einstein runs on the original MobiScribe at a speed of about 1000
>> iterations. It's quite slow and not very satisfying. I have a device here.
>> Are the newer devices any faster? I doubt it. Maybe the screen refresh is?
>> 
> ----------------------------------------------------------------------
> 
> http://newtontalk.net
> http://twitter.com/newtontalk