Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Modular Tutor Interface Using Automatic Pattern Recognition Engines

IP.com Disclosure Number: IPCOM000105465D
Original Publication Date: 1993-Aug-01
Included in the Prior Art Database: 2005-Mar-19
Document File: 4 page(s) / 145K

Publishing Venue

IBM

Related People

Bellegarda, J: AUTHOR [+5]

Abstract

A general procedure is suggested to provide users with feedback on the quality of their actions when writing, speaking or typing. The method relies on the automatic recognition of patterns produced in the course of these actions. The tutor interface is developed as a separate module with a prescribed input/output mechanism for communication with the pattern recognizer processor.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 36% of the total text.

Modular Tutor Interface Using Automatic Pattern Recognition Engines

      A general procedure is suggested to provide users with feedback
on the quality of their actions when writing, speaking or typing.
The method relies on the automatic recognition of patterns produced
in the course of these actions.  The tutor interface is developed as
a separate module with a prescribed input/output mechanism for
communication with the pattern recognizer processor.

      It was suggested in [*]  to use automatic speech (handwriting)
recognition to improve a quality of speech (handwriting) procedure.
This system requires among other things a solution to three different
tasks:

1.  The automatic recognition of the user's message

2.  To provide an automatic estimate of the quality of this message

3.  to provide an interaction with a user

      The solution of each of the tasks involves different
techniques.  The first task heavily depends on the use of automatic
pattern recognition technology and algorithms.  This technology is
different for the speech and handwriting approaches and very
different algorithms arise depending on the task at hand (e.g.,
discrete and cursive handwriting, or isolated and continuous speech)
and the peculiarities of the likely users ( children, adults), their
health (speech impaired, Parkinson disease ).  The implementation of
the automatic estimation system also depends on hardware and
operating system used.  But at the same time this second task is the
function of post decoding processing.  The third task is basically a
function of the estimate obtained in 2) and is independent on
different decoding and estimation algorithms.  The same interaction
(resp.  estimation) system could be used with different automatic
estimation (resp.  decoding) systems from 2) (resp.  1)).  This poses
the  problem on how to effectively separate these three tasks so that
these three systems could be designed independently .

      The suggested method is to develop the interaction (resp.
estimation) system as an independent module whose work is controlled
by specially designed parameters from the estimation (resp.  pattern
recognition).

      In order to implement this method the whole tutoring system
must be divided into three blocks with  prescribed input/output
mechanism for communication between these  blocks.

The first block should contain the automatic pattern recognition
system that proceeds as follows:

o   It receives as input a user's oral or handwritten digitized
    message.

o   This message M is then decoded using the following variant of
    maximum likelihood per symbol approach.

o   The decoded message S should consists of a string of units
    U1U2...Uk (e.g. phonemes or characters).

o   For each unit U the system can compute the likelihood of this
    unit U to be on the i-th place in the decoded message S. The most
    probable string of units would be those that consis...