Disambiguation (Predictive Texting) for AAC
Communication Matters Journal (2007)
Available from www.communicationmatters.org.uk
or
Abstract
Mobile phone style predictive texting is called 'disambiguation' and originated in the Assistive Technology and AAC fields. Such systems use 'restricted selection' keyboards to enter ambiguous text. Such keyboards (having restricted key sets) have been widely reported in the Human Computer Interaction field, and can provide efficient input rates by reducing the number of selection steps needed to identify a required word or option.Previous literature and existing theories related to the use of reduced selection entry systems will be reviewed in the context of AAC use. The historical context of the method will be presented as will it's links to other AAC techniques. The question as to why the method is not more extensively used in AAC will be discussed. In addition the development of disambiguation software for the PC and research into it's use in AAC will be presented.
Available from www.communicationmatters.org.uk
Page 1
Disambiguation (Predictive Texting) for AAC
Barnsley Assistive Technology Team
Disambiguation
- ‘Predictive Texting’ - for
AAC
Simon Judge
Mark Landeryou
Disambiguation
- ‘Predictive Texting’ - for
AAC
Simon Judge
Mark Landeryou
Page 2
Assistive Technology Team
Barnsley
Barnsley AT Team
Assistive Technology (AT) team,
covering 3 areas of S Yorkshire
Assess for and provide a wide variety of
AT
Run training and provide support on AT
Contribute to & run research and
development projects…
Barnsley
Barnsley AT Team
Assistive Technology (AT) team,
covering 3 areas of S Yorkshire
Assess for and provide a wide variety of
AT
Run training and provide support on AT
Contribute to & run research and
development projects…
Page 3
Assistive Technology Team
Barnsley
Overview
Some Theory: Introduction to
ambiguous entry methods
Some History: Literature review of
ambiguous entry techniques
Some Practicalities: Available systems
Some Research: Plans
Barnsley
Overview
Some Theory: Introduction to
ambiguous entry methods
Some History: Literature review of
ambiguous entry techniques
Some Practicalities: Available systems
Some Research: Plans
Page 4
Assistive Technology Team
Barnsley
Headlines
Talking about mobile phone style texting
BUT
This originated in the AAC/AT field!
Investigating disambiguation leads us 'past‘
other AAC/AT techniques
We have developed ambiguous keyboard
software and other software exists
Barnsley
Headlines
Talking about mobile phone style texting
BUT
This originated in the AAC/AT field!
Investigating disambiguation leads us 'past‘
other AAC/AT techniques
We have developed ambiguous keyboard
software and other software exists
Page 5
Assistive Technology Team
Barnsley
Ambiguity:
Ambiguity is usually bad, but:
Sometimes less choice is better (fewer steps)
Less choice often means less effort
As humans we are constantly ambiguous
We constantly 'disambiguate' what other
people are saying (particularly using
contextual information)
“having more than one possible meaning; not
clear.”
Barnsley
Ambiguity:
Ambiguity is usually bad, but:
Sometimes less choice is better (fewer steps)
Less choice often means less effort
As humans we are constantly ambiguous
We constantly 'disambiguate' what other
people are saying (particularly using
contextual information)
“having more than one possible meaning; not
clear.”
Page 6
Assistive Technology Team
Barnsley
Ambiguous Keyboards
An ambiguous keyboard:
where there is not a 1:1 mapping between key and character.
Or to be pedantic:
where there is not a 1:1 mapping between key and symbol
(since we might not use characters on our keys).
defabc.@
mnojklghi
xyztuvwpqrs
*_
abcdefg
hijklmnopq
rstuvwxyz
error_
Barnsley
Ambiguous Keyboards
An ambiguous keyboard:
where there is not a 1:1 mapping between key and character.
Or to be pedantic:
where there is not a 1:1 mapping between key and symbol
(since we might not use characters on our keys).
defabc.@
mnojklghi
xyztuvwpqrs
*_
abcdefg
hijklmnopq
rstuvwxyz
error_
Page 7
Assistive Technology Team
Barnsley
Disambiguation
Disambiguation is the process of removing
the ambiguity from the keyed entry.
This can be done a number of ways:
Using Codes: “coding”
Using a disambiguation process :
“disambiguating”
Let us consider an example:
Barnsley
Disambiguation
Disambiguation is the process of removing
the ambiguity from the keyed entry.
This can be done a number of ways:
Using Codes: “coding”
Using a disambiguation process :
“disambiguating”
Let us consider an example:
Page 8
Assistive Technology Team
Barnsley
Language
Ambiguous entry (example)
Suppose there are only 4 words in our
world:
This is our new language, we might also
have a corpus of how people use this
language.
Cat CanCodDog
Barnsley
Language
Ambiguous entry (example)
Suppose there are only 4 words in our
world:
This is our new language, we might also
have a corpus of how people use this
language.
Cat CanCodDog
Page 9
Assistive Technology Team
Barnsley
Ambiguous entry (example) – 4 keys
With 4 keys we can have a key per word
– so no ambiguity:
KeysCod CanCatDog
LanguageCat CanCodDog
Barnsley
Ambiguous entry (example) – 4 keys
With 4 keys we can have a key per word
– so no ambiguity:
KeysCod CanCatDog
LanguageCat CanCodDog
Page 10
Assistive Technology Team
Barnsley
Ambiguous entry (example) – 3 keys
3 keys – we have ambiguity…
LanguageCat CanCodDog
Keys
Barnsley
Ambiguous entry (example) – 3 keys
3 keys – we have ambiguity…
LanguageCat CanCodDog
Keys
Page 11
Assistive Technology Team
Barnsley
Some options
(bear with me!)
Have multiple concepts per key
Have multiple phonemes per key
Have multiple letters per key
Barnsley
Some options
(bear with me!)
Have multiple concepts per key
Have multiple phonemes per key
Have multiple letters per key
Page 12
Assistive Technology Team
Barnsley
Ambiguous entry (example) – 3 keys
Concepts - code the keys semantically:
Remind you of anything?
KeysAnimal andPurr
Object and
Bark
Container and
Wet
LanguageCat CanCodDog
Animal and
Purr
Object and
Bark
Container and
Wet
Barnsley
Ambiguous entry (example) – 3 keys
Concepts - code the keys semantically:
Remind you of anything?
KeysAnimal andPurr
Object and
Bark
Container and
Wet
LanguageCat CanCodDog
Animal and
Purr
Object and
Bark
Container and
Wet
Page 13
Assistive Technology Team
Barnsley
Ambiguous entry (example) – 3 keys
Coding the keys phonetically or with morphemes:
KeysCOD
AN
OG
D
AT
AN
OG
C
OD
D
AT
Cod LanguageCatCanDog
Barnsley
Ambiguous entry (example) – 3 keys
Coding the keys phonetically or with morphemes:
KeysCOD
AN
OG
D
AT
AN
OG
C
OD
D
AT
Cod LanguageCatCanDog
Page 14
Assistive Technology Team
Barnsley
Ambiguous entry (example) – 3 keys
Or use characters (graphemes) with 3 presses...
KeysC A N OD T
N O C A
Cod LanguageCatCanDog
N O D T
Barnsley
Ambiguous entry (example) – 3 keys
Or use characters (graphemes) with 3 presses...
KeysC A N OD T
N O C A
Cod LanguageCatCanDog
N O D T
Page 15
Assistive Technology Team
Barnsley
How about less than three keys?
The number of keys for a given language
defines the ‘disambiguation accuracy’
For English, this is surprising:
(~92% of English words are not ambiguous on a 12 key
keyboard [Witten, 1982])
Using 9 letter keys, there is unlikely to be more than 8
ambiguous words, often less than 3 [Sandnes et al.,]
The keying efficiency (the ratio of characters to key presses) is
typically in the range 80-90% [Arnott & Javed, 1992]
Our example is quite easy with 2 keys:
Barnsley
How about less than three keys?
The number of keys for a given language
defines the ‘disambiguation accuracy’
For English, this is surprising:
(~92% of English words are not ambiguous on a 12 key
keyboard [Witten, 1982])
Using 9 letter keys, there is unlikely to be more than 8
ambiguous words, often less than 3 [Sandnes et al.,]
The keying efficiency (the ratio of characters to key presses) is
typically in the range 80-90% [Arnott & Javed, 1992]
Our example is quite easy with 2 keys:
Page 16
Assistive Technology Team
Barnsley
Ambiguous entry (example) – 2 keys
KeysC A ND O T
D O T C A N
Cod LanguageCatCanDog
C A N D O T
Barnsley
Ambiguous entry (example) – 2 keys
KeysC A ND O T
D O T C A N
Cod LanguageCatCanDog
C A N D O T
Page 17
Assistive Technology Team
Barnsley
Ambiguous entry (example) – 1 key
One key?
KeyD O TC A N
Cod LanguageCatCanDog
?
Barnsley
Ambiguous entry (example) – 1 key
One key?
KeyD O TC A N
Cod LanguageCatCanDog
?
Page 18
Assistive Technology Team
Barnsley
Ambiguous entry
Unfortunately the English language has 600,000+
words, not 4!
How do we access all (or some) of these words from
a ‘restricted keyboard’??
In AAC we mostly use time (scanning) and Coding
(semantic compaction) to solve the problem
Baker (1982), “Minspeak”
Disambiguation is an alternative.
Barnsley
Ambiguous entry
Unfortunately the English language has 600,000+
words, not 4!
How do we access all (or some) of these words from
a ‘restricted keyboard’??
In AAC we mostly use time (scanning) and Coding
(semantic compaction) to solve the problem
Baker (1982), “Minspeak”
Disambiguation is an alternative.
Page 19
Assistive Technology Team
Barnsley
Disambiguating
DOG
Note the need for the
space key
(confirming)
... and the 'try again'
key – if the
disambiguation
doesn't work.
defabc.@
mnojklghi
wxyztuvpqrs
*_
Barnsley
Disambiguating
DOG
Note the need for the
space key
(confirming)
... and the 'try again'
key – if the
disambiguation
doesn't work.
defabc.@
mnojklghi
wxyztuvpqrs
*_
Page 20
Assistive Technology Team
Barnsley
Disambiguating
With 9 keys we can access a massive
dictionary
Approaching 1:1 key:character ratio
e.g. In an early paper, 92% words in
24,500 dictionary correctly disambiguated
Witten (1982), “Principles of Computer
Speech”
Barnsley
Disambiguating
With 9 keys we can access a massive
dictionary
Approaching 1:1 key:character ratio
e.g. In an early paper, 92% words in
24,500 dictionary correctly disambiguated
Witten (1982), “Principles of Computer
Speech”
Page 21
Assistive Technology Team
Barnsley
Information Theory
A normal 29 key keyboard is quite inefficient – we
don't transmit much information.
Shannon, 1948: “ A Mathematical Theory of
Communication”
With less than 29 keys:
Use information about the English language
Analyse a corpus of text use – frequency of use
Either at word level or parts of words (Ngrams)
Same principle Dasher is built on
Barnsley
Information Theory
A normal 29 key keyboard is quite inefficient – we
don't transmit much information.
Shannon, 1948: “ A Mathematical Theory of
Communication”
With less than 29 keys:
Use information about the English language
Analyse a corpus of text use – frequency of use
Either at word level or parts of words (Ngrams)
Same principle Dasher is built on
Page 22
Assistive Technology Team
Barnsley
Literature Review – Early Work
1981
Glasser, R.E. - “A telephone communication aid for the deaf
Johnson, A.B et al - “DTMF telecommunications for the deaf
and speech impaired”
1982
Witten, I - “Principles of Computer Speech”, word level
disambiguation
Baker, B - “Minspeak”
1985
Minneman, S.L - “ A simplified touch-tone telecommunications
aid for deaf and hearing impaired individuals”
Barnsley
Literature Review – Early Work
1981
Glasser, R.E. - “A telephone communication aid for the deaf
Johnson, A.B et al - “DTMF telecommunications for the deaf
and speech impaired”
1982
Witten, I - “Principles of Computer Speech”, word level
disambiguation
Baker, B - “Minspeak”
1985
Minneman, S.L - “ A simplified touch-tone telecommunications
aid for deaf and hearing impaired individuals”
Page 23
Assistive Technology Team
Barnsley
Literature Review - Timeline
1992
Arnot et al. - “Probabilistic Character Disambiguation for Reduced Keyboards Using
Small Text Samples”
1995
Tegic Launched
1998
Lesher Et Al (Enkidu) - “Optimal Character Arrangements for Ambiguous Keyboards”
2001
Mackenzie et al - “LetterWise: Prefix-based Disambiguation for mobile text input (2001)”
2002
Mackenzie et al - “Text entry for mobile computing: Models and methods, theory and
practice”
2003
Harbush et al - “Predictive and Highly Ambiguous Typing for a Severely Speech and
Motion Impaired User”
Barnsley
Literature Review - Timeline
1992
Arnot et al. - “Probabilistic Character Disambiguation for Reduced Keyboards Using
Small Text Samples”
1995
Tegic Launched
1998
Lesher Et Al (Enkidu) - “Optimal Character Arrangements for Ambiguous Keyboards”
2001
Mackenzie et al - “LetterWise: Prefix-based Disambiguation for mobile text input (2001)”
2002
Mackenzie et al - “Text entry for mobile computing: Models and methods, theory and
practice”
2003
Harbush et al - “Predictive and Highly Ambiguous Typing for a Severely Speech and
Motion Impaired User”
Page 24
Assistive Technology Team
Barnsley
Literature: Number of Keys
Tanaka-Ishii et al.
Looked at text entry using varying numbers of
keys with a dictionary. Their work suggest a
restricted 4 key set to be efficient.
With 4 keys speeds of 7-25wpm were achieved
(with a clear increase due to training).
Comparable to 10 key keyboards (7-25 wpm,
James and Reischel, 2001) . Good compared to
full keyboards (~30-40 wpm, typist single key
~23 wpm [Wiklund et al 1987] ).
Barnsley
Literature: Number of Keys
Tanaka-Ishii et al.
Looked at text entry using varying numbers of
keys with a dictionary. Their work suggest a
restricted 4 key set to be efficient.
With 4 keys speeds of 7-25wpm were achieved
(with a clear increase due to training).
Comparable to 10 key keyboards (7-25 wpm,
James and Reischel, 2001) . Good compared to
full keyboards (~30-40 wpm, typist single key
~23 wpm [Wiklund et al 1987] ).
Page 25
Assistive Technology Team
Barnsley
Literature: Key layouts
A number of authors have suggested improved
layouts, the principle improvements are
TOC moving keys that most frequently give errors
[Foulds, et al. 1987]
Levine, optimized arrangement to limit ambiguity [Levinie,
et al. 1987]
Interesting to note that arranging letters much out
of alphabetical order is often deemed a retrograde
step because of the additional difficulty in learning
to use the system.
Barnsley
Literature: Key layouts
A number of authors have suggested improved
layouts, the principle improvements are
TOC moving keys that most frequently give errors
[Foulds, et al. 1987]
Levine, optimized arrangement to limit ambiguity [Levinie,
et al. 1987]
Interesting to note that arranging letters much out
of alphabetical order is often deemed a retrograde
step because of the additional difficulty in learning
to use the system.
Page 26
Assistive Technology Team
Barnsley
Literature: AAC
Venkatagiri (1993)
Investigated reduced key set keyboards for AAC users
who have motion, vision, or search difficulties
He notes that reduced key keyboards having large keys
can be opened up to users who are unable to use direct
selection of full keyboards
Interesting comment that linear QWERTY keyboards are
limited to 0.77 wpm, compared to 2.5 wpm for most
efficient scanning (Venkatagiri, 1999), and AAC users
communicate at around 0.5-5 wp, (Vanderheiden, 1988)
Barnsley
Literature: AAC
Venkatagiri (1993)
Investigated reduced key set keyboards for AAC users
who have motion, vision, or search difficulties
He notes that reduced key keyboards having large keys
can be opened up to users who are unable to use direct
selection of full keyboards
Interesting comment that linear QWERTY keyboards are
limited to 0.77 wpm, compared to 2.5 wpm for most
efficient scanning (Venkatagiri, 1999), and AAC users
communicate at around 0.5-5 wp, (Vanderheiden, 1988)
Page 27
Assistive Technology Team
Barnsley
Literature: entry speeds
Silfverberg et al 2000
Described the application of Fitts's law to
predicting the rate of text entry using various
disambiguation schemes suitable for mobile
phones:
leads to accurate predictions of expert level text
entry rates, interesting question as to applicability
of this for AAC.
Barnsley
Literature: entry speeds
Silfverberg et al 2000
Described the application of Fitts's law to
predicting the rate of text entry using various
disambiguation schemes suitable for mobile
phones:
leads to accurate predictions of expert level text
entry rates, interesting question as to applicability
of this for AAC.
Page 28
Assistive Technology Team
Barnsley
AAC – Applications
Advantages (?) for literate (?) users:
with use of a small number of keys
with good cursor control
For example:
Someone with Dyspraxia, Ataxia, Tremor etc
Someone who might use a keyguard
Someone who wants to minimise range of
movements (e.g. Head pointer/ eye gaze)
Also could be the basis for some other systems?
Barnsley
AAC – Applications
Advantages (?) for literate (?) users:
with use of a small number of keys
with good cursor control
For example:
Someone with Dyspraxia, Ataxia, Tremor etc
Someone who might use a keyguard
Someone who wants to minimise range of
movements (e.g. Head pointer/ eye gaze)
Also could be the basis for some other systems?
Page 29
Assistive Technology Team
Barnsley
Available Systems:
Tapir
GazeTalk
UKO-II
Enkidu
GazeTalk…
Dkey (eventually)
Barnsley
Available Systems:
Tapir
GazeTalk
UKO-II
Enkidu
GazeTalk…
Dkey (eventually)
Page 30
Assistive Technology Team
Barnsley
Tapir
Cambridge Inference
Group, Piotir Zielinski
Mouse input –
specifically Eye Gaze
Novel disambiguation,
but similar to T9
Cross Platform:
Windows & Linux
Customisable??
Barnsley
Tapir
Cambridge Inference
Group, Piotir Zielinski
Mouse input –
specifically Eye Gaze
Novel disambiguation,
but similar to T9
Cross Platform:
Windows & Linux
Customisable??
Page 31
Assistive Technology Team
Barnsley
Enkidu – Impact XL
Both character and
word level
disambiguation
Editable pages, but
not layout.
Free download (not
many limitations!)
Barnsley
Enkidu – Impact XL
Both character and
word level
disambiguation
Editable pages, but
not layout.
Free download (not
many limitations!)
Page 32
Assistive Technology Team
Barnsley
UKO-II
Designed for an
individual, 3 switch
scanner & as
research project
Released as free
software
Tricky (ish) to setup
and configure…
(See link at end)
Barnsley
UKO-II
Designed for an
individual, 3 switch
scanner & as
research project
Released as free
software
Tricky (ish) to setup
and configure…
(See link at end)
Page 33
Assistive Technology Team
Barnsley
GazeTalk
ITU, Copenhagen
Mouse input &
Scanning, designed for
Gaze
Not exactly
disambiguation, but
based on it
Character positions
changed on predictions
Barnsley
GazeTalk
ITU, Copenhagen
Mouse input &
Scanning, designed for
Gaze
Not exactly
disambiguation, but
based on it
Character positions
changed on predictions
Page 34
Assistive Technology Team
Barnsley
Dkey - Implementation
Some existing systems, none for keyboard input
Developed a disambiguation keyboard for text entry
Spec:
Word level disambiguation
Big Dictionary (built from a number of freely available
dictionaries)
Word frequency
User editable dictionary
User definable key layout, number of keys
(Scanning)
Barnsley
Dkey - Implementation
Some existing systems, none for keyboard input
Developed a disambiguation keyboard for text entry
Spec:
Word level disambiguation
Big Dictionary (built from a number of freely available
dictionaries)
Word frequency
User editable dictionary
User definable key layout, number of keys
(Scanning)
Page 35
Assistive Technology Team
Barnsley
DKey
Barnsley
DKey
Page 36
Assistive Technology Team
Barnsley
Research Plans
User Studies to measure use of this system
Look at efficiency of input/output
Customisation of dictionary
Size of dictionary effects
Include people with disabilities
Establish method to launch software (open
source)
Barnsley
Research Plans
User Studies to measure use of this system
Look at efficiency of input/output
Customisation of dictionary
Size of dictionary effects
Include people with disabilities
Establish method to launch software (open
source)
Page 37
Assistive Technology Team
Barnsley
Summary
Disambiguation originated in AT
Offers a potential method of efficient
access for a number of users
Quite a bit of research… not much into
AT/AAC
Relates to other AAC methods
Further research into usefulness…
Barnsley
Summary
Disambiguation originated in AT
Offers a potential method of efficient
access for a number of users
Quite a bit of research… not much into
AT/AAC
Relates to other AAC methods
Further research into usefulness…
Page 38
Assistive Technology Team
Barnsley
Links
Summary at:
www.assistech.org.uk/doku.php/research:disambiguation
Barnsley AT Team:
www.barnsleyrd.nhs.uk
Barnsley
Links
Summary at:
www.assistech.org.uk/doku.php/research:disambiguation
Barnsley AT Team:
www.barnsleyrd.nhs.uk
Sign up today - FREE
Mendeley saves you time finding and organizing research. Learn more
- All your research in one place
- Add and import papers easily
- Access it anywhere, anytime
Start using Mendeley in seconds!
Readership Statistics
1 Reader on Mendeley
by Discipline
100% Engineering
by Academic Status
100% Researcher (at a non-Academic Institution)
by Country
100% United Kingdom