Browse Prior Art Database

Predictive Keyboard Optimized for Multiple Text Types

IP.com Disclosure Number: IPCOM000115746D
Original Publication Date: 1995-Jun-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 2 page(s) / 90K

Publishing Venue

IBM

Related People

Allard, DJ: AUTHOR [+3]

Abstract

Disclosed is a predictive keyboard that, under application program control, uses different sets of digraph and trigraph tables to optimize keyboard key predictions for specific types of text.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Predictive Keyboard Optimized for Multiple Text Types

      Disclosed is a predictive keyboard that, under application
program control, uses different sets of digraph and trigraph tables
to optimize keyboard key predictions for specific types of text.

      As touch-screen interfaces become smaller, emulating keyboard
input becomes more and more difficult.  One solution to on-screen
keyboard emulation for such interfaces is a predictive keyboard --
one that only displays a subset of the set of keyboard keys, based on
tables of digraph (letter pairs) and trigraph (letter triplets)
frequencies.  However, the type of text that users want to type can
have different underlying letter distributions, even within a
language,
making it impossible to provide optimal prediction with a single set
of
tables.  This disclosure provides a solution to this problem.

      To address the problem of on-screen keyboard emulation for
small touch- or stylus-based interfaces, previous researchers have
developed predictive keyboards -- keyboards that only offer the user
a small number of letters rather than the full set of a language's
letters.  The letters displayed are the most likely characters a user
would probably type, with likelihoods based on internal
English-language digraph and trigraph tables and the last few
characters the user typed.

      In the past, predictive keyboard researchers have suggested
that it would be possible to improve the predictive accuracy of
digraph and trigraph tables over time by recording the frequency of
trigram usage for a user and updating the digraph and trigraph tables
on the basis of that information.  This strategy would be good for a
user as long as that user types letters that come from a population
of letters that a single set of digraph and trigraph tables can
accurately model.  However, if users type from letter populations
that have markedly different digraph and trigraph distributions, then
an adaptive approach will not work as well as an approach that
switches among sets of digraph and trigraph tables optimized for
different types of text.

      For example, consider the text entry requirements for a
personal communicator.  Standard digraph and trigraph tables are
optimal for standard English text, but building the databases in a
personal communicato...