Gary made a diuthyrambic presentation of alt.rec.... Thanks Gary. (BTW,
I've credited your account as we said).
>>alt.rec? Sounds like a Usenet newsgroup.
>alt.rec (stands for ALTernate.RECognizer) is an HWR recognizer improvement
in fact, it is also based on Usenet names, hence the points of suspension
>for the printing HWR recognizer written by Paul Guyot of P&M Consulting (
>http://www.pnm-consulting.com/newton/) for the Newton Community.
[I removed some parts not to blush]
Actually, alt.rec....'s goal (in the final version, always remember that
all versions are alpha versions, and actually, 0.6a2 does not work [0.6a3
is already available]) is to offer an alternate recognizer for the Newton.
It tries to fits three lacks of the usual recognizers (not only on the
Newton, but in general):
a) recognition does not include grammatical & lexical (just Rosetta, since
Paragraph recognizer uses dictionaries) use
b) the Newton recognition's interface (i.e. correction popups, spell
checks, expansions) can be improved to allow fast HW input
c) the recognizers can be improved in some technical points (typically
linked with our researches in IA)
Unfortunately, point c cannot be very easily immproved until we haven't
discovered some essential points of the OS (including how to implement
properly our own recognizer)
Hence, the current version of alt.rec.... focuses on points a & b. Hence
also our researches about Newton dictionaries, which lead to DictMgr.
Therefore, improvement in recognition is today very felt with Rosetta only
(although with ParaGraph, alt.rec.... offers an intermediate position
between letter-by-letter and words only). But we'll add soon intelligent
tools (typically based on a simple Neuronal Network in the first step, then
a shell for grammatical interface, but that may come later) for lexical
use. In fact, the interface has been built from the beginning. We just have
to port our Neural Networks library (this also means optimizations, etc) &
test it for this application.
>It [ParaGraph]
>extensively uses dictionaries and such to interpret handwriting and compare
>what it can decipher to the dictionaries for potential matches. The biggest
>problem with the pre-2.0 Newtons is that they could not hold very large
>dictionaries, so the recognition was not very effective (it sucked!)
Moreover:
(a) it is time based and not horizontal position based (i.e. it can
recognize Hello if you write olleh from right to left): this is a problem
with the disapearing strokes
(b) it does not recognize accents on US/Int'l Newtons
(c) hence, but only when, like me, you are forced to use Rosetta, and you
separate letters, you realize that Rosetta is much better than Paragraph
WITH LEXICAL use (i.e. with alt.rec....)
>The standard Rosetta (printing) engine uses a letter-by-letter conversion
>process and doesn't use dictionaries at all. It has a much better algorithm
>for character recognition and is generally much better than the paragraph
>engine.
Totally agreeing.
>The software itself [alt.rec....] is a difficult piece to code and install.
Today, I changed the implementation to make it more compatible.
Technically, alt.rec.... patches Root method viewCorrectPopupScript as in
later releases, but now uses a keyboard to get the viewCaretChangedScript
message instead of using an idler object and then doing GetKeyView to
register views as they come. Besides, it does not change the view's protos
to patch viewChangedScript but it uses TieViews instead which is much
cleaner (hence, no more problems with Avi's Backdrop)
Thanks for your support,
Paul
---- P&M Consulting Newton Program http://www.pnm-consulting.com/newton/****************************************** This month's NewtonTalk brought to you by:
EVOTE.COM, the ultimate Political Junkie site on the 'Net.
The Clinton Administration, the George Bush 2000 Campaign, and almost every other major U.S. politician has said something nasty about us at some time. Find out why at: http://www.evote.com
******************************************
Need Subscribe/Unsubscribe info? Visit http://www.planetnewton.com
This archive was generated by hypermail 2b29 : Tue Feb 01 2000 - 00:00:45 EST