Re: NTLK Jot v Calligrapher v Newton

From: Paul Guyot (
Date: Tue Jun 06 2000 - 11:07:16 CDT

> > From what has been reported, Apple is holding on to Rosetta in order to
>> use it elsewhere. Of course, that's just a common sense guess based on the
>> simple fact that they're not selling it to anyone. The only thing I recall
>> was a recent report (forget where the link was to, MacOSRumors? Go2Mac?)
>> that trackpads on upcoming powerbooks would have text recognition of a
>> sort, meaning you could use them with a stylus. Who knows whether that
>> sort of thing will ever reach market.
>I have yet to see handwriting recognition as powerful and cool as Rosetta...
>Calligrapher is pretty good, but very much dependent on the size of the
>included word-list. Rosetta recognizes letters, regardless of language.

Sorry to say that folks, but Rosetta is primitive. It uses
Multi-layer perceptrons which were trained once, instead of allowing
further learning. (Rosetta does not learn on the contrary to
Paragraph). Paragraph's learning mechanism itself is primitive. (it
has several neurone networks per letter and change the frequence of
each, instead of modifying the network's weights on the fly).

The problem with MLP is the learning time. However, the Newton is
fast enough to get much better results with larger networks. But this
is not all.

The second mistake in Rosetta is the absence of post-process.
Paragraph relies heavily on it, while Rosetta & Graffiti don't even
use dictionaries. (hence alt.rec).

The third mistake with Rosetta is that, from what I have guessed when
reading the papers about it, there is no pre-treatment. Points are
stuffed as is in the networks. (that's also a problem with Graffiti).
Without pre-treatment, or small pre-treatment like scaling, it is
much easier to get results with a specific alphabet like Graffiti's.
Hence the plus of Rosetta is that they made a very good alphabet
without requiring to learn an alien one.

HWR technology evolves very fast today, like in fact any
form-treatment intelligent process. The previous uses were industrial
check of pieces, then military automatic treatment of videos, and HWR
will very probably make fast progress in the next years.
Pre-treatment is commonly used in other application. The most common
treatment is a simple FFT.
Post-treatment is currently studied in some centers including INRIA
in France. It is also linked with speech recognition & synthesis.

If Apple want to use Rosetta in another PDA, they will have to do that quickly.



P&M Consulting Newton Program
NewtonTalk brought to you by:

EVOTE.COM -- the ESPN of politics on the Internet! All the players, all the news, and the hottest analysis and features (plus 'toons!) anywhere.... visit today!

*************************************** Need Subscribe/Unsubscribe info?

Visit the NewtonTalk section at

This archive was generated by hypermail 2b29 : Sat Jul 01 2000 - 00:00:05 CDT