Character recognition (was Re: Return...)
Alan Kay
Alan.Kay at disney.com
Sat Nov 4 14:25:18 UTC 2000
Thanks Duane --
I looked at the patent and I don't anything particularly new,
especially with respect to their claims. Gabe Groner's 1960s recog
had all the features in Xeerox's and in Grafitti. (And BTW there are
any number of good free recognizers that have been derived from the
GRAIL one and from Ken Ledeen's, like a wonderful one by Steve
Purcell at MIT/Harvard in the 70s).
The reason I had not heard of Xerox's is that I left Xerox in
1980 and the patent in question was granted in 1997 to someone I
don't know.
Here is a pernicious situation, where the incompetence of the US
Patent organization has loosed yet another nuisance out into the
world. Because of the clear prior art from the sixties, most modern
character recognition will be found derivative of ideas that have
been in the public domain for decades. But why should companies still
be allowed to take actions in such cases that then will require time,
energy, court costs, etc., to merely get back to ground zero? You can
imagine that I have been involved as a deposition witness in many
many dispiriting disputes over ideas whose provenance is clear, but
that have been wrongly patented.
I believe that the entire US patent system has been broken to a
deep enough extent for a long enough time to be absolutely useless in
carrying out its charter functions. Fortunately, they weren't willing
to even start patenting programs until the very late 70s or maybe
even the early 80s. The first suits involving SW that I had to
testify at were in the early 80s, involved "look and feel", and the
deep legal precedents that were used were based on a ca. 1910 patent
on player piano roles(!) This is how bad the whole area is. It's like
radioactive pollution seeping into the drinking water.
Here is a wierd idea with regard to this particular subject. Gabe
Groner's recognizer fit into 2K of the 360/44 at RAND that was used
as a single-user (yes!) machine for GRAIL. I don't know where Gabe is
these days (or Ken Ledeen), but like the rest of us geezers from the
60s, they are likely still alive and kicking. Squeak can emulate the
various machine codes of the 360 much faster than it could run them
itself -- Dan's emulation of the Alto to make Smalltalk-72 run again
is 20 times faster than the original Alto.
So getting the original code from Gabe and emulating it would
give you a workable recognizer that CAN'T infringe those patents.
Come to think of it, somewhere we do have the Smalltalk-76
Ledeen-type recog code, and it CAN'T INFRINGE for the same reasons.
Moreover, I believe that any recog simply implemented from one
of these writeups, like Gabe's and like Sproull's of the Ledeen),
can't infringe -- though anyone can be paid to argue this, if
meaninglessly.
With regard to the current recognizer (and most recognizers). If your
tablet supplies subpixel resolution (the good ones do) and it is set
to bypass the "convert to mouse" SW, then you can draw very small
characters and have them recognized (you should antialias the strokes
also).
However, Nathanial's is so good that I've been using it instead
of the one-pager now in Squeak. With a smart recog like his, one
careful training session will fit most people without any further
training.
It would be a good idea to trace the provenance of Ken Perlin's
ideas. The particular idea of using direction strokes as meanings has
been around for a very long time, so this could be yet another
nuisance patent ...
Cheers,
Alan
At 5:35 PM -0800 11/3/00, Duane Maxwell wrote:
> >>Sorry to "poop in the punchbowl" again, but I've recently been reviewing
>>>the rulings in the Xerox v. Palm Computing case wherein Xerox attempted to
>>>assert its patent rights in single-stroke character recognition systems.
>>
>>Do you have a reference to this patent (I wasn't aware of any patent).
>
>The Xerox Unistroke patent is Patent No. 5,596,656, received January, 1997.
>The inventor is listed as David Goldberg. Xerox sued 3Com in April 1997,
>and the case was dismissed in June 2000. The patent was held valid, but
>3Com was found non-infringing due to certain technicalities, namely the
>occasional use of multiple strokes (for instance, X), and identical strokes
>being interpreted differently based on where they're made (ie. letters
>versus number areas). In my non-legal opinion, 3Com won because of
>mistakes that Xerox made developing the patent - assuming the patent had no
>prior art.
>
>http://www.delphion.com/details?pn=US05596656__
>
>I'm afraid I can't find the patent that Palm has, though I remember seeing
>it - it was apparently acquired from some external developer, back when
>they were with 3Com and it was still called US Robotics. The code was
>apparently originally developed and deployed for the Newton.
>
>>Not possible. The current one was derived very directly from the
>>original GRAIL recognizer at RAND done by Gabe Groner in 1965 (and
>>written up in a RAND report in 1966). Sponsored by ARPA and free to
>>all (as it should be). It was the original and truly great single
>>stroke recognizer. I have a great movie of it from the late sixties.
>
>I'm sure you're right, but at the moment, the patent is in force.
>
>>Patents are granted willy nilly these days. The patent office long
>>ago wimped out on trying to vet patents and have left matters up to
>>the courts (a terrible situation that gives rise to an infinity of
>>useless and usually meaningless disputes).
>
>Yes, we know :)
>
>> This is why there are a number of companies in Silicon Valley
>>that do nothing but demo prior art on old machines.
>
>I think many people would be happy if this group of patents were invalidated.
>
>(Re: QuikWriting):
>>Well, maybe I should try to talk to Ken about all of us being able to
>>use it. (I'm personally not a big fan of QuikWriting, but it would be
>>fun to have in Squeak.)
>
>It was a very quick hack to write (about two hours) and is somewhat more
>reliable than the current recognizer in that it doesn't need to be taught.
>Other than that, I would agree that it falls in the "isn't that neat"
>category with a number of other Squeak features. We developed it because
>we have some keyboardless tablets and didn't want to teach anything -
>besides which the current recognizer needs some "breathing room" that makes
>it unsuitable for filling in little text boxes.
>
>Good luck with Ken - although I think he assigned it to NYU and the last
>time I looked, they actively peddled licenses on their site.
>
>-- Duane
More information about the Squeak-dev
mailing list
|