Browse Prior Art Database

Method for Lowering Branching Factors in Forms Filling Applications

IP.com Disclosure Number: IPCOM000116045D
Original Publication Date: 1995-Jul-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 2 page(s) / 126K

Publishing Venue

IBM

Related People

Lukasik, R: AUTHOR [+2]

Abstract

Disclosed is a speech recognition system in which alphanumeric keys are used with database indices to limit the searches required to identify associated spoken data. The keys may be spoken, written, or typed. For example, a data entry person using the system could tell the system "DA David," or, "High school name is RE Redwood."

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 45% of the total text.

Method for Lowering Branching Factors in Forms Filling Applications

      Disclosed is a speech recognition system in which alphanumeric
keys are used with database indices to limit the searches required to
identify associated spoken data.  The keys may be spoken, written, or
typed.  For example, a data entry person using the system could tell
the system "DA David," or, "High school name is RE Redwood."

      In a typical forms filling application, a data entry person
supplies information, destined for a number of specific fields, to a
centralized data processing system.  This information may be read
orally, to be converted to text by a speech recognition system.  The
degree of difficulty of the speech recognition task is a function of
the data perplexity, which is a measure of the number of active, or
possible, competing words at a given time (1,2).  The task is
relatively simple when fields with numbers, such as zip codes, dates,
phone numbers, or account numbers, are being filled.  Pragmatic
information obtained from these fields may also be used to simplify
the filling of other fields by reducing the possible choices.  For
examples, telephone numbers may be used with reverse directories to
obtain names and address information, and zip-codes may be used to
limit searches for city and street names.  However, a typical form
also includes fields to be filled with information having high
perplexity, such as the fields for last name, first name, street
name, employer, high school name, college name, and occupation.

      In the presently disclosed system, names are sorted and
associated with their first one, two, or three letters.  The tables
may be constructed using standard grammar techniques, in which a
single call is used to find, for example, <alpha1> <alpha2>
<(alpha1)(alpha2) last name>, or in which a first call is used to
find <alpha1> <alpha2> while a second call is used to find
<(alpha1)(alpha2) last name>.  In general, a table is constructed for
each practical index, in this example, for each practical pair of
letters, including every possible name which is legitimately reached
using the index.  To reduce the ambiguity of the spoken information,
the user may be required to leave a slight pause between the
characters and the names, such as "EI" pause "Ellen."

      In a preferred version of the system, the competing, unselected
candidates are displayed on the screen, with the system automatically
proceeding to the next field and backing up when a correction command
is issued.  When the user detects an error, he says, "Select number
four," indicating the fourth alternative candidate in the prior
field, or, "alternate selection," followed by the correct name.  At
this point, the primary candidate has been eliminated and the
alternate selection window usually has a small number of choices.

      The recognition problems associated with using letters that
sound nearly alike are preferably minimized by...