# [NTLK] "Training" the newton's handwriting recognition

Jeremy O'Brien obrien654j at gmail.com
Thu Feb 4 21:06:51 EST 2010

```On Thu, Feb 04, 2010 at 07:28:33PM -0500, Larry Yaeger wrote:
> At 6:30 PM -0500 2/4/10, Don Zahniser wrote:
> >At 5:36 PM -0500 2/4/10, Jeremy O'Brien wrote:
> >>The newton I own was previously owned by another person, and I think it
> >>knows his handwriting better than mine, as it frequently thinks I am
> >>writing capital c's, s's, y's, and m's, when I am definitely not.
> >
> >You didn't specify, so I'll point out that the Printing HWR setting
> >does not learn your handwriting.
>
> At 6:34 PM -0500 2/4/10, Aaron Brigati wrote:
> >If you're using the Print recognizer on a 2.x Newton, it doesn't
> >learn. It's as good as it's going to get from day one.
> >
> >Only the Cursive recognizer learns.
>
> Just FYI...  Those statements are entirely correct except for one trivial exception:  The Print recognizer does continuously update it's estimate of your average character height, which helps with case disambiguation, which it sounds like Jeremy is having the most trouble with.  However, this adaptation happens on a short time scale, in case people alternate between writing large and small text, so there should be no lingering effect such as Jeremy is experiencing.
>
> If you are using the Cursive recognizer you already have good suggestions from Don and Aaron.
>
> If you are using the Print recognizer, I can suggest one tidbit that might help...  The Print recognizer is (overly, IMO) sensitive to the relative heights and positioning of adjacent characters.  So if you are writing the word "cat" and the top of your 'c' is ever so slightly higher than the top of the adjacent 'a', it is (much too) likely to decide you have written "Cat".  Knowing this, if you make yourself always write the first letter of a word starting with a lowercase letter just a tiny bit smaller or lower than the other letters, you will be much more likely to get what you want.
>
> Another team member developed the geometric context model, and while it was definitely a win overall, I think it still had a few quirks that never quite got ironed out.  Sorry about that.
>
> - larryy

That is very informational. Were you part of the newton team?

```